types of research studies cohort

Have a language expert improve your writing

Run a free plagiarism check in 10 minutes, generate accurate citations for free.

• Knowledge Base
• Methodology
• What Is a Cohort Study? | Definition & Examples

What Is a Cohort Study? | Definition & Examples

Published on February 24, 2023 by Tegan George .

A cohort study is a type of observational study that follows a group of participants over a period of time, examining how certain factors (like exposure to a given risk factor) affect their health outcomes. The individuals in the cohort have a characteristic or lived experience in common, such as birth year or geographic area.

While there are several types of cohort study—including open, closed, and dynamic—there are two that are particularly common: prospective cohort studies and retrospective cohort studies .

The initial cohort consisted of about 18,000 newborns. They were enrolled in the study shortly after birth, with regular follow-ups, medical examinations, and cognitive assessments to track their physical, social, and cognitive development.

Cohort studies are particularly useful for identifying risk factors for diseases. They can help researchers identify potential interventions to help prevent or treat the disease, and are often used in fields like medicine or healthcare research.

Table of contents

When to use a cohort study, examples of cohort studies, advantages and disadvantages of cohort studies, frequently asked questions.

Cohort studies are a type of observational study that can be qualitative or quantitative in nature. They can be used to conduct both exploratory research and explanatory research depending on the research topic.

In prospective cohort studies , data is collected over time to compare the occurrence of the outcome of interest in those who were exposed to the risk factor and those who were not. This can help ascertain whether the risk factor could be associated with the outcome.

In retrospective cohort studies , your participants must already possess the disease or health outcome being studied prior to joining. The study is then focused on analyzing the health outcomes of those who share the exposure to the risk factor over a period of time.

A cohort study could be a good fit for your research if:

• You have access to a large pool of research subjects and are comfortable and able to fund research stretching over a longer timeline.
• The relationship between the exposure and health outcome you’re studying is not well understood, and/or its long-term effects have not been thoroughly investigated.
• The exposure you’re studying is rare, or there are possible ethical considerations preventing you from a traditional experimental design .
• Cohort studies in general are more longitudinal in nature. They usually follow the group studied over a long period of time, investigating how certain factors affect their health outcomes.
• Case–control studies rely on primary research , comparing a group of participants already possessing a condition of interest to a control group lacking that condition in real time.

Receive feedback on language, structure, and formatting

Professional editors proofread and edit your paper by focusing on:

• Academic style
• Vague sentences
• Style consistency

See an example

Cohort studies are common in fields like medicine, epidemiology, and healthcare.

Cohort studies are a strong research method , particularly in epidemiology, health, and medicine, but they are not without their disadvantages.

Advantages of cohort studies

Advantages of cohort studies include:

• Cohort studies are better able to approach an estimation of causality than other types of observational studies. Due to their ability to establish temporality, multiple outcomes, and disease incidence over time, researchers are able to determine with more certainty that the exposure indeed preceded the outcome. This strengthens a claim for a cause-and-effect relationship between the variables of interest.
• Due to their long nature, cohort studies are a particularly good choice for studying rare exposures , such as exposure to a new drug or an environmental toxin. Other research designs aren’t able to incorporate the breadth and depth of the impact as broadly as cohort studies do.
• Because cohort studies usually rely on large groups of participants, they are better able to control for potentially confounding variables , such as age, gender identity, or socioeconomic status. Relatedly, the ability to use a sampling method that ensures a more representative sample of the population leads to findings that are typically much more generalizable , with higher internal validity and external validity .

Disadvantages of cohort studies

Disadvantages of cohort studies include:

• Cohort studies can be extremely time-consuming and expensive to conduct due to their long and intense nature.
• Cohort studies are at risk for biases inherent to long-term studies like attrition bias and survivorship bias , as participants are likely to drop out over time. Measurement errors like omitted variable bias and information bias can also confound your analysis, leading you to draw conclusions that may not be true.
• Like many other experimental designs , cohort studies can raise questions regarding ethical considerations . This is particularly the case if the exposure of interest is harmful, or if there is no known treatment for it. Prior to beginning your research, it is critical to ensure that participation in your study is fully voluntary, informed, and as safe as it can be for your research subjects.

The easiest way to remember the difference between prospective and retrospective cohort studies is timing. 

• A prospective cohort study moves forward in time, following a group of participants to track the development of an outcome of interest.
• A retrospective cohort study moves backward in time, first identifying a group of people who already possess the outcome of interest, and then looking backwards to assess their exposure to a risk factor.

A closed cohort study is a type of cohort study where all participants are selected at the beginning of the study, with no new participants added during any of the follow-up periods.

This approach is useful when the exposure being studied is rare, or when it isn’t practically or financially feasible to recruit new participants.

In a cohort study , the incidence refers to the number of new cases of a disease or health outcome that develop during the study period, while prevalence refers to the proportion of the population who have the disease or health outcome at a given point in time. Cohort studies are particularly useful for measuring incidence rates.

A dynamic cohort study is a type of cohort study where the participants are not fixed at the start of the study. Instead, new participants can be added over time if they become eligible to participate. This approach is useful when the study population is expected to change over time.

Sources in this article

We strongly encourage students to use sources in their work. You can cite our article (APA Style) or take a deep dive into the articles below.

George, T. (2023, February 24). What Is a Cohort Study? | Definition & Examples. Scribbr. Retrieved September 3, 2024, from https://www.scribbr.com/methodology/cohort-study/

Euser, A. M., Zoccali, C., Jager, K. J., & Dekker, F. W. (2009). Cohort Studies: Prospective versus Retrospective. Nephron Clinical Practice , 113 (3), c214–c217. https://doi.org/10.1159/000235241

Is this article helpful?

Tegan George

Other students also liked, what is a prospective cohort study | definition & examples, what is a retrospective cohort study | definition & examples, what is an observational study | guide & examples, what is your plagiarism score.

Cohort Study: Definition, Designs & Examples

Julia Simkus

Editor at Simply Psychology

BA (Hons) Psychology, Princeton University

Julia Simkus is a graduate of Princeton University with a Bachelor of Arts in Psychology. She is currently studying for a Master's Degree in Counseling for Mental Health and Wellness in September 2023. Julia's research has been published in peer reviewed journals.

Learn about our Editorial Process

Saul McLeod, PhD

Editor-in-Chief for Simply Psychology

BSc (Hons) Psychology, MRes, PhD, University of Manchester

Saul McLeod, PhD., is a qualified psychology teacher with over 18 years of experience in further and higher education. He has been published in peer-reviewed journals, including the Journal of Clinical Psychology.

Olivia Guy-Evans, MSc

Associate Editor for Simply Psychology

BSc (Hons) Psychology, MSc Psychology of Education

Olivia Guy-Evans is a writer and associate editor for Simply Psychology. She has previously worked in healthcare and educational sectors.

On This Page:

A cohort study is a type of longitudinal study where a group of individuals (cohort), often sharing a common characteristic or experience, is followed over an extended period of time to study and track outcomes, typically related to specific exposures or interventions.

In cohort studies, the participants must share a common factor or characteristic such as age, demographic, or occupation. A “cohort” is a group of subjects who share a defining characteristic.

Cohort studies are observational, so researchers will follow the subjects without manipulating any variables or interfering with their environment.

This type of study is beneficial for medical researchers, specifically in epidemiology, as scientists can use data from cohort studies to understand potential risk factors or causes of a disease.

Before any appearance of the disease is investigated, medical professionals will identify a cohort, observe the target participants over time, and collect data at regular intervals.

Weeks, months, or years later, depending on the duration of the study design, the researchers will examine any factors that differed between the individuals who developed the condition and those who did not.

They can then determine if an association exists between an exposure and an outcome and even identify disease progression and relative risk.

Retrospective

• A retrospective cohort study is a type of observational research that uses existing past data to identify two groups of individuals—those with the risk factor or exposure (cohort) and without—and follows their outcomes backward in time to determine the relationship.
• In a retrospective study , the subjects have already experienced the outcome of interest or developed the disease before starting the study.
• The researchers then look back in time to identify a cohort of subjects before developing the disease and use existing data, such as medical records, to discover any patterns.

Prospective

A prospective cohort study is a type of longitudinal research where a group of individuals sharing a common characteristic (cohort) is followed over time to observe and measure outcomes, often to investigate the effect of suspected risk factors.

In a prospective study , the investigators will design the study, recruit subjects, and collect baseline data on all subjects before they have developed the outcomes of interest.

• The subjects are followed and observed over a period of time to gather information and record the development of outcomes.

Determine cause-and-effect relationships

Because researchers study groups of people before they develop an illness, they can discover potential cause-and-effect relationships between certain behaviors and the development of a disease.

Provide extensive data

Cohort studies enable researchers to study the causes of disease and identify multiple risk factors associated with a single exposure. These studies can also reveal links between diseases and risk factors.

Enable studies of rare exposures

Cohort studies can be very useful for evaluating the effects and risks of rare diseases or unusual exposures, such as toxic chemicals or adverse effects of drugs.

Can measure a continuously changing relationship between exposure and outcome

Because cohort studies are longitudinal, researchers can study changes in levels of exposure over time and any changes in outcome, providing a deeper understanding of the dynamic relationship between exposure and outcome.

Limitations

Time consuming and expensive.

Cohort studies usually require multiple months or years before researchers are able to identify the causes of a disease or discover significant results. Because of this, they are often more expensive than other types of studies. Retrospective studies, though, tend to be cheaper and quicker than prospective studies as the data already exists.

Require large sample sizes

Cohort studies require large sample sizes in order for any relationships or patterns to be meaningful. Researchers are unable to generate results if there is not enough data.

Prone to bias

Because of the longitudinal nature of these studies, it is common for participants to drop out and not complete the study. The loss of follow-up in cohort studies means researchers are more likely to estimate the effects of an exposure on an outcome incorrectly.

Unable to discover why or how a certain factor is associated with a disease

Cohort studies are used to study cause-and-effect relationships between a disease and an outcome. However, they do not explain why the factors that affect these relationships exist. Experimental studies are required to determine why a certain factor is associated with a particular outcome.

The Framingham Heart Study

Studied the effects of diet, exercise, and medications on the development of hypertensive or arteriosclerotic cardiovascular disease, in a longitudinal population-based cohort.

The Whitehall Study

The initial prospective cohort study examined the association between employment grades and mortality rates of 17139 male civil servants over a period of ten years, beginning in 1967. When the Whitehall Study was conducted, there was no requirement to obtain ethical approval for scientific studies of this kind.

The Nurses’ Health Study

Researched long-term effects of nurses” nutrition, hormones, environment, and work-life on health and disease development.

The British Doctors Study

This was a prospective cohort study that ran from 1951 to 2001, investigating the association between smoking and the incidence of lung cancer.

The Black Women’s Health Study

Gathered information about the causes of health problems that affect Black women.

Millennium Cohort Study

Found evidence to show how various circumstances in the first stages of life can influence later health and development. The study began with an original sample of 18,818 cohort members.

The Danish Cohort Study of Psoriasis and Depression

Studied the association between psoriasis and the onset of depression.

The 1970 British Cohort Study

Followed the lives of around 17,000 people born in England, Scotland, and Wales in a single week of 1970.

Frequently Asked Questions

1. are case-control studies and cohort studies the same.

While both studies are commonly used among medical professionals to study disease, they differ.

Case-control studies are performed on individuals who already have a disease (cases) and compare them with individuals who share similar characteristics but do not have the disease (controls).

In cohort studies, on the other hand, researchers identify a group before any of the subjects have developed the disease. Then after an extended period, they examine any factors that differed between the individuals who developed the condition and those who did not.

2. What is the difference between a cross-sectional study and a cohort study?

Like case-control and cohort studies, cross-sectional studies are also used in epidemiology to identify exposures and outcomes and compare the rates of diseases and symptoms of an exposed group with an unexposed group.

However, cross-sectional studies analyze information about a population at a specific point in time, while cohort studies are carried out over longer periods.

3. What is the difference between cohort and longitudinal studies?

A cohort study is a specific type of longitudinal study. Another type of longitudinal study is called a  panel study  which involves sampling a cross-section of individuals at specific intervals for an extended period.

Panel studies are a type of prospective study, while cohort studies can be either prospective or retrospective.

Barrett D, Noble H. What are cohort studies? Evidence-Based Nursing 2019; 22:95-96.

Kandola, A.A., Osborn, D.P.J., Stubbs, B. et al. Individual and combined associations between cardiorespiratory fitness and grip strength with common mental disorders: a prospective cohort study in the UK Biobank. BMC Med 18, 303 (2020). https://doi.org/10.1186/s12916-020-01782-9

Marmot, M. G., Rose, G., Shipley, M., & Hamilton, P. J. (1978). Employment grade and coronary heart disease in British civil servants. Journal of Epidemiology & Community Health, 32(4), 244-249.

Rosenberg, L., Adams-Campbell, L., & Palmer, J. R. (1995). The Black Women’s Health Study: a follow-up study for causes and preventions of illness. Journal of the American Medical Women’s Association (1972), 50(2), 56-58.

Samer Hammoudeh, Wessam Gadelhaq and Ibrahim Janahi (November 5th 2018). Prospective Cohort Studies in Medical Research, Cohort Studies in Health Sciences, R. Mauricio Barría, IntechOpen, DOI: 10.5772/intechopen.76514. Available from: https://www.intechopen.com/chapters/60939

Setia M. S. (2016). Methodology Series Module 1: Cohort Studies. Indian journal of dermatology, 61(1), 21–25. https://doi.org/10.4103/0019-5154.174011

Zabor, E. C., Kaizer, A. M., & Hobbs, B. P. (2020). Randomized Controlled Trials. Chest, 158(1). https://doi.org/10.1016/j.chest.2020.03.013

Further Information

• Cohort Effect? Definition and Examples
• Barrett, D., & Noble, H. (2019). What are cohort studies?. Evidence-based nursing, 22(4), 95-96.
• The Whitehall Studies
• Euser, A. M., Zoccali, C., Jager, K. J., & Dekker, F. W. (2009). Cohort studies: prospective versus retrospective. Nephron Clinical Practice, 113(3), c214-c217.

• En español – ExME
• Em português – EME

Case-control and Cohort studies: A brief overview

Posted on 6th December 2017 by Saul Crandon

Introduction

Case-control and cohort studies are observational studies that lie near the middle of the hierarchy of evidence . These types of studies, along with randomised controlled trials, constitute analytical studies, whereas case reports and case series define descriptive studies (1). Although these studies are not ranked as highly as randomised controlled trials, they can provide strong evidence if designed appropriately.

Case-control studies

Case-control studies are retrospective. They clearly define two groups at the start: one with the outcome/disease and one without the outcome/disease. They look back to assess whether there is a statistically significant difference in the rates of exposure to a defined risk factor between the groups. See Figure 1 for a pictorial representation of a case-control study design. This can suggest associations between the risk factor and development of the disease in question, although no definitive causality can be drawn. The main outcome measure in case-control studies is odds ratio (OR) .

Figure 1. Case-control study design.

Cases should be selected based on objective inclusion and exclusion criteria from a reliable source such as a disease registry. An inherent issue with selecting cases is that a certain proportion of those with the disease would not have a formal diagnosis, may not present for medical care, may be misdiagnosed or may have died before getting a diagnosis. Regardless of how the cases are selected, they should be representative of the broader disease population that you are investigating to ensure generalisability.

Case-control studies should include two groups that are identical EXCEPT for their outcome / disease status.

As such, controls should also be selected carefully. It is possible to match controls to the cases selected on the basis of various factors (e.g. age, sex) to ensure these do not confound the study results. It may even increase statistical power and study precision by choosing up to three or four controls per case (2).

Case-controls can provide fast results and they are cheaper to perform than most other studies. The fact that the analysis is retrospective, allows rare diseases or diseases with long latency periods to be investigated. Furthermore, you can assess multiple exposures to get a better understanding of possible risk factors for the defined outcome / disease.

Nevertheless, as case-controls are retrospective, they are more prone to bias. One of the main examples is recall bias. Often case-control studies require the participants to self-report their exposure to a certain factor. Recall bias is the systematic difference in how the two groups may recall past events e.g. in a study investigating stillbirth, a mother who experienced this may recall the possible contributing factors a lot more vividly than a mother who had a healthy birth.

A summary of the pros and cons of case-control studies are provided in Table 1.

Table 1. Advantages and disadvantages of case-control studies.

Cohort studies

Cohort studies can be retrospective or prospective. Retrospective cohort studies are NOT the same as case-control studies.

In retrospective cohort studies, the exposure and outcomes have already happened. They are usually conducted on data that already exists (from prospective studies) and the exposures are defined before looking at the existing outcome data to see whether exposure to a risk factor is associated with a statistically significant difference in the outcome development rate.

Prospective cohort studies are more common. People are recruited into cohort studies regardless of their exposure or outcome status. This is one of their important strengths. People are often recruited because of their geographical area or occupation, for example, and researchers can then measure and analyse a range of exposures and outcomes.

The study then follows these participants for a defined period to assess the proportion that develop the outcome/disease of interest. See Figure 2 for a pictorial representation of a cohort study design. Therefore, cohort studies are good for assessing prognosis, risk factors and harm. The outcome measure in cohort studies is usually a risk ratio / relative risk (RR).

Figure 2. Cohort study design.

Cohort studies should include two groups that are identical EXCEPT for their exposure status.

As a result, both exposed and unexposed groups should be recruited from the same source population. Another important consideration is attrition. If a significant number of participants are not followed up (lost, death, dropped out) then this may impact the validity of the study. Not only does it decrease the study’s power, but there may be attrition bias – a significant difference between the groups of those that did not complete the study.

Cohort studies can assess a range of outcomes allowing an exposure to be rigorously assessed for its impact in developing disease. Additionally, they are good for rare exposures, e.g. contact with a chemical radiation blast.

Whilst cohort studies are useful, they can be expensive and time-consuming, especially if a long follow-up period is chosen or the disease itself is rare or has a long latency.

A summary of the pros and cons of cohort studies are provided in Table 2.

The Strengthening of Reporting of Observational Studies in Epidemiology Statement (STROBE)

STROBE provides a checklist of important steps for conducting these types of studies, as well as acting as best-practice reporting guidelines (3). Both case-control and cohort studies are observational, with varying advantages and disadvantages. However, the most important factor to the quality of evidence these studies provide, is their methodological quality.

• Song, J. and Chung, K. Observational Studies: Cohort and Case-Control Studies .  Plastic and Reconstructive Surgery.  2010 Dec;126(6):2234-2242.
• Ury HK. Efficiency of case-control studies with multiple controls per case: Continuous or dichotomous data .  Biometrics . 1975 Sep;31(3):643–649.
• von Elm E, Altman DG, Egger M, Pocock SJ, Gøtzsche PC, Vandenbroucke JP; STROBE Initiative.  The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies.   Lancet 2007 Oct;370(9596):1453-14577. PMID: 18064739.

Saul Crandon

Leave a reply cancel reply.

Your email address will not be published. Required fields are marked *

Save my name, email, and website in this browser for the next time I comment.

No Comments on Case-control and Cohort studies: A brief overview

Very well presented, excellent clarifications. Has put me right back into class, literally!

Very clear and informative! Thank you.

very informative article.

Thank you for the easy to understand blog in cohort studies. I want to follow a group of people with and without a disease to see what health outcomes occurs to them in future such as hospitalisations, diagnoses, procedures etc, as I have many health outcomes to consider, my questions is how to make sure these outcomes has not occurred before the “exposure disease”. As, in cohort studies we are looking at incidence (new) cases, so if an outcome have occurred before the exposure, I can leave them out of the analysis. But because I am not looking at a single outcome which can be checked easily and if happened before exposure can be left out. I have EHR data, so all the exposure and outcome have occurred. my aim is to check the rates of different health outcomes between the exposed)dementia) and unexposed(non-dementia) individuals.

Very helpful information

Thanks for making this subject student friendly and easier to understand. A great help.

Thanks a lot. It really helped me to understand the topic. I am taking epidemiology class this winter, and your paper really saved me.

Happy new year.

Wow its amazing n simple way of briefing ,which i was enjoyed to learn this.its very easy n quick to pick ideas .. Thanks n stay connected

Saul you absolute melt! Really good work man

am a student of public health. This information is simple and well presented to the point. Thank you so much.

very helpful information provided here

really thanks for wonderful information because i doing my bachelor degree research by survival model

Quite informative thank you so much for the info please continue posting. An mph student with Africa university Zimbabwe.

Thank you this was so helpful amazing

Apreciated the information provided above.

So clear and perfect. The language is simple and superb.I am recommending this to all budding epidemiology students. Thanks a lot.

Great to hear, thank you AJ!

I have recently completed an investigational study where evidence of phlebitis was determined in a control cohort by data mining from electronic medical records. We then introduced an intervention in an attempt to reduce incidence of phlebitis in a second cohort. Again, results were determined by data mining. This was an expedited study, so there subjects were enrolled in a specific cohort based on date(s) of the drug infused. How do I define this study? Thanks so much.

thanks for the information and knowledge about observational studies. am a masters student in public health/epidemilogy of the faculty of medicines and pharmaceutical sciences , University of Dschang. this information is very explicit and straight to the point

Very much helpful

Subscribe to our newsletter

You will receive our monthly newsletter and free access to Trip Premium.

Related Articles

Cluster Randomized Trials: Concepts

This blog summarizes the concepts of cluster randomization, and the logistical and statistical considerations while designing a cluster randomized controlled trial.

Expertise-based Randomized Controlled Trials

This blog summarizes the concepts of Expertise-based randomized controlled trials with a focus on the advantages and challenges associated with this type of study.

An introduction to different types of study design

Conducting successful research requires choosing the appropriate study design. This article describes the most common types of designs conducted by researchers.

Cohort Studies

Introduction

The characteristic feature of a cohort study is that the investigator identifies subjects at a point in time when they do not have the outcome of interest and compares the incidence of the outcome of interest among groups of exposed and unexposed (or less exposed) subjects. (We can refer to the groups being compared as exposure cohorts.) Cohorts may be identified retrospectively or prospectively, but in either case the outcome status needs to be established at least twice. It must be established that a cohort did not have the outcome of interest at the beginning of the observation period, and the cohort needs to be examined again to determine whether or not the outcome subsequently developed, i.e., the incidence in each of the exposure groups.

Learning Objectives

Upon successful completion of this section of the course, the student will be able to:

• Define what a cohort study is and explain its key features.
• Distinguish between "closed" and "open" cohorts.
• Define and distinguish among:

- Prospective cohort study

- Retrospective cohort study

- Ambidirectional study 

• Explain the advantages and disadvantages of the cohort design in general and the strengths and weaknesses of retrospective and prospective cohort studies.
• Explain the factors that should be considered in selecting subjects for a cohort study.
• Differentiate between a specific/special exposure cohort and a general cohort.
• Explain what is meant by the term "comparison group" .
• Explain the differences among the following types of comparison groups:

- An internal comparison group

- An external comparison group

- A general population comparison group

• Explain what the " healthy worker effect " is.
• Define " loss to follow-up " and explain what effects it may have on a study.

Prospective Versus Retrospective Cohort Studies

There are two fundamental types of cohort studies based on when and how the subjects are enrolled into the study:

Prospective Cohort Studies:

In prospective cohort studies the investigators conceive and design the study, recruit subjects, and collect baseline exposure data on all subjects, before any of the subjects have developed any of the outcomes of interest. The subjects are then followed into the future in order to record the development of any of the outcomes of interest. The follow up can be conducted by mail questionnaires, by phone interviews, via the Internet, or in person with interviews, physical examinations, and laboratory or imaging tests. Combinations of these methods can also be used.

Typically, the investigators have a primary focus, for example, to learn more about cardiovascular disease or cancer, but the data collected from the cohort over time can be used to answer many questions and test many possible determinants, even factors that they hadn't considered when the study was originally conceived.

The Framingham Heart Study, the Nurses Health Study, and the Black Women's Health Study are good examples of large, productive prospective cohort studies. In each of these studies, the investigators wanted to study risk factors for common chronic diseases. The investigators identified a cohort (group) of possible subjects who would be feasible to follow for a prolonged period. Eligible subjects had to meet certain criteria (inclusion criteria) to be included in the study as subjects. The investigators then determine the initial or "baseline" characteristics, behaviors, and other "exposures" of all subjects at the beginning of the study. Information is collected from all subjects in the same way using exactly the same questions and data collection methods for all subjects. They design the questions and data collection procedures very carefully in order to have accurate information about exposures before disease develops in any of the subjects.

For more information:

Link to Framingham Heart Study

Link to The Nurses Health Study

Link to The Black Women's Health Study

Of course, data analysis cannot take place until enough 'events' or 'outcomes' have occurred, so time must elapse, and the analyses will look at events that have occurred during the period of time from the beginning of the study until the time of the analysis or the end of the study. It goes without saying that analysis is always done retrospectively, because a span of time has to have elapsed before you can compare incidence. The thing that makes prospective cohort studies prospective is that they were designed prospectively, and subjects were enrolled and had baseline data collected before any of them developed any of the outcomes of interest. Determining baseline exposure status before disease events occur gives prospective studies an important advantage in reducing certain types of bias that can occur in retrospective cohort studies and case-control studies, though at the cost of efficiency.

After baseline information is collected, subjects in a prospective cohort study are then followed "longitudinally," i.e. over a period of time, usually for years. This enables the investigators to know when follow up began, if and when subjects become diseased, if and when they become lost to follow up, and whether their exposure status changed during the follow up period. By having individual data on these details for each subject, the investigators can compute and compare the incidence rates for each of the exposure groups.

The illustration below shows a hypothetical group of 12 subjects followed over a number of years. They were enrolled into the study at different times, and some of them became lost to follow up, i.e., they stopped responding to letters, emails and phone calls, so we don't know what happened to them; these are show by the horizontal follow up line stopping.

Three subjects developed the outcome of interest at the approximate dates show by the "X"s. The incidence rate was calculated by computing the disease free observation time for each subject, adding up the disease-free observation times for the entire group, and then dividing this into the number of events, as shown in the calculation below the time line.

Since the investigators asked about many exposures during baseline data collection, they can eventually use the data to study many associations between different exposures and disease outcomes. For example, one could identify smokers and non-smokers at baseline and compare their subsequent incidence of developing heart disease. Alternatively, one could group subjects based on their body mass index (BMI) and compare their risk of developing heart disease or cancer.

The data in the table below summarizes some of the results in a study is from the Nurses' Health Study in which they examined the association of body mass index (BMI) with heart disease. (Link to the article)

There were over 118,000 nurses in the study, and they divided the cohort int0 five exposure groups based on BMI. In this case they used the incidence rate of myocardial infarctions (MI, i.e., heart attacks) in the leanest women (BMI < 21) as a reference , against which they compared the incidence rates of MI in the other four groups. For example, the incidence rate of MI in the reference group (those with BMI < 21) was 23.1 per 100,000 person-years of disease-free observation time. The incidence rate in the heaviest group (BMI > 29) was 85.4 MIs per 100,000 person-years.

The Epi-Tools.XLS worksheet for cohort studies can compare either cumulative incidence (top section of the worksheet) or incidence rates like these (lower section of the worksheet). For example, if one were to compare the heaviest group (BMI > 29) to the women with BMI < 21 (the reference group), the Epi-Tools analysis would look like this 

Manson et al. also used the Nurses' Health Study (NHS) to examine the effect of exercise on cardiovascular disease. The NHS enrolled 121,700 female RNs in 1976, but they didn't begin to collect information on exercise until 1986. Since the original baseline data did not include information on exercise, the exercise study only used the women who had not yet developed any cardiovascular problems by 1986. So, the exercise study was restricted to the 72,448 subjects who were free of cardiovascular disease and cancer in 1986. In essence, the information on exercise and activity that they began collecting in 1986 represented a new baseline for this subset of the original cohort.

Link to the article by Manson et al.

Retrospective Cohort Studies

Retrospective studies also group subjects based on their exposure status and compare their incidence of disease. However, in this case both exposure status and outcome are ascertained retrospectively.

In essence, the investigators jump back in time to identify a useful cohort which was initially free of disease and 'at risk.' They then use whatever records are available to determine each subject's exposure status at the begin of the observation period, and they then ascertain what subsequently happened to the subjects in the two (or more) exposure groups. Retrospective cohort studies are also 'longitudinal,' because they examine health outcomes over a span of time. The distinction is that in retrospective cohort studies all of the cases of disease have already occurred before the investigators initiate the study. In contrast, exposure information is collected at the beginning of prospective cohort studies before any subjects have developed any of the outcomes or interest, and the 'at risk' period begins after baseline exposure data is collected and extends into the future.

Retrospective cohort studies are particularly useful for unusual exposures or occupational exposures. For example, if an investigator wanted to determine whether exposure to chemicals used in tire manufacturing was associated with an increased risk of death, one might find a tire manufacturing factory that had been in operation for several decades. One could potentially use employee health records to identify those who had had jobs which involved exposure to the chemicals in question (e.g., workers who actually manufactured tires) and non-exposed coworkers (e.g., clerical workers or sales personnel in the same company or, even better, workers also involved in manufacturing operations but with jobs that didn't involve exposure to the chemicals). One could then ascertain what had happened to all the subjects and compare the incidence of death in the exposed and non-exposed workers.  

Retrospective cohort studies like this are very efficient because they take much less time and cost much less than prospective cohort studies, but this advantage also creates potential problems. Sometimes exposure status is not clear when it is necessary to go back in time and use whatever data was available, because the data being used was not designed to be used in a study. Even if it was clear who was exposed to tire manufacturing chemicals based on employee records, it would also be important to take into account (or adjust for) other differences that could have influenced mortality (confounding factors). For example, in a study comparing mortality rates between workers exposed to solvents used in tire manufacture and an unexposed comparison group, it might be important to adjust for confounding factors such as smoking and alcohol consumption. However, it is unlikely that a retrospective cohort study would have accurate information on these other risk factors.

When an outbreak of Giardia (see this Link to CDC page on Giardia) occurred in Milton, MA , the Milton Health Department requested assistance from the epidemiologists in the MA Department of Public Health. (Kathleen MacVarish from the BUSPH Practice Office was the Health Agent in Milton who led the investigation.) The request for assistance was made some time after the start of the outbreak, and the outbreak was winding down by the time DPH began their study. The outbreak was clearly concentrated among members of the Wollaston Golf Club in Milton, MA , which had two swimming pools, one for adults and a wading pool for infants and small children. Given what they knew about the usual mechanisms by which Giardia is transmitted, the investigators thought that contamination of the kiddy pool by a child shedding Giardia into their stool was the most likely source. (NOTE) The study was conducted by getting most of the people in the cohort to complete a questionnaire in which one of the key questions was "Did you spend any time in the kiddy pool?" This outbreak clearly took place in a well-defined cohort (members of the club), and the investigators could determine how many people developed Giardia in each of the exposure groups (i.e., exposed to the kiddy pool or not). Moreover, they also knew how many respondents had been exposed to the kiddy pool and how many were not. In other words, they knew the denominators for the exposure groups, so they could calculate the cumulative incidence, risk difference, and the risk ratio. They found that people who had spent time in the kiddy pool had 9.0 more cases per 100 persons than those who spent time in the kiddy pool. The risk ratio was 3.27. Because the investigation started after the cases had already occurred, DPH's study of Giardia in Milton is an example of a retrospective cohort study.

An Ambidirectional Cohort Study

A cohort study may also be   ambidirectional , meaning that there are both retrospective and prospective phases of the study. Ambidirectional studies are much less common than purely prospective or retrospective studies, but they are conceptually consistent with and share elements of the advantages and disadvantages of both types of studies. The Air Force Health Study (AFHS) - also known as the Ranch Hand Study - was initiated by the U. S. Air Force in 1979 to assess the possible health effects of military personnel's exposure to Agent Orange and other chemical defoliants sprayed during the Vietnam War. The study was conducted comparing:

• 1,098 pilots exposed to dioxin in Vietnam (Operation Ranch Hand)
• 1,549 men who flew cargo missions in Southeast Asia during the same time

This is an "ambidirectional" study, because it had both a retrospective component and a prospective component. Some of the problems suspected to be caused by Agent Orange would have occurred shortly after exposure (e.g., skin rashes). These were addressed by looking at the cohort retrospectively to see if the exposed pilots had had more problems than the controls. Other problems (e.g., infertility & cancer) might not surface until some time after the exposure. Therefore, the cohort was followed prospectively to see if they had a greater incidence of these problems. The reports that emerged from the study suggested links between Agent Orange exposure and nine distinct diseases: chloracne, Hodgkin's disease, multiple myeloma, non- Hodgkin's lymphoma, porphyria cutanea tarda, respiratory cancers (lung, bronchus, larynx and trachea), soft-tissue sarcoma, acute and subacute peripheral neuropathy, and prostate cancer. 

Closed (Fixed) versus Open Cohorts

A closed cohort is one with fixed membership. Once the cohort is defined by enrolling subjects and follow up begins, no one can be added. The number of subjects may decline because of death or loss to follow up, but no additional subjects are added. As a result, closed cohorts always get smaller over time. Citizens of Japan who were exposed to radiation when atomic bombs were dropped on Hiroshima and Nagasaki during the second World War, would be considered members of a fixed or closed cohort that was defined by an event. Ashengrau and Seage would classify the bombing victims as a "fixed cohort" and make a distinction between a fixed cohort and a closed cohort. They define a closed cohort as similar to a fixed cohort except that a closed cohort is one that has no losses to follow up, for example, a cohort of people who attended a luncheon that resulted in an outbreak of Salmonellosis.

In contrast, an open cohort is dynamic, meaning that members can leave or be added over time. Rothman gives the example of a state cancer registry. Subjects are continually added when they are diagnosed with cancer, so new subjects are continually added. Subjects can also leave the cohort by moving to a new state or dying. Another example of an open or dynamic cohort would be students at Boston University.

These descriptions should sound familiar, because they essentially parallel the descriptions of fixed and dynamic populations from the Measures of Disease Frequency module. The great majority of cohort studies are conducted in closed (or fixed) cohorts, because it is more difficult to establish eligibility and track people in an open cohort, since they can enter and leave at any time. This problem becomes greater as the size of the cohort gets larger and/or the study continues for a longer period of time. Note that the retrospective cohort study of Giardia in Milton was an open cohort (members of the golf club), but the population was relatively small and time period very short.  

Selection of Subjects

General population cohorts versus special exposure cohorts.

For common risk factors, (e.g., smoking, obesity) investigators may enroll a general population cohort , e.g.,

• The general population (e.g., the Framingham Heart Study) or
• A particular subset of the general population,, e.g., nurses or doctors, because they provide reliable information & they are easier to follow-up.

Once a general population cohort is enrolled, investigators will ascertain their baseline exposures to a large number of exposures of interest and possible confounding factors that they may need to adjust for in the analyses.

For uncommon risks , investigators use special exposure cohorts , e.g.,

• Soldiers exposed to dioxin (agent orange) in Vietnam
• Survivors of the bombing of Hiroshima and Nagasaki
• Residents of the Love Canal area of Niagara, NY who were exposed to chemical wastes

The Comparison Group

The ideal comparison group in a cohort study would be a group that was exactly the same as the exposed group, except that they would be unexposed. This is referred to as the "counterfactual ideal," because it is impossible for the same person to be both exposed and unexposed at the same time. Consequently, the best one can do is to select a comparison group that differs with respect to the exposure of interest but is a similar as possible with respect to other factors that might influence the outcome. There are two key things that are essential in selecting the comparison group in a cohort study:

• The unexposed (or less exposed) comparison group should be as similar as possible with respect to other factors that could influence the outcome being studied (possible confounding factors).
• Information collection should be as accurate & as comparable as possible in all groups in order to avoid biasing the association.

There are three general types of comparison groups for cohort studies.

• An internal comparison group
• A comparison cohort
• The general population

Internal Comparison Group

An i nternal comparison group consists of unexposed members of the same cohort. This is generally the best comparison group, because the subjects are comparable in many respects. For example, as noted earlier, the Nurses' Health Study, used the cohort to study a possible association between obesity and myocardial infarction. Subjects from the cohort were categorized into one of five levels of body mass index, and the group with the lowest BMI was used as an internal comparison group or reference group, against which the other categories were compared.

External Comparison Group

When it isn't possible to take a well-defined cohort and divide it into exposure groups, sometimes one can identify an external comparison cohort. This type of comparison group was used when researchers wanted to look at occupational exposure to disulfide in rayon factory workers. Because virtually all workers in the plant were exposed to disulfide, it was not possible to use an internal comparison group from the same plant. Instead, they selected a group of people employed in a paper mill as an external comparison cohort. Both groups had similar education, age, socioeconomic status, and gender distribution. However, they may have differed with respect to other important confounding factors.

 General Population as a Comparison Group

The third possibility is to use the  general population as a comparison group. This is used occasionally in situations where only a small percentage of the population is exposed, e.g., with an unusual occupational exposure. However, the general population may differ from the exposed work force in many ways, including overal health.

One study looked at mortality rates of workers in the rubber industry and compared them to the general population of the US. There are several problems with this. 1) Some of the general population will have had the exposure (same occupation); 2) the general population includes people who are unable to work because of illness or disability. Employed workers tend to be healthier than the general population. This is a well-documented phenomenon know as the "healthy worker effect." Rates of disease and death tend to be higher in the general population than in the employed work force because the general population includes many people who are too sick or disabled to work. As a result, even if the exposure was, in fact, associated with higher mortality, the magnitude of association would be underestimated because of the inherently higher mortality rate in the general population (which includes both employed and unemployed workers).   Although this is not a problem with all diseases, the general population generally exhibits the greatest departure from the counterfactual ideal, and therefore is less widely used today than in the past. Another notable disadvantage of the general population comparison group is that data on important confounders are almost never available.  

The use of the general population as a comparison group is most likely to be seen today in descriptive epidemiology, particularly when there are many categories of exposure with only a small number of outcomes per category.   For example, the Massachusetts Cancer Registry reports the cancer rate in each of 351 cities and towns using the overall rate in the general population as a comparison.   Similarly, descriptions of occupational mortality based on death certificate data may have hundreds of different occupations and also use the general population as a comparison group.

Studies of this type sometimes use a standardized mortality ratio (SMR) as the measure of association. Data from the general population provide overall rates of mortality in the population. These rates are then used to predict how many deaths would be expected in the cohort under study. The SMR is the ratio of observed deaths in the cohort to the number of deaths expected. The SMR is interpreted much like a risk ratio. For example, an SMR=1.2 indicates 1.2 times the risk in the general population or a 20% increase in risk. (Note that sometimes the SMR is multiplied x 100; if so, SMR=120 would also indicate a 20% increase in risk. If population rates are available by age, gender, and race, then SMRs can be adjusted or "standardized" to control for confounding by these factors. This is a method sometimes referred to as "indirect standardization."  

A similar analytical approach is used to compute standardized incidence ratios (SIR). For example, the Massachusetts Cancer Registry uses the general population of Massachusetts as a comparison group in order to examine whether the incidence of specific cancers differs in individual communities compared to the entire state's incidence. In this setting the measure of association that is used is a standardized incidence ratio (SIR). SIRs can also be interpreted much like a risk ratio, although they are typically multiplied by 100, so that SIR=120 would indicate an incidence that was 20% greater than that in the overall population.

Link to the Massachusetts Cancer Registry

Link to BUSPH learning module on Standardized Rates and page on Standardized Incidence Ratios

Follow Up in Cohort Studies

Selection bias from enrollment procedures rarely occurs in cohort studies, because the outcomes have not yet occurred at the time when subjects are enrolled, so a potential participant's eventual outcome status is unknown and therefore can not influence . However, selection bias can occur in a prospective cohort study as a result of   differences in retention during the follow-up period after enrollment.   When the observation period spans many years (in either retrospective or prospective cohort studies) it can be difficult to track subjects for the entire study. Subjects may disappear as a result of death, relocation, or (in prospective studies) loss of interest in the study. Studies with follow up rates of less than 60% will generally be seen as having limited validity, but even losses of 20% can introduce bias if the reasons for loss are related to both exposure status and outcome status.  

Losses to follow-up can introduce bias (a deviation of the observed value of the measure of association from the value that would have been observed in the absence of bias) if there are differences in likelihood of loss to follow-up that are related to exposure status   and   outcome. In general, large prospective cohort studies are doing well if they can maintain follow-up of 80-90% of their sample for long periods.

As an illustration of how bias can be introduced with loss to follow-up, consider the following example. Suppose investigators were prospectively studying the association between use of oral contraceptives and development of thromboembolism (TE), i.e., blood clots in veins of the lower extremities or pelvis that can break loose and become lodged in the branches of the pulmonary artery]. Suppose, 20/10,000 OC users developed thromboembolism, but only 10/10,000 controls did, i.e., OC users really had a 2-fold greater risk. If roughly 4,000 subjects were lost to follow-up in each group, and if 12 of the 20 subjects who developed thromboembolism in the OC group became lost to follow-up, but only 2 subjects with thromboembolism were lost in the control group, the differential loss to follow-up would make it appear that rates of thromboembolism were similar, and the estimate of association (risk ratio) would be biased.The bias that can result from this differential follow up is a type of selection bias.

The enrollment of subjects in a prospective cohort study like this would not introduce selection bias, because the outcome has not yet occurred. However, retention of subjects may be differentially related to exposure and outcome, and this has a similar effect that can bias the results. In the hypothetical cohort study below investigators compared the incidence of thromboembolism (TE) in 10,000 women on oral contraceptives (OC) and 10,000 women not taking OC. TE occurred in 20 subjects taking OC and in 10 subjects not taking OC, so the true risk ratio was (20/10,000) / (10/10,000) = 2.

This unbiased data would give a risk ratio as follows:

However, suppose there were substantial loses to follow-up in both groups, and a greater tendency to loose subjects taking oral contraceptives who developed thromboembolism. In other words, there was differential loss to follow up with loss of 12 diseased subjects in the group taking oral contraceptives, but loss of only 2 subjects with thromboembolism in the unexposed group. This might result in a contingency table like the one shown below.

This biased data would give a risk ratio as follows:

So, in this scenario both exposure groups lost about 40% of their subjects during the follow up period, but there was a greater loss of diseased subjects in the exposed group than in the unexposed group, and it was this differential loss to followup that biased the results.

A study with substantial loss to follow does not have to produce a biased estimate of an association, but it certaintly raises concerns about the accuracy of the estimate. If the losses among the groups being compared are non-differntial, then the estimate will not be biased by the losses. Since there is no way of predictiing the effects of loss to follow up, researchers do their best to reduce it by maintaining contact with participants at regular intervals, collecting contact information from friends or relatives that would know how to reach a participant should s/he move, using the National Death Index and other databases to track the vital status of participants who do not respond to attempts at contact, as well as other strategies. Most of these strategies are only applicable to prospective cohort studies, because all the follow-up time has already occurred in a retrospective cohort study before the investigators get involved. Carefully done prospective cohort studies will go to extraordinary lengths to maintain high follow-up rates.

Unlike the bias that can occur from differential loss to follow up, selection bias at enrollment rarely occurs in cohort studies, because the outcomes have not yet occurred at the time when subjects are enrolled, so a potential participant's eventual outcome status is unknown and therefore can not influence their participation. This is even more unlikely in prospective than retrospective cohort studies, although even in the latter the cohort is almost always created based on information that was recorded prior to the development of the outcome. 

Non-participation in a Prospective Cohort Study

Ordinarily, some of the individuals invited to be subjects in a prospective cohort study refuse to participate. This can produce bias in retrospective cohort studies and case-control studies, because exposure status and outcomes have already occurred at the time of enrollment. However, non-participations will not bias a prospective cohort study in which the outcomes of interest have not yet occurred. To illustrate suppose 100,000 subjects are invited to be in a study on the association between smoking and risk of heart attack, but 50,000 refuse to participate. Only 20% of those who agree are smokers versus 30% in those who refuse. The frequency of smoking in the study subjects is not representative of its prevalence in the population, but you can still compare smokers and non-smokers with respect to the incidence of myocardial infarction. In a situation like this, substantial non-participation could detract from the generalizability of the findings. For example, if there were a high rate of non-participation among African-Americans, it might not be appropriate to extrapolate the findings to this subset of the population.

Selection Bias in Retrospective Cohort Studies 

A similar type of bias can occur in retrospective cohort studies if subjects in one of the exposure groups are more or less likely to be selected if they had the outcome of interest.

Consider a hypothetical investigation of an occupational exposure (e.g., an organic solvent) that occurred 15-20 years ago in factory. Over the years there were suspicions that working eith the solvent led to adverse health events, but no definitive data existed. Eventually, a retrospective cohort study was conducted using the employee health records. If all records had been retained the results might have looked like those shown in the first contingency table below.

However, suppose that many of the old records had been lost or discarded, but,given the suspicions about the effects of the solvent, the records of employees who had worked with the solvents and subequently had health problems were more likely to be retained. Consequently, record retention was 99% among workers who were exposed and developed health problems , but recorded retention was only 80% for all other workers. This scenario would result in data shown in the next contingency table.

Differential loss of records results in selection bias and an overestimate of the association in this case, although depending on the scenario, this type of selection bias could also result in an underestimate of an associaton.

Advantages & Disadvantages of Cohort Studies

Clarity of Temporal Sequence (Did the exposure precede the outcome?): Cohort studies more clearly indicate the temporal sequence between exposure and outcome, because in a cohort study, subjects are known to be disease-free at the beginning of the observation period when their exposure status is established. In case-control studies, one begins with diseased and non-diseased people and then acertains their prior exposures. This is a reasonable approach to establishing past exposures, but subjects may have difficulty remembering past exposures, and their recollection may be biased by having the outcome (recall bias).

Allow Calculation of Incidence: Cohort studies allow you to calculate the incidence of disease in exposure groups, so you can calculate:

• Absolute risk (incidence)
• Relative risk (risk ratio or rate ratio)
• Risk difference
• Attributable proportion (attributable risk %)

Facilitate Study of Rare Exposures: While a cohort design can be used to investigate common exposures (e.g., risk factors for cardiovascular disease and cancer in the Nurses' Health Study), they are particularly useful for evaluating the effects of rare or unusual exposures, because the investigators can make it a point to identify an adequate number of subjects who have an unusual exposure, e.g.,

• Exposure to toxic chemicals (Agent Orange)
• Adverse effects of drugs (e.g., thalidomide) or treatments (e.g., radiation treatments for ankylosing spondylitis)
• Unusual occupational exposures (e.g., asbestos, or solvents in tire manufacturing, )

Allow Examination of Multiple Effects of a Single Exposure

Avoid Selection Bias at Enrollment: Cohort studies, especially prospective cohort studies, reduce the possibility that the results will be biased by selecting subjects for the comparison group who may be more or less likely to have the outcome of interest, because in a cohort study the outcome is not known at baseline when exposure status is established. Nevertheless, selection bias can occur in retrospective cohort studies (since the outcomes have already occurred at the time of selection), and it can occur in prospective cohort studies as a result of differential loss to follow up.

The "Air Force Health Study" on agent orange illustrates these advantages.

• It was clear that the exposure preceded adverse outcomes among exposed subjects who developed problems.
• It was used to evaluate the effects of an unusual risk factors (agent orange).
• It allowed direct calculation of incidence rates.
• It enabled the investigators to study multiple outcomes of this single unusual exposure.
• The prospective component of the study was not biased by knowledge of outcome status, because the outcomes hadn't occurred at the time of enrollment.

Link to a video on Agent Orange from the the New York Times

Disadvantages of Prospective Cohort Studies

• You may have to follow large numbers of subjects for a long time.
• They can be very expensive and time consuming.
• They are not good for rare diseases.
• They are not good for diseases with a long latency.
• Differential loss to follow up can introduce bias.

Disadvantages of Retrospective Cohort Studies

• As with prospective cohort studies, they are not good for very rare diseases.
• If one uses records that were not designed for the study, the available data may be of poor quality.
• There is frequently an absence of data on potential confounding factors if the data was recorded in the past.
• It may be difficult to identify an appropriate exposed cohort and an appropriate comparison group.
• Differential losses to follow up can also bias retrospective cohort studies.

Analysis of a Cohort Study Using an Excel Spreadsheet

The video below (21 min.) will demonstrate how to analyze a large cohort study for which data has been stored in an Excel spreadsheet.

A Critical Thinking Exercise

Moderate Exercise or Just Take a Statin?

This exercise is based on the following two journal articles that focus on strategies to prevent cardiovascular disease. Links to the full articles are provided below.

• Manson et al.: A prospective study of walking as compared with vigorous exercise in the prevention of coronary heart disease in women. N. Engl. J. Med. 1999;341:650-8.
• Ridker PM, Danielson E, et al.: Rosuvastatin to prevent vascular events in men and women with elevated C-reactive protein. N Engl J Med 2008; 359:2195-2207.

Compare and contrast the potential utility of each of these interventions as strategies to reduce cardiovascular disease in a population. In your discussion you should address issues of feasibility, cost, adherence, and risk versus benefit. Would you favor one intervention over the other? Why?

Cohort Studies: The Key to Longitudinal Research Success

Introduction

What is meant by cohort study, types of cohort studies, the importance of cohort studies in public health and policy.

• Advantages & disadvantages of cohort studies

The pursuit of scientific knowledge often involves tracking changes over time, examining causes and effects, or exploring relationships between variables.

One of the most effective methodologies for doing so in fields like medicine, psychology, sociology, and education is the cohort study. This research design allows for in-depth, longitudinal investigations that can reveal valuable insights unattainable through other research methods.

The cohort study remains a cornerstone in longitudinal research , offering a robust framework for answering research questions that demand extensive time-based observations.

In this article, we will explore what cohort studies are, the types of cohort studies, and the various advantages and disadvantages of employing this research methodology.

A cohort study is an observational research method that involves following a specific group of people, known as a cohort, over a defined period. This form of study is commonly used in various scientific fields to examine the relationship between different variables and outcomes, particularly when studying the long-term effects or trends associated with a certain exposure, behavior, or condition.

A "cohort" is a group of individuals who share a common characteristic or experience within a defined period. For example, birth cohorts consist of individuals born in the same year or within a range of years. Examining a birth cohort can be particularly useful for studying lifetime trajectories and generational differences, as they allow researchers to observe how a specific group's health, behavior, and other variables change over time. The concept of a cohort and its operationalization in research is thus foundational to this type of study.

What is a cohort study best used for?

Cohort studies are ideal for observing the longitudinal impacts of different factors. For instance, they can be used to study the long-term health effects of certain diets, the societal impacts of educational policies, or the progression of diseases. Cohort studies are particularly valuable for:

• Understanding the temporal sequence of events : Since cohort studies observe participants over an extended period, they can clarify which variable precedes another, lending weight to causal interpretations.
• Exploring rare or complex phenomena : Some conditions or outcomes are rare and may require a longer observation period to gather sufficient data for meaningful analysis .
• Gauging the effects of multiple variables : Cohort studies can track a wide range of variables simultaneously, offering a multifaceted view of the phenomenon under study.

What is an example of a cohort study?

Perhaps one of the most renowned cohort studies is the Framingham Heart Study. Initiated in 1948 in the town of Framingham, Massachusetts, this study set out to identify the common factors that contribute to cardiovascular disease.

Starting with an initial set of several thousand adult participants, the study expanded over the years to include second and third generations from the same families. The study has provided unprecedented insights into the risk factors for cardiovascular disease, including the role of cholesterol, blood pressure, and smoking.

It also broke ground by incorporating both men and women as well as individuals from diverse age groups, thereby giving a more complete picture of cardiovascular health across a broad spectrum of the population. The Framingham Heart Study serves as an exemplar of how cohort studies can offer deep insights into critical public health issues over an extended period.

By focusing on a dedicated cohort and tracking multiple variables over time, this study has yielded invaluable longitudinal data that has greatly influenced public health policies and medical practices.

Understanding the different types of cohort studies is essential for choosing the right approach for your research. Each type comes with its own set of advantages and challenges, affecting everything from the study's timeline to its reliability. Here, we'll examine dynamic, open, and closed cohort studies, as well as prospective and retrospective cohort studies.

Dynamic, open, and closed cohort study

Before diving into prospective and retrospective studies, it's important to note that cohort studies can also be classified based on the flexibility of their cohort membership.

In a dynamic cohort study, new participants can be added over the course of the study. This is often useful when studying conditions or behaviors that have high turnover rates, such as employment in a particular industry.

An open cohort study is a variation of the dynamic cohort study. Here, subjects can enter or leave the study at different times. This is particularly useful in long-term studies where attrition rates could be high.

In contrast, a closed cohort study starts with a fixed population that is followed over time. No new participants are added, and those who leave the study, often due to death or withdrawal, are not replaced. Closed cohort studies are advantageous when studying a very specific population or condition.

Prospective cohort study

A prospective cohort study involves selecting your cohort and then following them into the future, collecting data as you go. This is often considered the gold standard of cohort studies for several reasons:

• Causality : By following the cohort over time, prospective studies offer a better framework for establishing causal relationships between variables.
• Control over variables : Researchers have more control over the variables they collect, ensuring that data is consistent and tailored to the study's aims.
• Reduced bias : The forward-looking nature of prospective studies can reduce certain types of bias, such as recall bias.

However, prospective cohort studies do come with drawbacks:

• Time-consuming : These studies often take years, if not decades, to complete.
• Resource-intensive : The long duration and extensive data collection requirements make prospective studies costly.
• Attrition : Over time, you may lose participants due to various reasons, impacting the study's validity.

Retrospective cohort study

Retrospective cohort studies, by contrast, look backward in time. Researchers use existing data to trace back the outcomes and exposures among their cohort. The benefits of this approach include:

• Speed : These studies can often be completed much faster since all the data already exists.
• Cost-effective : Utilizing existing data reduces the costs associated with data collection.
• Feasibility : For some research questions , the events of interest may have already occurred, making a retrospective approach the only viable option.

However, they're not without disadvantages:

• Data quality : Researchers have to rely on the available data, which might not be as comprehensive or as tailored to their research question.
• Limited control : The inability to control how and what data was collected can limit the study’s scope.
• Bias risks : The use of existing data can introduce various forms of bias , including selection bias and information bias.

Understanding the nuances between these different types of cohort studies is crucial for any researcher planning to embark on a longitudinal study . Whether you opt for a dynamic, open, or closed cohort design, or choose between a prospective or retrospective approach, your decision will impact the study's length, complexity, and overall outcomes.

Turn to ATLAS.ti for research projects big and small

From pilot research to longitudinal studies, our powerful interface does it all. Download a free trial today.

The impact of cohort studies extends beyond the academic sphere into public health and policy, making them indispensable tools in shaping the well-being of societies. The ability of a cohort study to capture complex interplay between variables over time provides researchers and policymakers with unique insights that other study designs may not offer.

For example, findings from cohort studies have been instrumental in highlighting the adverse health effects of smoking, leading to widespread public health campaigns and legislative actions to reduce tobacco use.

Similarly, data from cohort studies have significantly influenced nutritional guidelines, emphasizing the importance of a balanced diet and regular exercise to mitigate the risk of chronic diseases like diabetes and cardiovascular disorders.

But the reach of cohort studies is not confined to healthcare. They have been used to assess the long-term impacts of educational programs, contributing valuable data that help reform curriculum and teaching methodologies.

Cohort studies can also be applied to environmental issues; long-term data can provide compelling evidence of the impact of pollution or climate change on health, thereby influencing policy decisions related to environmental protection and sustainable development.

However, the influence of cohort studies on policy is not without its challenges. The lengthy duration of many cohort studies means that findings may not become actionable until years or even decades after the research begins. Additionally, while a cohort study can offer strong suggestions of causality, they do not provide the definitive proof that comes from randomized controlled trials. This can sometimes make it difficult to leverage cohort study data in policy debates that require unequivocal evidence.

Nonetheless, the cumulative impact of cohort studies on public health and policy is profound. They offer a nuanced understanding of long-term effects and relationships between variables, providing a strong foundation for interventions and policies designed to improve quality of life over the long term.

Advantages & disadvantages of cohort studies

A cohort study can be a powerful tool in the research arsenal for various reasons, yet it also comes with a unique set of limitations. Understanding both is vital for researchers contemplating the use of this method in their work.

Longitudinal data

One of the key strengths of cohort studies is their ability to establish temporal sequences and, consequently, stronger suggestions of causality. Unlike cross-sectional studies , which offer only a snapshot in time, cohort studies track changes over extended periods.

This allows researchers to identify which variables precede others and offers a stronger foundation for drawing causal inferences. Furthermore, cohort studies are exceptional for studying the development of diseases and conditions that manifest over a long duration, or for understanding the lifelong impacts of certain exposures or interventions.

However, the longitudinal nature of cohort studies is both a strength and a weakness. Following participants over an extended period can be logistically complex and financially taxing. The investment in time and resources is often significant, which can be a barrier for researchers with limited funding.

Managing cohorts

The longer the study, the greater the risk of participant attrition, which can compromise the results. Participants may move, lose interest, or pass away, making it challenging to maintain a stable study cohort over time.

Moreover, cohort studies often require large sample sizes, especially when studying rare outcomes. As the sample size grows, so does the complexity of managing the data and the cost of the study. Therefore, the trade-off between the study's comprehensiveness and its feasibility becomes a key concern.

Quality of data collection and analysis

Data quality is another area where cohort studies both excel and face challenges. On one hand, researchers have the opportunity to carefully plan their data collection methods, optimizing for quality and relevance to the research question .

On the other hand, especially in retrospective cohort studies, researchers are sometimes limited to using existing data. This can introduce challenges such as data inconsistency, as researchers have no control over how the original data was collected.

Another consideration is the risk of bias . While prospective cohort studies are generally less susceptible to certain biases like recall bias, they are not entirely immune to errors in measurement or interpretation. Retrospective cohort studies, however, are often more susceptible to these issues because researchers rely on pre-existing data, which may contain unrecognized biases.

Cohort studies offer a robust framework for investigating complex questions over time but come with their own set of methodological and logistical challenges. The choice to use this approach should be carefully considered in the context of the research question, the available resources, and the potential limitations that could affect the study's outcome and interpretation.

Make sense of your data with ATLAS.ti

Cutting-edge data analysis tools make research faster and easier than ever. See how with a free trial.

Cookies on this website

We use cookies to ensure that we give you the best experience on our website. If you click 'Accept all cookies' we'll assume that you are happy to receive all cookies and you won't see this message again. If you click 'Reject all non-essential cookies' only necessary cookies providing core functionality such as security, network management, and accessibility will be enabled. Click 'Find out more' for information on how to change your cookie settings.

Study designs

This short article gives a brief guide to the different study types and a comparison of the advantages and disadvantages.

See also  Levels of Evidence  

These study designs all have similar components (as we’d expect from the PICO):

• A defined population (P) from which groups of subjects are studied
• Outcomes (O) that are measured

And for experimental and analytic observational studies:

• Interventions (I) or exposures (E) that are applied to different groups of subjects

Overview of the design tree

Figure 1 shows the tree of possible designs, branching into subgroups of study designs by whether the studies are descriptive or analytic and by whether the analytic studies are experimental or observational. The list is not completely exhaustive but covers most basics designs.

Figure: Tree of different types of studies (Q1, 2, and 3 refer to the three questions below)

> Download a PDF by Jeremy Howick about study designs

Our first distinction is whether the study is analytic or non-analytic. A  non-analytic  or  descriptive  study does not try to quantify the relationship but tries to give us a picture of what is happening in a population, e.g., the prevalence, incidence, or experience of a group. Descriptive studies include case reports, case-series, qualitative studies and surveys (cross-sectional) studies, which measure the frequency of several factors, and hence the size of the problem. They may sometimes also include analytic work (comparing factors “” see below).

An  analytic  study attempts to quantify the relationship between two factors, that is, the effect of an intervention (I) or exposure (E) on an outcome (O). To quantify the effect we will need to know the rate of outcomes in a comparison (C) group as well as the intervention or exposed group. Whether the researcher actively changes a factor or imposes uses an intervention determines whether the study is considered to be observational (passive involvement of researcher), or experimental (active involvement of researcher).

In  experimental  studies, the researcher manipulates the exposure, that is he or she allocates subjects to the intervention or exposure group. Experimental studies, or randomised controlled trials (RCTs), are similar to experiments in other areas of science. That is, subjects are allocated to two or more groups to receive an intervention or exposure and then followed up under carefully controlled conditions. Such studies controlled trials, particularly if randomised and blinded, have the potential to control for most of the biases that can occur in scientific studies but whether this actually occurs depends on the quality of the study design and implementation.

In  analytic observational  studies, the researcher simply measures the exposure or treatments of the groups. Analytical observational studies include case””control studies, cohort studies and some population (cross-sectional) studies. These studies all include matched groups of subjects and assess of associations between exposures and outcomes.

Observational studies investigate and record exposures (such as interventions or risk factors) and observe outcomes (such as disease) as they occur. Such studies may be purely descriptive or more analytical.

We should finally note that studies can incorporate several design elements. For example, a the control arm of a randomised trial may also be used as a cohort study; and the baseline measures of a cohort study may be used as a cross-sectional study.

Spotting the study design

The type of study can generally be worked at by looking at three issues (as per the Tree of design in Figure 1):

Q1. What was the aim of the study?

• To simply describe a population (PO questions) descriptive
• To quantify the relationship between factors (PICO questions) analytic.

Q2. If analytic, was the intervention randomly allocated?

• No? Observational study

For observational study the main types will then depend on the timing of the measurement of outcome, so our third question is:

Q3. When were the outcomes determined?

• Some time after the exposure or intervention? cohort study (‘prospective study’)
• At the same time as the exposure or intervention? cross sectional study or survey
• Before the exposure was determined? case-control study (‘retrospective study’ based on recall of the exposure)

Advantages and Disadvantages of the Designs

Randomised Controlled Trial

An experimental comparison study in which participants are allocated to treatment/intervention or control/placebo groups using a random mechanism (see randomisation). Best for study the effect of an intervention.

Advantages:

• unbiased distribution of confounders;
• blinding more likely;
• randomisation facilitates statistical analysis.

Disadvantages:

• expensive: time and money;
• volunteer bias;
• ethically problematic at times.

Crossover Design

A controlled trial where each study participant has both therapies, e.g, is randomised to treatment A first, at the crossover point they then start treatment B. Only relevant if the outcome is reversible with time, e.g, symptoms.

• all subjects serve as own controls and error variance is reduced thus reducing sample size needed;
• all subjects receive treatment (at least some of the time);
• statistical tests assuming randomisation can be used;
• blinding can be maintained.
• all subjects receive placebo or alternative treatment at some point;
• washout period lengthy or unknown;
• cannot be used for treatments with permanent effects

Cohort Study

Data are obtained from groups who have been exposed, or not exposed, to the new technology or factor of interest (eg from databases). No allocation of exposure is made by the researcher. Best for study the effect of predictive risk factors on an outcome.

• ethically safe;
• subjects can be matched;
• can establish timing and directionality of events;
• eligibility criteria and outcome assessments can be standardised;
• administratively easier and cheaper than RCT.
• controls may be difficult to identify;
• exposure may be linked to a hidden confounder;
• blinding is difficult;
• randomisation not present;
• for rare disease, large sample sizes or long follow-up necessary.

Case-Control Studies

Patients with a certain outcome or disease and an appropriate group of controls without the outcome or disease are selected (usually with careful consideration of appropriate choice of controls, matching, etc) and then information is obtained on whether the subjects have been exposed to the factor under investigation.

• quick and cheap;
• only feasible method for very rare disorders or those with long lag between exposure and outcome;
• fewer subjects needed than cross-sectional studies.
• reliance on recall or records to determine exposure status;
• confounders;
• selection of control groups is difficult;
• potential bias: recall, selection.

Cross-Sectional Survey

A study that examines the relationship between diseases (or other health-related characteristics) and other variables of interest as they exist in a defined population at one particular time (ie exposure and outcomes are both measured at the same time). Best for quantifying the prevalence of a disease or risk factor, and for quantifying the accuracy of a diagnostic test.

• cheap and simple;
• ethically safe.
• establishes association at most, not causality;
• recall bias susceptibility;
• confounders may be unequally distributed;
• Neyman bias;
• group sizes may be unequal.

Types of Study Design

• 📖 Geeky Medics OSCE Book
• ⚡ Geeky Medics Bundles
• ✨ 1300+ OSCE Stations
• ✅ OSCE Checklist PDF Booklet
• 🧠 UKMLA AKT Question Bank
• 💊 PSA Question Bank
• 💉 Clinical Skills App
• 🗂️ Flashcard Collections | OSCE , Medicine , Surgery , Anatomy
• 💬 SCA Cases for MRCGP

To be the first to know about our latest videos subscribe to our YouTube channel 🙌

Table of Contents

Suggest an improvement

• Hidden Post Title
• Hidden Post URL
• Hidden Post ID
• Type of issue * N/A Fix spelling/grammar issue Add or fix a link Add or fix an image Add more detail Improve the quality of the writing Fix a factual error
• Please provide as much detail as possible * You don't need to tell us which article this feedback relates to, as we automatically capture that information for you.
• Your Email (optional) This allows us to get in touch for more details if required.
• Which organ is responsible for pumping blood around the body? * Enter a five letter word in lowercase
• Name This field is for validation purposes and should be left unchanged.

Introduction

Study designs are frameworks used in medical research to gather data and explore a specific research question .

Choosing an appropriate study design is one of many essential considerations before conducting research to minimise bias and yield valid results .

This guide provides a summary of study designs commonly used in medical research, their characteristics, advantages and disadvantages.

Case-report and case-series

A case report is a detailed description of a patient’s medical history, diagnosis, treatment, and outcome. A case report typically documents unusual or rare cases or reports  new or unexpected clinical findings .

A case series is a similar study that involves a group of patients sharing a similar disease or condition. A case series involves a comprehensive review of medical records for each patient to identify common features or disease patterns. Case series help better understand a disease’s presentation, diagnosis, and treatment.

While a case report focuses on a single patient, a case series involves a group of patients to provide a broader perspective on a specific disease. Both case reports and case series are important tools for understanding rare or unusual diseases .

Advantages of case series and case reports include:

• Able to describe rare or poorly understood conditions or diseases
• Helpful in generating hypotheses and identifying patterns or trends in patient populations
• Can be conducted relatively quickly and at a lower cost compared to other research designs

Disadvantages

Disadvantages of case series and case reports include:

• Prone to selection bias , meaning that the patients included in the series may not be representative of the general population
• Lack a control group, which makes it difficult to conclude  the effectiveness of different treatments or interventions
• They are descriptive and cannot establish causality or control for confounding factors

Cross-sectional study

A cross-sectional study aims to measure the prevalence or frequency of a disease in a population at a specific point in time. In other words, it provides a “ snapshot ” of the population at a single moment in time.

Cross-sectional studies are unique from other study designs in that they collect data on the exposure and the outcome of interest from a sample of individuals in the population. This type of data is used to investigate the distribution of health-related conditions and behaviours in different populations, which is especially useful for guiding the development of public health interventions .

Example of a cross-sectional study

A cross-sectional study might investigate the prevalence of hypertension (the outcome) in a sample of adults in a particular region. The researchers would measure blood pressure levels in each participant and gather information on other factors that could influence blood pressure, such as age, sex, weight, and lifestyle habits (exposure).

Advantages of cross-sectional studies include:

• Relatively quick and inexpensive to conduct compared to other study designs, such as cohort or case-control studies
• They can provide a snapshot of the prevalence and distribution of a particular health condition in a population
• They can help to identify patterns and associations between exposure and outcome variables, which can be used to generate hypotheses for further research

Disadvantages of cross-sectional studies include:

• They cannot establish causality , as they do not follow participants over time and cannot determine the temporal sequence between exposure and outcome
• Prone to selection bias , as the sample may not represent the entire population being studied
• They cannot account for confounding variables , which may affect the relationship between the exposure and outcome of interest

Case-control study

A case-control study compares people who have developed a disease of interest ( cases ) with people who have not developed the disease ( controls ) to identify potential risk factors associated with the disease.

Once cases and controls have been identified, researchers then collect information about related risk factors , such as age, sex, lifestyle factors, or environmental exposures, from individuals. By comparing the prevalence of risk factors between the cases and the controls, researchers can determine the association between the risk factors and the disease.

Example of a case-control study

A case-control study design might involve comparing a group of individuals with lung cancer (cases) to a group of individuals without lung cancer (controls) to assess the association between smoking (risk factor) and the development of lung cancer.

Advantages of case-control studies include:

• Useful for studying rare diseases , as they allow researchers to selectively recruit cases with the disease of interest
• Useful for investigating potential risk factors for a disease, as the researchers can collect data on many different factors from both cases and controls
• Can be helpful in situations where it is not ethical or practical to manipulate exposure levels or randomise study participants

Disadvantages of case-control studies include:

• Prone to selection bias , as the controls may not be representative of the general population or may have different underlying risk factors than the cases
• Cannot establish causality , as they can only identify associations between factors and disease
• May be limited by the availability of suitable controls , as finding appropriate controls who have similar characteristics to the cases can be challenging

Cohort study

A cohort study follows a group of individuals (a cohort) over time to investigate the relationship between an exposure or risk factor and a particular outcome or health condition. Cohort studies can be further classified into prospective or retrospective cohort studies.

Prospective cohort study

A prospective cohort study is a study in which the researchers select a group of individuals who do not have a particular disease or outcome of interest at the start of the study.

They then follow this cohort over time to track the number of patients who develop the outcome . Before the start of the study, information on exposure(s) of interest may also be collected.

Example of a prospective cohort study

A prospective cohort study might follow a group of individuals who have never smoked and measure their exposure to tobacco smoke over time to investigate the relationship between smoking and lung cancer .

Retrospective cohort study

In contrast, a retrospective cohort study is a study in which the researchers select a group of individuals who have already been exposed to something (e.g. smoking) and look back in time (for example, through patient charts) to see if they developed the outcome (e.g. lung cancer ).

The key difference in retrospective cohort studies is that data on exposure and outcome are collected after the outcome has occurred.

Example of a retrospective cohort study

A retrospective cohort study might look at the medical records of smokers and see if they developed a particular adverse event such as lung cancer.

Advantages of cohort studies include:

• Generally considered to be the most appropriate study design for investigating the temporal relationship between exposure and outcome
• Can provide estimates of incidence and relative risk , which are useful for quantifying the strength of the association between exposure and outcome
• Can be used to investigate multiple outcomes or endpoints associated with a particular exposure, which can help to identify unexpected effects or outcomes

Disadvantages of cohort studies include:

• Can be expensive and time-consuming to conduct, particularly for long-term follow-up
• May suffer from selection bias , as the sample may not be representative of the entire population being studied
• May suffer from attrition bias , as participants may drop out or be lost to follow-up over time

Meta-analysis

A meta-analysis is a type of study that involves extracting outcome data from all relevant studies in the literature and combining the results of multiple studies to produce an overall estimate of the effect size of an intervention or exposure.

Meta-analysis is often conducted alongside a systematic review and can be considered a study of studies . By doing this, researchers provide a more comprehensive and reliable estimate of the overall effect size and their confidence interval (a measure of precision).

Meta-analyses can be conducted for a wide range of research questions , including evaluating the effectiveness of medical interventions, identifying risk factors for disease, or assessing the accuracy of diagnostic tests. They are particularly useful when the results of individual studies are inconsistent or when the sample sizes of individual studies are small, as a meta-analysis can provide a more precise estimate of the true effect size.

When conducting a meta-analysis, researchers must carefully assess the risk of bias in each study to enhance the validity of the meta-analysis. Many aspects of research studies are prone to bias , such as the methodology and the reporting of results. Where studies exhibit a high risk of bias, authors may opt to exclude the study from the analysis or perform a subgroup or sensitivity analysis.

Advantages of a meta-analysis include:

• Combine the results of multiple studies, resulting in a larger sample size and increased statistical power, to provide a more comprehensive and precise estimate of the effect size of an intervention or outcome
• Can help to identify sources of heterogeneity or variability in the results of individual studies by exploring the influence of different study characteristics or subgroups
• Can help to resolve conflicting results or controversies in the literature by providing a more robust estimate of the effect size

Disadvantages of a meta-analysis include:

• Susceptible to publication bias , where studies with statistically significant or positive results are more likely to be published than studies with nonsignificant or negative results. This bias can lead to an overestimation of the treatment effect in a meta-analysis
• May not be appropriate if the studies included are too heterogeneous , as this can make it difficult to draw meaningful conclusions from the pooled results
• Depend on the quality and completeness of the data available from the individual studies and may be limited by the lack of data on certain outcomes or subgroups

Ecological study

An ecological study assesses the relationship between outcome and exposure at a population level or among groups of people rather than studying individuals directly.

The main goal of an ecological study is to observe and analyse patterns or trends at the population level and to identify potential associations or correlations between environmental factors or exposures and health outcomes.

Ecological studies focus on collecting data on population health outcomes , such as disease or mortality rates, and environmental factors or exposures, such as air pollution, temperature, or socioeconomic status.

Example of an ecological study

An ecological study might be used when comparing smoking rates and lung cancer incidence across different countries.

Advantages of an ecological study include:

• Provide insights into how social, economic, and environmental factors may impact health outcomes in real-world settings , which can inform public health policies and interventions
• Cost-effective and efficient, often using existing data or readily available data, such as data from national or regional databases

Disadvantages of an ecological study include:

• Ecological fallacy occurs when conclusions about individual-level associations are drawn from population-level differences
• Ecological studies rely on population-level (i.e. aggregate) rather than individual-level data; they cannot establish causal relationships between exposures and outcomes, as the studies do not account for differences or confounders at the individual level

Randomised controlled trial

A randomised controlled trial (RCT) is an important study design commonly used in medical research to determine the effectiveness of a treatment or intervention . It is considered the gold standard in research design because it allows researchers to draw cause-and-effect conclusions about the effects of an intervention.

In an RCT, participants are randomly assigned to two or more groups. One group receives the intervention being tested, such as a new drug or a specific medical procedure. In contrast, the other group is a control group and receives either no intervention or a placebo .

Randomisation ensures that each participant has an equal chance of being assigned to either group, thereby minimising selection bias . To reduce bias, an RCT often uses a technique called blinding , in which study participants, researchers, or analysts are kept unaware of participant assignment during the study. The participants are then followed over time, and outcome measures are collected and compared to determine if there is any statistical difference between the intervention and control groups.

Example of a randomised controlled trial

An RCT might be employed to evaluate the effectiveness of a new smoking cessation program in helping individuals quit smoking compared to the existing standard of care.

Advantages of an RCT include:

• Considered the most reliable study design for establishing causal relationships between interventions and outcomes and determining the effectiveness of interventions
• Randomisation of participants to intervention and control groups ensures that the groups are similar at the outset, reducing the risk of selection bias and enhancing internal validity
• Using a control group allows researchers to compare with the group that received the intervention while controlling for confounding factors

Disadvantages of an RCT include:

• Can raise ethical concerns ; for example, it may be considered unethical to withhold an intervention from a control group, especially if the intervention is known to be effective
• Can be expensive and time-consuming to conduct, requiring resources for participant recruitment, randomisation, data collection, and analysis
• Often have strict inclusion and exclusion criteria , which may limit the generalisability of the findings to broader populations
• May not always be feasible or practical for certain research questions, especially in rare diseases or when studying long-term outcomes

Dr Chris Jefferies

• Yuliya L, Qazi MA (eds.). Toronto Notes 2022. Toronto: Toronto Notes for Medical Students Inc; 2022.
• Le T, Bhushan V, Qui C, Chalise A, Kaparaliotis P, Coleman C, Kallianos K. First Aid for the USMLE Step 1 2023. New York: McGraw-Hill Education; 2023.
• Rothman KJ, Greenland S, Lash T. Modern Epidemiology. 3 rd ed. Philadelphia: Lippincott Williams & Wilkins; 2008.

Other pages

• Product Bundles 🎉
• Join the Team 🙌
• Institutional Licence 📚
• OSCE Station Creator Tool 🩺
• Create and Share Flashcards 🗂️
• OSCE Group Chat 💬
• Newsletter 📰
• Advertise With Us

Join the community

• Type 2 Diabetes
• Heart Disease
• Digestive Health
• Multiple Sclerosis
• Diet & Nutrition
• Health Insurance
• Public Health
• Patient Rights
• Caregivers & Loved Ones
• End of Life Concerns
• Health News
• Thyroid Test Analyzer
• Doctor Discussion Guides
• Hemoglobin A1c Test Analyzer
• Lipid Test Analyzer
• Complete Blood Count (CBC) Analyzer
• What to Buy
• Editorial Process
• Meet Our Medical Expert Board

What Is a Cohort Study?

A cohort study often looks at 2 (or more) groups of people that have a different attribute (for example, some smoke and some don't) to try to understand how the specific attribute affects an outcome. The goal is to understand the relationship between one group's shared attribute (in this case, smoking) and its eventual outcome.

 pixelfit/Getty Images

Cohort Study Design

There are two categories of evidence-based human medical research:

Experimental research: This involves a controlled process through which each participant in a clinical trial is exposed to some type of intervention or situation—like a drug, vaccine, or environmental exposure. Sometimes there is also a control group that is not exposed for comparison. The results come from tracking the effects of the exposure or intervention over a set period of time.

Observational research: This is when there is no intervention. The researchers simply observe the participants' exposure and outcomes over a set period of time in an attempt to identify potential factors that could affect a variety of health conditions.

Cohort studies are longitudinal, meaning that they take place over a set period of time—frequently, years—with periodic check-ins with the participants to record information like their health status and health behaviors.

They can be either:

• Prospective: Start in the present and continue into the future
• Retrospective: Start in the present, but look to the past for information on medical outcomes and events

Purpose of Cohort Studies

The purpose of cohort studies is to help advance medical knowledge and practice, such as by getting a better understanding of the risk factors that increase a person's chances of getting a particular disease.

Participants in cohort studies are grouped together based on having a shared characteristic—like being from the same geographic location, having the same occupation, or having a diagnosis of the same medical condition.

Each time the researchers check-in with participants in cohort trials, they're able to measure their health behaviors and outcomes over a set period of time. For example, a study could involve two cohorts: one that smokes and the other that doesn't. As the data is collected over time, the researchers would have a better idea of whether there appears to be a link between a behavior—in this case, smoking—and a particular outcome (like lung cancer, for example).  

Strengths of Cohort Studies

Much of the medical profession's current knowledge of disease risk factors comes from cohort studies. In addition to showing disease progression, cohort studies also help researchers calculate the incidence rate, cumulative incidence, relative risk, and hazard ratio of health conditions.  

• Size : Large cohort studies with many participants usually give researchers more confident conclusions than small studies.
• Timeline : Because they track the progression of diseases over time, cohort studies can also be helpful in establishing a timeline of a health condition and determining whether specific behaviors are potential contributing factors to disease.  
• Multiple measures : Often, cohort studies allow researchers to observe and track multiple outcomes from the same exposure. For example, if a cohort study is following a group of people undergoing chemotherapy, researchers can study the incidence of nausea and skin rashes in the patients. In this case, there is one exposure (chemotherapy) and multiple outcomes (nausea and skin rashes).  
• Accuracy : Another strength of cohort studies—specifically, prospective cohort studies—is that researchers might be able to measure the exposure variable, other variables, and the participants' health outcomes with relative accuracy.
• Consistency : Outcomes measured in a study can be done uniformly.

Retrospective cohort studies have their own benefits, namely that they can be conducted relatively quickly, easily, and cheaply than other types of research.

Weaknesses of Cohort Studies

While cohort studies are an essential part of medical research, they are not without their limitations.

These can include:

• Time: Researchers aren't simply bringing participants into the lab for one day to answer a few questions. Cohort studies can last for years—even decades—which means that the costs of running the study can really add up.
• Self-reporting: Even though retrospective cohort studies are less costly, they come with their own significant weakness in that they might rely on participants' self-reporting of past conditions, outcomes, and behaviors. Because of this, it can be more difficult to get accurate results.  
• Drop-out: Given the lengthy time commitment required to be a part of a cohort study, it's not unusual for participants to drop out of this type of research. Though they have every right to do that, having too many people leave the study could potentially increase the risk of bias.
• Behavior alteration: Another weakness of cohort studies is that participants may alter their behavior in ways they wouldn't otherwise if they were not part of a study, which could alter the results of the research.
• Potential for biases: Even the most well-designed cohort studies won't achieve results as robust as those reached via randomized controlled trials. This is because by design—i.e. people put into groups based on certain shared traits—there is an inherent lack of randomization.  

A Word From Verywell

Medicines, devices, and other treatments come to the market after many years of research. There's a long journey between the first tests of early formulations of a drug in a lab, and seeing commercials for it on TV with a list of side effects read impossibly quickly.

Think about the last time you had a physical. Your healthcare provider likely measured several of your vital signs and gave you a blood test, then reported back to you about the various behaviors you may need to change in order to reduce your risk of developing certain diseases. Those risk factors aren't just guesses; many of them are the result of cohort studies.

Song JW, Chung KC. Observational studies: cohort and case-control studies .  Plast Reconstr Surg . 2010;126(6):2234-2242. doi:10.1097/PRS.0b013e3181f44abc.

Barrett D, Noble H. What are cohort studies? Evidence-Based Nursing . 2019;22(4):95-96. doi:10.1136/ebnurs-2019-103183

Wang X, Kattan MW. Cohort studies: design, analysis, and reporting .  CHEST . 2020;158(1):S72-S78. doi: 10.1016/j.chest.2020.03.014.

Setia MS. Methodology series module 1: cohort studies.   Indian J Dermatol . 2016;61(1):21-25. doi:10.4103/0019-5154.174011.

By Elizabeth Yuko, PhD Yuko has a doctorate in bioethics and medical ethics and is a freelance journalist based in New York.

How to Identify Different Types of Cohort Studies

The most important characteristics that you should look for to identify a cohort are the following:

• It is an observational study (the investigator is an observer and does not intervene)
• It follows participants over time (several months, or even years)
• It compares the incidence of the outcome (i.e. the number of participants who developed that outcome over the follow-up period) between exposed and unexposed groups

The objective of a cohort study is to estimate if being exposed to a certain risk factor (or treatment) influences the risk of developing the outcome.

The participants

A cohort study follows participants belonging to:

• 2 groups: This is the classic case where a cohort follows a group of participants exposed to a certain risk factor and another group of unexposed participants. A cohort can also follow 2 groups regardless to their exposure status (for example males and females)
• a different level of exposure to a risk factor
• or an age category, etc.
• A group of people with the same occupation, job, or any subgroup of interest: In this case the cohort starts with a group of participants which their exposure status will be determined later
• A sample from the general population, chosen at random or not (convenience sample or made of people willing to participate)

No matter how these participants were chosen, the sample should always represent the target population. Otherwise the results won’t generalize well and the study would be a waste of time and resources.

Note that the number of participants in a cohort can be either fixed or dynamic:

• Fixed: the number cannot change once the participants are recruited at the start of the study
• Dynamic: participants can be added to the study within the follow-up period

The outcome

The outcome of interest can be:

• An event (like initiating smoking or stopping smoking)
• A score (or a change in the score) on a particular test (IQ test, blood pressure, etc.)

The follow-up period

Sometimes it is tricky to identify if the study follows participants over time or not.

To illustrate this point, I’ll give you an example.

Suppose you read the following sentence in the methods section of a study:

“Participants are surveyed over a period of two weeks to record their emotional status and caloric intake”.

The question is:

Are these participants “followed” over a period of 2 weeks? Is this a cohort study?

In fact, anytime we have to collect data for any kind of study (cross-sectional or cohort) it cannot be instantaneous, it has to happen over some period of time. In a cross-sectional study, we can collect data over a period of many days and average them out in order to reduce the noise in our data (an example would be blood pressure measurements). Still, this is not a cohort study!

A cohort will follow participants over much longer periods of time, like months or years and the measurement of exposure will precede the measurement of the outcome (unlike a cross-sectional study where exposure and outcome are both measured at the same time).

Characteristics of different types of cohort studies

We have 3 different types of cohorts and they differ by whether or not they use historical data and how they use it:

• Prospective cohort
• Retrospective cohort
• Mixed cohort

1. Prospective cohort

The prospective cohort (or simply cohort ) begins by identifying the exposure status of participants. It then follows them over time and the outcome will occur in the future with respect to the starting date of the study. (see figure below)

Key identifier:

At the start of the study, the outcome has not occurred yet.

Gives the most freedom in designing and planning a cohort.

Limitations:

Takes a long time (as we have to wait for the outcome to occur), and is more expensive relative to the other types of cohort designs.

2. Retrospective cohort

The retrospective cohort uses historical data about individuals who were followed in the past before the study even started (perhaps they participated in a past study). At the start of the retrospective cohort, the investigators will not follow participants over time, instead they will be determining which participants developed the outcome.

Otherwise, same as with a prospective cohort, we will be comparing the incidence of the outcome between the exposed and unexposed groups.

The follow-up period occurred before the study started.

The objective of a retrospective cohort is to shorten the follow-up time and get the results sooner and therefore cheaper than with a prospective cohort.

The study duration will be as long as it takes to determine who has the outcome and who has not.

Relies on past measurements that may be subject to bias or may not be optimal for our specific study.

This design may be subject to temporal bias — difficulty in determining whether the exposure occurred before the outcome or not, which may invalidate any conclusion about causality.

3. Mixed cohort

A mixed cohort is a combination of both prospective and retrospective designs. The investigator uses historical data to determine the exposure, AND then follows the participants over time before determining who developed the outcome of interest and who didn’t.

The exposure occurred in the past and is determined using historical data, and the outcome will occur in the future (with respect to the start date of the study) after a period of follow-up.

Advantages:

The follow-up period is shorter than that of a prospective cohort, so we get the results sooner and cheaper compared to a prospective design.

And because the outcome will be measured in the future, we will not be relying on past measurements of the outcome.

Limitation:

By using past exposure data, we will not have the freedom of design as with a prospective cohort.

And because we still need to follow participants over time, the duration of the study will be longer compared to a retrospective design.

Note: How to distinguish between prospective and retrospective cohorts?

In a prospective cohort, we can still ask participants for information about things that happened in the past. This does not make the design retrospective or mixed. The biggest difference between prospective and retrospective designs is that in a prospective design, the follow-up period occurs after the start of the study and in a retrospective design, the follow-up period occurred before the study even started.

• David Celentano, Moyses Szklo. Gordis Epidemiology . 6th edition. Elsevier; 2018.
• Szklo M, Nieto FJ. Epidemiology: Beyond the Basics . 4th edition. Jones & Bartlett Learning; 2018.
• Hackshaw A. A Concise Guide to Observational Studies in Healthcare . 1st Edition. Wiley-Blackwell; 2015.

Further reading

• Cohort vs Cross-Sectional Study
• Cohort vs Randomized Controlled Trials
• Case Report: A Beginner’s Guide with Examples
• Experimental vs Quasi-Experimental Design
• Understand Quasi-Experimental Design Through an Example

Your Account

Manage your account, subscriptions and profile.

MyKomen Health

ShareForCures

In Your Community  

In Your Community

View resources and events in your local community.

Change your location:

Susan G. Komen®

One moment can change everything.

Types of Research Studies

Epidemiology studies.

Epidemiology is the study of the patterns and causes of disease in people.

The goal of epidemiology studies is to give information that helps support or disprove an idea about a possible link between an exposure (such as alcohol use) and an outcome (such as breast cancer) in people.

The 2 main types of epidemiology studies are:

• Observational studies ( prospective cohort or case-control )

Randomized controlled trials

Though they have the same goal, observational studies and randomized controlled trials differ in:

• The way they are conducted
• The strengths of the conclusions they reach

Observational studies

In observational studies, the people in the study live their daily lives as they choose. They exercise when they want, eat what they like and take the medicines their doctors prescribe. They report these activities to researchers.

There are 2 types of observational studies:

Prospective cohort studies

Case-control studies.

A prospective cohort study follows a large group of people forward in time.

Some people will have a certain exposure (such as alcohol use) and others will not.

Researchers compare the different groups (for example, they might compare heavy drinkers, moderate drinkers, light drinkers and non-drinkers) to see which group is more likely to develop an outcome (such as breast cancer).

In a case-control study, researchers identify 2 groups: cases and controls.

• Cases are people who already have an outcome (such as breast cancer).
• Controls are people who do not have the outcome.

The researchers compare the 2 groups to see if any exposure (such as alcohol use) was more common in the history of one group compared to the other.

In randomized controlled trials (randomized clinical trials), researchers divide people into groups to compare different treatments or other interventions.

These studies are called randomized controlled trials because people are randomly assigned (as if by coin toss) to a certain treatment or behavior.

For example, in a randomized trial of a new drug therapy, half the people might be randomly assigned to a new drug and the other half to the standard treatment.

In a randomized controlled trial on exercise and breast cancer risk, half the participants might be randomly assigned to walk 10 minutes a day and the other half to walk 2 hours a day. The researchers would then see which group was more likely to develop breast cancer, those who walked 10 minutes a day or those who walked 2 hours a day.

Many behaviors, such as smoking or heavy alcohol drinking, can’t be tested in this way because it isn’t ethical to assign people to a behavior known to be harmful. In these cases, researchers must use observational studies.

Patient series

A patient series is a doctor’s observations of a group of patients who are given a certain treatment.

There is no comparison group in a patient series. All the patients are given a certain treatment and the outcomes of these patients are studied.

With no comparison group, it’s hard to draw firm conclusions about the effectiveness of a treatment.

For example, if 10 women with breast cancer are given a new treatment, and 2 of them respond, how do we know if the new treatment is better than standard treatment?

If we had a comparison group of 10 women with breast cancer who got standard treatment, we could compare their outcomes to those of the 10 women on the new treatment. If no women in the comparison group responded to standard treatment, then the 2 women who responded to the new treatment would represent a success of the new treatment. If, however, 2 of the 10 women in the standard treatment group also responded, then the new treatment is no better than the standard.

The lack of a comparison group makes it hard to draw conclusions from a patient series. However, data from a patient series can help form hypotheses that can be tested in other types of studies.

Strengths and weaknesses of different types of research studies

When reviewing scientific evidence, it’s helpful to understand the strengths and weaknesses of different types of research studies.

Case-control studies have some strengths:

• They are easy and fairly inexpensive to conduct.
• They are a good way for researchers to study rare diseases. If a disease is rare, you would need to follow a very large group of people forward in time to have many cases of the disease develop.
• They are a good way for researchers to study diseases that take a long time to develop. If a disease takes a long time to develop, you would have to follow a group of people for many years for cases of the disease to develop.

Case-control studies look at past exposures of people who already have a disease. This causes some concerns:

• It can be hard for people to remember details about the past, especially when it comes to things like diet.
• Memories can be biased (or influenced) because the information is gathered after an event, such as the diagnosis of breast cancer.
• When it comes to sensitive topics (such as abortion), the cases (the people with the disease) may be much more likely to give complete information about their history than the controls (the people without the disease). Such differences in reporting bias study results.

For these reasons, the accuracy of the results of case-control studies can be questionable.

Cohort studies

Prospective cohort studies avoid many of the problems of case-control studies because they gather information from people over time and before the events being studied happen.

However, compared to case-control studies, they are expensive to conduct.

Nested case-control studies

A nested case-control study is a case-control study within a prospective cohort study.

Nested case-control studies use the design of a case-control study. However, they use data gathered as part of a cohort study, so they are less prone to bias than standard case-control studies.

All things being equal, the strength of nested case-control data falls somewhere between that of standard case-control studies and cohort studies.

Randomized controlled trials are considered the gold standard for studying certain exposures, such as breast cancer treatment. Similar to cohort studies, they follow people over time and are expensive to do.

Because people in a randomized trial are randomly assigned to an intervention (such as a new chemotherapy drug) or standard treatment, these studies are more likely to show the true link between an intervention and a health outcome (such as survival).

Learn more about randomized clinical trials , including the types of clinical trials, benefits, and possible drawbacks.

Overall study quality

The overall quality of a study is important. For example, the results from a well-designed case-control study can be more reliable than those from a poorly-designed randomized trial.

Finding more information on research study design

If you’re interested in learning more about research study design, a basic epidemiology textbook from your local library may be a good place to start. The National Cancer Institute also has information on epidemiology studies and randomized controlled trials.

Animal studies

Animal studies add to our understanding of how and why some factors cause cancer in people.

However, there are many differences between animals and people, so it makes it hard to translate findings directly from one to the other.

Animal studies are also designed differently. They often look at exposures in larger doses and for shorter periods of time than are suitable for people.

While animal studies can lay the groundwork for research in people, we need human studies to draw conclusions for people.

All data presented within this section of the website come from studies done with people.

Joining a research study

Research is ongoing to improve all areas of breast cancer, from prevention to treatment.

Whether you’re newly diagnosed, finished breast cancer treatment many years ago, or even if you’ve never had breast cancer, there may be breast cancer research studies you can join.

If you have breast cancer, BreastCancerTrials.org in collaboration with Susan G. Komen® offers a custom matching service that can help find a studies that fit your needs. You can also visit the National Institutes of Health’s website to find a breast cancer treatment study.

If you’re interested in being part of other studies, talk with your health care provider. Your provider may know of studies in your area looking for volunteers.

Learn more about joining a research study .

Learn more about clinical trials .

Learn what Komen is doing to help people find and participate in clinical trials .

Updated 12/16/20

TOOLS & RESOURCES

In Your Own Words

How has having breast cancer changed your outlook?

Share Your Story or Read Others

Log in using your username and password

• Search More Search for this keyword Advanced search
• Latest content
• Current issue
• Write for Us
• BMJ Journals

You are here

• Volume 22, Issue 4

What are cohort studies?

• Article Text
• Article info
• Citation Tools
• Rapid Responses
• Article metrics

• http://orcid.org/0000-0003-4308-4219 David Barrett 1 ,
• Helen Noble 2
• 1 Faculty of Health Sciences , University of Hull , Hull , UK
• 2 School of Nursing and Midwifery , Queen’s University Belfast , Belfast , UK
• Correspondence to Dr David Barrett, Faculty of Health Sciences, University of Hull, Hull HU6 7RX, UK; D.I.Barrett{at}hull.ac.uk

https://doi.org/10.1136/ebnurs-2019-103183

Statistics from Altmetric.com

Request permissions.

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

• statistics and research methods

In 1951, Richard Doll and Austin Bradford-Hill commenced a ground-breaking research project by writing to all registered doctors in the UK to ask about their smoking habits. The British Doctors Study recruited and followed-up over 40 000 participants, monitoring mortality rates and causes of death over the subsequent years and decades. Even by the time of the first set of preliminary results in 1954, there was evidence to link smoking with lung cancer and increased mortality. 1 Over the following decades, the study provided further definitive evidence of the health risks from smoking, and was extended to explore other causes of death (eg, heart disease) and other behavioural variables (eg, alcohol intake).

The Doctors Health Survey is one of the largest, most ambitious and best-known cohort studies and demonstrates the value of this approach in supporting our understanding of disease risk. However, as a method, cohort studies can have much wider applications. This article provides an overview of cohort studies, identifying the opportunities and challenges they present to researchers, and the role they play in developing the evidence base for nursing and healthcare more broadly.

Cohort studies are a type of longitudinal study —an approach that follows research participants over a period of time (often many years). Specifically, cohort studies recruit and follow participants who share a common characteristic, such as a particular occupation or demographic similarity. During the period of follow-up, some of the cohort will be exposed to a specific risk factor or characteristic; by measuring outcomes over a period of time, it is then possible to explore the impact of this variable (eg, identifying the link between smoking and lung cancer in the British Doctors Study.) Cohort studies are, therefore, of particular value in epidemiology, helping to build an understanding of what factors increase or decrease the likelihood of developing disease.

Though the most high-profile types of cohort studies are usually related to large epidemiological research studies, they are not the only application of this method. Within nursing research, cohort studies have focused on the progress of nurses through their education and careers. Li et al —as part of the European NEXT study group—recruited almost 6500 female nurses who, at the time of recruitment, had no intention to leave the profession. The study followed the cohort up for a year, identifying that 8% developed the intention to leave nursing, often due to issues such as poor salary or limited promotion prospects. 4

Usually, cohort studies should adopt a purely observational approach. However, some research is labelled as a cohort study while exploring the effectiveness of specific interventions. For example, Lansperger et al explored nurse practitioner (NP)-led critical care in a large university hospital in the USA. They collected data on all patients who were admitted to the intensive care unit over a 3-year period. Patients from this cohort were cared for by teams led by either doctors or NPs, and outcomes (primarily 90-day mortality) were monitored. By comparing the groups, the researchers established that outcomes were similar regardless of whether patient care was led by a doctor or an NP. 5

Strengths and weaknesses of cohort studies

Cohort studies are an effective and robust method of establishing cause and effect. As they are usually large in size, researchers are able to draw confident conclusions regarding the link between risk factors and disease. In many cases, because participants are often free of disease at the commencement of the study, cohort studies are particularly useful at identifying the timelines over which certain behaviours can contribute to disease.

However, the nature of cohort studies can cause challenges. Collecting prospective data on thousands of participants over many years (and sometimes decades) is complex, time-consuming and expensive. Participants may drop out, increasing the risk of bias; equally, it is possible that the behaviour of participants may alter because they are aware that they are part of a study cohort. The analysis of data from these large-scale studies is also complex, with large numbers of confounding variables making it difficult to link cause and effect. Where cohort (or ‘cohort-like’) studies link to a specific intervention (as in the case of the Lansperger et al study into nursing practitioner-led critical care 5 ), the lack of randomisation to different arms of the study makes the approach less robust than randomised controlled trials.

One way of making a cohort study less time-consuming is to carry it out retrospectively. This is a more pragmatic approach, as it can be completed more quickly using historical data. For example, Wray et al used a retrospective cohort study to identify factors that were associated with non-continuation of students on nursing programmes. By exploring characteristics in five previous cohorts of students, they were able to identify that factors such as being older and/or local were linked to higher levels of continuation. 6

However, this retrospective approach increases the risk of bias in the sampling of the cohort, with greater likelihood of missing data. Retrospective cohort studies are also weakened by the fact that the data fields available are not designed with the study in mind—instead, the researcher simply has to make use of whatever data are available, which may hinder the quality of the study.

Reporting and critiquing of cohort studies

When reporting a cohort study, it is recommended that STROBE guidance 7 is followed. STROBE is an international, collaborative enterprise which includes experts with experience in the organisation and of dissemination of observational studies, including cohort studies. The aim is to STrengthen the Reporting of OBservational studies in Epidemiology. The STROBE checklist for cohort studies - available at https://www.strobe-statement.org/fileadmin/Strobe/uploads/checklists/STROBE_checklist_v4_combined.pdf - includes detail related to the introduction/methods/results/discussion of the study.

Critical appraisal of any cohort study is essential to identify the strengths and weaknesses of the study and to determine the usefulness and validity of the study findings. Components of critical appraisal in relation to cohort studies include evaluation of the study design in relation to the research question, assessment of the methodology, suitability of statistical methods used, conflicts of interest and how relevant the research is to practice. 8–10

Cohort studies are the cornerstone of epidemiological research, providing an understanding of risk factors for disease based on findings in thousands of participants over many years. Disease prevention guidelines used by nurses and other healthcare professionals across the globe are based on the evidence from high-profile studies, such as the British Doctors Study, the Framingham Heart Study and the Nurses’ Health Study. However, cohort studies offer opportunities outside epidemiology: in nursing research, the approach is useful in exploring areas such as factors that influence students’ progression through their programme or nurses’ progression through their career.

This approach to research does bring with it some important challenges—often related to their size, complexity and longevity. However, with careful planning and implementation, cohort studies can make valuable contributions to the development of evidence-based healthcare.

• Colditz GA ,
• Philpott SE ,
• Hankinson SE
• Galatsch M ,
• Siegrist J , et al
• Landsperger JS ,
• Semler MW ,
• Wang L , et al
• Aspland J ,
• Barrett D , et al
• von Elm E ,
• Altman DG ,
• Egger M , et al
• Rochon PA ,
• Gurwitz JH ,
• Sykora K , et al
• Critical Appraisal Skills Programme

Competing interests None declared.

Patient consent for publication Not required.

Provenance and peer review Commissioned; internally peer reviewed.

Read the full text or download the PDF:

• Open access
• Published: 07 September 2024

Lung cancer survival by immigrant status: a population-based retrospective cohort study in Ontario, Canada

• Arlinda Ruco 1 , 2 , 3 , 4 , 5 ,
• Aisha K. Lofters 5 , 6 , 7 , 8 ,
• Hong Lu 8 ,
• Nancy N. Baxter 6 , 8 , 9 , 10 ,
• Sara Guilcher 8 , 11 ,
• Alexander Kopp 8 ,
• Mandana Vahabi 8 , 12 &
• Geetanjali D. Datta 13 , 14 , 15  

BMC Cancer volume  24 , Article number:  1114 ( 2024 ) Cite this article

Metrics details

Lung cancer is one of the most common cancers and causes of cancer death in Canada. Some previous literature suggests that socioeconomic inequalities in lung cancer screening, treatment and survival may exist. The objective of this study was to compare overall survival for immigrants versus long-term residents of Ontario, Canada among patients diagnosed with lung cancer.

This population-based retrospective cohort study utilized linked health administrative databases and identified all individuals (immigrants and long-term residents) aged 40 + years diagnosed with incident lung cancer between April 1, 2012 and March 31, 2017. The primary outcome was 5-year overall survival with December 31, 2019 as the end of the follow-up period. We implemented adjusted Cox proportional hazards models stratified by age at diagnosis, sex, and cancer stage at diagnosis to examine survival.

Thirty-eight thousand seven hundred eighty-eight individuals diagnosed with lung cancer were included in our cohort including 7% who were immigrants. Immigrants were younger at diagnosis and were more likely to reside in the lowest neighbourhood income quintile (30.6% versus 24.5%) than long-term residents. After adjusting for age at diagnosis, neighbourhood income quintile, comorbidities, visits to primary care in the 6 to 30 months before diagnosis, continuity of care, cancer type and cancer stage at diagnosis, immigrant status was associated with a lower hazard of dying 5-years post-diagnosis for both females (0.7; 95% CI 0.6–0.8) and males (0.7; 95% CI 0.6–0.7) in comparison to long-term residents. This trend held in adjusted models stratified by cancer stage at diagnosis. For example, female immigrants diagnosed with early stage lung cancer had a hazard ratio of 0.5 (95% CI 0.4–0.7) in comparison to long-term residents.

Overall survival post diagnosis with lung cancer was better among Ontario immigrants versus long-term residents. Additional research, potentially on the protective effects of immigrant enclave and the intersection of immigrant status with racial/ethnic identity, is needed to further explore why better overall survival for immigrants remained.

Peer Review reports

Lung cancer is one of the most common cancers and causes of cancer death in Canada [ 1 , 2 ]. It is estimated that there were 30,000 new cases in 2022 accounting for approximately one quarter of all cancer deaths [ 2 ]. The probability of developing lung cancer is estimated at 7%, with 1 in 15 Canadians expected to be diagnosed with the disease during their lifetime [ 1 ]. Stage of diagnosis is strongly associated with survival, with stage IV survival estimated at less than 10% over 5 years [ 2 ]. Unfortunately, about 70% of lung cancer cases in Canada are diagnosed at a late stage when treatment is likely to be less effective and studies suggest that socioeconomic status (SES) may play a role [ 1 ].

Socioeconomic inequalities in lung cancer screening, treatment and survival have been documented [ 3 , 4 , 5 ]. A recent overview of 8 systematic reviews conducted from 2010–2021 representing many high-income geographical regions including North America, the United Kingdom (UK), Scandinavia, Australia, New Zealand, and Korea observed that socioeconomic inequalities present in several lung cancer outcomes including those related to diagnosis, treatment and survival [ 3 ]. For example, one of the reviews found that lower SES was associated with a lower likelihood of receiving cancer treatment (OR 0.79; 95% CI 0.73–0.86) [ 4 ]. Another review conducted by Finke et al., [ 6 ] found that those with lower income had higher mortality rates (HR 1.13; 95% CI 1.08–1.19). This finding was also consistent for those who lived in lower SES neighbourhoods [ 6 ].

Ontario, Canada’s most populous province, has over 14 million residents, with an estimated 30% of the population being foreign-born [ 7 ]. Immigrants are more likely to have lower income than non-immigrants, [ 8 ] and therefore it may be important to explore whether immigration status may confer additional risk. Previous studies have also reported on the “healthy immigrant effect,” where those who immigrate to a country tend to be in better health that native-born residents which may help to explain some differences by immigration status, though this effect is said to diminish over time with longer time since landing [ 9 ]. While the healthy immigrant effect has been observed in prior work, previous studies have also found this not to be true. For example, we previously found no difference in lung cancer diagnostic stage between immigrants and non-immigrants [ 10 ]. It is unknown if inequalities in lung cancer survival between immigrants and non-immigrants are present in Canada. The objective of this study was to compare survival for immigrants versus long-term residents of Ontario among individuals diagnosed with lung cancer between 2012–2017.

Study design and setting

This population-based retrospective cohort study used linked health administrative databases from Ontario, Canada’s most populous and multi-ethnic province. Ontario provides universal coverage for medically-necessary hospital care and physician services through the Ontario Health Insurance Plan (OHIP) with almost all Ontario residents being beneficieries (coverage approximately 95%) [ 11 ]. Given the universal coverage, health administrative databases in the province are comprehensive in capturing most health care delivered in the jurisdictions. The databases are held at ICES (formerly known as the Institute for Clinical Evaluative Sciences). ICES is an independent, non-profit research institute whose legal status under Ontario’s health information privacy law allows it collect and analyze health care and demographic data, without consent, for health system evaluation and improvement. The study was approved by the Research Ethics Board at Unity Health Toronto (REB # 19–072) and reporting of the study is consistent with the RECORD statement (Appendix A) [ 12 ].

Data sources

Several data sources were used to construct the analytic dataset. The datasets were linked using unique encoded identifiers and analyzed at ICES. The Ontario Cancer Registry (OCR) was used to ascertain cancer diagnosis, tumour characteristics and survival. OCR contains information on approximately 95% of cancers diagnosed in the province since 1964. The Immigration, Refugees and Citizenship Canada – Permanent Resident (IRCC-PR) database was used to ascertain immigrant status and contains information on all permanent residents since 1985 including country from which the person emigrated and country of birth. The Registered Persons Database (RPDB) contains demographic information on all individuals who are beneficiaries of OHIP from April 1991 onwards. Physician billing data were obtained from OHIP and information on patient-provider rostering was obtained through the Client Agency Program Enrolment database. The Discharge Abstract Database (DAD) and National Ambulatory Care Reporting System (NACRS) and the Same Day Surgery (SDS) database contain information for all inpatient hospitalizations and ambulatory/outpatient hospital services and were used to obtain healthcare utilization information. The Ontario Registrar General – Deaths (ORG) Database was used to confirm mortality.

Study cohort

We identified all individuals (immigrants and long-term residents) who were diagnosed with incident lung cancer at age 40 years and older between April 1, 2012 and March 31, 2017. Long-term residents included anyone not recorded in the IRCC-PR database. We excluded those with invalid identifiers, missing sex, living in rural areas, who had previously been diagnosed with lung cancer prior to April 1st, 2012, who were not residents of Ontario, whose date of last contact with the healthcare system was more than 3 years ago (and thus had limited available data), and those whose cancer was stage 0/in situ. Individuals were followed from study entry until death, last date of OHIP eligibility or until the end of the follow-up period (December 31, 2019), whichever occurred first.

The primary outcome of interest was 5-year overall survival. We selected this outcome over cancer-specific survival due to the small sample size of the immigrant group. We obtained demographic characteristics for individuals in our cohort including age at lung cancer diagnosis (continuous and categorical: 40–64, 65–74, 75 +), sex (male, female), socioeconomic status based on the median neighbourhood income quintile (Q1-lowest to Q5-highest), and comorbidities using the Johns Hopkins ACG® System (version 10) Aggregated Diagnosis Groups (ADGs) [ 13 ]. We also stratified based on cancer stage at diagnosis (early stage (I & II), late stage (III & IV)) and collected information on cancer type (adenocarcinoma, small cell, squamous cell, other).

Immigrant characteristics included immigrant category (economic class, family class, resettled refugee and protected persons, other immigrants), region of origin based on the World Bank regions (East Asia and Pacific, Europe and Central Asia, Latin America and the Caribbean, Middle East and North Africa, South Asia, Sub-Saharan Africa, USANZ), and years since landing.

We assessed the number of primary care visits for each patient and their continuity of care was captured through the Usual Provider of Care (UPC) Index, [ 14 ] which uses primary care physician billing claims in the 6 to 30 months prior to the cancer diagnosis date to determine the proportion of visits to the most-often visited primary care provider. Primary care visits in the 6 months prior to diagnosis may be related to the diagnosis and thus not reflective of usual care. This index is calculated only for patients with at least 3 visits; high continuity of care is defined as > 0.75 [ 14 ].

We used descriptive statistics to summarize the study cohort and baseline characteristics stratified by immigrant status. We compared 5-year overall survival by sociodemographic and clinical characteristics between immigrants and long-term residents using Chi-Square tests and standardized differences. Standardized difference scores are intuitive indexes which measure the effect size between two groups. Compared to a t-test or Wilcoxon rank-sum test, they are independent of sample size. Their use can be recommended for comparing baseline covariates [ 15 ]. Kaplan–Meier curves were produced for 5-year overall survival by age, stage group and immigrant status.

We used Cox proportional hazards models to model the outcome in a time-to-event analysis. Models were censored at 5 years. We explored results stratified by age group, sex, and cancer stage to examine the risk of mortality post-diagnosis for immigrants in relation to long-term residents. Age was included as a continuous variable in the models. We also adjusted models for neighbourhood income quintile, comorbidities, primary care provider visits in the 6 to 30 months before diagnosis, continuity of care, stage at diagnosis and cancer type and present adjusted hazard ratios with 95% confidence intervals and associated p -values. Significance was determined at the p  = 0.05 level with 2-sided p -values. We also conducted a sensitivity analysis including those diagnosed with non-small cell lung cancer only, because of the differences in aggressiveness of the disease. Observations with any missing values were minimal and excluded from the analysis. All analyses were completed using SAS software (version 9.4).

A total of 38,788 individuals were diagnosed with lung cancer between April 1, 2012 and March 31, 2017 including 2,696 immigrants (7%) and 36,092 long-term residents (93%) (Fig.  1 ). Baseline characterteristics of the study cohort have been previously published [ 10 ]. Approximately 49.3% of the cohort was female and 26.5% of the total cohort was diagnosed with early-stage lung cancer. Overall, immigrants were younger at diagnosis (68 years versus 72 years) and were more likely to reside in the lowest neighbourhood income quintile (30.6% versus 24.5%). Immigrants in our cohort also had a lower median number of co-morbidities in comparison to long-term residents (7 (IQR 5–10) versus 8 (IQR 5–11)). Stage of diagnosis did not differ by immigrant status. Mean number of PCP visits in the 6 to 30 months prior to diagnosis was 9.4 ± 8.5 for immigrants and 9.0 ± 8.7 for non-immigrants. A total of 12.2% of individuals had at least 5 years of follow-up. However, among those without 5 years follow-up 84% died within 5 years. Median follow-up time (IQR) was similar among immigrants (5 years with IQR 4–6) and long-term residents (5 years with IQR 4–6).

Cohort creation flow chart

A total of 34,394 individuals had cancer stage information available ( n  = 4,394 with missing stage). Those diagnosed at an early stage were more likely to be males for immigrants (50.7%) and females for long-term residents (53.6%). Those diagnosed at a late stage had fewer comorbidities and were more likely to have small cell carcinoma for both groups. The median number of PCP visits in the 6 to 30 months prior to diagnosis was higher among those diagnosed with early stage for immigrants (9 versus 7 visits) and long-term residents (8 versus 6 visits). The stage of diagnosis was not associated with immigrant class, region of origin or years in Canada.

Table 1 includes immigration characteristics stratified by cancer stage at diagnosis for immigrants with this information available ( n  = 2,405). Most immigrants (81%) had immigrated more than 10 years previous to diagnosis with the median years since landing being 20 years prior (interquartile range 12–24). The most common immigrant category was sponsored family class followed by economic class and resettled refugees and protected persons. The most common region of origin was East Asia and Pacific followed by Europe and Central Asia (Table  1 ).

Table 2 presents the number of deaths in the 5-years after diagnosis by immigrant status, sex, age group and cancer stage. Briefly, 1,571 immigrants died in the cohort compared to 24,721 long-term residents. Figure  2 displays the 5-year overall survival Kaplan Meier curve by immigrant status (Fig.  2 A) stratified by age group (Fig.  2 B) and cancer stage at diagnosis (Fig.  2 C). Overall, survival varied by immigrant status and was higher for immigrants (32.2%) versus long-term residents (20.6%) (Fig.  2 ). Survival rate decreased with increasing age including 26.5% for those aged 40–64, 24.2% for those aged 65–74 and 12.5% for those 75 + . Immigrants had better overall survival even when stratifying by age group and cancer stage.

Kaplan Meier curves with 95% confidence intervals for overall 5-year survival by immigrant status and stratified by age group and cancer stage at diagnosis. A Overall 5-year survival by immigrant status B Overall 5-year survival by immigrant status and age category. C Overall 5-year survival by immigrant status and stage at diagnosis

Table 3 presents the multivariable time-to-event model for the hazard of dying 5-years post diagnosis by immigrant status stratified by cancer stage at diagnosis, sex and age group. Immigrant status was associated with better overall survival for both females (0.7; 95% CI 0.6–0.8) and males (0.7; 95% CI 0.6–0.7) in comparison to long-term residents. This was consistent even when stratified by cancer stage at diagnosis. For example, female immigrants diagnosed with early stage lung cancer had a hazard ratio of 0.5 (95% CI 0.4–0.7) in comparison to long-term residents. A similar pattern was observed for males. However, when stratified by age category female immigrants aged 65–74 diagnosed with early-stage disease did not have a significantly different hazard ratio compared to their long-term resident counterparts. Immigrants diagnosed with late-stage disease had better overall survival for both females and males and when stratified by age group.

Results from the sensitivity analysis including only those diagnosed with non-small cell lung cancer showed that survival varied by immigrant status with 37% of immigrants and 26% of long-term residents surviving at least 5 years (data not shown). Similar to the overall results, in fully adjusted models, immigrant status was either not associated with survival or protective for longer survival in all age, sex, and stage strata.

Our results show that immigrants to Ontario had better overall survival after diagnosis in comparison to long-term residents. Specifically, we found that 5-year overall survival was 32.2% for immigrants compared to only 20.6% for long-term residents. Immigrant status was associated with a lower hazard of dying for both females (0.7; 95% CI 0.6–0.8) and males (0.7; 95% CI 0.6–0.7) in comparison to long-term residents. This trend held true even when stratified by cancer stage at diagnosis.

The observed difference in overall survival by immigrant status for lung cancer has also been observed for other types of cancers including colorectal cancer [ 16 ]. These findings may be partially explained by the healthy immigrant effect, a phenomenon where those who immigrate to a country are in better health than native-born residents. Canadian immigration policies do rely on the health of immigrants as a condition for entry in many cases [ 17 ]. Support for the healthy immigrant effect has been reported including for chronic diseases such as cancers and within the Canadian context and in prior work comparing survival among immigrants and those who are native-born [ 9 , 17 , 18 ]. A study comparing differences in cancer survival between immigrants and non-immigrants in Norway, found that non-Western immigrants had better lung cancer survival compared to Norwegians (HR 0.78; 95% CI 0.71–0.85) [ 19 ]. However, it has also been observed that the advantage of the healthy immigrant effect diminishes [ 9 , 17 , 18 ] or that immigrants converge to non-immigrants [ 9 , 18 ] over time though we did not assess this in our study. Diminishing of the healthy immigrant effect may be a reflection of increasing age and thus increased comorbidities, as we observed increasing hazard ratios with increasing age for both immigrants and long-term residents in our study. Of note, immigrants had fewer comorbidities than non-immigrants regardless of age.

Differences in smoking prevalence have also been found among immigrants and non-immigrants which may help to explain some of our findings related to better survival among immigrants. A recent study exploring disparities in cigarette smoking in Canada found that the relative risk of smoking was higher among males versus females (RR 1.39) and that male and female non-immigrants were more likely than their immigrant counterparts to smoke (RR 1.26 for male non-immigrants and 2.93 for female non-immigrants) [ 20 ].

Future research should consider the heterogeneity that exists within immigrant populations including the intersection with race. For example, immigrants in our study had immigrated from a number of regions including East, Central or South Asia and Europe. It would be important to explore how health outcomes may differ between immigrants who are racialized versus not. Some prior work suggests that immigrants who are racialized may experience worse health outcomes as a result of racial discrimination and socioeconomic disadvantage [ 21 ]. Additionally, many immigrant enclaves exist in Canada in big cities like Toronto. As such, the potential protective effects of immigrant enclaves and immigrant-based residential segregation should be considered in future work [ 22 ].

The results of this study should be interpreted considering study strengths and limitations. Limitations of this study include those common to studies using health administrative databases including limitations in data availability and coding errors. In this case, the IRCC-PR database contains information on all permanent residents only since 1985. As a result, we were unable to ascertain if some of the long-term residents in our analyses were indeed immigrants from prior to 1985. We also did not have information on any incident lung cancer diagnoses if these occurred while an individual was living in another country or if they are Canadian-born but used to live in another province. Additionally, we excluded those living in rural areas as the number of immigrant lung cancer patients residing in rural areas is very small. Given the limited number of immigrants living in rural areas, we don’t believe that inclusion of those living in rural regions would have materially impacted the results of our study. However, this could be an area for further research to examine outcomes among those living in rural areas. We also excluded those with no contact with the healthcare system in the last 3 years. However, we anticipate those with limited contact with the healthcare system are likely to be healthier and our estimates are likely conservative of the true effect. We also recognize that the data set used for this study had follow-up until 2019, before the COVID-19 pandemic. As such, it is possible that our findings may not be the same as seen today given the large influx of refugees since the pandemic and other global pressures forcing individuals around the world to immigrate. Strengths of this work include use of databases that contain a complete census of cancers in the province given the context of a universal healthcare system. We were also able to ascertain survival by stage of diagnosis for the majority of people diagnosed with lung cancer during our study period. The results of our study may have some limited generalizability to other Canadian provinces with similar population demographics including a large number of immigrants and similar set-up of healthcare services such as British Columbia.

Conclusions

In conclusion, overall survival 5-years post diagnosis with lung cancer is better among Ontario immigrants versus long-term residents. In a system of universal healthcare, additional strategies to address previously documented socioeconomic inequalities especially for those living in the lowest income neighbourhoods are needed given the social gradient of cigarette smoking, one of the biggest risk factors for lung cancer. Additional research, potentially on the protective effects of immigrant enclave and the intersection of immigrant stautus with racial/ethnic identity, is needed to further explore why better overall survival for immigrants remained despite the reported diminishing impact of the healthy immigrant effect.

Availability of data and materials

The dataset from this study is held securely in coded form at ICES. While legal data sharing agreements between ICES and data providers (e.g., healthcare organizations and government) prohibit ICES from making the dataset publicly available, access may be granted to those who meet pre-specified criteria for confidential access, available at www.ices.on.ca/DAS (email: [email protected]). The full dataset creation plan and underlying analytic code are available from the authors upon request, understanding that the computer programs may rely upon coding tem-plates or macros that are unique to ICES and are therefore either inaccessible or may require modification.

Abbreviations

Aggregated Diagnosis Groups

Confidence Interval

Discharge Abstract Database

Hazard Ratio

Institute for Clinical Evaluative Sciences

Interquartile range

Immigration, Refugees and Citizenship Canada – Permananet Resident

National Ambulatory Care Reporting System

Ontario Cancer Registry

Ontario Health Insurance Plan

Ontario Registrar General – Deaths

Primary Care Provider

Registered Persons Database

Socioeconomic Status

United Kingdom

Usual Provider Index

United States, Australia and New Zealand

Canadian Cancer Statistics Advisory Committee in collaboration with the Canadian Cancer Society Statistics Canada and the Public Health Agency of Canada. Canadian Cancer Statistics: A 2022 special report on cancer prevalence. Toronto: Canadian Cancer Society; 2022. Available from: https://cancer.ca/Canadian-Cancer-Statistics-2022-EN . Accessed 10 Feb 2023.

Brenner DR, Poirier A, Woods RR, Ellison LF, Billette JM, Demers AA, et al. Projected estimates of cancer in Canada in 2022. Can Med Assoc J. 2022;194(17):E601 LP-E607.

Article   Google Scholar  

Forrest LF, Adams J, Wareham H, Rubin G, White M. Socioeconomic inequalities in lung cancer treatment: systematic review and meta-analysis. PLoS Med. 2013;10(2):e1001376.

Article   PubMed   PubMed Central   Google Scholar  

Redondo-Sánchez D, Petrova D, Rodríguez-Barranco M, Fernández-Navarro P, Jiménez-Moleón JJ, Sánchez M-J. Socio-Economic inequalities in lung cancer outcomes: an overview of systematic reviews. Cancers (Basel). 2022;14(2):398.

Article   PubMed   Google Scholar  

Sosa E, D’Souza G, Akhtar A, Sur M, Love K, Duffels J, et al. Racial and socioeconomic disparities in lung cancer screening in the United States: a systematic review. CA Cancer J Clin. 2021;71(4):299–314.

Finke I, Behrens G, Weisser L, Brenner H, Jansen L. Socioeconomic Differences and Lung Cancer Survival-Systematic Review and Meta-Analysis. Front Oncol. 2018;8:536.

Statistics Canada. 2021 Census of Population. Available from: https://www12.statcan.gc.ca/census-recensement/2021/as-sa/fogs-spg/page.cfm?topic=9&lang=E&dguid=2021A000235 Accessed 10 Feb 2023.

Statistics Canada. 2016 Census of Population. Available from: https://www12.statcan.gc.ca/census-recensement/2016/dp-pd/index-eng.cfm Accessed 10 Feb 2023.

De Maio FG. Immigration as pathogenic: a systematic review of the health of immigrants to Canada. Int J Equity Health. 2010;9:27. https://doi.org/10.1186/1475-9276-9-27 .

Lofters AK, Gatov E, Lu H, Baxter NN, Guilcher SJT, Kopp A, Vahabi M, Datta GD. Lung cancer inequalities in stage of diagnosis in Ontario, Canada. Cur Onc. 2021;28(3):1946–56. https://doi.org/10.3390/curroncol28030181 .

Chan B, Anderson GM, et al. Trends in physician fee-for-service billing patterns. In: Goel V, Williams JI, Anderson GM, et al., editors. Patterns of Health Care in Ontario. The ICES Practice Atlas. 2nd ed. Ottawa, Ontario: Canadian Medical Association; 1996. p. 247.

Google Scholar  

Benchimol EI, Smeeth L, Guttmann A, Harron K, Moher D, Petersen I, et al. The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement. PLOS Med. 2015;12(10):e1001885. https://doi.org/10.1371/journal.pmed.1001885 .

The Johns Hopkins University. About the ACG System. Available from: https://www.hopkinsacg.org/about-the-acg-system/ Accessed on 10 Feb 2023.

Breslau N, Reeb KG. Continuity of care in a university-based practice. J Med Educ. 1975;50(10):965–9.

PubMed   Google Scholar  

Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med. 2009;28(25):3083–107. https://doi.org/10.1002/sim.3697 . PMID:19757444;PMCID:PMC3472075.

Coughlin SS. Social determinants of colorectal cancer risk, stage, and survival: a systematic review. Int J Colorectal Dis. 2020;35:985–95. https://doi.org/10.1007/s00384-020-03585-z .

Government of Canada. Medical inadmissibility. Available from: https://www.canada.ca/en/immigration-refugees-citizenship/services/immigrate-canada/inadmissibility/reasons/medical-inadmissibility.html Accessed 14 Feb 2023.

Cheung MC, Earle CC, Fischer HD, Camacho X, Liu N, Saskin R, Shah BR, Austin PC, Singh S. Impact of immigration status on cancer outcomes in Ontario, Canada. J Oncol Pract. 2017;13(7):e602–12. https://doi.org/10.1200/JOP.2016.019497 . Epub 2017 Jun 12 PMID: 28605254.

Thøgersen H, Møller B, Robsahm TE, Babigumira R, Aaserud S, Larsen IK. Differences in cancer survival between immigrants in Norway and the host population. Int J Cancer. 2018;143(12):3097–105. https://doi.org/10.1002/ijc.31729 . Epub 2018 Oct 3 PMID: 29987865.

Chaiton M, Callard C. Mind the Gap: Disparities in Cigarette Smoking in Canada. Tob Use Insights. 2019;12:1179173X19839058. https://doi.org/10.1177/1179173X19839058 . PMID: 30944522; PMCID: PMC6437323.

Engelman M, Ye LZ. The immigrant health differential in the context of racial and ethnic disparities: the case of diabetes. Adv Med Sociol. 2019;19:147–71. https://doi.org/10.1108/S1057-629020190000019008 . PMID:31057317;PMCID:PMC6494443.

Hiebert D. Ethnocultural Minority Enclaves in Montreal, Toronto and Vancouver. Montreal: Institute for Research on Public Policy; 2015.  https://oaresource.library.carleton.ca/ethnominoritystudy-no52.pdf .

Download references

This study was supported by ICES, which is funded by an annual grant from the Ontario Ministry of Health (MOH) the Ministry of Long-Term Care (MLTC). This study also received funding from the Canadian Institutes for Health Research (CIHR) FRN 162506. However, the analyses, conclusions, opinions and statements expressed herein are solely of the authors and do not reflect those of the funding or data sources; no endorsement is intended or should be inferred. This document used data adapted from the Statistics Canada Postal CodeOM Conversion File, which is based on data licensed from Canada Post Corporation, and/or data adapted from the Ontario Ministry of Health Postal Code Conversion File, which contains data copied under license from ©Canada Post Corporation and Statistics Canada. Parts of this mate-rial are based on data and/or information compiled and provided by the Ontario Ministry of Health, Canadian Institute for Health Information (CIHI), Ontario Health (OH), and Immigration, Refugees and Citizenship Canada (IRCC) current to May 2017. However, the analyses, conclusions, opinions and statements expressed herein are solely of the authors and do not reflect those of the funding or data sources; no endorsement is intended or should be inferred.

Author information

Authors and affiliations.

Interdisciplinary Health Program, St. Francis Xavier University, Antigonish, NS, Canada

Arlinda Ruco

Beatrice Hunter Cancer Research Institute, Halifax, NS, Canada

VHA Home HealthCare, Toronto, ON, Canada

Nova Scotia Health, Halifax, NS, Canada

Peter Gilgan Centre for Women’s Cancers , Women’s College Hospital, Toronto, ON, Canada

Arlinda Ruco & Aisha K. Lofters

Li Ka Shing Knowledge Institute – St. Michael’s Hospital, Unity Health Toronto, Toronto, ON, Canada

Aisha K. Lofters & Nancy N. Baxter

Department of Family and Community Medicine, University of Toronto, Toronto, ON, Canada

Aisha K. Lofters

ICES: Institute for Clinical Evaluative Sciences, Toronto, ON, Canada

Aisha K. Lofters, Hong Lu, Nancy N. Baxter, Sara Guilcher, Alexander Kopp & Mandana Vahabi

Melbourne School of Population and Global Health, University of Melbourne, Melbourne, Australia

Nancy N. Baxter

Department of Surgery, University of Toronto, Toronto, ON, Canada

Leslie Dan Faculty of Pharmacy, University of Toronto, Toronto, ON, Canada

Sara Guilcher

Daphne Cockwell School of Nursing, Toronto Metropolitan University, Toronto, ON, Canada

Mandana Vahabi

Department of Social and Preventive Medicine, Université de Montréal, Montréal, QC, Canada

Geetanjali D. Datta

Research Center of the University of Montreal Hospital Center (CR-CHUM), Montréal, QC, Canada

Cancer Research Center for Health Equity, Cedars-Sinai Medical Center, Los Angeles, CA, USA

You can also search for this author in PubMed   Google Scholar

Contributions

Conceptualization, AL, GDD.; Methodology AL, GDD, HL, NNB, SG, AK, MV; Analysis HL, AK; Interpretation AR, AL, HL, NNB, SG, AK, MV, GDD; Resources, AL, GDD; Writing – Original Draft Preparation, AR, AL, GDD; Writing – Review & Editing, AR, AL, HL, NNB, SG, AK, MV, GDD; Supervision, AL, GDD; Funding Acquisition, AL, GDD. All authors have read and agreed to the published version of the manuscript.

Corresponding author

Correspondence to Geetanjali D. Datta .

Ethics declarations

Ethics approval and consent to participate.

This study was approved by the Research Ethics Board at Unity Health Toronto (REB #19–072). ICES is a prescribed entity under Ontario’s Personal Health Information Protection Act (PHIPA). Section 45 of PHIPA authorizes ICES to collect personal health information, without consent, for the purpose of analysis or compiling statistical information with respect to the management of, evaluation or monitoring of, the allocation of resources to or planning for all or part of the health system. Projects that use data collected by ICES under Sect. 45 of PHIPA, and use no other data, are exempt from REB review. The use of the data in this project is authorized under Sect. 45 and approved by ICES’ Privacy and Legal Office.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary Information

Supplementary material 1., rights and permissions.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/ .

Reprints and permissions

About this article

Cite this article.

Ruco, A., Lofters, A.K., Lu, H. et al. Lung cancer survival by immigrant status: a population-based retrospective cohort study in Ontario, Canada. BMC Cancer 24 , 1114 (2024). https://doi.org/10.1186/s12885-024-12804-7

Download citation

Received : 09 February 2024

Accepted : 13 August 2024

Published : 07 September 2024

DOI : https://doi.org/10.1186/s12885-024-12804-7

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Lung cancer
• Retrospective cohort study

ISSN: 1471-2407

• Submission enquiries: [email protected]
• General enquiries: [email protected]

Log in using your username and password

• Search More Search for this keyword Advanced search
• Latest content
• For authors
• Browse by collection
• BMJ Journals

You are here

• Volume 14, Issue 9
• Mixed-methods protocol for the WiSSPr study: Women in Sex work, Stigma and psychosocial barriers to Pre-exposure prophylaxis in Zambia
• Article Text
• Article info
• Citation Tools
• Rapid Responses
• Article metrics

• http://orcid.org/0000-0002-4570-6686 Ramya Kumar 1 , 2 ,
• http://orcid.org/0000-0003-4076-0170 Deepa Rao 3 ,
• http://orcid.org/0000-0002-8189-0732 Anjali Sharma 1 ,
• Jamia Phiri 1 ,
• Martin Zimba 4 ,
• Maureen Phiri 4 ,
• Ruth Zyambo 5 ,
• Gwen Mulenga Kalo 5 ,
• Louise Chilembo 5 ,
• Phidelina Milambo Kunda 6 ,
• Chama Mulubwa 1 ,
• Benard Ngosa 1 ,
• http://orcid.org/0000-0001-5208-7468 Kenneth K Mugwanya 7 ,
• Wendy E Barrington 8 ,
• http://orcid.org/0000-0002-3629-3867 Michael E Herce 1 , 9 ,
• http://orcid.org/0000-0001-9968-7540 Maurice Musheke 1
• 1 Centre for Infectious Disease Research in Zambia , Lusaka , Zambia
• 2 Epidemiology , University of Washington School of Public Health , Seattle , Washington , USA
• 3 University of Washington School of Public Health , Seattle , Washington , USA
• 4 Zambia Sex Workers Alliance , Lusaka , Zambia
• 5 Tithandizeni Umoyo Network , Lusaka , Zambia
• 6 Lusaka District Health Office , Zambia Ministry of Health , Lusaka , Zambia
• 7 Epidemiology, Global Health , University of Washington School of Public Health , Seattle , Washington , USA
• 8 Epidemiology; Child, Family, and Population Health Nursing; Health Systems and Population Health , University of Washington School of Public Health , Seattle , Washington , USA
• 9 Institute for Global Health and Infectious Diseases , University of North Carolina at Chapel Hill , Chapel Hill , North Carolina , USA
• Correspondence to Dr Ramya Kumar; ramya.kumar.mlk{at}gmail.com

Introduction Women engaging in sex work (WESW) have 21 times the risk of HIV acquisition compared with the general population. However, accessing HIV pre-exposure prophylaxis (PrEP) remains challenging, and PrEP initiation and persistence are low due to stigma and related psychosocial factors. The WiSSPr (Women in Sex work, Stigma and PrEP) study aims to (1) estimate the effect of multiple stigmas on PrEP initiation and persistence and (2) qualitatively explore the enablers and barriers to PrEP use for WESW in Lusaka, Zambia.

Methods and analysis WiSSPr is a prospective observational cohort study grounded in community-based participatory research principles with a community advisory board (CAB) of key population (KP) civil society organi sations (KP-CSOs) and the Ministry of Health (MoH). We will administer a one-time psychosocial survey vetted by the CAB and follow 300 WESW in the electronic medical record for three months to measure PrEP initiation (#/% ever taking PrEP) and persistence (immediate discontinuation and a medication possession ratio). We will conduct in-depth interviews with a purposive sample of 18 women, including 12 WESW and 6 peer navigators who support routine HIV screening and PrEP delivery, in two community hubs serving KPs since October 2021. We seek to value KP communities as equal contributors to the knowledge production process by actively engaging KP-CSOs throughout the research process. Expected outcomes include quantitative measures of PrEP initiation and persistence among WESW, and qualitative insights into the enablers and barriers to PrEP use informed by participants’ lived experiences.

Ethics and dissemination WiSSPr was approved by the Institutional Review Boards of the University of Zambia (#3650-2023) and University of North Carolina (#22-3147). Participants must give written informed consent. Findings will be disseminated to the CAB, who will determine how to relay them to the community and stakeholders.

• MENTAL HEALTH
• HIV & AIDS
• EPIDEMIOLOGIC STUDIES
• Health Equity
• QUALITATIVE RESEARCH
• SOCIAL MEDICINE

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/ .

https://doi.org/10.1136/bmjopen-2023-080218

Statistics from Altmetric.com

Request permissions.

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

STRENGTHS AND LIMITATIONS OF THIS STUDY

The Women in Sex work, Stigma and PrEP (WiSSPr) study uses a mixed-methods approach which is ideal for intersectional stigma research because it allows quantitative research to be grounded in the lived experiences of people, while ensuring that aspects of stigma that emerge at the intersections of identities are measured in testable ways.

Qualitative aim enrolls peer navigators to capture the perspectives of women who are at the unique interface of recipients of care as sex workers themselves, and supporters of health service delivery.

Uses core principles of community-based participatory research which value key populations as equal contributors to the knowledge production process.

Limitations include an inability to longitudinally assess the alignment of pre-exposure prophylaxis (PrEP) adherence and persistence with HIV risk, and limitations in measuring PrEP adherence by self-report and pharmacy dispensations instead of by drug biomarkers.

Introduction

Women engaging in sex work (WESW) are a key population (KP) that experiences an unacceptably high risk of HIV infection. In 2019, the Joint United Nations Programme on HIV/AIDS (UNAIDS) estimate WESW have 21 times the risk of HIV acquisition compared with the general population of adults aged 15 – 49 years old. 1 In Southern and East Africa, KPs and their sexual partners account for 25% of all new HIV infections. 2 To reduce the burden of HIV in Africa, HIV prevention strategies tailored to the unique needs of WESW are critical to safeguarding their health, as well as the health of people in their sexual networks. 3 4

While HIV pre-exposure prophylaxis (PrEP) is highly effective in preventing HIV infection, its real-world efficacy is closely linked to adherence, which is a complex process for WESW. A systematic review of PrEP usage and adherence among WESW reveals complex interrelationships between individual perceptions of HIV risk, social support and fear of healthcare provider stigma. 5 WESW may experience multiple stigmatised identities, conditions or behaviours, such as participating in sex work, having a substance use disorder, and taking HIV prevention medication. 6

Zambia has a generalised HIV epidemic, and the capital city of Lusaka is a major regional transit hub attracting WESW from the region. Approximately 3,396 live in Lusaka with over half (53%) living with HIV, underscoring the need to urgently tailor prevention strategies for this population. 7 WESW in Zambia are subject to violence and discrimination in the form of verbal, physical and sexual abuse from strangers, acquaintances, clients, intimate partners and even law enforcement. 8 Surveys among WESW in Zambia have identified healthcare provider stigma and discrimination, as well as a lack of confidential care as main barriers to HIV prevention services at public health facilities. 7 9 Therefore, a better understanding of the multiple stigmas that WESW experience is a critical first step to designing interventions to meet their HIV prevention needs.

In recent years, Zambia has made significant progress in reaching WESW and providing them with comprehensive HIV prevention services. Since May 2019, the PEPFAR-funded Key Population Investment Fund (KPIF) has been successfully engaging with KP in Lusaka Province and providing them with community-based HIV prevention and treatment services. KPIF is implemented by the Centre for Infectious Disease Research in Zambia (CIDRZ) in partnership with the Zambian Ministry of Health (MoH), US Centers for Disease Control and Prevention and importantly, key population civil society organisations (KP-CSOs). A key objective of the KPIF programme is to improve PrEP initiation, persistence and adherence for HIV-negative WESW. For this study, we propose to leverage existing KPIF infrastructure to enhance study feasibility and ensure its real-world relevance to achieving this key objective.

Although PrEP initiations are high in the KPIF programme, they may not accurately reflect PrEP effectiveness. 10 A systematic review of 41 studies found high discontinuation rates at 1 month. 11 Despite WHO recommendations and national PrEP guidelines for regular HIV testing and follow-up visits, maintaining client engagement with PrEP has been challenging. 12 13 This has resulted in a lack of data on short-term PrEP persistence among WESW in Zambia. Assessing the percentage of clients who do not return for their first follow-up visit is crucial for determining PrEP effectiveness. Current prevention strategies do not adequately address the multiple stigmas and psychosocial stress that hinder PrEP persistence.

Specific objectives

The Women in Sex work, Stigma and PrEP (WiSSPr) mixed-methods study aims to (1) measure the association between multiple stigmas on PrEP initiation and persistence among HIV-negative adult WESW and (2) qualitatively explore the enablers and barriers (interpersonal, psychosocial and structural) to initiating and persisting on PrEP. The qualitative aim will complement and contextualise 14–16 findings from the quantitative results. We hypothesize that WESW with high levels of any type of stigma will be less likely to initiate and persist on PrEP.

Conceptual framework

Interview guides will be informed by the Community, Opportunity, Motivation – Behaviour (COM-B) framework to assess how these components drive engagement with PrEP services. 17 18 The COM-B model is commonly used in HIV prevention because it offers a framework to guide the development and implementation of targeted interventions, thereby enhancing the efficacy and reach of HIV prevention programmes. 19 This framework will guide us to identify deficits in knowledge or skills (Capability), environmental and social contexts (Opportunity), and personal motivations and attitudes (Motivation). This integrated approach ensures that all relevant aspects of behaviour change are considered, leading to more effective and sustainable health outcomes.

Directed acyclic graph

Directed acyclic graphs (DAG) visually synthesise a priori knowledge about the hypothesised relationships between variables of interest, helping to identify causal pathways and potential confounders that could bias the results. We propose confounders based on their known association with stigmas and PrEP persistence, using evidence from published studies addressing similar questions. Controlling for the following variables will be sufficient to block any unconditionally open, non-causal backdoor paths that could lead to confounding: age, community hub, duration of sex work, and education ( figure 1 ).

• Download figure
• Open in new tab
• Download powerpoint

Directed acyclic graph illustrating the causal effect of stigma on PrEP persistence. PrEP, pre-exposure prophylaxis.

Methods and analysis

Study design.

We will use a prospective observational cohort study design with mixed methods to characterise PrEP outcomes for HIV-negative WESW in Lusaka, Zambia. Trained research assistants will administer a one-time, 75-item psychosocial survey to participants and follow them prospectively in the electronic medical record. For the qualitative aim, we will conduct in-depth interviews (IDIs) with WESW to get perspectives of prevention services with peer navigators who are both recipients of care and supporters of health service delivery.

Mixed-methods integration

We will use the NIH ‘Best Practices for Mixed Methods’ guidelines to design, analyse and interpret qualitative and quantitative data in mixed-methods research. 20 Specifically, we will employ a convergent parallel design that collects both qualitative and quantitative data concurrently and separately, prioritising both the quantitative and qualitative strands equally but keeping them independent during analysis. We will interpret the extent to which the two sets of results converge, diverge, relate to each other and/or combine to create a better understanding in response to the study’s overall purpose. 20

Study setting

The study population is composed of adult WESW who are living or working within the catchment areas of two community hubs located within urban Lusaka. Based on CIDRZ’s prior published work, we anticipate that the study population will be comprised largely (63%) of younger women (18 – 29 years old). 10

Study exposures and outcomes

Table 1 identifies the primary outcomes of PrEP initiation and persistence from pharmacy dispensations records in the last 90 days for survey participants. Several studies have accessed this data from the national electronic medical record system SmartCare. 21 22 CIDRZ is a key Smartcare implementing partner and routinely leveraging this data to optimise service delivery for KP in KPIF in order to better understand outcomes for HIV treatment and prevention in the national HIV programme. 23–28 Table 2 identifies the independent variables of interest including sociodemographic history, intersectional stigma (everyday discrimination scale), 29 substance use (ASSIST), 30 depressive symptoms (Patient Health Questionnaire, PHQ), 31 as well as sex work, HIV and PrEP-related stigmas and resulting discrimination using established questionnaires. 32–34 The qualitative outcomes are insights into the enablers and barriers to PrEP use informed by participants’ lived experiences according to the COM-B model.

• View inline

WiSSPr study outcomes

WiSSPr study independent variables

Sample size

We determined the minimum sample size using Demidenko’s method for logistic regression with binary interactions, informed by effect size and variance data from Witte et al ’s study on PrEP acceptability among women in Uganda. 35–37 Sample size considerations are based on our primary outcome of PrEP initiation and informed by preliminary programmatic data that formed assumptions about baseline HIV prevalence and estimated PrEP initiations. Each site tests an average of 200 WESW per month, which will allow an estimated 800 women to be tested during the 2-month enrolment period. We project approximately 56% (448) will test HIV-negative, and of these, we estimate 403 (90%) will be eligible, and 350 (87%) will agree to initiate PrEP. Due to time and resource limitations, we seek to enroll a sample of 300 eligible WESW. Assuming 5% of participant medical records cannot be found, a total cohort of 285 PrEP users would allow us to estimate the prevalence ratio of stigma on PrEP initiation of 1.98 or higher (positive association), or 0.50 or lower (negative association) at 80% power with a significance level of 0.05. We aim to recruit 18 participants for IDIs, based on prior research with this population and qualitative methodology guidelines suggesting that 6 – 10 interviews per subgroup are sufficient to reach thematic saturation 14 20

Participant recruitment

The study will start in July 2023. WiSSPr will recruit 300 participants from a convenience sample of WESW who are receiving HIV services from two community-based hubs which have been functioning as MoH drop-in wellness centres since October 2021. All HIV testing and prevention services at these community hubs are led by teams of KP and MoH staff. Outreach activities take place in venues where WESW socialise, such as brothels, bars, or the home of a KP. Recruitment activities will take place during these outreach activities. KPIF programming leverages KP social networks to mobilise WESW for recruitment into the study. A total of 18 participants, including 6 peer navigators, 6 WESW who discontinue PrEP after initiation, and 6 WESW who continue on PrEP, will be purposively sampled for IDIs, or until we achieve thematic saturation. 38 Qualitative data collection will take place at least 30 days after the quantitative recruitment begins, in order to sample women who initiate a 30 day supply of PrEP but do not return to pick up another refill. Figure 2 outlines the WiSSPr study recruitment process.

The WiSSPr study flow diagram summarises the stages of participant recruitment and follow-up. PrEP, pre-exposure prophylaxis; WiSSPr, Women in Sex work, Stigma and PrEP.

Recruitment will end when 300 participants have been enrolled for the survey and 18 participants enrolled for interviews. PrEP event data will be abstracted from SmartCare approximately 3 months after the final participant’s enrollment. Study activities, including qualitative data collection, data quality control and assurance, and data analysis, are anticipated to continue until the planned end of the study in September 2024.

We will engage the community advisory board (CAB) in collaborative decision-making on: (1) how best to conduct outreach to venues that WESW frequent, (2) how to engage leaders in the sex work community to inform them about this study, and (3) to encourage WESW participation in a way that minimises social harms. Box 1 identifies the inclusion and exclusion criteria for the study. Written informed consent in English or local languages (ChiNyanja or IchiBemba) will be obtained before enrollment. As an added measure of protection for this marginalised population, participants must complete an informed consent quiz to ensure that they understand the potential risks of study participation. Participants will receive the Zambia Kwacha equivalent of US$5 per survey and interview as compensation for their time.

Inclusion and exclusion criteria

Cohort inclusion and exclusion criteria are as follows:

Inclusion criteria: (1) identify as a cis-gendered or transgendered woman, (2) age ≥ 18 years, (3) earns a significant amount of income from exchanging sex for money or goods in the last 3 months, (4) HIV-negative status and eligible for PrEP according to national guidelines, (5) not planning to transfer care to another site within the next 30 days, (6) speaks English or ChiNyanja or IchiBemba and (7) willing and able to provide written informed consent

Exclusion criteria: (1) do not identify as a woman, (2) age < 18 years old, (3) has not earned a significant amount of income from exchanging sex for money or goods or has earned for < 3 months, (4) HIV-positive status or status is unknown or ineligible for PrEP, (5) planning to transfer care to another site within the next 30 days, (6) unable to speak English or ChiNyanja or IchiBemba and (7) not willing or able to provide written informed consent

In-depth interviews will be conducted with cohort members, as well as peer navigators. The inclusions and exclusion criteria for peer navigators is as follows:

Inclusion criteria: (1) age ≥ 18 years old, (2) history working as a peer health navigator, (3) history of providing HIV services to women engaging in sex work, (4) speaks English or ChiNyanja or IchiBemba and (5) willing and able to provide written informed consent.

Exclusion criteria: (1) age < 18 years, (2) does not have a history working as a peer health navigator, (3) does not have a history of providing HIV services to women engaging in sex work, (4) unable to speak English or ChiNyanja or IchiBemba and (5) not willing or able to provide written informed consent.

Quantitative data collection

A team of 3–5 trained research assistants will administer a tablet-based survey ( online supplemental file 1 ) for quicker data entry, real-time quality control and logic checks to reduce data entry errors and immediate data backup compared with paper. Surveys, estimated to take 60 min each, will be conducted in English, ChiNyanja or IchiBemba, based on participant preference. The survey tool will be piloted with CAB members and peer navigators. Patient medical records are routinely entered by KPIF programme staff into a secure, standardised electronic data capture system, from which we will extract relevant deidentified data using the participants’ SmartCare ID numbers.

Supplemental material

Qualitative data collection.

We will use a semi-structured interview guide ( online supplemental file 1 ) with open-ended questions and probes to explore specific themes related to HIV prevention and intersectional stigma. This guide allows some flexibility for participants to follow topics of interest to them. The themes we will explore are informed by the COM-B conceptual framework which include perceived and enacted stigma, the impact of intersectional stigmas on health service utilisation service needs, enablers such as psychosocial support or the trustworthiness of the healthcare system. The guide also includes modules on PrEP where the interviewer will explain oral and long-acting injectable PrEP and assess participants perceptions of the advantages and disadvantages and willingness to use these different PrEP options. Participants will be asked about their own perceptions as well as their perceived opinions of their peers, as this approach has yielded richer responses in previous studies. 39 Interviews are estimated to take 60 minutes and will be conducted in English, ChiNyanja, or IchiBemba in a private location at a community safe space or other similarly secure location determined by participant preference. We will request permission to audio record interviews for transcription and translation. All interviews will be conducted by a single trained interviewer. The interview guides will be piloted with CAB members before implementation.

Data management

SmartCare serves as a repository of clinical data for WESW accessing KPIF services. A secure server will be used to store encrypted study data, including the study database. Quantitative data collected on tablets will be transmitted to the server at the end of each day. To ensure data safety, there will be daily backups, and data will also be stored on secure drives.

All IDIs will be audio recorded. Audio recordings will be transcribed verbatim and then translated into English in a single step by qualified research staff. The audio recordings will not be marked with any identifying information. Instead, interviewers will use unique participant codes to label the audio recordings. No personal identifiers will be used, and any identifiers inadvertently mentioned during interviews will be purged from the transcripts prior to analysis.

All medical records that contain participant identities are treated as confidential in accordance with the Zambian Data Protection Act. All study documents related to the participants will only include an assigned participant code. Only research staff will have access to linkable information, which will be kept strictly confidential. All records will be archived in a secure storage facility for 3 years after the completion of the study per local regulatory guidelines, after which time all electronic data will be deleted from project servers and hard drives, and all paper-based records will be disposed of.

Quantitative data analysis

We will conduct univariable analyses to examine whether there are differences in the levels of stigma, discrimination, depressive symptoms and substance use disorders among those who initiate PrEP versus those who do not, stratified by community hub. We will report the prevalences of HIV and PrEP stigmas, discrimination due to intersectional stigma identified by the Everyday Discrimination scale, depression and suicidal ideation identified by PHQ, and substance use disorders identified by ASSIST. We will sum all items within a screener to a total score before collapsing data into categorical variables. For cases where missing data are more limited (approximately < 5%), for single items and measures, we will use mean imputation to derive a score. If there is substantial missingness (> 10%) then we will use missing data methods such as multiple imputation.

A PHQ-9 score ≥ 10 is commonly used in primary care settings as a cut-off for probable major depression. 40 PHQ-9 cut-off scores of 5, 10, 15 and 20 will be categorised as mild, moderate, moderately severe and severe depression, respectively. The ASSIST gives 10 risk scores for tobacco, alcohol, cannabis, cocaine, amphetamine-type stimulants, inhalants, sedatives, hallucinogens, opioids and other drugs. The score is higher the more frequently the participant reports using substances. For alcohol use, we will use cut-offs of 11 and 27 for moderate and high risk of substance use disorder. For all other substances cut-offs of 4, and 27 for moderate and high risk. 30

PrEP initiation will be calculated using the total number of individuals initiated on PrEP over the total number of HIV-negative individuals who were enrolled and eligible for PrEP. We refer to the complement of discontinuation as PrEP persistence. 41 We define immediate discontinuation for those who initiate a 30 day supply of PrEP and do not return for any refills over the 108 day observation period in alignment with national antiretroviral therapy (ART) programme guidelines on continuity of care and management of missed appointments. 21 42 We will calculate a medication possession ratio (MPR) of total days with medication in patient possession to the observation period, as a measure of engagement in services and report both the MPR and IQR ( table 1 ).

We will use Stata (V.16.1, StataCorp) for analysis, reporting descriptive statistics to characterise the study population and bivariate associations between key exposures and immediate discontinuation with Pearson’s χ 2 statistics. We will fit Poisson regression models, which will estimate prevalence ratios of discrimination, PrEP stigma and HIV stigma on immediate discontinuation of PrEP over a 3-month follow-up period, controlling for confounders identified by the DAG. Adjusted prevalence ratio estimates will be reported with 95% CIs and p-values at the alpha = 0.05 significance level.

Qualitative data analysis

We will analyse the qualitative data using established analytical software (NVivo, QSR International, Melbourne, Australia) through deductive reasoning based on our conceptual model and inductive reasoning to identify major and minor themes emerging from audio recordings and transcripts. The process of eliciting themes will involve familiarisation with interview transcripts and noting emergent themes, adapting our conceptual framework as necessary, performing open coding, developing a codebook, performing data reduction, data display using matrices and/or tables, and interpretation to map out relationships in the data. Two coders will review these data, independently identify emergent themes, and confer to agree on final coding and findings. We will apply established qualitative research principles in our analyses, including negative case analysis and respondent validation. 43 44

Participant attitudes and preferences relating to elements of future stigma-reduction intervention, psychosocial support provision and long-acting injectable PrEP will be described qualitatively. We will strive for critical reflexivity by outlining our point of view in relation to the interviewees of the study during data collection and will state how positionality and context may have affected the findings. The credibility and trustworthiness of qualitative data will be assured through member-checking by participants themselves. 45

Ethics and dissemination

WiSSPr was approved by the Institutional Review Boards of the University of Zambia (#3650 -2023) and University of North Carolina, the Zambia National Health Research Authority and the Lusaka Provincial and District Health Offices. A final study notification will be sent on completion of the study, or in the event of early termination. Participants are free to withdraw from the study at any time without affecting their right to medical care.

The study findings will be disseminated to KP community members, providers, researchers and policy-makers. The CAB will review preliminary results and advise on meaningful dissemination to the KP community, National AIDS Council, National HIV and Mental Health Technical Working Groups, investigators and stakeholders. The information will be presented at conferences or published in peer-reviewed journals. Participants’ personal information will not be included in any publications.

Patient and public involvement

We will use principles of community-based participatory research (CBPR) to ensure patient and public involvement in this study. CBPR is a research paradigm that focuses on relationships between academic and community partners, with principles of co-learning, mutual benefit and long-term commitment. 46 CBPR incorporates community theories, participation, and practices into the research efforts and plays a role in expanding the reach of implementation science to influence practice and policies for eliminating health disparities. 46 47

To collaboratively develop this study with clients and the public, we will use CBPR principles and create a CAB with Lusaka District Health Office and two KP-CSOs working in the study sites: Zambia Sex Workers Alliance and Tithandizeni Umoyo Network. As a study team, our first priority is to develop trust with people engaging in sex work. Trust development is a construct of CBPR and has also emerged as a synthesising theory. 48 49 Trust types are ordered along a relative continuum from least (trust deficit) to most (critical reflective) trust which reflects an ability to discuss and move on after a misstep. 48 Given the historical marginalisation and stigmatisation of WESW in Zambia, we anticipate a trust deficit and have allocated time and budget to nurture and develop trust along this continuum. We will build trust through ‘role-based trust’ as researchers, ‘proxy trust’ from the reputation of CIDRZ and KP CSO team members’ work with KPs in Zambia, and ultimately aim to establish ‘critical reflective’ trust.

The research questions and outcome measures were developed in collaboration with the CAB, ensuring they reflect the priorities, experiences and preferences of the sex worker community. Input from the CAB helped tailor the study to address the most pressing issues identified by the community. The study team will work with the CAB to adapt the study within complex systems of organisational and cultural context and knowledge. Collaborative decision-making will occur prior to the study launch, throughout the recruitment period, and during dissemination. The CAB will provide feedback on the potential burden of the intervention and the time required for participation, so that the study minimises inconvenience and respected participants’ time constraints. All partners will decide what it means to have a ‘collaborative, equitable partnership’ and how to make that happen. 50 The CAB will advise on which community hub to recruit from first, and how to work with community leaders to adapt study standard operating procedures to not disrupt service implementation at study sites. They will also advise on how to minimise potential risks to participants, including ways to reduce emotional distress and ensure physical safety. Participants experiencing emotional distress will be referred for psychosocial support with evidence-based mental health therapy specialised for those with depression and substance abuse, with the KPIF providing transportation and a peer navigator accompanying them to the facility providing these services. The CAB will be actively involved in planning the dissemination of study results to participants and the wider community, helping decide what information to share, the timing of the dissemination and the most appropriate formats for communicating the findings.

The WiSSPr study is significant as it addresses the limitations of HIV interventions that focus solely on HIV-related stigma, without considering co-occurring stigmas linked to other identities or conditions. This study will inform the design of PrEP service delivery programmes for WESW in Zambia and the region. Understanding stigmas and related psychosocial factors is crucial for developing effective, evidence-based stigma-reduction interventions for WESW in Africa. Our long-term goal is to optimise person-centred HIV prevention by implementing inclusive, affirming practices for individuals facing multiple barriers.

Strengths of this study include (1) a mixed-methods approach which grounds quantitative research in the lived experiences of people and measures aspects of stigma that emerge at the intersections of identities, (2) qualitative data from peer navigators capturing perspectives of women at the unique interface of being recipients of care as sex workers as well as direct supporters of health service delivery, and (3) incorporation of core principles of CBPR which value KP-CSOs as equal contributors to the knowledge production process.

Several methodological limitations are also inherent in the study. First, we are unable to longitudinally assess the alignment of PrEP adherence and persistence with HIV risk. We will be limited to measuring PrEP adherence by self-report and pharmacy dispensations instead of by biomarkers of tenofovir use. Secondly, recruitment might fall short at some sites, necessitating expansion to additional community outreach venues leveraging our network of KPs. Lastly, cohort studies may have bias, due to recall and social desirability bias of self-reported measures, and missing data.

Ethics statements

Patient consent for publication.

Not applicable.

Acknowledgments

The authors would like to acknowledge the infrastructure support provided by the Centre for Infectious Disease Research in Zambia (CIDRZ) and the Key Populations Investment Fund (KPIF) programme. The authors would also like to thank peer navigators and leaders in the sex work community for their assistance in developing the study approach and recruiting study participants.

• ↵ Global hiv & aids statistics — fact sheet . 2023 . Available : https://www.unaids.org/en/resources/fact-sheet
• Gerbase A , et al
• Hontelez JAC ,
• Veraart A , et al
• Ghayda RA ,
• Yang JW , et al
• Stangl AL ,
• Leddy AM , et al
• Population Council
• Esterhuizen T ,
• Meerkotter A
• Zambia Key Population Investment Fund
• Stankevitz K ,
• Lloyd J , et al
• Sandelowski M
• Johnston M ,
• Francis J , et al
• van Stralen MM ,
• Schaefer R ,
• Gregson S ,
• Fearon E , et al
• Meissner H ,
• Creswell J ,
• Klassen AC , et al
• Heilmann E ,
• Itoh M , et al
• Sikazwe I ,
• Musheke M ,
• Chiyenu K , et al
• Namwase AS , et al
• Barradas DT , et al
• Chipungu J ,
• Smith HJ , et al
• Mwango LK ,
• Stafford KA ,
• Blanco NC , et al
• Williams A ,
• Savory T , et al
• Luhanga D , et al
• Williams DR ,
• Neighbors HW ,
• Kroenke K ,
• Spitzer RL ,
• Williams JBW
• Stockton MA ,
• Stewart C , et al
• Kraemer J ,
• Oga E , et al
• Demidenko E
• Filippone P ,
• Ssewamala FM , et al
• Palinkas LA ,
• Horwitz SM ,
• Green CA , et al
• Mantsios A ,
• Muraleetharan O ,
• Donastorg Y , et al
• Costantini L ,
• Pasquarella C ,
• Odone A , et al
• Mhlophe H ,
• Comins C , et al
• Republic of Zambia Ministry of Health
• Dhiman VK , et al
• Wallerstein NB ,
• Wallerstein N ,
• Lucero JE ,
• Boursaw B ,
• Eder MM , et al
• Duran B , et al
• Minkler M ,
• Wallerstein N

MEH and MM are joint senior authors.

X @idlidosa2, @kenmugwanya, @webarrington

Contributors RK, DR, AS, MM, MH, KKM and WB conceived and designed the study. RK, DR, AS, MM, MH, JP, MZ, MP, RZ, GMK, LC, PMK, CM and BN created the interview guides and survey. All authors revised drafts and gave final approval for publication. MM is the guarantor of the study and accepts full responsibility for the finished work and the conduct of the study, had access to the data and controlled the decision to publish.

Funding The study is being supported by the NIH Fogarty Global Health Fellowship awarded by the NIH Fogarty International Center Grant #D43TW009340.

Competing interests None declared.

Patient and public involvement Patients and/or the public were involved in the design, or conduct, or reporting, or dissemination plans of this research. Refer to the Methods section for further details.

Provenance and peer review Not commissioned; externally peer reviewed.

Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.

Read the full text or download the PDF:

• - Google Chrome

Intended for healthcare professionals

• My email alerts
• BMA member login
• Username * Password * Forgot your log in details? Need to activate BMA Member Log In Log in via OpenAthens Log in via your institution

Search form

• Advanced search
• Search responses
• Search blogs
• Risk of dementia after...

Risk of dementia after initiation of sodium-glucose cotransporter-2 inhibitors versus dipeptidyl peptidase-4 inhibitors in adults aged 40-69 years with type 2 diabetes: population based cohort study

Linked editorial.

SGLT-2 inhibitors and dementia

• Related content
• Peer review

This article has a correction. Please see:

• Risk of dementia after initiation of sodium-glucose cotransporter-2 inhibitors versus dipeptidyl peptidase-4 inhibitors in adults aged 40-69 years with type 2 diabetes: population based cohort study - September 02, 2024
• Anna Shin , senior researcher 1 ,
• Bo Kyung Koo , professor 2 ,
• Jun Young Lee , professor 3 ,
• 1 Medical Research Collaborating Centre, Department of Internal Medicine, Seoul National University Bundang Hospital, Seongnam, Korea
• 2 Division of Endocrinology, Department of Internal Medicine, Seoul National University College of Medicine, SMG-SNU Boramae Medical Centre, Seoul, Korea
• 3 Department of Psychiatrics, Seoul National University College of Medicine, SMG-SNU Boramae Medical Centre, Seoul, Korea
• 4 Department of Internal Medicine, Seoul National University College of Medicine, Seoul National University Bundang Hospital, Seongnam, Korea
• Correspondence to: E H Kang kangeh{at}snubh.org
• Accepted 4 July 2024

Objective To compare the risk of dementia associated with sodium-glucose cotransporter-2 (SGLT-2) inhibitors versus dipeptidyl peptidase-4 (DPP-4) inhibitors in adults aged 40-69 years with type 2 diabetes.

Design Population based cohort study.

Setting Korean National Health Insurance Service data, 2013-21.

Participants 110 885 propensity score matched pairs of adults with type 2 diabetes aged 40-69 years who were initiators of either an SGLT-2 inhibitor or a DPP-4 inhibitor.

Main outcome measures The primary outcome was new onset dementia. Secondary outcomes were dementia requiring drug treatment and individual types of dementia, including Alzheimer’s disease and vascular dementia. Control outcomes were genital infections (positive), and osteoarthritis related clinical encounters and cataract surgery (negative). Hazard ratios and 95% confidence intervals (CIs) were estimated using Cox models. Follow-up time stratified analyses (>2 years and ≤2 years) and subgroup analyses by age, sex, concomitant use of metformin, and baseline cardiovascular risk were performed.

Results 110 885 propensity score matched pairs of initiators of an SGLT-2 inhibitor or a DPP-4 inhibitor were followed-up for a mean 670 (standard deviation 650) days, generating 1172 people with newly diagnosed dementia: incidence rate 0.22 per 100 person years in initiators of SGLT-2 inhibitors and 0.35 per 100 person years in initiators of DPP-4 inhibitors, with hazard ratios of 0.65 (95% CI 0.58 to 0.73) for dementia, 0.54 (0.46 to 0.63) for dementia requiring drugs, 0.61 (0.53 to 0.69) for Alzheimer’s disease, and 0.48 (0.33 to 0.70) for vascular dementia. The hazard ratios for the control outcomes were 2.67 (2.57 to 2.77) for genital infections, 0.97 (0.95 to 0.98) for osteoarthritis related encounters, and 0.92 (0.89 to 0.96) for cataract surgery. When calibrated for residual confounding measured by cataract surgery, the hazard ratio for dementia was 0.70 (0.62 to 0.80). The association was greater for more than two years of treatment (hazard ratio of dementia 0.57, 95% CI 0.46 to 0.70) than for two years or less (0.52, 0.41 to 0.66) and persisted across subgroups.

Conclusion SGLT-2 inhibitors might prevent dementia, providing greater benefits with longer treatment. As this study was observational and therefore prone to residual confounding and informative censoring, the effect size could have been overestimated. Randomised controlled trials are needed to confirm these findings.

Introduction

Dementia concerns damage to the brain parenchyma, resulting in a permanent degradation of higher cortical functions, mood, and even behaviour. 1 According to a World Health Organization (WHO) report in 2021, the number of people with dementia globally is expected to reach 78 million by 2030. 2 Despite the severe consequences of dementia, the success rate of the development for dementia drugs has been markedly low in the past two decades, leaving only extremely limited options for disease modifying treatment. 3 Evidence has, however, emerged to support the importance of modifiable risk factors for dementia, including diabetes. 4 According to a pooled analysis, type 2 diabetes is associated with a 60% greater risk of dementia, 5 predisposing such people to both Alzheimer’s disease and vascular dementia. 6 The mechanisms linking type 2 diabetes and dementia are multifactorial, involving insulin resistance, hypoglycaemic episodes, and vascular compromise. 7 In line with this, meta-analyses on observational studies have shown that certain antiglycaemic drugs may have neuroprotective effects in people with diabetes. 8 9 10

Sodium-glucose cotransporter-2 (SGLT-2) inhibitors are a newer class of antiglycaemic drugs that inhibit reabsorption of glucose in the proximal tubule. Key randomised controlled trials have shown significant cardiorenal protection from use of SGLT-2 inhibitors beyond glucose lowering effects. 11 SGLT-2 inhibitors are now considered one of the drug repurposing candidates for disease modifying treatment of dementia. 12 Recent evidence suggests neuroprotective effects of SGLT-2 inhibitors based on penetration of the drug through the blood-brain barrier, SGLT-2 expression in brain tissue, and direct inhibition of acetylcholinesterase, as well as indirect cardiometabolic benefits. 13

Previous observational studies have suggested better preservation of cognitive function among people with type 2 diabetes treated with SGLT-2 inhibitors than other treatments, including dipeptidyl peptidase-4 (DPP-4) inhibitors, 14 15 16 17 another newer class of antiglycaemic drugs found to have no effect on cognitive performance in recent randomised controlled trials compared with sulfonylurea and placebo. 18 19 The methodological approaches of these observational studies were often limited, however, and did not meet the active comparator new user design, leaving concerns about confounding or bias. 15 16 A recent well designed study on residents in Ontario, Canada compared new users of SGLT-2 inhibitors with new users of DPP-4 inhibitors and found that the former were associated with a 20-34% reduced risk of dementia among people older than 66 years. 14 The effects on younger populations and specific types of dementia (eg, Alzheimer’s disease, vascular dementia) were not, however, examined. Moreover, it is unclear whether different patient characteristics such as concomitant treatment or comorbidity status would modify such drug effects. We therefore compared the risk of dementia among adults with diabetes younger than 70 years who initiated an SGLT-2 inhibitor or DPP-4 inhibitor using the nationally representative Korea National Health Insurance Service database.

Data source

We conducted a cohort study using data from the Korea National Health Insurance Service database during 2013-21. This database covers the entire population of Korea and provides longitudinal patient data, including personal characteristics, ICD-10 (international classification of diseases, 10th revision) diagnosis codes, procedures, prescription and dispensing records (drug names, prescription and dispensing dates, days’ supply, dose, and route of administration), and type of healthcare utilisation (outpatient, inpatient, or emergency department). 20

Study design and population

We emulated a target trial for the outcomes of interest (see supplemental table S1 for the framework of the target trial emulation) using a propensity score matched active comparator new user cohort study design (see supplemental figure S1 for the detailed study design).

Adults aged 40-69 years with an ICD-10 code for type 2 diabetes who had initiated an SGLT-2 inhibitor or DPP-4 inhibitor were eligible for inclusion in the study (see supplemental figure S2 for the participant selection process and supplemental table S2 for ICD-10 codes used in this selection process). To implement a new user active comparator design, we only included initiators of the two competitive study drugs, an SGLT-2 inhibitor and a DPP-4 inhibitor, who had not been dispensed either drug for at least 365 days (the baseline period) before the first dispensing date of the study drug (the index date). To be included, individuals were required to be free of any dementia and related drugs ever before the index date. We also excluded those with ICD-10 diagnosis codes for type 1 diabetes mellitus, HIV, or end stage renal disease (or dialysis service) during the baseline period, and those who concomitantly used glucagon-like peptide-1 receptor agonists or thiazolidinedione on the index date.

Outcome measurement

Our primary outcome was incident dementia based on ICD-10 diagnosis codes in a primary position recorded on inpatient or outpatient claims (see supplemental table S3 for ICD-10 codes used to define outcomes). 21 To improve specificity of outcome ascertainment, we examined dementia defined by the diagnosis codes along with dispensing of dementia drugs (donepezil, rivastigmine, galantamine, or memantine) as a secondary outcome. In Korea, dementia drugs are reimbursed by the Rare and Intractable Diseases programme, where beneficiaries should qualify for a diagnosis certificate of dementia based on brain imaging and cognitive function testing. Other secondary outcomes were individual types of dementia (eg, Alzheimer’s disease, vascular dementia) in a primary position.

Control outcomes

To assess reproducibility of established relations and unmeasured systematic bias, we also compared the risk of positive and negative control outcomes between the two treatment groups (see supplemental table S3). Given the higher risk of genital infections associated with SGLT-2 inhibitors compared with DPP-4 inhibitors in randomised controlled trials, we examined genital infections as a positive control outcome. 22 We also examined osteoarthritis related encounters and cataract surgery as negative control outcomes. A null association with treatment is expected for appropriate negative control outcomes, which share unmeasured confounders with the outcome and are unaffected by treatment. 23 As with dementia, osteoarthritis and cataract are degenerative diseases of older people. Therefore, osteoarthritis related encounters and cataract surgery would share with dementia unmeasured confounders such as frailty, lifestyle, and healthcare system usage patterns associated with ageing, and cataract surgery would also share smoking and alcohol consumption. 24 25 Osteoarthritis related encounters would be expected for symptomatic or advanced osteoarthritis. Thus we considered such encounters to be minimally affected by the study drugs despite mild weight reduction effect of SGLT-2 inhibitors. 11 Also, two meta-analyses reported a null association between the development of cataract and treatment with SGLT-2 inhibitors. 26 27 Using a deviation from the null association between a negative control outcome and treatment, we estimated corrected hazard ratios and corresponding 95% confidence intervals (CIs) adjusting for residual confounding. 23 28

We identified covariates related to diabetes severity and risk of dementia for the 365 day pre-index baseline period (see supplemental table S2 for ICD-10 codes used to ascertain covariates). The covariates included personal characteristics, sociodemographic factors, complications from diabetes (retinopathy, nephropathy, neuropathy, and diabetic foot), classes and number of antiglycaemic drugs, risk factors for dementia (ie, cardiometabolic risk factors, hearing loss, head trauma, fracture history, mood or mental disorders, and anticholinergic drugs), other comorbidities and related drugs, Charlson-Deyo comorbidity index, 29 and healthcare service use patterns such as hospital admissions, emergency department visits, and outpatient clinic visits.

Statistical analysis

We used propensity score matching to account for confounding. The propensity score was estimated for each comparison using a multivariable logistic regression model that included >110 baseline covariates (see supplemental table S4 for the full list). Nearest neighbour matching for SGLT-2 inhibitor versus DPP-4 inhibitor was done in a ratio of 1:1, with a caliper of 0.025 on the propensity score scale. Balance between covariates after propensity score matching was considered to have been achieved when the absolute standardised difference was <0.1 between the two treatment groups. 30 Propensity score matched incidence rates of primary and secondary outcomes were calculated per 100 person years.

We primarily used Cox proportional hazard models to estimate the hazard ratios and corresponding 95% CIs. Owing to the discrete difference in mortality between the two treatments, 11 we also presented hazard ratios (95% CIs) from Fine-Gray models, adjusting for competing risk of death. 31 The proportional hazard assumption was tested by adding the interaction term between treatment and follow-up time in the model. When the interaction was statistically significant, we performed a follow-up time stratified analysis to examine the time varying treatment effect. We sorted propensity score matched study participants into two groups according to their follow-up times (≤2 years or >2 years), then estimated a matched set stratified hazard ratio (95% CI) within the two groups.

In our primary as treated analysis, patients were followed from the day after the index date up to the first occurrence of the censoring events (outcome event, disenrollment, death, end of database (31 December 2021), or treatment change through discontinuation, switching, or adding). Drug discontinuation was defined as no dispensing within 90 days from the expected refill date. The expected refill date was calculated by adding days’ supply to the last dispensing date of the study drug. Participants who discontinued the study drug were followed up until the last expected refill date plus a 30 day grace period. Although switching between different SGLT-2 inhibitors or between different DPP-4 inhibitors was not a censoring event, adding or switching to other classes of antiglycaemic treatments resulted in immediate censoring. We performed an intention-to-treat analysis as our secondary analysis, where participants were followed up until censoring events except for treatment change to deal with concerns of informative censoring.

Sensitivity analyses —Firstly, to avoid reverse causation from delayed diagnosis of dementia, we started follow-up after 365 days from the index date in both as treated and intention-to-treat analyses (up to three years and the whole follow-up). Secondly, we applied a grace period of 180 or 365 days for the censoring by treatment change to capture delayed diagnoses made after the change of treatment. Thirdly, to eliminate the effect of hypoglycaemic episodes during treatment, analyses were done excluding those who concurrently used drugs with hypoglycaemia potential (insulin, sulfonylurea, or glinides) on the index date. Fourthly, we adjusted for the duration of diabetes mellitus for those who had an ascertainable type 2 diabetes diagnosis date, defined as the first date of an ICD-10 code for type 2 diabetes diagnosis in the primary position free of such codes for at least 365 days before the diagnosis date. Lastly, we utilised the entirety of new users of SGLT-2 inhibitors and DPP-4 inhibitors using propensity score based fine stratification and weighting to achieve greater generalisability. 32

Subgroup analyses —Prespecified propensity score matched subgroup analyses were done based on participants’ age (≥60 years and <60 years), sex, concurrent metformin use, and baseline cardiovascular risk. The estimation of propensity score and matching were done separately for individual subgroups. The subgroup with high cardiovascular risk was defined as men aged ≥50 years and women aged ≥55 years who had at least one diagnosis of angina, myocardial infarction, stroke, or peripheral vascular disease during the one year pre-index period. 20 We tested interaction terms between the treatment and individual stratifying factors.

Patient and public involvement

This study analysed secondary data without patient involvement. Patients were not invited to be involved in the study design, development of outcomes, interpretation of the results, or drafting of the manuscript. The primary barrier against patient and public involvement was use of an administrative database, which requires a specific study design and pharmacoepidemiological method to ensure internal validity, leaving minimal potential for the patient and public to be engaged.

Baseline patient characteristics

Supplemental figure S2 shows the selection process of the study cohort. We identified 112 663 new users of SGLT-2 inhibitors and 847 999 new users of DPP-4 inhibitors who were free of known dementia and did not use either of the study drugs at baseline. Before propensity score matching, most baseline covariates, including diabetes complications and number of antiglycaemic drugs, were overall relatively well balanced, reflecting the effectiveness of the active comparator new user design ( table 1 , also see supplemental table S4 for the distribution of the full list of covariates between the two groups). Some covariates showed imbalance, with standardised differences >0.1, particularly cardiovascular comorbidities, which were more prevalent among initiators of SGLT-2 inhibitors than among initiators of DPP-4 inhibitors (16.8% v 10.6% for angina pectoris, 3.1% v 1.6% for myocardial infarction, 7.8% v 4.2% for heart failure, 66.6% v 59.8% for hypertension, 78.8% v 70.9% for hyperlipidaemia). After propensity score matching in a 1:1 ratio, 110 885 pairs of initiators of SGLT-2 inhibitors and DPP-4 inhibitors were included in the analysis (mean age 61.9 years, 55.7% men) ( table 1 , also see supplemental table S4). All propensity score matched baseline covariates, including psychiatric disorders, cardiovascular diseases, other comorbidities, use of drugs with anticholinergic activity, and use of other drugs, were well balanced (standardised differences <0.1). The study participants’ mean comorbidity score was 2.4 (standard deviation (SD) 1.8). Cardiometabolic factors were highly common, with 66.5% of participants having hypertension and 78.6% having hyperlipidaemia. Established cardiovascular diseases were observed in 16.7% of participants with angina, 6.4% with stroke, and 3.1% with myocardial infarction. The most common oral antiglycaemic agents used during the baseline period were biguanide (52.2%), followed by sulfonylurea (27.8%) and thiazolidinedione (8.2%). The most common index SGLT-2 inhibitor was dapagliflozin (58.6%), followed by empagliflozin (35.4%), and the most common index DPP-4 inhibitors were gemigliptin (22.7%), linagliptin (22.4%), and sitagliptin (20.4%) (see supplemental table S5).

Select baseline characteristics of propensity score matched cohort. Values are number (percentage) unless stated otherwise

• View inline

Comparative risk of dementia between initiators of SGLT-2 inhibitors and DPP-4 inhibitors

The mean follow-up time of patients was 670 (SD 650) days, with 612 (SD 613) days for initiators of SGLT-2 inhibitors and 728 (SD 679) days for initiators of DPP-4 inhibitors (see supplemental table S6 for distribution of censoring events). A total of 1172 participants with newly diagnosed dementia were identified, with incidence rates per 100 person years of 0.22 for initiators of SGLT-2 inhibitors and 0.35 for initiators of DPP-4 inhibitors. The corresponding hazard ratio was 0.65 (95% CI 0.58 to 0.73; table 2 ). The lowered risk of dementia associated with use of SGLT-2 inhibitors compared with DPP-4 inhibitors was similarly observed for secondary outcomes: hazard ratio 0.54 (0.46 to 0.63) for dementia requiring drugs, 0.61 (0.53 to 0.69) for Alzheimer’s disease, and 0.48 (0.33 to 0.70) for vascular dementia. The results were consistent with those of intention-to-treat analyses: 0.65 (0.60 to 0.71) for dementia, 0.60 (0.54 to 0.67) for dementia requiring drugs, 0.63 (0.57 to 0.69) for Alzheimer’s disease, and 0.62 (0.49 to 0.79) for vascular dementia. Estimates for the Fine-Gray models were also similar. We found a 2.67-fold risk (95% CI 2.57-fold to 2.77-fold) of genital infections associated with SGLT-2 inhibitors versus DPP-4 inhibitors. The hazard ratios for association between treatment and negative control outcomes were 0.97 (95% CI 0.95 to 0.98) for osteoarthritis related encounters and 0.92 (0.89 to 0.96) for cataract surgery. When corrected using the association between treatment and cataract surgery, the hazard ratios for dementia increased by about 7.7% (see supplemental table S7), to 0.70 (0.62 to 0.80).

Comparative risk of dementia between initiators of SGLT-2 inhibitors and DPP-4 inhibitors in main propensity score matched cohort

Follow-up time stratified analysis

A significant interaction (P<0.05) was observed between treatment and follow-up time for all outcomes except vascular dementia in the as treated analysis. The Kaplan-Meier curve diverged more in the later follow-up period for these outcomes ( fig 1 ), indicating that the effect would be greater with longer treatment. According to the follow-up time stratified analyses (46 767 propensity score matched pairs treated for two or less years, 16 827 pairs treated for more than two years; see supplemental table S8 for the distribution of baseline covariates for individual stratified groups), the magnitude of association modestly increased with more than two years of treatment compared with two years or less for these outcomes (see supplemental table S9): hazard ratio for more than two years versus two years or less of treatment was 0.52 (95% CI 0.41 to 0.66) v 0.57 (0.46 to 0.70) for dementia, 0.41 (0.29 to 0.57) v 0.45 (0.33 to 0.61) for dementia requiring drugs, and 0.48 (0.37 to 0.63) v 0.53 (0.41 to 0.68) for Alzheimer’s disease.

Kaplan-Meier curves for dementia-free survival comparing propensity score matched initiators of SGLT-2 inhibitors with initiators of DPP-4 inhibitors. CI=confidence interval; DPP-4=dipeptidyl peptidase-4; SGLT-2=sodium-glucose cotransporter-2

• Download figure
• Open in new tab
• Download powerpoint

Sensitivity analyses

The results were highly consistent even after accounting for the 365 day lag time from the index date ( table 3 ), with hazard ratios in as treated analyses of 0.57 (0.48 to 0.68) for dementia, 0.48 (0.38 to 0.61) for dementia requiring drugs, 0.55 (0.45 to 0.67) for Alzheimer’s disease, and 0.46 (0.26 to 0.80) for vascular dementia. In the intention-to-treat analyses with lag time applied, the hazard ratios were 0.80 (0.75 to 0.86) for dementia, 0.84 (0.77 to 0.91) for dementia requiring drugs, 0.80 (0.74 to 0.86) for Alzheimer’s disease, and 0.80 (0.66 to 0.98) for vascular dementia.

Lag time analyses on comparative risk of dementia in main propensity score matched cohort, with follow-up starting after 365 days from index date

For as treated analyses with longer grace periods after treatment change, a slightly increased incidence rate of dementia was noted in both treatment groups but to a greater degree among initiators of SGLT-2 inhibitors, with a hazard ratio of 0.72 (0.65 to 0.80) for dementia for a grace period of 180 days and 0.76 (0.69 to 0.83) for a grace period of 365 days (see supplemental table S10). Decreased incidence rates of genital infections were also noted among initiators of SGLT-2 inhibitors.

The results were consistent regardless of concurrent use of a drug with hypoglycaemic potential (see supplemental tables S11 and S12), with a hazard ratio of 0.69 (0.60 to 0.80) for dementia. The duration of type 2 diabetes was identified for 45 088 propensity score matched pairs (1008 v 925 days for initiators of SGLT-2 inhibitors and DPP-4 inhibitors, respectively, with a standardised difference of 0.10). Consistent results were observed after adjusting for duration of type 2 diabetes (see supplemental tables S13 and S14), with a hazard ratio of 0.60 (0.50 to 0.72) for dementia. We also observed similar results in propensity score based fine stratification weighted analyses (see supplemental tables S15 and S16), with a hazard ratio of 0.68 (0.62 to 0.75) for dementia.

Subgroup analysis

Supplemental table S17 presents the baseline characteristics of the subgroups. The lower risk associated with SGLT-2 inhibitors was overall consistent across subgroups stratified by age, sex, concurrent metformin use, and baseline cardiovascular risk ( fig 2 , also see supplemental table S18). However, statistical significance was not achieved for the subgroups with relatively small outcome numbers (eg, those aged <60 years). We did not find any interaction between the treatment and individual stratifying factors.

Comparative risk of dementia between initiators of sodium-glucose cotransporter-2 inhibitors and initiators of dipeptidyl peptidase-4 inhibitors in individual propensity score matched subgroups (as treated analysis). CI=confidence interval

This large population based cohort study among adults aged 40-69 years with type 2 diabetes found a 35% reduced risk of dementia associated with use of SGLT-2 inhibitors compared with DPP-4 inhibitors. This finding persisted regardless of dementia type and across subgroups of populations with diverse characteristics. Highly consistent results over a range of secondary and sensitivity analyses supported the robustness of our study findings. Our findings also suggest that the treatment effect of SGLT-2 inhibitors escalated with time.

Relevance of study design to internal validity

An active comparator new user design is a powerful pharmacoepidemiological approach that effectively copes with both measured and unmeasured confounding in observational studies. 33 One of the key advantages of this approach would be that similar disease (type 2 diabetes in our example) severity and related comorbidity profile can be expected between the two treatment groups because the participants in both groups are at the beginning of a similar stage of a given treatment. International guidelines had equally recommended SGLT-2 inhibitors and DPP-4 inhibitors as second line treatment until December 2018 34 when the revised guideline preferentially recommended use of SGLT-2 inhibitors in the presence of atherosclerotic cardiovascular disease, heart failure, or chronic kidney disease. 35 This approach also ensures that selection bias associated with depletion of susceptible people (to inefficacy or intolerance, or both) is avoided, allowing all individuals initiating the study drug to contribute to the follow-up from the start of the treatment. In this context, our study design offered greater internal validity than in previous studies. 15 16

Interpretation of results and comparison with other studies

We observed a known association between a positive control outcome and treatment. 22 The association for osteoarthritis related encounters was close to null (hazard ratio 0.97, 95% CI 0.95 to 0.98), which achieved statistical significance owing to excess power from a highly frequent outcome. A slight deviation (0.92, 0.89 to 0.96) from the null association was observed for cataract surgery. A bias measure (7.7% increased hazard ratio) based on this deviation indicated that the association between treatment and dementia was largely unexplained solely by residual confounding.

In preclinical studies, SGLT-2 inhibitors have shown direct neuroprotective effects through multiple pathways. 13 36 37 38 These drugs exhibited anticholinergic activity, 13 prevented ultrastructural changes of neurovascular units associated with cognitive decline in mice with diabetes, 36 and ameliorated amyloid β deposition and tau phosphorylation in the brain tissue of mice with Alzheimer’s disease and type 2 diabetes. 37 Diurnal catabolism induced by SGLT-2 inhibitors restored autophagy by downregulating the mTOR (mechanistic target of rapamycin) pathway, which is chronically activated in Alzheimer’s disease. 38 Based on these preclinical findings, SGLT-2 inhibitors may delay the progression of dementia in people with type 2 diabetes both for Alzheimer’s disease and for vascular dementia, independent of the cardiorenal benefits exerted by SGLT-2 inhibitors.

A considerable effect estimate found within a relatively short period (≤2 years) of follow-up needs attention. Dementia develops through a continuum of accumulated molecular and structural changes. 7 Heterogeneous states of disease progression yet to reach definitive dementia are likely to exist among people with type 2 diabetes at baseline or even after applying a one year lag time. This is likely true since mild cognitive impairment, a transitional state between normal ageing and dementia, 7 is prevalent among 12-18% and 23% of people aged ≥60 years in the US and Korea, respectively, with 10-15% of the annual conversion to dementia. 39 40 Notably, mild cognitive impairment is 1.4~2.0 times more prevalent among people with type 2 diabetes with accelerated progression. 41 42 43 44 Because the time span between mild cognitive impairment and dementia has already been shortened, and progression is particularly rapid among people with type 2 diabetes, early risk reduction against dementia could be seen in the presence of effective treatment (see supplementary figure S3 for a schematic explanation). This scenario also complies with the finding that the cognitive benefits of SGLT-2 inhibitor use versus non-use were better noted for those with mild cognitive impairment than with normal cognitive function at baseline. 17 Moreover, the visible action of SGLT-2 inhibitors versus DPP-4 inhibitors was rapid, based on the time elapsed until the first statistically significant result as early as day 5 for the benefits on death and worsening heart failure. 45

A recent prospective cohort study found that use of SGLT-2 inhibitors for more than three years improved cognitive function scores compared with non-use. 46 Although this finding suggests that longer treatment might generate more benefits, the study was subject to confounding by indication and immortal time bias owing to the comparison between users (eg, prevalent users) and non-users of SGLT-2 inhibitors. 33 Our study comparing new users of two competing drugs, SGLT-2 inhibitors and DPP-4 inhibitors, further supports favourable results for early initiation of the drug and prolonged treatment.

We observed attenuated results with lag time applied in intention-to-treat analyses and with longer grace periods. Since incidence rates of genital infections continually decreased among users of SGLT-2 inhibitors in these analyses, loss of treatment effect associated with misclassification of drug use played a role in driving the results towards null. Initiators of SGLT-2 inhibitors, however, were more frequently censored by treatment change than initiators of DPP-4 inhibitors. Because patients with risk factors for treatment change (non-adherence, inefficacy, or adverse events) can be more prone to develop dementia than patients without these risk factors, informative censoring may have overestimated the results in our as treated analysis. Nevertheless, the overall results between as treated and intention-to-treat analyses were similar ( table 2 ), suggesting non-substantial informative censoring.

In subgroup analyses, we observed highly consistent results, but did not find an interaction between treatment and individual characteristics of the study population. Unlike the expectation that SGLT-2 inhibitors might be associated with greater benefits against the risk of vascular dementia than Alzheimer’s disease, the magnitude of association was accompanied by widely overlapping 95% CIs between the two types of dementia for all analyses. Thus, it is not surprising to observe no interaction between treatment and baseline cardiovascular risk. A recent meta-analysis also reported that the pooled beneficial association between dementia and use of SGLT-2 inhibitors versus other antiglycaemic treatments was not affected by cardiovascular diseases. 10 These findings suggest that the underlying mechanisms are not limited to cardiorenal pathways, possibly involving direct neuroprotective pathways observed in preclinical studies. 13 36 37 38 According to previous studies on metformin monotherapy versus no treatment, metformin was not associated with incident dementia. 47 48 Based on these findings, concurrent use of metformin is unlikely to interact with SGLT-2 inhibitors in modifying the risk of dementia.

Strengths and limitations of this study

Several important strengths of this study deserve comment. Firstly, we used rigorous pharmacoepidemiological approaches, in particular we adopted an active comparator new user design and extensive propensity score matching. 33 The diagnosis codes in the primary position and applying disease specific drugs would increase the specificity of the outcome. The sensitivity analyses and control outcomes add relevant internal validity to this study. Secondly, compared with a previous study, 14 we included relatively younger people (aged 40-69 years) with type 2 diabetes, broadening the target population of benefits associated with use of SGLT-2 inhibitors. Thirdly, we used a nationally representative database, providing high generalisability. Fourthly, we performed comprehensive analyses for time varying comparisons of SGLT-2 inhibitors versus DPP-4 inhibitors, diverse subgroups, and individual types of dementia, presenting highly consistent results.

This study also has limitations. Firstly, owing to the observational nature of our study, it is inherently subject to residual or unmeasured confounding. Although we balanced many proxies of type 2 diabetes severity and comorbidities and used negative control outcomes, direct test results on serum glucose levels, renal function, severity of other comorbidities, health behaviours (eg, smoking and alcohol consumption), and duration of type 2 diabetes were not fully ascertainable from the claims data. Secondly, diagnoses of dementia are commonly delayed, rendering studies on dementia risk particularly susceptible to informative censoring, reverse causation, and outcome misclassification, which may have resulted in overestimation of our results. Thirdly, our study did not provide exact mechanisms of neuroprotection.

Conclusions

This large population based cohort study found that initiation of SGLT-2 inhibitors was associated with a 35% lower risk of dementia compared with initiation of DPP-4 inhibitors in people with type 2 diabetes aged 40-69 years. This association was similarly observed for Alzheimer’s disease and vascular dementia and was also consistent across subgroups. We observed a greater association with treatment duration longer than two years. These findings underscore the need for future randomised controlled trials.

What is already known on this topic

Despite increasing numbers of people with dementia globally, current options for disease modifying treatments are limited

Type 2 diabetes substantially predisposes people to Alzheimer’s disease and vascular dementia through multiple pathways

A previous study suggested a decreased risk of dementia associated with sodium-glucose cotransporter-2 (SGLT-2) inhibitors versus dipeptidyl peptidase-4 (DPP-4) inhibitors among people with type 2 diabetes aged >66 years

What this study adds

This large population based cohort study among people with type 2 diabetes aged 40-69 years found a 35% lower risk of dementia associated with use of SGLT-2 inhibitors compared with DPP-4 inhibitors

This finding persisted regardless of dementia type and across subgroups of diverse population characteristics such as age, sex, concomitant use of metformin, and baseline cardiovascular risk

The treatment effect of SGLT-2 inhibitors compared with DPP-4 inhibitors increased with time

Ethics statements

Ethical approval.

The Institutional Review Board of the Seoul National University Bundang Hospital exempted the study protocol (X-2206-762-901) and waived written patient consent based on the fully deidentified database.

Data availability statement

Patient level data are not publicly allowed according to data use agreement. Aggregate level data can be requested from the corresponding author.

Acknowledgments

We thank Joongyub Lee (Department of Preventive Medicine, Seoul National University College of Medicine, Seoul, Korea) for comments on the pharmacoepidemiological design of this study.

Contributors: EHK conceived and designed the study and drafted the manuscript. AS did the analyses. EHK and AS had full access to all the data in the study and take responsibility for the integrity of the data and accuracy of the data analysis. All authors interpreted the data, critically revised the manuscript for important intellectual content, and approved the final version. EHK is the guarantor. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.

Funding: This study was supported by the Korea Health Industry Development Institute (KHIDI)-AZ Diabetes Research programme (grant No 08-2022-0261). The funders had no role in considering the study design or in the collection, analysis, interpretation of data, writing of the report, or decision to submit the article for publication.

Competing interests: All authors have completed the ICMJE uniform disclosure form at www.icmje.org/disclosure-of-interest/ and declare: support from the Korea Health Industry Development Institute (KHIDI)-AZ Diabetes Research programme; no financial relationships with any organisations that might have an interest in the submitted work in the previous three years; no other relationships or activities that could appear to have influenced the submitted work. EHK receives grants from Celltrion and SK Chemicals to Seoul National University Bundang Hospital for unrelated studies. BKK receives grants from KHIDI to SMG-SNU Boramae Medical Centre for unrelated studies.

Transparency: The guarantor (EHK) affirms that this manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and, if relevant, registered) have been explained.

Dissemination to participants and related patient and public communities: The use of deidentified data preclude direct dissemination to participants, and we have no plans to involve patients in the dissemination of study results. Study findings will be disseminated by all coauthors through their own institutions. The article will be distributed within the corresponding author’s institution.

Provenance and peer review: Not commissioned; externally peer reviewed.

This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/ .

• American Psychiatric Association
• ↵ World Health Organization. Fact sheets of dementia. Geneva, Switzerland: WHO; 2021. www.who.int/news-room/fact-sheets/detail/dementia .
• Gauthier S ,
• Livingston G ,
• Huntley J ,
• Sommerlad A ,
• Chatterjee S ,
• Peters SA ,
• Woodward M ,
• Biessels GJ ,
• Staekenborg S ,
• Brunner E ,
• Scheltens P
• Shaaban CE ,
• Kuate Defo A ,
• Pisaturo A ,
• Daskalopoulou SS
• Gunawan PY ,
• Gunawan PA ,
• Hariyanto TI
• Heerspink HJL ,
• Cuthbertson DJ ,
• Wilding JPH
• Katsenos AP ,
• Iskander C ,
• Verhagen C ,
• Janssen J ,
• CARMELINA Investigators
• Taylor DH Jr . ,
• Plassman BL
• Lipsitch M ,
• Tchetgen Tchetgen E ,
• Lindblad BE ,
• Håkansson N ,
• Philipson B ,
• Richardson DB ,
• Laurier D ,
• Schubauer-Berigan MK ,
• Sundararajan V ,
• Henderson T ,
• Muggivan A ,
• Austin PC ,
• Rothman KJ ,
• Bateman BT ,
• Hernandez-Diaz S ,
• Huybrechts KF
• Yoshida K ,
• Solomon DH ,
• Inzucchi SE ,
• Bergenstal RM ,
• Davies MJ ,
• D’Alessio DA ,
• Fradkin J ,
• Hayden MR ,
• Hierro-Bujalance C ,
• Infante-Garcia C ,
• Del Marco A ,
• Esterline RL ,
• Oscarsson J ,
• ↵ 2022 Alzheimer’s disease facts and figures . Alzheimers Dement 2022 ; 18 : 700 - 89 . pmid: 35289055 OpenUrl CrossRef PubMed
• Luchsinger JA ,
• Winkler A ,
• Cukierman T ,
• Gerstein HC ,
• Williamson JD
• Patorno E ,
• Everett BM ,
• Morley JE ,
• Scherrer JF ,

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Indian J Psychol Med
• v.44(2); 2022 Mar

Research Design: Cohort Studies

Chittaranjan andrade.

1 Dept. of Clinical Psychopharmacology and Neurotoxicology, National Institute of Mental Health and Neurosciences, Bangalore, Karnataka, India.

In a cohort study, a group of subjects (the cohort) is followed for a period of time; assessments are conducted at baseline, during follow-up, and at the end of follow-up. Cohort studies are, therefore, empirical, longitudinal studies based on data obtained from a sample; they are also observational and (usually) naturalistic. Analyses can be conducted for the cohort as a whole or for subgroups amongst which comparisons can be drawn. Because there is no randomization to the subgroups of interest, cause and effect relationships cannot be determined, and relationships between variables must be stated as associations that may or may not be influenced by confounding. The cohort that is studied can be prospectively or retrospectively defined, and each method has its advantages and disadvantages. These and other issues are explained with the help of examples. A special note is made of cohort studies in Indian psychiatry.

Previous articles in this series on research design discussed classifications in research design 1 and prospective and retrospective, cross- sectional and longitudinal studies. 2 This article examines a specific research design that is increasingly being employed in medicine and psychiatry: the cohort study.

Cohort Study: General Description

A cohort is a group of subjects. In a cohort study, the cohort is made up of subjects who meet the study selection criteria. Identification of the cohort, or recruitment, occurs across a period of time. The cohort so identified is followed for a further period of time. The study usually ends on a set date or when the desired endpoint has been reached. Assessments are conducted at recruitment, during follow-up, and at the end of follow-up.

Examples of Cohort Studies

• All children born in a hospital during a two-year period are followed until each child reaches the age of 18 years; the main objective of the study is to examine the incidence and predictors of different childhood-onset psychiatric disorders.
• All patients newly diagnosed with schizophrenia are followed for 10 years to examine how different sociodemographic (e.g., male vs. female), clinical (e.g., short vs. long duration of untreated psychosis), and treatment (e.g., first vs. second generation antipsychotics) characteristics influence the course and outcome of the disorder.
• All doctors in a geographical area are followed until death to examine how their lifestyle behaviors and professional exposures influence their risk of developing cancer, cardiovascular disease, and dementia.

The Framingham Heart Study, the Nurses Health Study, and the Nun Study are examples of well-known cohort studies. In the field of mental health, the Adolescent Brain Cognitive Development (ABCD) study is a large ongoing cohort study that will follow a cohort of >10,000 children from childhood to adult life; the purpose of the study is to understand the determinants of cognitive, social, emotional, and physical development. 3

Characteristics of Cohort Studies

Examined from the perspective of research design, cohort studies are empirical because they collect and examine data. They are sample-based because a group of individuals is studied. They are always longitudinal because there is a follow-up, but can be prospectively or retrospectively defined, as was clarified in an earlier article. 2 There is no randomization to subgroups of interest and no blinding of subjects or raters to the characteristics of interest. They can be uncontrolled, with outcomes examined in a group of persons defined by a single characteristic (see example 1 in the previous section), or they can be quasi-controlled, with subgroups being compared (see example 2 in the previous section). 4 They are usually observational and naturalistic, though interventions can be factored into the study protocol. For readers unfamiliar with these terms, “observational” means that the investigators merely record data; they do not intervene in patient care. “Naturalistic” means that patients receive whatever treatment their health care professionals consider necessary and appropriate, or, in other words, treatment as usual.

Retrospective Cohort Studies

As an example of a retrospective cohort study, a cohort can be defined to comprise all children born in a health care system between 1980 and 1990. The health care system can be a single hospital, a group of hospitals, an insurance database, or any other database that maintains records of medical information. The data of the children identified are examined and extracted from paper or electronic charts until, say, 2020, to ascertain what maternal, gestational, and early postnatal characteristics, recorded during and after the pregnancy, predict adult mental health.

If large electronic health care databases exist, as they do, e.g., in the Scandinavian countries and the USA, a very large cohort, with tens or hundreds of thousands of subjects, can quickly be identified and followed up in the database. Information on the variables of interest can be extracted in a relatively short period of time with little effort and expense. Large cohorts, so identified, will have high statistical power to examine hypotheses of interest.

Retrospectively defined cohorts are, however, compromised in quality because chart data or health care database data tend to be casually rather than accurately recorded; different health care personnel would have obtained and recorded the data for different subjects in the cohort with potentially poor inter-rater reliability, and data for important independent variables may be unavailable for many or all subjects. These advantages and disadvantages are common to all retrospective studies.

Prospective Cohort Studies

As an example of a prospective cohort study, pregnant women can be recruited across the course of two years; relevant participant and gestational data can be recorded, and the children who are born can be followed for, say, the next 30 years to determine what maternal, gestational, and early postnatal characteristics, recorded during and after the pregnancy, predict adult cognitive and mental health.

An advantage of prospective cohort studies is that all relevant variables can be thought of in advance, and data related to these variables can be accurately measured and recorded by trained study staff. Disadvantages are that prospective cohort studies are expensive to conduct and take long to complete; in fact, the investigators who analyze the data may be the successors of those who started the study. For practical reasons related to expense and effort, prospective cohorts are mostly smaller than retrospective cohorts. These advantages and disadvantages are common to all prospective studies.

Theoretical and Practical Issues

Cohort studies are complicated by many issues. Characteristics that are recorded at the baseline, such as medications and drug doses, or smoking and drinking variables, or dietary variables, may change during the course of the study. Repeated follow-up visits need to be scheduled. Subjects may drop out for various reasons, including death. Different subjects are followed up for different durations of time. Most importantly, when comparing subgroups of interest, such as outcomes in children gestationally exposed vs. unexposed to antidepressant drugs, because subjects were not randomized to their respective groups, many confounding variables can complicate the analysis of data and the interpretation of the analysis. To overcome these problems, subjects can be propensity score-matched to reduce confounding, 5 and Cox proportional hazards regression, which yields a hazard ratio, can be used to adjust for confounders and to take into consideration the varying duration of follow-up as well as the time of occurrence of the outcome of interest. Whereas randomized controlled trials (RCTs) follow the CONSORT guidelines and systematic reviews and meta-analyses follow the PRISMA guidelines, cohort studies follow the STROBE guidelines.

Cohort studies can generate useful epidemiological data. However, when examining relationships between variables, because there is no randomization, the results of analyses can only be considered as associations. As an example, gestational exposure to antidepressant drugs may be associated with the development of autism spectrum disorder in the offspring, but whether the association represents a cause– effect relationship cannot be stated. Cause and effect relationships, as studied in RCTs, cannot be determined from observational data generated in cohort studies. However, cohort studies can be used to examine hypotheses that cannot be examined in RCTs, as in the example cited above; this is because RCTs are hard to conduct in conditions such as pregnancy. Cohort studies also generate large quantities of data that can be studied from different perspectives.

Implications for Research in India

Long-term funding is hard to obtain in India, and so few cohort studies in psychiatry have been published from this country. An outstanding example of a prospective cohort study is the Madras Longitudinal Study, initiated in 1981. 6 Another important example is the more recent Thirthalli Cohort. 7 There are many other examples of prospective cohorts, such as the Prospective Assessment of Maternal Mental Health Study, 8 but these are considerably shorter studies. India does not yet have a digitized health care structure that would allow retrospective cohorts to be constructed.

Declaration of Conflicting Interests: The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding: The author received no financial support for the research, authorship, and/or publication of this article.

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings
• My Bibliography
• Collections
• Citation manager

Save citation to file

Email citation, add to collections.

• Create a new collection
• Add to an existing collection

Add to My Bibliography

Your saved search, create a file for external citation management software, your rss feed.

• Search in PubMed
• Search in NLM Catalog
• Add to Search

Use of menopausal hormone therapy before and after diagnosis and ovarian cancer survival-A prospective cohort study in Australia

Affiliations.

• 1 Population Health Program, QIMR Berghofer Medical Research Institute, Brisbane, Australia.
• 2 School of Public Health, University of Queensland, Brisbane, Australia.
• 3 Department of Gynaecological Oncology, Westmead Hospital, Westmead, Australia.
• 4 Centre for Cancer Research, The Westmead Institute for Medical Research, Westmead, Australia.
• 5 The Daffodil Centre, The University of Sydney, A Joint Venture with Cancer Council NSW, Sydney, Australia.
• 6 Consumer Representative, Brisbane, Australia.
• 7 Consumer Representative, Melbourne, Australia.
• 8 Queensland Centre for Gynaecological Cancers, Royal Brisbane and Women's Hospital, Brisbane, Australia.
• 9 Department of Medical Oncology, Prince of Wales Hospital and Prince of Wales Clinical School UNSW Sydney, Sydney, Australia.
• 10 Gynaecological Oncology Unit, Mercy Hospital for Women, Melbourne, Australia.
• PMID: 39222307
• DOI: 10.1002/ijc.35154

Menopausal hormone therapy (MHT) use before ovarian cancer diagnosis has been associated with improved survival but whether the association varies by type and duration of use is inconclusive; data on MHT use after treatment, particularly the effect on health-related quality of life (HRQOL), are scarce. We investigated survival in women with ovarian cancer according to MHT use before and after diagnosis, and post-treatment MHT use and its association with HRQOL in a prospective nationwide cohort in Australia. We used Cox proportional hazards regression to estimate hazard ratios (HR) and 95% confidence intervals (CI) and propensity scores to reduce confounding by indication. Among 690 women who were peri-/postmenopausal at diagnosis, pre-diagnosis MHT use was associated with a significant 26% improvement in ovarian cancer-specific survival; with a slightly stronger association for high-grade serous carcinoma (HGSC, HR = 0.69, 95%CI 0.54-0.87). The associations did not differ by recency or duration of use. Among women with HGSC who were pre-/perimenopausal or aged ≤55 years at diagnosis (n = 259), MHT use after treatment was not associated with a difference in survival (HR = 1.04, 95%CI 0.48-2.22). Compared to non-users, women who started MHT after treatment reported poorer overall HRQOL before starting MHT and this difference was still seen 1-3 months after starting MHT. In conclusion, pre-diagnosis MHT use was associated with improved survival, particularly in HGSC. Among women ≤55 years, use of MHT following treatment was not associated with poorer survival for HGSC. Further large-scale studies are needed to understand menopause-specific HRQOL issues in ovarian cancer.

Keywords: menopausal hormone therapy; ovarian cancer; propensity score; quality of life; survival.

© 2024 The Author(s). International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC.

PubMed Disclaimer

• Australian Institute of Health and Welfare. Cancer in Australia 2019. Cancer Series no.119. 2019.
• Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2021. CA Cancer J Clin. 2021;71:7‐33.
• Collaborative Group On Epidemiological Studies Of Ovarian C, Beral V, Gaitskell K, et al. Menopausal hormone use and ovarian cancer risk: individual participant meta‐analysis of 52 epidemiological studies. Lancet. 2015;385:1835‐1842.
• Lee AW, Wu AH, Wiensch A, et al. Estrogen plus progestin hormone therapy and ovarian cancer: a complicated relationship explored. Epidemiology. 2020;31:402‐408.
• Lee AW, Ness RB, Roman LD, et al. Association between menopausal estrogen‐only therapy and ovarian carcinoma risk. Obstet Gynecol. 2016;127:828‐836.

Grants and funding

• GNT1025142/National Health and Medical Research Council
• GNT1073898/National Health and Medical Research Council
• GNT1120431/National Health and Medical Research Council
• GNT1173346/National Health and Medical Research Council

LinkOut - more resources

Full text sources.

• Citation Manager

NCBI Literature Resources

MeSH PMC Bookshelf Disclaimer

The PubMed wordmark and PubMed logo are registered trademarks of the U.S. Department of Health and Human Services (HHS). Unauthorized use of these marks is strictly prohibited.

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

• View all journals
• Explore content
• About the journal
• Publish with us
• Sign up for alerts
• Open access
• Published: 06 September 2024

Primary succession of Bifidobacteria drives pathogen resistance in neonatal microbiota assembly

• Yan Shao   ORCID: orcid.org/0000-0002-8662-0504 1 ,
• Cristina Garcia-Mauriño 2 ,
• Simon Clare 1 ,
• Nicholas J. R. Dawson 1 ,
• Andre Mu   ORCID: orcid.org/0000-0002-0853-9743 1 ,
• Anne Adoum 1 ,
• Katherine Harcourt 1 ,
• Junyan Liu 1 ,
• Hilary P. Browne   ORCID: orcid.org/0000-0002-1305-2470 1 ,
• Mark D. Stares 1 ,
• Alison Rodger 2 ,
• Peter Brocklehurst 3 ,
• Nigel Field 2 &
• Trevor D. Lawley   ORCID: orcid.org/0000-0002-4805-621X 1  

Nature Microbiology ( 2024 ) Cite this article

52 Altmetric

Metrics details

• Metagenomics
• Microbial ecology

Human microbiota assembly commences at birth, seeded by both maternal and environmental microorganisms. Ecological theory postulates that primary colonizers dictate microbial community assembly outcomes, yet such microbial priority effects in the human gut remain underexplored. Here using longitudinal faecal metagenomics, we characterized neonatal microbiota assembly for a cohort of 1,288 neonates from the UK. We show that the pioneering neonatal gut microbiota can be stratified into one of three distinct community states, each dominated by a single microbial species and influenced by clinical and host factors, such as maternal age, ethnicity and parity. A community state dominated by Enterococcus faecalis displayed stochastic microbiota assembly with persistent high pathogen loads into infancy. In contrast, community states dominated by Bifidobacterium , specifically B. longum and particularly B. breve , exhibited a stable assembly trajectory and long-term pathogen colonization resistance, probably due to strain-specific functional adaptions to a breast milk-rich neonatal diet. Consistent with our human cohort observation, B. breve demonstrated priority effects and conferred pathogen colonization resistance in a germ-free mouse model. Our findings solidify the crucial role of Bifidobacteria as primary colonizers in shaping the microbiota assembly and functions in early life.

Similar content being viewed by others

Metagenomic insights of the infant microbiome community structure and function across multiple sites in the United States

Human milk oligosaccharides modify the strength of priority effects in the Bifidobacterium community assembly during infancy

Bacterial colonization reprograms the neonatal gut metabolome

Human gut microbiota colonization commences immediately at birth when neonates are exposed to microorganisms from the surrounding environment and maternal sources (for example, gut 1 , 2 , 3 , 4 , 5 , vagina 2 , 3 , 4 , skin 3 , 4 , breast milk 3 , 6 ). We recently reported in the UK Baby Biome Study (BBS) that maternal transmission of primary colonizers, such as commensal Bifidobacterium and Bacteroides species, is disrupted in caesarean-section (CS) and antibiotic-exposed births, instead predisposing the neonatal gut microbiota (NGM) to colonization by antibiotic resistant healthcare-associated pathogens 1 . This observation suggests the possibility of ‘priority effects’ in human gut microbiota assembly, which posits the arrival order of primary colonizer species determines the outcome of the microbiota assembly during a primary ecological succession (from sterile to complex communities) 7 , 8 . The NGM represents the earliest window of opportunity for intervention with probiotics or prebiotics to prevent or restore impaired microbiota development. However, little is known about the ecological priority effects in the NGM assembly due to a lack of high-resolution, longitudinal human microbiome data from the neonatal period (that is, the first month of life).

To comprehensively examine NGM assembly dynamics, we expanded on phase 1 of our BBS cohort (BBS1) 1 , 9 with an additional 688 neonatal participants (primarily day 7) in phase 2 (BBS2), effectively doubling our sampling effort. A large-scale, longitudinal metagenomic characterization of the combined BBS dataset, comprising 2,387 gut microbiota samples from 1,288 healthy UK neonates (≤1 month), enabled us to study neonatal microbiota assembly with unparalleled scale and resolution (Extended Data Fig. 1a,b and Supplementary Tables 1 – 3 ). To investigate the origin and both short-term and long-term stability of the NGM primary colonizers, we utilized three subgroups from the expanded BBS2 cohort. These included (1) 183 neonate–mother pairs (representing 14% of participants), referred to as investigating ‘maternal transmission’; (2) 359 participants with longitudinal sampling within the neonatal period (median = 3 samples per participant on days 4, 7 and 21; representing 28% of participants), referred to as investigating ‘neonatal longitudinal colonization’; and (3) 302 participants with paired samples taken both in the neonatal period and later in infancy (at 8.75 ± 1.98 months; representing 23% of participants), referred to as investigating ‘infancy persistence’ (Extended Data Fig. 1c ).

Complementing the increased sample size, we have also updated extensive, high-quality clinical and sociodemographic metadata harmonized from BBS clinical record forms and hospital electronic records (Methods), thereby facilitating robust statistical and epidemiological assessment of primary succession patterns. Most neonates in this cohort (84.5%, N  = 836) were at least partially breastfed by their mothers, with 44.1% being exclusively breastfed ( N  = 436). A large majority of participants at the time of infancy sampling were still being breastfed (86.2%; N  = 199), with very few fully weaned (0.87%; N  = 2). Only 11.3% ( N  = 123) received postnatal antibiotics during the first week of life (Supplementary Table 4 ).

Three community states in the neonatal gut microbiota

To delineate the primary succession patterns of the NGM, we sought to identify the primary colonizers driving gut microbial community structure during the neonatal period. Applying partitioning around medoids (PAM) clustering to 1,904 BBS neonatal gut metagenomes at the species level revealed an optimal clustering of three within the NGM, hereafter referred to as ‘NGM community states’ 10 (Fig. 1a and Extended Data Fig. 2a,b ). These three community states were further validated by another widely used microbial community typing method: the Dirichlet multinomial mixture (DMM) modelling framework (Extended Data Fig. 2c,d ). Both the PAM and DMM-based approaches showed strong concordance in community state assignments (Cramér’s V correlation of 0.726; Extended Data Fig. 2e ) and core species compositions (Extended Data Fig. 2f ). Notably, these three community states were consistently observed across the three main sampling points in the BBS cohort (days 4, 7 and 21), underscoring their representativeness of the neonatal period, irrespective of the timing of sample collection (Extended Data Fig. 3 ).

a , Principal coordinates analysis (PCoA) plots of 1,904 neonatal gut metagenomes sampled within the first 30 days of life and clustered using the PAM algorithm on the basis of species-level JSD. Three distinct NGM community states (optimal number clusters k  = 3) were identified via PAM clustering. The inset pie chart displays the proportion of the three NGM community states, each labelled according to its primary driver species, namely B. breve (BB, green; N  = 336, 17.6% of the samples), E. faecalis (EF, purple; N  = 827, 43.4% of the samples) and B. longum (BL, orange; N  = 741, 38.9% of the samples). Ellipses encapsulate 67% of the samples within each respective cluster. b , Top 10 driver species contributing to variation observed in the ordination space, as ranked by effect size (‘envfit’ R 2 , false discovery rate (FDR)-corrected two-sided test, P  < 0.05). c , d , Each NGM community state is dominated by a single driver species, as measured by the high relative abundance of the driver species ( c ) and the low alpha diversity ( d ) across the three NGM community states (FDR-corrected, two-sided Wilcoxon test). Boxplot centre line and red point indicate the median and mean, respectively; box limits indicate the upper and lower quartiles; and whiskers indicate 1.5× the interquartile range (BB n  = 336, EF n  = 827, BL n  = 741).

Three bacterial species, Bifidobacterium longum subsp. longum (BL), Bifidobacterium breve (BB) and Enterococcus faecalis (EF) acted as the taxonomic drivers for each community state (Fig. 1b and Extended Data Figs. 2g and 4 ). Each species dominated their respective NGM community states with a relative mean abundance of 56.5% for BB, 21.7% for EF and 27.2% for BL (Fig. 1c ). Henceforth, they are referred to as NGM driver species with acronyms indicating each respective community state.

The observed single-species dominance of either B. breve , B. longum or E. faecalis in very early life can also be consistently observed in other cohorts, albeit underreported owing to the previous undersampling during the neonatal period (the largest sample size being <100). Evidence for this comes from diverse populations and methodologies, including 16S gene or quantitative PCR (qPCR)-based observations in Norway 11 ( N  = 87) and Denmark 12 ( N  = 16), as well as shotgun metagenomic surveys of neonates across industrialized urban populations similar to the UK BBS cohort in Europe (Sweden 13 ), Asia (Israel 14 ) and North America (the TEDDY cohort 15 , 16 ) (Extended Data Fig. 5 ). Importantly, the NGM community states observed across industrialized cohorts are paralleled in non-industrialized populations. In a peri-urban cohort in South Asia (Bangladesh 17 ), although B. breve continues to be a primary NGM driver species, the community states typically driven by B. longum and E. faecalis in industrialized settings are instead represented by closely similar species: B. infantis (closely related to B. longum ) and Escherichia coli (sharing facultative anaerobic and opportunistic pathogenic traits with E. faecalis ). Collectively, these cross-study validations strengthen the generalizability of our results in neonatal populations from different geographical regions and lifestyles beyond the UK, and using different methodologies.

Of note, B. longum subsp. infantis ( B. infantis ), which is closely related to BL and often used as an infant probiotic, was not identified as a driver species. It was rarely detected (~2% prevalence based on 0.5% relative abundance) in the BBS neonates 14 . The near absence of B. infantis in our UK neonatal cohort aligns with findings from other Western industrialized countries, including a recent meta-analysis 14 of cohorts from Israel, Sweden, Finland, Estonia, Italy and the USA 18 , where there is little evidence of B. infantis naturally colonizing the gut microbiota of healthy, full-term infants. This underscores the importance of distinguishing between closely related species that exhibit very different host colonization patterns.

Applying metagenomic strain tracking analysis on the ‘maternal transmission’ subset, only B. longum exhibited evidence of maternal transmission, with all evaluable BL neonates (15 out of 15) harbouring the exact same B. longum strain found in their mothers’ gut microbiota. This result, consistent with a recent global meta-analysis 19 , strongly indicates the maternal gut microbiota as the main source of the BL community state (Extended Data Fig. 6 ). While we could have overlooked maternal transmission of very low-abundance B. breve and E. faecalis below the metagenomic strain detection limit, we consider it more likely that they originate from unsampled maternal (for example, B. breve in breast milk 20 , 21 ) or environmental sources (for example, E. faecalis in the hospital birth environment 22 , 23 ) previously implicated as potential sources of these species in the NGM.

The abundant dominance of single driver species was particularly pronounced in community state BB, in which B. breve constituted over half of the NGM by mean relative abundance, and exhibited the lowest microbial richness and evenness, as reflected by the alpha (Shannon) diversity (Fig. 1d ). In comparison, the other two NGM community states, BL and EF, had higher microbial diversity, and other moderately abundant species frequently co-occurred with the driver species (Extended Data Fig. 2f,g ); B. longum with commensal E. coli , Bacteroides and other Bifidobacterium species; E. faecalis with environment and skin-associated Streptococcus , Staphylococcus spp., as well as healthcare-associated opportunistic pathogens Enterococcus , Klebsiella , Enterobacter spp. and C. perfringens . Notably, these less-dominant species in EF were also known signatures of hospital CS birth not only in this UK cohort 1 but also in cohorts from North America 24 , 25 , Latin America 24 and Europe 13 , 24 , 26 .

Factors influencing the acquisition of the NGM community states

To determine the perinatal factors influencing the acquisition of each NGM community state, we performed epidemiological analyses using 20 high-quality clinical and sociodemographic metadata variables ( N  = 1,108 eligible participants; Fig. 2 and Supplementary Table 5 ). After adjusting for potential confounders in multivariate fixed-effect logistic regression models, we found that the acquisition of an EF community state was independently associated with being born via CS birth (compared to vaginal delivery (VD); adjusted odds ratio (AOR) = 2.30 [95% CI 1.34–3.95], P  = 0.003; 70.5/23.6/40.0% among EF/BL/BB, respectively) and with the mother receiving intrapartum antibiotics during labour (AOR = 3.69 [95% CI 2.11–6.42], P  < 0.001; 80.8/32.7/46.3% among EF/BL/BB, respectively). Conversely, being born via CS birth and labour antibiotics exposure were negatively associated with BL acquisition (AOR for CS vs VD = 0.36 [95% CI 0.21–0.64], P  < 0.001; AOR for receiving antibiotics during labour = 0.46 [95% CI 0.26–0.79], P  = 0.005, respectively).

a – c , Multivariate associations between clinical and sociodemographic variables and each week-1 NGM community state. Three different models were built: EF vs non-EF ( a ), BL vs non-BL ( b ) and BB vs non-BB ( c ). Likelihood ratio tests (two-sided) were used to calculate P values (without FDR correction), with P  ≤ 0.05 in the multivariate models displayed. Odds ratios (OR) are plotted on a log 10 scale. For details of univariate and multivariate analyses, refer to Supplementary Tables 5 and 6 . The week-1 NGM community state was identified for each eligible participant using the earliest available sample from week 1, either on day 4 ( N  = 64) or day 7 ( N  = 1,044).

Interestingly, several intrinsic host factors including sex (male with BB), maternal ethnicity (Asian with EF and BB), age (<30 and ≥40 with EF and BL, respectively) and parity (first time giving birth with EF) were also independently associated with specific community states. For example, mothers identifying as Asian (compared with white participants) were more likely to acquire BB (AOR = 2.11 [95% CI 1.32–3.38], P  = 0.006) but less likely to acquire EF (AOR = 0.63 [95% CI 0.41–0.95], P  = 0.04; 9.0/12.1/19.5% among EF/BL/BB, respectively). It is noteworthy that BB is the only community state that was exclusively influenced by host factors and independent of any clinical factors including mode of birth and antibiotics, which may suggest a distinct route of BB acquisition that remains unaffected by the perturbations associated with hospital births. These observations align with the hypotheses that maternal factors, such as genetic determinants of breast milk composition (for example, secretor status of the mothers) 27 , a history of previous pregnancies or cohabitation with children 28 , as well as cross-cultural differences in infant-care-associated behaviours 7 may influence the vertical transmission of maternal microbiota.

Neither postnatal antibiotics nor breastfeeding exposure, whether immediately after birth or within the first week of life, appeared to predispose neonates to any specific community state. This lack of association is probably attributed to the uniformly high-levels of antibiotic-free status (84.7/90.8/89.5% among EF/BL/BB, respectively) and breastfeeding rates (79.1/81.8/88.6% among EF/BL/BB, respectively) during the earliest postnatal window sampled in this cohort. The absence of an association between breastfeeding and EF also aligns with previous reports that, despite its antimicrobial properties, breast milk alone does not inhibit E. faecalis growth in vitro 29 , 30 .

Priority effects in NGM community state stability

We reasoned that the three primary colonizers as NGM drivers could benefit from priority effects, which would be evident through the exclusion of, or replacement by, later-arriving species in the NGM. To search for evidence of such priority effects, we sought to examine the stability and temporal signals of both the NGM community states and their driver species in the ‘neonatal longitudinal’ subset, stratified by birth modes. Most VD neonates who initially acquired a Bifidobacterium -dominated community state (either 92% for BB or 89% for BL, 79% or 72% by considering transient switches between day 4 and 7) during week 1 retained their community state when resampled in week 3 (Fig. 3a and Extended Data Fig. 7a ). By contrast, EF was the most unstable community state, with less than half of the neonates (29% in VD and 39% in CS) remaining in their early EF community state during the neonatal period (EF vs BB AOR 16.2 [95%CI 3.84–68.10], EF vs BL AOR 13.89 [4.02–48.02]; P  < 0.001; Supplementary Table 6 ). Irrespective of birth mode, BB proved more stable than EF (pairwise chi-square test, corrected P  < 0.001), while the sample size was insufficient to be confident about the relative stability of BL in CS neonates (65% versus 48% for EF; pairwise chi-square test, corrected P  = 0.52).

a , b , Stability of NGM community states ( a ) and levels of three species driving NGM community states ( b ) (week 1, based on the earlier sample of day 4 or 7) in neonates longitudinally sampled from weeks 1 to 3 (day 21, total N  = 306; VD N  = 140; CS N  = 166). The proportion of community states that remained consistent from weeks 1 to 3 is depicted as a percentage of their initial sample size in week 1 (labelled in black). Participants starting with BB or BL on week 1 were significantly more likely to retain their community state in week 3 compared with those with EF (pairwise chi-square tests with FDR correction, P  < 0.001). c – e , Persistence of the dominant abundance of driver species of NGM community states in week 1 ( c , d ) or week 3 ( e ) in the paired longitudinal samples obtained later at week 3 ( c ) and in infancy ( d , e ). f , Persistent carriage of week-1 driver species in paired longitudinal samples obtained later in infancy. Species carriage is defined using a threshold of 0.1% relative abundance. Sample sizes of participants longitudinally sampled for weeks 1 and 3 shown in a – c are: total N  = 306; VD N  = 26/39/75 among BB/EF/BL, respectively; CS N  = 26/114/26 among BB/EF/BL, respectively; for week 1 and infancy (also referred to as the ‘infancy persistence’ group) shown in d and f : total N  = 302; VD N  = 27/43/90 among BB/EF/BL, respectively; CS N  = 17/108/17 among BB/EF/BL, respectively; and for week 3 and infancy shown in e : total N  = 146; VD N  = 12/11/43 among BB/EF/BL, respectively; CS N  = 17/37/26 among BB/EF/BL, respectively. Colour represents NGM community states or driver species: BB and B. breve in green; EF and E. faecalis in purple; BL and B. longum in orange. Boxplots as in Fig. 1 . Statistical differences in abundance between time points ( a ), species ( c – e ) and carriage frequency ( f ) were determined using paired t -tests, Wilcoxon tests and chi-square tests (all two-sided) with FDR correction, respectively.

The stability of the underlying driver species closely mirrored the observed community state dynamics. In contrast to E. faecalis , which rapidly declined throughout the stochastic assembly trajectory of the early community state EF, both B. breve and B. longum retained their high abundance within their respective community states throughout the 3-week neonatal sampling window (Fig. 3b and Extended Data Fig. 7b ). Notably, both species, as late-arriving secondary colonizers (that is, colonized NGM only in week 3), exhibited signs of competitively excluding E. faecalis in CS neonates who initially acquired the EF community state (Fig. 3b ). This competitive exclusion effect seemed most pronounced for B. breve ; in contrast to B. longum , it was able to colonize VD neonates at increasing levels as a late-arriving species (Extended Data Fig. 7b ). Among the primary colonizers that dominated the NGM in the first week, B. breve is the only species conferring durable colonization dominance (relative to the other driver species), which persisted as far as the final neonatal period sampling point at week 3 ( P  < 0.001 in VD and CS; Fig. 3c ).

The stability of the two Bifidobacterium species is also reflected at the strain level (Extended Data Fig. 6 ); most of the neonates retained the same B. longum (79.5%, N  = 35/44 BL neonates) or B. breve (75%, N  = 24/32 BB neonates) strain they initially acquired throughout the neonatal period, in contrast to 62.3% for E. faecalis ( N  = 43/69 EF neonates; the denominators represent longitudinally sampled individuals with detectable strain sharing events).

Together, as primary colonizers, both Bifidobacterium species benefit from priority effects, maintaining a stable NGM assembly trajectory owing to their ability to confer durable species dominance and inhibit the later arrival of opportunistic pathogens such as E. faecalis . In particular, B. breve exhibits stronger priority effects between the two species (that is, only as a primary colonizer), as well as strong deterministic exclusion of E. faecalis (that is, as either a primary or a secondary colonizer).

Stability of NGM driver species into infancy

We also assessed the longer-term engraftment of the NGM driver species in participants resampled 6–12 months beyond the neonatal period using the ‘infancy persistence’ subset. Remarkably, the relative dominance of B. breve (over the other driver species, in VD, P  < 0.05; Fig. 3d ) also extended into infancy when there was still no significant difference in breastfeeding rates between early NGM community states (BB/EF/BL: 88.4%/89.6%/80.5%, chi-square test, P  = 0.18). In addition, the long-term competitive exclusion effect of B. breve was evident in CS neonates who either retained or transitioned into BB (primarily from EF) by week 3. These long-term stability patterns were exclusively observed for B. breve , with its abundance in infancy being almost double in neonates who previously had a BB community state compared with those with other community states (Fig. 3e ).

Although NGM driver species rarely retained their differential abundance later in infancy (except B. breve ), the frequency of carriage for all three driver species was consistently higher in infants stratified by their corresponding NGM community states (Fig. 3f ). As many as 93% of VD (or 77% of CS) neonates with week-1 community state of BB still carried B. breve , compared with 58% and 66% (or 65% of CS) of VD neonates with week-1 community states EF and BL, respectively (pairwise chi-square tests, P  < 0.001). While levels of E. faecalis in community state EF waned over time to non-differential levels later in infancy, neonatal acquisition of EF remains a predisposing factor for longer-term carriage of E. faecalis . This opportunistic pathogen species was still detected in higher proportions (44%) in neonates from the EF community state during their first week (relative to 37–41% in BB and 35–38% in BL) when resampled later in infancy, regardless of their birth mode (pairwise chi-square tests, P  < 0.001; Fig. 3f ).

EF state enriched with virulence and antibiotic resistance genes

To determine the functional differences among NGM community states, we leveraged their driver species as proxies for functional analyses, using 1,249 high-quality isolate ( N = 133) and metagenome-assembled genomes ( N  = 1,116) generated from the corresponding community state samples (BB N  = 297, EF N  = 561, BL N  = 391; Supplementary Table 7 ). We found a striking difference between Bifidobacterium spp. and E. faecalis functional profiles in antimicrobial resistance (AMR) and virulence potential. Importantly, all E. faecalis strain genomes recovered from neonates with EF community states encoded known virulence factors including 70% predicted to produce the toxin cytolysin 31 . By contrast, both Bifidobacterium driver species genomes displayed markedly reduced levels of AMR and virulence-associated genes, with a burden 10- to a 100-fold less than in EF (median 17 versus 0; Fig. 4a ). Further AMR gene screening of the entire gut resistome within each community state revealed a higher carriage of high-risk AMR genes, such as CTX-M-15 linked to extended-spectrum beta-lactamase (ESBL), in both BL and EF community states (Fig. 4b ). This underscores the notable pathogenic potential of ESBL-carrying Enterobacteriaceae pathogens co-occurring in non-BB community states. These findings align with our risk factor analyses (Fig. 2 ), which identified maternal antibiotics exposure during labour (to some VD and all CS neonates) as a strong risk for the acquisition of an EF community state that bears increased risk of AMR and virulence.

a , Counts of detected AMR and virulence genes in driver species genomes, with median values enclosed in brackets. Wilcoxon test (two-sided) with FDR correction; number of genomes (isolates in brackets): BB N  = 297 (30), EF N  = 561 (54) and BL N  = 391 (49). b , Carriage of high-risk AMR genes associated with ESBL in the day-7 NGM community state samples based on raw metagenomic assemblies (BB N  = 207, EF N  = 498, BL N  = 444). The x axis shows the most clinically prevalent ESBL genes belonging to CTX-M, OXA, SHV and TEM families. c , Proportion of species genomes, indicated by a colour gradient, predicted to utilize HMOs or their primary downstream products, lactose and fucose. The actual proportions are labelled for genotypes that are not completely present. The predictions are based on the presence of both the gene and its encoded transporters required for utilization of each substrate. 2′-fucosyllactose (2′-FL) liberates lactose and fucose which are also present in breast milk. Utilizations of LNnT, LNT and LNB will all liberate lactose. d , NGM driver species BB confers pathogen colonization resistance in vivo. The boxplot depicts the relative abundance of BB compared to the opportunistic pathogen species EF or K. oxytoca (KO). The x axis represents three experimental groups co-colonized as follows: (1) BB type strain DSM 20213 (2′-FL + ) with EF; (2) BB natural variant D19 (2′-FL − , isolated from a BBS neonate) with EF; and (3) BB type strain (2′-FL + ) with KO (D63). The BB genotype (2′-FL +/− ) indicates whether the strain encodes the α- l -fucosidase (GH95) enzyme encoding for 2′-FL metabolism. In each co-colonization group, one group of mice also received a 2′-FL supplement (50 mg ml −1  per day) in their daily drinking water. The y axis for BB co-colonization with KO is shown on a log scale. Each experimental condition included 3–5 mice per cage and 3 technical replicate cages. Statistical differences between treatment groups were determined using a t -test with Welch’s correction (two-sided). Boxplot centre line indicates the median, box limits indicate upper and lower quartiles, and whiskers indicate 1.5× the interquartile range.

Pathogen resistance of B. breve via metabolic adaptation to HMOs

At the genome-wide functional level, we observed distinct metabolic landscapes of NGM community states based on KEGG orthologues (Extended Data Fig. 8a ), particularly in metabolic repertoire of carbohydrate-active enzymes (Extended Data Fig. 8b ). Both Bifidobacterium community states, in contrast to EF, exhibited an enrichment in carbohydrate-active enzymes associated with metabolizing human milk oligosaccharides (HMOs) abundant and exclusively found in human breast milk. By contrast, EF predominantly possesses genes tailored for utilizing complex dietary glycans such as mannan and chitin, as well as those like starch and cellulose that are commonly found in a plant-based diet usually consumed later in life (Extended Data Figs. 8b and 9 ).

Compared with the limited HMO metabolic capability of the BL community state, BB is capable of utilizing the all the major HMO substrates including lacto- N -tetraose (LNT), lacto- N -neotetraose (LNnT) and lacto- N -biose (LNB), as well as the primary end-products of HMO metabolism l -fucose and d -lactose, which are naturally present in human breast milk (Fig. 4c ). Interestingly, among the three community states, only BB—comprising nearly all B. breve genomes (97.6%, N  = 290/297)—encode the enzyme (α- l -fucosidase, GH95 or GH29) required for metabolizing the most abundant HMO component 2′-fucosyllactose (2′-FL). Although these B. breve strains lack known transporters for importing 2′-FL for intracellular metabolism, previous in vitro experiments have shown that similar strains are capable of growing on 2′-FL 32 , 33 . Therefore, B. breve might be able to metabolize 2′-FL via a previously uncharacterized pathway. In contrast, such capability is extremely rare among BL (5.0%, N  = 19/391) and completely absent in EF (Fig. 4c ). Notably, the species-level variations in HMO utilization observed in the study strains are representative of BB/BL/EF species, exhibiting patterns consistent with those previously reported 34 . These patterns are not influenced by breastfeeding rates in this neonatal cohort, which are uniformly high and statistically indistinguishable among the community states (79.1%, 81.8% and 88.6% for EF, BL and BB, respectively).

Given that opportunistic pathogens including E. faecalis, E. faecium, Klebsiella oxytoca, K. pneumoniae, Enterobacter cloacae and Clostridium perfringens , which are enriched in the EF community state, lack the capability to metabolize HMOs and their by-products, we hypothesize that B. breve ’s versatility in utilizing these predominant neonatal dietary components substantially enhances its fitness against opportunistic pathogens in vivo. Considering that all neonates in the study would have been exposed to the same level of HMOs through a predominantly breast milk-based diet, regardless of their community state, we reason that the metabolic capability to utilize HMOs, including but not limited to 2′-FL, not only contributes to the dominance and stability of the BB community state but also enables B. breve to outcompete pathogenic species that cannot utilize HMOs. Supporting our hypothesis, we demonstrate in a gnotobiotic mouse model, co-colonized with B. breve and the opportunistic pathogen driver species E. faecalis , that B. breve dominates, and this dominance is amplified by dietary 2′-FL supplementation (Fig. 4d ). The 2′-FL-mediated pathogen resistance in vivo phenotype of B. breve also extends to the Gram-negative enteropathogen K. oxytoca , albeit to a lesser extent. Importantly, the anti-pathogen effect was absent in mice colonized with a natural B. breve variant isolated from a BBS neonate lacking the α- l -fucosidase (GH95) enzyme necessary for 2′-FL metabolism. These findings suggest that B. breve ’s strain-specific and gene-dependent utilization of HMOs could have a crucial role in enhancing resistance to pathogen colonization by inhibiting pathogen growth.

In presumably the largest neonatal gut metagenome study ever undertaken, we discovered three distinct NGM community states in over 1,000 healthy, full-term neonates drawn from the general UK population, representing diverse ethnicities and sociodemographic backgrounds. Factors that may influence the maternal gut microbiota, such as maternal age, ethnicity and parity, as well as events that influence its vertical transmission to the neonatal gut during the perinatal period (for example, CS and maternal antibiotics), serve as independent determinants of the acquisition of primary colonizers. The presence of a highly unstable community state (EF) with AMR-enriched opportunistic pathogens underscores the hospital environments and practices, such as maternal antibiotics during labour and elective CS births, as important risk factors 1 , 35 , 36 , 37 , 38 . Although antibiotics after birth and breastfeeding are known important factors shaping the later infant-stage microbiome development 13 , 15 , 39 , 40 , these postnatal factors had no observable effect on very early NGM dynamics on either the acquisition or the switching of the NGM community states. Together, our findings highlight that the NGM assembly outcome is highly dependent on the succession of primary colonizer species, with prenatal and perinatal factors associated with birth exerting profound influences.

Although the early-life microbiota is thought to be highly dynamic as reflected by high inter-individual variation 1 , here we describe an undisturbed, native primary succession pattern in microbiota assembly driven by a single Bifidobacterium species. B. longum is strongly linked to factors that promote maternal gut microbiota transmission at birth, such as vaginal delivery and absence of antibiotics. While B. breve seems unaffected by these factors, its independent association with maternal ethnicity (Asian) could be linked to the mother’s FUT2 secretor status, which determines the presence of 2′-FL and other HMOs in breast milk and is reportedly more common in Asian participants than in white participants 41 . The pattern of exclusive dominance by either B. breve or B. longum during very early life could also be observed in other cohorts across geographically diverse populations 11 , 12 , 13 , 14 , 15 , 16 . Earlier neonatal cohorts, limited by their smaller sample sizes ( N  < 100 compared with N  > 1,000 in this study) and lack of longitudinal samplings, were unable to report such patterns as distinctly and conclusively as we have in this study. Given that de novo identification of optimal community state clusters is sample size dependent 10 , our expanded BBS dataset—nearly 10 times larger than the previously largest neonatal dataset 13 —provided us with the statistical power to report a distinct tripartite NGM community structure. This includes a previously undescribed at-risk community state (EF) harbouring AMR-carrying opportunistic pathogens, and presumably for the first time, the epidemiological and longitudinal dynamics signatures of each NGM community state. Our findings provide crucial evidence that can guide the rational selection of species and strains for infant interventional trials, as well as the development of next-generation microbiota-based therapeutics. Future studies can stratify infants by their earliest gut community states to examine potential associations with longer-term health outcomes.

Both Bifidobacterium community states can drive deterministic and stable assembly trajectories in vivo through optimized utilization of HMOs exclusively present in human breast milk, the predominant diet during the neonatal period. Our human and in vivo data are in agreement with recent observations based on in vitro experiments 42 , 43 , showing that B. breve is functionally better adapted to an HMO-rich diet in very early life and dominate NGM through priority effects. Here we further demonstrated, in human and mouse, the functional impact of B. breve priority effects, resulting in stronger colonization resistance against AMR-enriched pathogens, including E. faecalis and K. oxytoca .

While the exact origins of opportunistic pathogens such as E. faecalis contributing to EF remain to be confirmed, their strong association with disruptions of natural birth (for example, CS and antibiotics) and their ubiquitous presence in the hospital birth environment 22 , 23 strongly indicate the hospital operating room as the most likely source, with exposure further exacerbated by the lack of maternal microbiota transmission that frequently occurs during natural birth. Although the EF perturbation patterns appear to be largely transient, with the neonatal microbiota naturally recovering from a delayed colonization trajectory 1 , 44 , inadequate pathogen clearance could persist into infancy. Along with the short-term exposure to high AMR and virulence, early acquisition of pathogens represents increased risk for infection susceptibility due to the immature immune system in very early life 45 . Also, the delayed or lack of exposure to commensal B. breve or/and B. longum as a primary colonizer in the critical neonatal window of immunity 45 and neurological 46 development could potentially result in neurodevelopment and immune-mediated disorders later in childhood 47 . Epidemiological evidence from other independent birth cohorts indicates that a non- Bifidobacterium (for example, EF) community state may predispose neonates to an increased risk of neurological disorders 48 and respiratory diseases (for example, asthma and atopy 49 , 50 ), including respiratory infections 26 , 51 , later in childhood.

Bifidobacterium spp. are known to achieve bifidogenic effects through the provision of HMOs, with a notable focus on B. infantis and its probiotic application as a specialized HMO-utilizing species. Despite its prevalence and dominance in infants from low- to middle-income and non-industrialized settings 17 , 52 , B. infantis is notably absent in this UK cohort and other Western cohorts, suggesting that it may no longer be naturally colonizing newborns in Western, industrialized populations. Its notable absence indicates a potential lack of a reservoir for B. infantis to establish itself as a primary colonizer, despite the considerable selective advantage that extensive exposure to HMOs during the neonatal period would presumably provide. Our results demonstrate that an HMO functional niche could be filled by other species ( B. breve or B. longum ) capable of metabolizing HMO if they are prevalent in the perinatal microbial species pool. The findings of strain-dependent utilization of HMOs, including but not limited to 2′-FL, and colonization resistance phenotypes of B. breve further highlight that the success of primary succession is probably dependent on both the species prevalence and strain-level functional variation.

Maternal seeding of microbial metabolizers of the specialized bioactives in breast milk probably represents an evolutionally conserved strategy to prime human gut microbiota assembly with primary colonizers with the highest likelihood for priority effects, such as B. breve and, to a lesser extent, B. longum . While both species have been associated with maternal origins 53 , strain transmission analyses from both our work as well as that of others 19 have identified only B. longum as the most frequently transmitted species from the mother’s gut. Although B. breve did not appear to originate from the maternal gut microbiota, we cannot rule out the possibility of vertical transmission of very low-abundance B. breve strains. Recent cultivation-based evidence has confirmed that such transmission can occur below the limits of metagenomic strain detection 54 . Other unsampled maternal or environmental sources could also be involved in seeding B. breve . One likely source is breast milk microbiota, where B. breve has been detected and implicated in the entero-mammary pathway—a retrograde mechanism for milk inoculation 21 . Future research should investigate the global strain reservoir and transmission patterns of Bifidobacterium species, especially for the poorly understood B. breve . Considering the limited success of probiotic-derived B. infantis strains in natural engraftment of neonatal gut microbiota in both industrialized and non-industrialized populations 18 , 55 , comprehensive strain-level functional characterizations of naturally prevalent and stable primary colonizers, such as B. breve , are vital. This effort will expedite the discovery of infant probiotics that are better optimized for local populations.

Study population

The Baby Biome Study (BBS) participants were recruited at the Barking, Havering and Redbridge University Hospitals NHS Trust, the University Hospitals Leicester NHS Trust and the University College London Hospitals NHS Foundation Trust from May 2014 to December 2017. The study was approved by the NHS London City and East Research Ethics Committee (REC reference 12/LO/1492). Mothers provided written informed consent for their participation and the participation of their children in the study. The study was performed in compliance with all relevant ethics regulations.

Whole-genome sequencing and analysis

The study participants, drawn from a general population of women giving birth in hospitals in the UK without any clinical inclusion or exclusion criteria as per the BBS study protocol 56 , are predominantly healthy, full-term neonates. The study dataset comprised 2,387 metagenomes, with 1,679 from the previously published 1 BBS phase 1 (BBS1) and 708 new neonatal gut metagenomes in BBS phase 2 (BBS2), totalling 1,288 participants. The aim of BBS2 was to sequence all the remaining neonatal samples collected from the original BBS study. The study sample size was predicated on detecting differences by mode of birth rather than providing statistical power to discern differences in microbial community states. The sampling and data processing protocols, ranging from sample collection to sequence data generation, quality control (low-quality trimming and human decontamination) and processing, remained unchanged from those previously described 1 for BBS1. In brief, faecal samples were collected at home by parents from neonates in the first 3 weeks of life (primarily on days 4, 7 and 21) and later in infancy. Paired maternal faecal samples were taken at the hospital around the time of birth. Most new samples in BBS2 were collected on day 7 of life. The only change was an institute-wide upgrade in the Illumina sequencing platform, transitioning from HiSeq 2500-v4 (2 ×125 bp) to HiSeq 4000 (2 ×151 bp). A multiplexing strategy was employed to ensure that the target depth remained consistent with BBS1. While the upgraded sequencing platform has resulted in a marginal increase in sequencing depth for BBS2 (from 19.3 to 20.4 million reads per sample post-quality control, calculated with seqkit (v.2.4.0) 57 , P  < 0.001, two-sided t -test), it did not impact either the community state assignment ( P  = 0.4731, likelihood ratio test via multinomial logistic regression) or the recovery of high-quality genomes (proportion of the total genome bins) for NGM driver species ( P  = 0.9716, Mantel–Haenszel chi-squared test, stratified by species).

Read-based taxonomic classification was performed against the Genome Taxonomy Database (GTDB, RS207) representative bacterial and archaeal species genomes ( N  = 65,703) using bowtie2 (v.2.3.5) 58 and inStrain (v.1.3.0) 59 ‘profile’ with the recommended ‘–database mode’ and 50% genome breadth (covered by ≥1 read) cut-off, as previously described 52 , 59 . The R package phyloseq (v.1.12.0) 60 was used for metagenomic data analysis, and results were processed and visualized using tidyverse (v.2.0.0) in RStudio (v.4.1.0).

Strain sharing analysis was performed using StrainPhlAn4 (ref. 61 ), following the workflow and species-specific strain identity thresholds previously described 19 . Where appropriate, multiple testing corrections were applied to all statistical tests using the Benjamini–Hochberg FDR method with a significance threshold of 5%, unless otherwise specified.

Cultivation and whole genome sequencing of the NGM species isolates were performed using the previously established workflow 1 for BBS1. In brief, the NGM species in driver NGM samples were cultured from corresponding frozen faecal samples using selective media: Bifidobacterium selective media (Sigma-Aldrich) for B. longum and B. breve , and Enterococcus selective agar (Sigma-Aldrich) for E. faecalis . Purified bacterial isolates were sequenced on the Illumina HiSeq X or NovaSeq 6000 system (2 ×151 bp), and assembled and quality-controlled using shovill (v.1.1.0; https://github.com/tseemann/shovill ) and CheckM2 (ref. 62 ), respectively.

Clinical and sociodemographic metadata sources and management

Participant data were collected using a clinical record form at enrolment by the BBS research midwives or from available clinical records at birth. Hospital maternity electronic records with pregnancy and perinatal clinical information were obtained directly from the hospital trusts, and databases containing the variables of interest were merged. Variables were harmonized where possible across different databases. For discrepancies, data from the BBS clinical record forms were given priority, and hospital electronic data were used to complete missing data. At the time of stool sample collection, mothers completed a short form on feeding mode and antibiotic exposure. A total of 20 variables were included in the final analyses on the basis of clinical relevance, quality of data and completeness ( N  = 6 maternal, N  = 8 perinatal or at time of delivery, N  = 5 postnatal, N  = 1 at the time of stool sample collection variables). Ten variables had no missing or <1% missing data. Four had between <1% and 5% missing data (index of multiple deprivation (IMD), maternal smoking, prolonged rupture of membranes (PROM) and neonatal labour antibiotics after birth), two had between 5% and 15% missing data (maternal ethnicity and feeding mode at the time of stool sample collection), and one had >30% missing data (skin to skin).

We used participant postcode to determine IMD 63 , which provides a measure of socioeconomic status that is calculated as an area-level relative deprivation score that we organized into quintiles from 1 (least deprived) to 5 (most deprived). The score considers seven individually weighted domains (income, employment, education, health, crime, barriers to housing and services, and living environment). Prophylactic antibiotics were administered to all mothers undergoing caesarean section in this cohort, as well as to newborns displaying risk factors or clinical indicators of early-onset neonatal infection, in accordance with local trust policies and UK national guidelines at the time 64 , 65 . To our knowledge, no participants were given antibiotics for treating bloodstream infections of E. faecalis . Skin-to-skin contact is defined as contact of mother and baby immediately after birth at least for 1 h or until the next feeding 66 . Feeding mode at the time of stool sample collection was determined through a questionnaire that included three categories: exclusive breastfeeding, exclusive bottle feeding, or both (that is, mixed feeding). For comparisons involving (non)exclusive breastfeeding, the latter two categories were merged into a single ‘non-exclusive breastfeeding’ category.

Statistical analyses

No statistical methods were used to pre-determine sample sizes, but this study already represents the largest dataset of longitudinal faecal metagenomes ( n  = 1,904; n  = 2,387 including infancy samples) of newborns ( n  = 1,288). No data were excluded unless they failed quality control steps. Microbiome data collection and analysis were not randomized or performed blind to the conditions of the experiments, as this is an observational study. Biological counting experiments were blinded by another person other than the experimenter before being counted to avoid experimental bias. For mouse experiments, treatments were randomized by cage by researchers blinded to treatment conditions. Unless otherwise stated, non-parametric statistical tests were performed unless tests for normality and equal variances showed that these assumptions were met.

For the epidemiological analyses of NGM community states in the first week of life, BBS participants with sufficient metadata were explored (90.4%, N  = 1,108 of 1,225 participants with week-1 sampling). The week-1 NGM community state was determined for each eligible participant by using the earliest available sample from week 1, collected either on day 4 ( N  = 64) or day 7 ( N  = 1,044).

To ascertain risk factors for specific NGM community states: BB versus non-BB, EF versus non-EF, and BL versus non-BL, univariate analyses using fixed-effect logistic regression models were initially performed. Subsequent multivariate models were constructed, also using fixed-effect logistic regression, and included only participants with complete datasets while excluding variables with over 15% missing data. Likelihood ratio tests were employed to calculate all P values. A hierarchical framework was applied in building the multivariate models. Variables were organized in a sequential order into either distal (maternal) or more proximal categories (delivery, postnatal care and the first week of life). Variables were considered potential confounders if they occurred simultaneously with or before exposure variables 67 . Within each category, all variables from that category or previous categories were incorporated into the model to account for confounding.

Sensitivity analyses were conducted to identify factors associated with NGM community state, switching between weeks 1 and 3. This included a subset of ‘neonatal longitudinal’ participants with sufficient metadata (87.6%, N  = 268 of 306, corresponding to Fig. 2a ). Both univariate and subsequent multivariate analyses were conducted using fixed-effect logistic regression in the same manner as described above. Multivariate models were further adjusted for the week-1 community state (that is, EF, BB or BL) to discern whether any associations were driven by the baseline community state. There was no strong evidence of association, other than for the baseline community state itself. These analyses could not extend to independent community states switches due to insufficient sample size. All analyses were conducted using Stata (v.17.0).

Community state assignment

The NGM community state assignment was applied to all neonatal samples ( N  = 1,904) using two popular methods, namely, the original clustering-based PAM method described in ref. 68 and the probabilistic modelling-based Dirichlet multinomial mixtures (DMM) approach described previously 69 . In accordance with the original protocols, PAM clustering was applied to the species-level relative abundance distance measured by the Jensen–Shannon divergence (JSD) using the R packages ‘cluster’ (v.2.1.4) and ‘vegan’ (v.2.6.4), and DMM models were fitted on the species-level relative abundance matrix, modelled by the Dirichlet multinomial distribution, using the R package ‘DirichletMultinomial’ (v.1.4). For both methods, the optimal number of clusters of three was determined on the basis of the Calinski–Harabasz index for PAM clustering and the model fit score based on Laplace approximation for DMM. The community states were named according to the top taxonomic driver (species) that contributed the most to microbial community variation (‘envfit’ R 2 , P  < 0.05) in PAM and to each Dirichlet component (cluster) in DMM. The strength of association between the PAM and the DMM-based community states was 0.726 (Cramer’s V correlation). For downstream analyses, the PAM-based community state assignment was selected because it maximized both the sample size of community states BB and BL (Extended Data Fig. 2e ) and the mean relative abundance of the driver species in the respective community state ( B. breve in BB, E. faecalis in EF; Extended Data Fig. 2f ).

To validate the single-species dominance in external neonatal cohorts, the same workflow for community state type assignment was independently applied to four public gut metagenomic datasets with a comparable sampling window (<6 months) to the BBS cohort, including partial or exclusive sampling of the neonatal period (0–1 month). The earliest sampling windows were from cohorts derived from diverse geographical populations and lifestyles, including Sweden 13 , 42 ( PRJEB6456 , days 4–12, N  = 37), Israel 14 ( PRJNA994433 , weeks 1–24, N  = 60), the USA (TEDDY cohort 15 , 16 , PRJNA400115 , months 2–6, N  = 69) and Bangladesh 17 ( PRJNA806984 , months 0–2, N  = 234).

Metagenome assembly and functional analyses

Quality-controlled, raw paired-end reads were first assembled with SPAdes (v.3.13.5) 70 with the option –meta. Unassembled reads were then filtered out by mapping raw reads back to metaSPAdes 71 -assembled contigs using bwa-mem (v.0.7.17) 72 , followed by re-assembly with MEGAHIT (v.1.1.3) 73 using default parameters. Subsequently, the metaSPAdes and MEGAHIT assemblies were combined, sorted and short contigs (<1,500 bp) removed. The resulting assemblies were then independently binned with MetaBAT 2 (v.2.13) 74 , MaxBin2 (v.2.2.4) 75 and CONCOCT (v.0.4) 76 using default parameters and a minimum contig length threshold of 1,500 bp (option –minContig 1500). The depth of contig coverage required for the binning was inferred by mapping the raw reads back to their assemblies with bwa-mem (v.0.7.17) and then calculating the corresponding read depths of each individual contig with samtools 77 (‘samtools view -Sbu’ followed by ‘samtools sort’) together with the ‘jgi_summarize_bam_contig_depths’ function from MetaBAT 2.

Thereafter, individual genome bin sets produced by three binning programs were consolidated into a refined bin set consisting of the best version of each bin based on the most optimal genome completion and contamination metrics among all seven versions of hybridized bin sets (MetaBAT 2, MaxBin2, CONCOCT, MetaBAT 2 + MaxBin2, MetaBAT 2 + CONCOCT, MaxBin2 + CONCOCT, MetaBAT 2 + MaxBin2 + CONCOCT) as estimated by CheckM (v.1.0.7) 78 using the metaWRAP (v.1.2) 79 ‘bin_refinement’ pipeline 79 . In total, 22,668 prokaryotic metagenome-assembled genomes (MAGs) met the criteria of having >50% completeness and <5% contamination, as determined by CheckM2 (ref. 62 ). These MAGs were subsequently taxonomically assigned using the GTDB 80 (R214) taxonomy with GTDB-Tk (v.2.3.0) 81 .

For genome analyses of the three NGM driver species, data were derived from samples as either cultivated isolate genomes or metagenome-assembled genomes (MAGs) when cultured strains were unavailable. Only near-complete, high-quality MAGs were used in the functional analyses ( N  = 1,116). All genomes met strict quality control criteria, which included ≥90% completeness, ≤5% contamination, an N50 value of ≥10 kb, passing the GUNC test, an average contig length of ≥5 kb and ≤500 contigs, as previously described 82 . Genome annotation of metabolic function was performed using DRAM (v.1.4.5) 83 , which integrates annotations from multiple databases, including Pfam, KEGG (KOfam), UniProt, dbCAN (carbohydrate-active enzymes) and MEROPS (peptidases). The functional gene counts from KEGG and CAZy annotations were used to generate a PCA plot using the R package ‘pcaMethods’, employing conventional singular value decomposition with imputation. The genome-based prediction of HMO substrate utilization was based on KEGG and CAZy annotations mapped against a list of manually curated relevant genes and pathways as described recently 42 , 43 . The genes corresponding to HMO substrates (enzymes; transporters) were: 2′FL (GH95 and/or GH29, FL1_Blon0341-0343 and/or FL2_Blon2202-2204), lactose (GH2, LacS), fucose (FumC/D/E/F/G, FucP), LNT (GH42 or GH136, GltABC), LNnT (GH20, Bbr_1554) and LNB (GH112, GltABC). In silico screening of AMR and virulence factor genes was performed at the species level with species MAGs and at the sample level with raw metagenome assemblies as input for ABRicate against the NCBI AMRFinderPlus and VFDB databases as previously described 1 . The AMR genes encoding for the extended-spectrum β-lactamase (ESBL) phenotype were annotated using the curated antibiotic subclass of the NCBI Pathogen Detection Reference Gene Catalog (as of 1 October 2023).

Bacterial strains and reagents

The bacterial strains used in this study were either part of the in-house (HMIL) culture collection cultivated from the BBS faecal samples or requested from public collections (DSMZ). Specific strains were: B. breve strains (type strain DSM 20213 and D19 isolated from a BBS neonate), E. faecalis (D13 isolated from a BBS neonate) and K. oxytoca (D63 isolated from a BBS neonate). Purified HMO 2′FL (GlyCare 2FL 9000, batch 20156002) and LNnT were purchased from Glycom, DSM.

Mouse experiment

Wild-type C57BL/6N mice were maintained under germ-free conditions at the Wellcome Sanger Institute Home Office-approved facility, with all procedures carried out in accordance with the UK Animals (Scientific Procedures) Act of 1986 under Home Office approval (PPL no. 80/2643). Germ-free mice were housed under a 12 h light/12 h dark cycle, ambient temperature and humidity condition in positive-pressure isolators (Bell), with faeces tested by culture, microscopy and PCR to ensure sterility. Consumables were autoclaved at 121 °C for 15 min before introduction into the isolators. For experimentation, 6-week-old mice of both sexes were randomly assigned to treatment groups. Cages were opened in a vaporized hydrogen peroxide-sterilized, class II cabinet (Bioquell), with mono-colonized gnotobiotic lines generated by oral gavage on day 1 ( B. breve ) at the concentration of 10 9  colony-forming units (c.f.u.) per ml and day 4 (challenged by opportunistic pathogen species E. faecalis or K. oxytoca at the concentration of 10 4  c.f.u. per ml). Materials were prepared in Dulbecco’s PBS at 100 mg ml −1 immediately before administration under anaerobic conditions (10% H, 10% CO 2 , 80% N) in a Whitley DG250 workstation at 37 °C. Mice were maintained in sterile ISOcages (Tecniplast) and housed on ISOrack for the period of the experiment.

Control groups of mice colonized with BB, EF or KO without any treatment were also included to confirm mono-colonization. One of the two groups of the co-colonized mice (for both BB + EF and BB + KO experiments) were exposed to 2′-FL via daily drinking water (50 mg ml −1  per day) throughout the experiment. Faecal samples were collected on each oral gavage day and plated to test for contamination. Mice were killed on day 11 (7 days post inoculation on day 4), with faecal samples collected and plated for colony count on yeast extract casitone fatty acids (YCFA) aerobically (to select for E. faecalis or K. oxytoca ) and YCFA with mupirocin (to select for Bifidobacterium spp.) media under anaerobic conditions. YCFA is a complex, broad-range medium 84 . Each experimental condition included 3–5 mice per cage and 3 technical replicate cages.

DNA was extracted from faeces using FastDNA Spin Kit for Soil (MPBio) according to manufacturer instructions, and DNA eluted into 100 µl of double-distilled H 2 O. Eluted DNA was then diluted 1:50 and qPCR performed using SYBR Green chemistry (Thermo Fisher). The absolute bacterial load in each faecal sample was determined by qPCR using a calibration curve generated with genomic DNA and taxon-specific primer sequences ( E. faecalis , F: 5′-CCCTTATTGTTAGTTGCCATCATT-3′, R: 5′-ACTCGTTGTACTTCCCATTGT-3′; Bifidobacterium spp., F: 5′-CTCCTGGAAACGGGTGG-3′, R: 5′-GGTGTTCTTCCCGATATCTACA-3′; K. oxytoca , F: 5′-GGACTACGCCGTCTATCGTCAAG-3′, R: 5′- TAGCCTTTATCAAGCGGATACTGG-3′). As previously described 85 , the relative abundance of each target species was estimated by normalizing to those of a universal bacterial 16S primer (F: 5′-GTGSTGCAYGGYTGTCGTCA-3′, R: 5′-ACGTCRTCCMCACCTTCCTC-3′).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability

Shotgun metagenomic sequencing data (after quality trimming and human decontamination) of the entire Baby Biome Study cohort have been deposited to the European Nucleotide Archive under study accession number ERP115334 . Bacterial genome assemblies for the three species analysed have been deposited in Zenodo at https://doi.org/10.5281/zenodo.12667210 (ref. 86 ). Sample metadata and participant-level clinical metadata of de-identified study participants are provided in the Supplementary Tables. The raw faecal samples and bacterial isolates are available from the corresponding authors upon request.

Code availability

All software used to perform these analyses is publicly available. Software tools used are listed in the main text and Methods.

Shao, Y. et al. Stunted microbiota and opportunistic pathogen colonization in caesarean-section birth. Nature 574 , 117–121 (2019).

Article   CAS   PubMed   PubMed Central   Google Scholar  

Mitchell, C. M. et al. Delivery mode affects stability of early infant gut microbiota. Cell Rep. Med. 1 , 100156 (2020).

Bogaert, D. et al. Mother-to-infant microbiota transmission and infant microbiota development across multiple body sites. Cell Host Microbe 31 , 447–460 (2023).

Article   CAS   PubMed   Google Scholar  

Ferretti, P. et al. Mother-to-infant microbial transmission from different body sites shapes the developing infant gut microbiome. Cell Host Microbe 24 , 133–145 (2018).

Yassour, M. et al. Strain-level analysis of mother-to-child bacterial transmission during the first few months of life. Cell Host Microbe 24 , 146–154 (2018).

Fehr, K. et al. Breastmilk feeding practices are associated with the co-occurrence of bacteria in mothers’ milk and the infant gut: the CHILD cohort study. Cell Host Microbe 28 , 285–297 (2020).

Sprockett, D., Fukami, T. & Relman, D. A. Role of priority effects in the early-life assembly of the gut microbiota. Nat. Rev. Gastroenterol. Hepatol. 15 , 197–205 (2018).

Article   PubMed   PubMed Central   Google Scholar  

Debray, R. et al. Priority effects in microbiome assembly. Nat. Rev. Microbiol. 20 , 109–121 (2022).

Mäklin, T. et al. Strong pathogen competition in neonatal gut colonisation. Nat. Commun. 13 , 7417 (2022).

Costea, P. I. et al. Enterotypes in the landscape of gut microbial community composition. Nat. Microbiol. 3 , 8–16 (2018).

Avershina, E. et al. Bifidobacterial succession and correlation networks in a large unselected cohort of mothers and their children. Appl. Environ. Microbiol. 79 , 497–507 (2013).

Laursen, M. F. & Roager, H. M. Human milk oligosaccharides modify the strength of priority effects in the Bifidobacterium community assembly during infancy. ISME J . 17, 2452–2457 (2023).

Bäckhed, F. et al. Dynamics and stabilization of the human gut microbiome during the first year of life. Cell Host Microbe 17 , 690–703 (2015).

Article   PubMed   Google Scholar  

Ennis, D., Shmorak, S., Jantscher-Krenn, E. & Yassour, M. Longitudinal quantification of Bifidobacterium longum subsp. infantis reveals late colonization in the infant gut independent of maternal milk HMO composition. Nat. Commun. 15 , 894 (2024).

Stewart, C. J. et al. Temporal development of the gut microbiome in early childhood from the TEDDY study. Nature 562 , 583–588 (2018).

Vatanen, T. et al. The human gut microbiome in early-onset type 1 diabetes from the TEDDY study. Nature 562 , 589–594 (2018).

Vatanen, T. et al. A distinct clade of Bifidobacterium longum in the gut of Bangladeshi children thrives during weaning. Cell 185 , 4280–4297.e12 (2022).

Casaburi, G. et al. Metagenomic insights of the infant microbiome community structure and function across multiple sites in the United States. Sci. Rep. 11 , 1472 (2021).

Valles-Colomer, M. et al. The person-to-person transmission landscape of the gut and oral microbiomes. Nature 614 , 125–135 (2023).

Martín, R. et al. Isolation of bifidobacteria from breast milk and assessment of the bifidobacterial population by PCR-denaturing gradient gel electrophoresis and quantitative real-time PCR. Appl. Environ. Microbiol. 75 , 965–969 (2009).

Kordy, K. et al. Contributions to human breast milk microbiome and enteromammary transfer of Bifidobacterium breve . PLoS ONE 15 , e0219633 (2020).

Brooks, B. et al. Microbes in the neonatal intensive care unit resemble those found in the gut of premature infants. Microbiome 2 , 1 (2014).

Brooks, B. et al. Strain-resolved analysis of hospital rooms and infants reveals overlap between the human and room microbiome. Nat. Commun. 8 , 1814 (2017).

Song, S. J. et al. Naturalization of the microbiota developmental trajectory of Cesarean-born neonates after vaginal seeding. Med 2 , 951–964.e5 (2021).

Dos Santos, S. J. et al. Maternal vaginal microbiome composition does not affect development of the infant gut microbiome in early life. Front. Cell. Infect. Microbiol. 13 , 303 (2023).

Google Scholar  

Reyman, M. et al. Impact of delivery mode-associated gut microbiota dynamics on health in the first year of life. Nat. Commun. 10 , 4997 (2019).

Lewis, Z. T. et al. Maternal fucosyltransferase 2 status affects the gut bifidobacterial communities of breastfed infants. Microbiome 3 , 13 (2015).

Martin, R. et al. Early-life events, including mode of delivery and type of feeding, siblings and gender, shape the developing gut microbiota. PLoS ONE 11 , e0158498 (2016).

Schlievert, P. M., Kilgore, S. H., Seo, K. S. & Leung, D. Y. Glycerol monolaurate contributes to the antimicrobial and anti-inflammatory activity of human milk. Sci. Rep. 9 , 14550 (2019).

Sweeney, E. et al. The effect of breastmilk and saliva combinations on the in vitro growth of oral pathogenic and commensal microorganisms. Sci. Rep. 8 , 15112 (2018).

Coburn, P. S. & Gilmore, M. S. The Enterococcus faecalis cytolysin: a novel toxin active against eukaryotic and prokaryotic cells. Cell. Microbiol. 5 , 661–669 (2003).

Bunesova, V., Lacroix, C. & Schwab, C. Fucosyllactose and l -fucose utilization of infant Bifidobacterium longum and Bifidobacterium kashiwanohense . BMC Microbiol. 16 , 248 (2016).

Ruiz-Moyano, S. et al. Variation in consumption of human milk oligosaccharides by infant gut-associated strains of Bifidobacterium breve . Appl. Environ. Microbiol. 79 , 6040–6049 (2013).

Sakanaka, M. et al. Varied pathways of infant gut-associated Bifidobacterium to assimilate human milk oligosaccharides: prevalence of the gene set and its correlation with bifidobacteria-rich microbiota formation. Nutrients 12 , 71 (2019).

Azad, M. B. et al. Impact of maternal intrapartum antibiotics, method of birth and breastfeeding on gut microbiota during the first year of life: a prospective cohort study. BJOG 123 , 983–993 (2016).

Tapiainen, T. et al. Impact of intrapartum and postnatal antibiotics on the gut microbiome and emergence of antimicrobial resistance in infants. Sci. Rep. 9 , 10635 (2019).

Nogacka, A. et al. Impact of intrapartum antimicrobial prophylaxis upon the intestinal microbiota and the prevalence of antibiotic resistance genes in vaginally delivered full-term neonates. Microbiome 5 , 93 (2017).

Li, W. et al. Vertical transmission of gut microbiome and antimicrobial resistance genes in infants exposed to antibiotics at birth. J. Infect. Dis. 224 , 1236–1246 (2021).

Bokulich, N. A. et al. Antibiotics, birth mode, and diet shape microbiome maturation during early life. Sci. Transl. Med. 8 , 343ra82 (2016).

Yassour, M. et al. Natural history of the infant gut microbiome and impact of antibiotic treatment on bacterial strain diversity and stability. Sci. Transl. Med. 8 , 343ra81 (2016).

Azad, M. B. et al. Human milk oligosaccharide concentrations are associated with multiple fixed and modifiable maternal characteristics, environmental factors, and feeding practices. J. Nutr. 148 , 1733–1742 (2018).

Ojima, M. N. et al. Priority effects shape the structure of infant-type Bifidobacterium communities on human milk oligosaccharides. ISME J. 16 , 2265–2279 (2022).

Lou, Y. C. et al. Infant microbiome cultivation and metagenomic analysis reveal Bifidobacterium 2′-fucosyllactose utilization can be facilitated by coexisting species. Nat. Commun. 14 , 7417 (2023).

Podlesny, D. & Fricke, W. F. Strain inheritance and neonatal gut microbiota development: a meta-analysis. Int. J. Med. Microbiol. 311 , 151483 (2021).

Olin, A. et al. Stereotypic immune system development in newborn children. Cell 174 , 1277–1292.e14 (2018).

Bethlehem, Ra. I. et al. Brain charts for the human lifespan. Nature 604 , 525–533 (2022).

Torow, N. & Hornef, M. W. The neonatal window of opportunity: setting the stage for life-long host–microbial interaction and immune homeostasis. J. Immunol. 198 , 557–563 (2017).

Beghetti, I. et al. Early-life gut microbiota and neurodevelopment in preterm infants: any role for Bifidobacterium ? Eur. J. Pediatr. 181 , 1773–1777 (2022).

Depner, M. et al. Maturation of the gut microbiome during the first year of life contributes to the protective farm effect on childhood asthma. Nat. Med. 26 , 1766–1775 (2020).

Fujimura, K. E. et al. Neonatal gut microbiota associates with childhood multisensitized atopy and T cell differentiation. Nat. Med. 22 , 1187–1191 (2016).

Alcazar, C. G.-M. et al. The association between early-life gut microbiota and childhood respiratory diseases: a systematic review. Lancet Microbe 3 , e867–e880 (2022).

Olm, M. R. et al. Robust variation in infant gut microbiome assembly across a spectrum of lifestyles. Science 376 , 1220–1223 (2022).

Browne, H. P., Shao, Y. & Lawley, T. D. Mother–infant transmission of human microbiota. Curr. Opin. Microbiol. 69 , 102173 (2022).

Feehily, C. et al. Detailed mapping of Bifidobacterium strain transmission from mother to infant via a dual culture-based and metagenomic approach. Nat. Commun. 14 , 3015 (2023).

Barratt, M. J. et al. Bifidobacterium infantis treatment promotes weight gain in Bangladeshi infants with severe acute malnutrition. Sci. Transl. Med. 14 , eabk1107 (2022).

Bailey, S. R. et al. A pilot study to understand feasibility and acceptability of stool and cord blood sample collection for a large-scale longitudinal birth cohort. BMC Pregnancy Childbirth 17 , 439 (2017).

Shen, W., Sipos, B. & Zhao, L. SeqKit2: a Swiss army knife for sequence and alignment processing. iMeta 3 , e191 (2024).

Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9 , 357–359 (2012).

Olm, M. R. et al. inStrain profiles population microdiversity from metagenomic data and sensitively detects shared microbial strains. Nat. Biotechnol. 39 , 727–736 (2021).

McMurdie, P. J. & Holmes, S. phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS ONE 8 , e61217 (2013).

Blanco-Míguez, A. et al. Extending and improving metagenomic taxonomic profiling with uncharacterized species using MetaPhlAn 4. Nat. Biotechnol . 41 , 1633–1644 (2023).

Chklovski, A., Parks, D. H., Woodcroft, B. J. & Tyson, G. W. CheckM2: a rapid, scalable and accurate tool for assessing microbial genome quality using machine learning. Nat. Methods 20 , 1203–1212 (2023).

Ministry of Housing, Communities and Local Government. English indices of deprivation 2019 (GOV.UK, 2019).

Caesarean Birth NICE guideline [NG192] (NICE, 30 January 2024); https://www.nice.org.uk/guidance/ng192

Neonatal Infection: Antibiotics for Prevention and Treatment NICE guideline [NG195] (NICE, 19 March 2024); https://www.nice.org.uk/guidance/ng195

Widström, A., Brimdyr, K., Svensson, K., Cadwell, K. & Nissen, E. Skin‐to‐skin contact the first hour after birth, underlying implications and clinical practice. Acta Paediatr. 108 , 1192–1204 (2019).

Victora, C. G., Huttly, S. R., Fuchs, S. C. & Olinto, M. T. The role of conceptual frameworks in epidemiological analysis: a hierarchical approach. Int. J. Epidemiol. 26 , 224–227 (1997).

Arumugam, M. et al. Enterotypes of the human gut microbiome. Nature 473 , 174–180 (2011).

Holmes, I., Harris, K. & Quince, C. Dirichlet multinomial mixtures: generative models for microbial metagenomics. PLoS ONE 7 , e30126 (2012).

Bankevich, A. et al. SPAdes: a new genome assembly algorithm and its applications to single-cell sequencing. J. Comput. Biol. 19 , 455–477 (2012).

Nurk, S., Meleshko, D., Korobeynikov, A. & Pevzner, P. A. metaSPAdes: a new versatile metagenomic assembler. Genome Res. 27 , 824–834 (2017).

Li, H. & Durbin, R. Fast and accurate long-read alignment with Burrows–Wheeler transform. Bioinformatics 26 , 589–595 (2010).

Li, D., Liu, C.-M., Luo, R., Sadakane, K. & Lam, T.-W. MEGAHIT: an ultra-fast single-node solution for large and complex metagenomics assembly via succinct de Bruijn graph. Bioinformatics 31 , 1674–1676 (2015).

Kang, D. D. et al. MetaBAT 2: an adaptive binning algorithm for robust and efficient genome reconstruction from metagenome assemblies. PeerJ 7 , e7359 (2019).

Wu, Y.-W., Tang, Y.-H., Tringe, S. G., Simmons, B. A. & Singer, S. W. MaxBin: an automated binning method to recover individual genomes from metagenomes using an expectation-maximization algorithm. Microbiome 2 , 26 (2014).

Alneberg, J. et al. Binning metagenomic contigs by coverage and composition. Nat. Methods 11 , 1144–1146 (2014).

Li, H. et al. The sequence alignment/map format and SAMtools. Bioinformatics 25 , 2078–2079 (2009).

Parks, D. H., Imelfort, M., Skennerton, C. T., Hugenholtz, P. & Tyson, G. W. CheckM: assessing the quality of microbial genomes recovered from isolates, single cells, and metagenomes. Genome Res. 25 , 1043–1055 (2015).

Uritskiy, G. V., DiRuggiero, J. & Taylor, J. MetaWRAP—a flexible pipeline for genome-resolved metagenomic data analysis. Microbiome 6 , 158 (2018).

Parks, D. H. et al. A complete domain-to-species taxonomy for Bacteria and Archaea. Nat. Biotechnol. 38 , 1079–1086 (2020).

Chaumeil, P.-A., Mussig, A. J., Hugenholtz, P. & Parks, D. H. GTDB-Tk v2: memory friendly classification with the genome taxonomy database. Bioinformatics 38 , 5315–5316 (2022).

Nayfach, S., Shi, Z. J., Seshadri, R., Pollard, K. S. & Kyrpides, N. C. New insights from uncultivated genomes of the global human gut microbiome. Nature 568 , 505–510 (2019).

Shaffer, M. et al. DRAM for distilling microbial metabolism to automate the curation of microbiome function. Nucleic Acids Res. 48 , 8883–8900 (2020).

Duncan, S. H., Hold, G. L., Harmsen, H. J., Stewart, C. S. & Flint, H. J. Growth requirements and fermentation products of Fusobacterium prausnitzii , and a proposal to reclassify it as Faecalibacterium prausnitzii gen. nov., comb. nov. Int. J. Syst. Evol. Microbiol. 52 , 2141–2146 (2002).

Forster, S. C. et al. Identification of gut microbial species linked with disease variability in a widely used mouse model of colitis. Nat. Microbiol. 7 , 590–599 (2022).

Shao, Y. Bacterial genomes of the Baby Biome Study. Zenodo https://doi.org/10.5281/zenodo.12667210 (2024).

Download references

Acknowledgements

This work was funded by the Wellcome Trust and the Wellcome Sanger Institute (WT101169MA, 206194 and 220540/Z/20/A). For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. We thank the participating families for their time and contribution to Baby Biome Study; and the research midwives at recruiting hospitals for recruitment and clinical metadata collection.

Author information

Authors and affiliations.

Host–Microbiota Interactions Laboratory, Wellcome Sanger Institute, Hinxton, UK

Yan Shao, Simon Clare, Nicholas J. R. Dawson, Andre Mu, Anne Adoum, Katherine Harcourt, Junyan Liu, Hilary P. Browne, Mark D. Stares & Trevor D. Lawley

Institute for Global Health, University College London, London, UK

Cristina Garcia-Mauriño, Alison Rodger & Nigel Field

Birmingham Clinical Trials Unit, University of Birmingham, Birmingham, UK

Peter Brocklehurst

You can also search for this author in PubMed   Google Scholar

Contributions

Y.S. and T.D.L. conceived and designed the study. Y.S. coordinated the experiments and performed computational analyses with assistance from A.M. Y.S. and M.D.S. cultured bacteria strains and performed DNA extraction. S.C. performed germ-free mouse experiments with assistance from N.J.R.D., A.A., K.H., J.L. and H.P.B. A.R., P.B., N.F. and T.D.L. conceived and designed the Baby Biome Study and obtained funding. N.F., A.R. and P.B. managed participant recruitment and sample collection, and coordinated the clinical metadata collection. C.G.-M. curated the clinical metadata and undertook the clinical epidemiological analyses with N.F. Y.S. and T.D.L. wrote the manuscript with inputs from H.P.B., A.M., C.G.-M., A.R., P.B. and N.F.

Corresponding authors

Correspondence to Yan Shao or Trevor D. Lawley .

Ethics declarations

Competing interests.

T.D.L. is the co-founder and CSO of Microbiotica. The other authors declare no competing interests.

Peer review

Peer review information.

Nature Microbiology thanks the anonymous reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended data fig. 1 overview of sampling in the baby biome study..

(a-b) Shotgun metagenomes of 2,387 faecal samples from 1,288 neonatal subjects across Phase 1 (Shao et al. 1 ) and Phase 2 (this paper). The majority of samples (80% or 1,904) are from the neonatal period (b) , primarily taken on day 4 (N=360), day 7 (N=1,149), and day 21 (N=350). (a) Rows represent subjects with paired maternal samples (for ‘maternal transmission’ analysis), longitudinal samples taken during the neonatal period (for ‘neonatal longitudinal’ analysis), and samples from the infancy period (for ‘infancy persistence analysis’). These relationships are indicated by lines linking the samples, with summarised proportions in (c) .

Extended Data Fig. 2 Consistency of NGM community state assignment across typing methods.

(a-d) Identification of three NGM community states using both (a) Partitioning Around Medoids (PAM) clustering of JSD, with statistical support from the Calinski-Harabasz (CH) index, and (c) Dirichlet Multinomial Mixture (DMM) modelling using the Laplace approximation. ( b, d ) PCoA plots, representing 1,904 neonatal gut metagenomes, are color-coded by community state assignments, and based on species-level Bray-Curtis distances. (e-g) PAM-based and DMM-based community state assignment concordance: (e ) Correlation between community state assignments shown with a Cramér's V correlation of 0.726. The proportions of community states assigned by each method are labelled. The breakdown of community states BB/EF/BL in PAM is 336/827/741, and in DMM is 252/1097/555. (f) Overlap in the dominant core species (≥1% mean abundance) in each community state, grouped at the genus level, with exceptions for the driver species B. breve , E. faecalis and B. longum . PAM-based assignment was chosen in downstream analyses given the higher relative abundances of these driver species in their respective community states (versus DMM-based assignment): B. breve 67.9% vs 56.5% (p<0.001), E. faecalis 21.7% vs 16.8% (p<0.001), and B. longum 27.25% vs 29.90% (p=0.24). Wilcoxon-test (two-sided) with FDR correction. ( g ) The top 10 driver species for each DMM-based community state are displayed, ranked by their assignment strength, as indicated on the y-axis.

Extended Data Fig. 3 Consistency of NGM community state assignment across neonatal time points.

Identification of three NGM community states using PAM-based clustering across three major time points in the neonatal period (day 4, N=360; day 7, N=1,149; day 21, N=350). PCoA plots, are color-coded by community state assignments and based on species-level JSD. Ellipses encapsulate 67% of the samples within each respective cluster.

Extended Data Fig. 4 Abundance and co-occurrence of the NGM community state driver species.

PCoA plots depicted in Fig. 1 , with arrows, illustrate the scale and direction of core NGM species (>1% mean abundance) driving the formation of NGM community states (clusters). The length of the arrows is scaled to reflect the degree of contribution to the variation in NGM composition, with the arrow points towards increasing species abundance. Species that frequently co-occur with the NGM driver species within their respective community states share the same arrow direction.

Extended Data Fig. 5 Validation of NGM community states and driver species across geographies and lifestyles.

All three NGM community states and the driver species ( B. breve, B. longum or B. infantis, and E. faecalis ) were independently detected in infant gut metagenomic cohorts (0–6 months) from diverse geographical regions and lifestyles. These include Europe (Sweden, days 4–12, N=37), the United States (TEDDY cohort, months 2–6, N=69), the Middle East (Israel, weeks 1–24, N=60), and South Asia (Bangladesh, months 0–2, N=234). In the Bangladeshi cohort, which is a non-industrialised and non-urban population, the B. infantis and E. coli -driven clusters are representative of the B. longum (closely related to B. infantis ) and E. faecalis (also facultative anaerobe opportunistic pathogen) community states, respectively. The analysis and visualization methods are consistent with those described in Fig. 1a, b .

Extended Data Fig. 6 Strain-level dynamics and stability across NGM species.

(a) Frequency of study participants detected with the same strains (in grey, otherwise in white) from their mother's faecal samples across NGM community states. To delineate transmission trends, the chart is categorized by birth mode and the three NGM driver species. Frequency of strain-sharing event (for example, maternal transmission in mother-baby pair or strain persistence within-individual longitudinal samples) is presented as raw counts of detectable strain sharing events normalized by the total number of subjects per birth mode and NGM community state (week 1). (b) Bar plots counting strain-sharing events across three settings: (Left) Maternal transmission in mother-infant dyads (183 subjects; 167 transmissions from 213 evaluated species-sample pairs). (Middle) Neonatal persistence via neonatal longitudinal sampling (359 subjects; 700 transmissions from 938 evaluated pairs). (Right) Infancy persistence from neonatal into infancy period (302 subjects; 464 transmissions from 920 evaluated pairs). When longitudinal samples were considered, strain sharing events were considered only once per subject per setting, using the time point with highest counts. Only species with ≥20 strain-sharing events detected across three settings are shown. Three community state driver species are highlighted in boxes. Transmission patterns often align with phylogeny: Actinomycetota/Actinobacteria (pink) and Bacteroidota/Bacteroidetes (green) typically transmit maternally during vaginal birth and persist into infancy. Conversely, Bacillota/Firmicutes (purple) and Pseudomonadota/Proteobacteria (orange) show lower maternal transmission rates and reduced neonatal persistence. Notable outliers include E. coli and B. breve . The size of bubbles represents the transmissibility of each species, which is its ratio of detected to potential strain-sharing events, as determined by StrainPhlAn4. Only subject pairs with sequencing depth sufficient for StrainPhlAn strain-level analyses are displayed; data points not shown are non-evaluable.

Extended Data Fig. 7 Colonisation dynamics in neonatal longitudinal samples.

(a) Overview of NGM community states of all subjects individually sampled on major neonatal period sampling points day 4, 7 or 21, stratified by birth mode. In VD, N=176/602/156 on day 4, 7, and 21, respectively; In CS, N=184/547/194 on day 4, 7, and 21, respectively. Total samples N=1859. (b) Longitudinal shifts in NGM community states and the levels of driver species from week 1 to week 3, based on subjects longitudinally sampled across days 4, 7, and 21, N=234; VD, N=111; CS, N=123). Community states that remained consistent from first (day 4) to the final neonatal longitudinal sampling (day 7 or/and 21), is depicted as a percentage of their starting pool size (labelled in black). Subjects that began with either BB or BL community state on day 4 were significantly more likely to remain in the same community state on day 7 and 21, compared to those that began with EF (pairwise chi-squared tests with FDR correction, q-values < 0.01). However, this trend was not observed as early as day 7 (global chi-squared test, p=0.7043). The colour scheme represents the community states or driver species: BB and B. breve in green; EF and E. faecalis in purple; BL and B. longum in orange. Statistical differences in species abundance between longitudinal samples was determined using paired ANOVA test (two-sided) with FDR correction. Boxplot center line and red point indicate the median and mean, respectively; box limits indicate the upper and lower quartiles; and whiskers indicate 1.5× the interquartile range.

Extended Data Fig. 8 Species-driven functional divergence in NGM community states.

(a-b) Principal Component Analysis (PCA) of community state driver enterotype species genomes. Groupings are based on the presence of genes tied to the full metabolic repertoire using (a) KEGG orthologs (KOfams) and carbon metabolism via (b) Carbohydrate-Active enZYmes (CAZymes). Each dot denotes an individual strain: B. longum (BL, N=342) in orange, B. breve (BB, N=267) in green, and E. faecalis (EF, N=507) in blue. Ellipses encapsulate the 95% confidence intervals. Arrows showcase the contribution of select CAZy genes to principal components (details in Extended Data Fig. 8 ). CAZy genes for human milk oligosaccharides (HMOs) utilisation are highlighted in red.

Extended Data Fig. 9 Carbon metabolism of NGM community state driver species.

(a) A heatmap displays the clustering of carbohydrate-active enzymes (CAZymes) across the genomes of three driver species. Genes are coloured based on their corresponding carbohydrate substrate categories. (b-d) Volcano plots depict differentially enriched CAZymes in each driver species, comparing (b) BB vs. EF, (c) BL vs. EF, and (d) BB vs. BL. The effect size represents the difference in the proportion of genes between species. P-values are adjusted using Fisher's exact test (two-sided) with FDR correction. Genes related to HMO metabolism are marked in red. Significantly enriched genes are labelled for clarity. Arrows at the top indicate the direction of species enrichment in each comparison. B. longum (BL, N=342) is shown in orange, B. breve (BB, N=267) in green, and E. faecalis (EF, N=507) in blue.

Supplementary information

Reporting summary, supplementary data.

Table of contents tab. Supplementary Table 1. Sample_accession. ENA accessions of the 2,387 samples of entire BBS cohort analysed in this manuscript: BBS1 (N = 1,679) and BBS2 (N = 708). 2. Neonatal_subject. Clinical metadata of the BBS neonates included in the statistical analysis (N = 1,108), cohort characteristics described in Extended Data Table 1. 3. Neonatal_sample. Sample metadata (age/day and community states) of the neonatal samples included in the analyses. N = 1,904. 4. Epi_metadata_summary. Descriptive table of the BBS neonatal population with available metadata (N = 1,108/1,288, 90%). 5. Epi_result_neonatal_state. Clinical and sociodemographic variables associated with the acquisition of NGM community state measured in the first week of life (N = 1,108). 6. Epi_result_neonatal_switch. Clinical and sociodemographic variables associated with NGM community state switching between week 1 and week 3 (N = 306). 7. Genome_accession. Sample accessons of the species genomes generated from the BBS samples and analysed in the functional analyses.

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Shao, Y., Garcia-Mauriño, C., Clare, S. et al. Primary succession of Bifidobacteria drives pathogen resistance in neonatal microbiota assembly. Nat Microbiol (2024). https://doi.org/10.1038/s41564-024-01804-9

Download citation

Received : 22 April 2024

Accepted : 05 August 2024

Published : 06 September 2024

DOI : https://doi.org/10.1038/s41564-024-01804-9

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

Quick links

• Explore articles by subject
• Guide to authors
• Editorial policies

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.