Cohort studies are the analytical design of observational studies that are epidemiologically used to identify and quantify the relationship between exposure and outcome. Due to the longitudinal design, cohort studies have several advantages over other types of observational studies. The purpose of this chapter is to cover the various characteristics of prospective cohort studies. This chapter is divided into three main sections. In the first we introduce the concept and ranking of cohort studies, as well as the advantages and disadvantages. In the second we focus on the design of cohort studies, mainly its prospective aspect, and the distinguishing features from the retrospective type. The section also covers the essential characteristics of a cohort study design and its varied applications in medical research. In the third we go over examples of prospective studies in the medical field. For each, an overview of the study design is given, along with a random selection of study findings/impact, strengths and weaknesses.
- observational study
- cohort study
- prospective cohort study
- longitudinal study
- study design
- medical research
1. Cohort studies
The term “cohort” originates from Latin “cohors” . A term that was used in the military back in Roman times, which referred to a unit that is comprised of 300–600 men, of which each 10 cohorts were named a legion . In the field of epidemiology, Frost was the first to introduce the term “cohort study” back in 1935 . Cohort refers to a group of individuals that share a common factor or a defining characteristic [4, 5], or in other words, cohort is a certain component of a specific population that can be measured and followed throughout time . Cohort studies are classified under the non-experimental type of studies , which are observational by default .
A cohort study follows people as groups, two or more, from exposure to outcome [2, 8]. The two groups would be categorized based on their exposure status to “exposed” and “unexposed” [4, 9, 10]. If there were multiple groups then these would be categorized either by the type or level of exposure . The main characteristic of a cohort study is that it follows participants in a forward manner, from the presence of the exposure to the presence of the outcome [2, 9, 10, 11]. Or as De Rango describes it: using a longitudinal pattern, a cohort study, follows a group or groups of individuals over time in order to ascertain the incidence of a predetermined outcome after being exposed to a certain factor, whether being a risk factor, medication, or intervention . Cohort studies can either be prospective (concurrent) or retrospective (non-concurrent) .
1.2. Ranking of cohort studies
Researchers agree that cohort studies, as related to the hierarchy of evidence, rank below meta-analysis, systematic review and randomized controlled trial, but rank higher than case–control studies, cross sectional studies, case series/reports [13, 14, 15, 16]. As newer models or classifications of the hierarchy of evidence have emerged, where meta-analysis and systematic reviews have been removed from the hierarchy and repositioned as a magnifying glass or a lens through which evidence from other types of studies can be viewed or scrutinized; cohort studies remain below randomized controlled trials and higher than the other types . Cohort studies provide information on the relationship between exposure and outcome when a randomized controlled trial is not possible to conduct for whatever reason [6, 15].
1.3. Advantages of cohort studies
Cohort studies are the design of choice when randomization is not practical or ethical [6, 18]. They are also useful in the study of infections  and for hypothesis generation . Due to the design of cohort studies, and since temporal sequence is present, both incidence rate and cumulative incidence can be calculated [2, 8, 20, 21, 22]. They also allow for the measurement of relative risk (RR) [2, 8, 23], hazard ratio , and attributable risk [8, 23]. Furthermore, they allow for the study of multiple outcomes that can be associated with a single type of exposure [2, 20] or multiple exposures . Additionally, they allow for the study of rare exposures [2, 18, 20]. Finally, cohort studies have lower risk of encountering survivor bias , and recall bias [9, 21]. Survivor bias occurs when focusing only on those who survived or made it through a certain criteria or point, and ignoring those that didn’t, such as studying rapidly fatal diseases .
1.4. Disadvantages of cohort studies
Among the disadvantages of cohort studies is selection bias, which may occur when the participants are not representative of the population or of the patient grouping that they fall under. This in turn will influence how well or not the results can be generalized to the rest of the population, in what is known as external validity [2, 12, 18, 24, 25]. This will be covered later in section three of this chapter under aspects of cohort studies. Another disadvantage is that causation cannot be established from cohort studies [18, 20], as it would require an experimental design in order to determine any causal effect . However, due to the longitudinal design of cohort studies, they may aid in studying a certain causal hypothesis . A third disadvantage is that they require a large sample size, which might pose an issue when dealing with outcomes that take a long time to develop . Finally, cohort studies cannot be used to study rare outcomes .
2. Prospective cohort studies
2.1. Types of cohort studies
Cohort studies are either prospective or retrospective [1, 2, 18]. In the former, the researcher would assess exposure at baseline and then follow the person over time in order to determine the outcome such as the development of a disease [9, 18, 20, 21, 26]. In the latter, the order is reversed, as a cohort is established after the follow up has been conducted, or the outcome has developed, and exposure is then assessed in a retrospective manner [9, 18, 20, 21, 27]. Merrill indicates that the outcome status at the start of the study is what determines the overall study type. If the outcome has not yet developed then it is a prospective study, and if the outcome has already developed then it is a retrospective study . Cohort studies can also be classified based on whether or not participants are replaced once they are lost. If those that drop out or are lost to follow up are replaced with new participants, then this would be classified as a dynamic or an open cohort. In the case that those lost do not get replaced, then it would be classified as a fixed or closed cohort [4, 20].
2.1.1. Prospective cohort studies
Prospective cohort studies, as the name indicates, observes a group of people after being exposed to a certain factor in order to investigate the outcome, following the natural sequence of time, starting with the present and looking forward in time [12, 18, 20], which in turn provides true risk (absolute) estimates for the groups under investigation . It is considered the gold standard among observational studies . Under this type of study, the researcher would have control over data collection methodology, as well as the overall cohort study set up, which gives prospective cohort studies an advantage over retrospective cohort studies . Further advantages and disadvantages of prospective cohort studies are discussed below.
220.127.116.11. Advantages of prospective cohort studies
Euser et al. highlight the major advantage of prospective cohort studies as being accurate in regards to the information collected about exposures, endpoints, and confounders . Others list the following as advantages of prospective cohort studies; first: the exposure has already been measured before the outcome has occurred, which allows for the assessment of temporal sequence . This allows for the calculation of incidence and the determination of the disease process [2, 12, 20, 23]. Second: elimination of recall bias, as there is no need for any recollection of information since the data is being collected in a prospective manner . However, Kip et al. reported that recall bias can pose an issue in prospective cohort studies if the exposure is self-reported, brief, and requires multiple measurements, such as stress episodes . Third: It allows for the study of exposures were randomization is not practical or ethical . Fourth: it allows for the study of rare exposures . Fifth: it allows for the study of multiple outcomes [20, 26].
18.104.22.168. Disadvantages of prospective cohort studies
Among the disadvantages of prospective cohort studies is the loss to follow up, which is common among cohort studies. This can ultimately lead to differential loss to follow up among those exposed and unexposed, which in turn can complicate the interpretation of the results [2, 7, 12, 18, 24]. Another disadvantage is that they are time consuming if follow up periods are far apart. This would be resource consuming as well, which would make prospective cohort studies not suitable for the study of outcomes that take long time to develop [18, 20, 24, 26]. A third disadvantage is that they are expensive to conduct [18, 20, 30]. The third section of this chapter is dedicated to providing examples of prospective cohort studies.
2.1.2. Retrospective cohort studies
As previously described, retrospective cohort studies, also known as historic  or historical  cohorts, use data that has already been collected, such as databases of healthcare records, in order to investigate the association between the exposure and the outcome [22, 24, 26, 28]. Although the outcome has already occurred, the design of retrospective cohort studies is similar to those of prospective cohort studies . They also have similar advantages and disadvantages [26, 28]. Hess indicates that retrospective studies in general are useful as pilot studies for future prospective studies .
Retrospective cohort studies have advantages and disadvantages. They are time efficient and cheap since the data has been collected previously and is available for scrutiny [18, 20, 26]. Additionally, since the exposure has already been measured before the outcome has occurred, this allows for the assessment of temporal sequence . However, retrospective cohort studies use information that has been collected in the past for another objective other than the current study , and in some cases, collected for a purpose that is not related to medical research . Due to this factor, the investigator lacks control over the collection of data [24, 26, 27]. Additionally, the measurement of exposure and outcome might be inconsistent or inaccurate, which can become a source of bias [24, 27, 28, 31, 32].
Examples of retrospective cohort studies:
High plasma phosphate as a risk factor for decline in renal function and mortality in pre-dialysis patients [18, 33]. In this study, Voormolen et al. followed the clinical course among incident pre-dialysis patients, using medical charts, to study the decline in kidney function and its association with plasma phosphate levels [18, 33].
Assessment of female sex as a risk factor in atrial fibrillation in Sweden: nationwide retrospective cohort study [28, 34]. In this study, Friberg et al. investigated gender differences in the incidence of stroke among those with atrial fibrillation using the Swedish hospital discharge registry [28, 34].
Outcomes of care by hospitalists, general Internists, and family physicians . In this study, Lindenauer et al. collected data from various hospitals in the USA, and compared the outcome of patients treated by the three types of care provider .
2.1.3. Aspects of cohort studies
Validity is the epidemiological assessment to the lack of systematic error [4, 11]. There are two types of validity: internal validity and external validity [4, 11, 25]. Internal validity refers to the inferences made from the study that are related to the same source population [4, 5, 11, 25, 36], as to whether or not the study has measured what it had originally planned on measuring [25, 36]. For an example, if the exposure caused the observed change in the outcome, then the study would be considered to have high internal validity . On the other hand, if the observed change in the outcome was caused by a systematic error (bias), then the study would be considered to have low internal validity . Threats or violations to internal validity will be discussed later in this section under bias.
External validity refers to the degree to which the study results can be generalized to other populations [4, 5, 11, 25, 36]. For example, if the study participants were not representative of the general population, then the study results cannot be generalizable to others . The highest level of external validity occurs when the results can be generalized to three other domains: other populations, other environments, and other times . External validity can be improved by using random selection .
It is essential to have internal validity in order to establish external validity; that is the study must have internal validity in the first place in order to have external validity [4, 11]. For an example, if the exposure caused the observed change in the outcome, then the results can be generalizable to others. If the observed change was caused by any other factor, then the results cannot be generalized to others [4, 11]. Based on the validity hierarchy, cohort studies are considered to have low internal validity, while the external validity is high [11, 16].
Bias is a study systematic error in the design, conduct, or analysis that can be categorized into three main categories: selection bias, information bias, and confounding [4, 25, 38]. Selection bias occurs when the sample chosen for the study is not obtained randomly, so that the sample chosen is no longer representative of the overall population [4, 25, 38, 39]. This type of bias includes three types: attrition bias, non-respondent bias, and the healthy entrant effect . Attrition bias, or loss to follow up bias, occurs due to dropouts or death, which can be encountered in studies with long follow up durations (prospective) . Non-respondent bias occurs when those that respond are different than those that don’t respond. For example, nonsmokers are more likely to return questionnaires about smoking than smokers are . The healthy entrant effect or the healthy worker effect occurs when there are differences between those that are exposed and those that are not exposed. For an example, when comparing working individuals to the general population, as workers are more likely to be healthier than the general population. In order to avoid this type of bias, it is recommended to use two similar groups, such as using two groups of working individuals .
Information bias (measurement bias) , occurs when the data obtained is being recorded inaccurately [4, 25, 38, 39, 40]. This type of bias can be differential (nonrandom) or nondifferential (random) as related to the outcome [4, 9, 23, 25]. The former is dependent on other variables and leads to overestimation or underestimation of any possible association, while the latter is independent from other variables and leads to underestimation of any possible association [4, 9, 23], and if the exposure was dichotomous, this type leads to bias towards the null . Non differential is more commonly encountered in cohort studies . Information bias can be reduced by using standardized assessment tools that have been validated . Information bias is also known as classification bias, observation bias , or misclassification bias .
Confounding: confounding is a distortion of the effect [4, 25] that may lead to overestimation or underestimation of an effect, or even reversing the direction of an effect . A confounding factor is a risk factor that is associated with the exposure and influences the outcome, however, is not related to the causation sequence [4, 25, 39]. Unlike selection and information bias, confounding can be controlled for prior to study initiation, or after study completion . Controlling for confounding factors can be accomplished through: restriction, matching, stratification, and using multivariate techniques [23, 25, 27].
Restriction would involve excluding those with the confounding factor [23, 25]. If the confounding factor is categorical, then participants that fall within that category would be excluded , such as if smoking was considered to be a confounding factor, then those that smoke would be excluded . If the confounding factor was continuous, such as age, then a range of that variable would be used to restrict the confounding . Matching would involve choosing two groups that are similar to each other as much as possible [23, 25, 41], such as matching by gender or age . Matching can be either individual matching or frequency matching. The former involves matching on an individual participant level, while the latter refers to matching on a group level . Overmatching may occur when matching is being used, which may reflect on the statistical efficiency, validity, or cost efficiency of the study . After the completion of the study, and during the analysis stage, stratification can be used to control for confounding by dividing the groups into several subgroups that are based on the confounding factor [23, 25, 39, 41]. Multivariate techniques are also used during the analysis stage and allow for the control of multiple factors [25, 39, 41].
22.214.171.124. Exposure and risk
Exposure must be determined using a clear and accurate definition [2, 22], which in some cases may involve levels of exposure . This helps in eliminating possible selection bias . The challenge becomes greater when there are multiple exposure assessments over an extended period of time . The validity and the cost are two important aspects that must be taken into consideration when selecting an exposure measurement tool .
Both groups, those exposed and those that are not exposed should be at risk of eventually developing the outcome at some stage . The exclusion criteria should exclude those that are not at risk of developing the outcome . For an example, a study investigating the role of antipsychotics in the development of diabetes, should exclude those with diabetes to start with, since they are not at risk . This helps in eliminating possible selection bias .
Outcomes should have a clear and specific definition from the beginning of the study [2, 22], which must be measurable as well [2, 22]. Outcomes should also be measured in a similar manner across all participants [2, 22]. This helps in eliminating possible information bias . It is recommended to use measurement tools that have been previously validated when dealing with secondary data, and to blind those who are assessing the outcome when dealing with primary data .
The comparison group or controls (unexposed group) should be similar to the exposed group in all possible aspects, but differ in regards to the exposure itself . Three types of controls can be used, with the first being the most preferable: internal comparisons, other external cohorts, and the general population .
126.96.36.199. Follow up
To avoid loss to follow up and its consequent effects on the validity of the study results; measures should be taken in order to minimize the attrition rate [2, 22, 24, 27, 42]. Some of these actions include excluding those that are at high risk of not committing to the study, providing incentives for participation, collecting personal information that would allow or facilitate future contact, and maintaining ongoing contact on regular basis during the conduction period of the study [2, 23, 24, 27]. The maximum acceptable limit for loss to follow up is 20% [23, 24, 42].
Precision is based on the absence of random error or chance [4, 11]. This random variation can be due to the sample itself, or how it was selected, or how it was measured [4, 11]. Standard deviations and confidence intervals are useful in determining the precision of a study, as a large standard deviation or a wide confidence interval would indicate low precision . Random error or variation can be reduced by increasing the sample size [4, 27, 43], improving how you sample and how you measure, in addition to using the appropriate statistical methods .
188.8.131.52. Analysis of data
The main statistical term or product of cohort studies is the relative risk or risk ratio [6, 21], which represents the risk of developing the outcome among those that are exposed in relation to those that are not exposed . An RR that is equivalent to 1 indicates an absence of any type of association. An RR that is greater than 1 would indicate that there is a positive correlation between the exposure and risk of developing a disease. An RR that is smaller than 1 would indicate the presence of a protective effect between the exposure and the outcome . Other outcome measures include: hazard ratios, survival curves, and life-table rates . Some of the common statistical analysis involving cohort studies include: analysis of variance (ANOVA), multivariate analysis of variance (MANOVA), mixed effect regression model, and generalized estimating equation models .
The reporting of prospective cohort studies should follow the STROBE guidelines , which also apply to other observational studies [41, 44]. This acronym stands for: Strengthening the Reporting of Observational Studies in Epidemiology. These guidelines were designed by a group of international scholars including journal editors, epidemiologists, statisticians and researchers in order to set universal standards when reporting observational studies. It is comprised of a 22 item checklist that precisely dictates what should be reported under each section of an article [44, 45, 46, 47]. Sessler and Imrey indicate that the most crucial ones are related to the study: objectives, methodology, definitions, source of data, statistical analysis, participants, and results . Further information can be found at
Bookwala et al. outlined three main factors that aid in evaluating prospective cohort studies in their article titled “the three-minute appraisal of a prospective cohort study”. These are related to (1) comparison groups selection; (2) the impact of confounding variables; (3) type of analytical strategy used . Finally, the equator network (which is supported by the University of Oxford, UK, and aims to improve the quality and transparency of health research) provides guidelines and instructions for the reporting of various kinds of studies. These can be found at
3. Examples of prospective cohort studies
3.1. The Framingham Heart Study
The Framingham heart study, initiated in 1948 by The National Heart Institute (currently the National Heart, Lung, and Blood Institute) , is considered to be the longest, ongoing, prospective cohort study in the history of the USA . Others view it as a live model that illustrates the cohort design . The study was based on the hypothesis that arteriosclerosis and hypertensive cardiovascular disease are the result of several causation factors combined rather than an individual factor . Based on this, the aim of the study was to investigate the factors that contribute to the development of cardiovascular disease (CVD) by following a large cohort of individuals over a long period of time . Back then in 1951, when the first article about the study was published, little was known about arteriosclerosis and hypertensive cardiovascular disease .
The original cohort included 5209 participants, ages 30–62 years, that were recruited at the beginning of the study in the town of Framingham, Massachusetts, USA . The same cohort has been followed since initiation every two years for physical, laboratory, and lifestyle examinations . The second generation, the offspring cohort, was recruited in 1971 and included 5124 participants. While 1994 witnessed the enrollment of the first Omni cohort (n = 506), in order to diversify the study population. More recently in 2002, the third generation cohort (n = 4095) was enrolled, while in 2003 the new offspring cohort (n = 103), and the second Omni group (n = 410) was enrolled . The study continues to follow these cohorts every 2–6 years . This multi generation, multi ethnicity, enrollment design aided significantly in the study of genetics in relation to a wide range of factors and illnesses [51, 54].
Based on the Framingham study data, since initiation and through November 2017, a total of 3561 articles have been published so far . The accumulation of knowledge that has risen from this study has shed the light on cardiovascular disease risk factors [50, 51, 56], by further expanding on our understanding of chronic illnesses such as diabetes, obesity, metabolic syndrome and nonalcoholic fatty liver disease [51, 57]. Such risk factors include high blood pressure, high cholesterol levels, smoking, obesity, diabetes, and physical inactivity [50, 57].
The study was the basis of which the Framingham risk score was built on . Initially published by Wilson et al. in 1998 , it allows for the calculation of a 10 year risk estimate of developing coronary heart disease (CHD) based on the levels of different variables [56, 58]. This would allow for the undertaking of preventive measures . Later on in 2002, the Adult Treatment Panel of the National Cholesterol Education Program used the risk score as a foundation for its risk calculator .
3.1.2. Study findings
The study website (
3.1.3. Strengths and weaknesses
In addition to what had been previously discussed regarding the benefits of the prospective design of the study, a high retention rate is among the strengths of the Framingham Heart Study as participants continue to return for their follow up visits despite the years . Among the weaknesses is that the study was conducted in one population residing in one locality , which in turn reflects on the ability to generalize findings to other populations . Another weakness is that the study cohort was not randomly selected, as investigators had to use volunteers in order to obtain the necessary sample. The final cohort ended up being more healthy when compared to the general population [7, 60].
3.2. The Nurses’ Health Study (NHS)
This National Institutes of Health (NIH) funded study started in 1976 , and as of today includes more than 275,000 participants and counting, as the Nurses’ Health Study 3 is still recruiting subjects . The study looks into the risk factors that have been implicated in major chronic diseases among women . Initially, the study focused on heart disease, cancer, smoking, and contraceptive methods . As the study evolved, it investigated many other lifestyle factors, characteristics, and diseases [61, 63].
The original cohort of the study has been followed up on by mail every two years, with a minimum response rate of 90% . The second cohort, under NHS 2, was enrolled in 1989 and included 116,430 women. These also were followed up on using mail every two years. A food frequency questionnaire was added in 1991 and was mailed out every four years, with a response rate of 85–90%. Later on blood and urine samples were collected from participants . The third cohort, under NHS 3, was enrolled in 2010 and is still enrolling, with a goal of diversifying the study population to include other ethnic backgrounds .
3.2.2. Study findings
The study website (
3.2.3. Strengths and weaknesses
With focus on women, it is considered to be the longest and largest running prospective cohort study that investigates the role of lifestyle on health . Among the strengths of this study is that it included multiple assessments of the various lifestyle characteristics and exposure factors [63, 80], in turn, it also contributed to the methodology of lifestyle assessment in general, which has been used in other studies [63, 80]. Additionally, it allowed for the calculation of mortality rates . As for the weaknesses, white women dominated the original cohort, which reflects on the generalizability of the study results [4, 63].
3.3. The Caerphilly Prospective Study (CAPS)
Also known as the Caerphilly Heart Disease Study, this study was conducted in Caerphilly, South Wales, UK, and focused on ischemic heart disease (IHD) in relation to hormones, hemostatic factors, and lipids . As the study evolved, other investigations were included which looked into cognitive function, stroke and hearing problems .
The study included four phases. In the first phase, 2512 males, ages 45–59 years, were recruited in 1979. The procedures included blood tests, electrocardiogram (ECG), clinical history, lifestyle and IHD related questionnaires . The second phase ran from 1984 to 1988 and included 447 males. An audiometry test was added to the list of investigations that were included in the first phase . Phase 3 took place from 1989 to 1993 and added a cognitive function test and a bleeding time test . Phase 4 was conducted from 1993 to 1997, which included the audiometry and cognitive function tests originally included in the second and third phases, respectively . Follow up was conducted at a later stage through mail. The study has accumulated in a total of 150 studies and counting .
3.3.2. Study findings
Among the findings of the Caerphilly Prospective study; Elwood et al. showed that adopting a healthy lifestyle was associated with lower rates of chronic disease, as well as less cognitive impairment and dementia . In other findings, Mertens et al. reported an inverse association between CVD and adopting a healthy diet , while Bolton et al. reported an inverse association between mid-life lung function and arterial stiffness among men . Additional findings can be found elsewhere [85, 86, 87, 88, 89, 90, 91].
The three sections of this chapter covered the two types of cohort studies. Observational studies in general and cohort studies in specific are a good source of information when an experiment is not feasible. Prospective cohort studies provide valuable information when studying the relationship between exposure and outcome. As with any type of study, prospective cohort studies come with advantages and disadvantages that need to be taken into consideration when interpreting the results of these studies.
Conflict of interest
The author(s) declare no conflict of interest.
|ANOVA||analysis of variance|
|CAPS||Caerphilly Prospective Study|
|CHD||coronary heart disease|
|IHD||ischemic heart disease|
|MANOVA||multivariate analysis of variance|
|NHS||Nurses’ Health Study|
|NIH||National Institutes of Health|