This paper continues the series on the observational study designs, focusing on the cohort design. The word ‘cohort’ was adopted from the Roman term of 300 to 600 fighting soldiers who march together (Hood, 2009; Hulley, 2013). The epidemiology community-initiated using ‘cohort’ during the 1930s to mean a “designated group which are followed or traced over a period of time “(Hood, 2009, p. E2). The term is currently defined as a group of people with pre-defined common characteristic(s) (i.e., smokers, exposure to lead in drinking water, ICU nurses) followed longitudinally with periodic measurements to determine the incidence of specific health outcomes or events (Alexander, 2015; Hulley, 2013; Song & Chung, 2010). Since cohort studies are observational, study participants are monitored, and study interventions are not provided. This paper describes the prospective and retrospective cohort designs, examines the strengths and weaknesses, and discusses methods to report the results.
The cohort study design is an excellent method to understand an outcome or the natural history of a disease or condition in an identified study population (Mann, 2012; Song & Chung, 2010). Since participants do not have the outcome or disease at study entry, the temporal causality between exposure and outcome(s) can be assessed using this design (Hulley, 2013; Song & Chung, 2010). A vital feature of a cohort study is selecting the study participants based on mutual characteristics such as geographic location, birth year, or occupation (Song & Chung, 2010). Cohorts are also selected based on exposure and non-exposure status (Setia, 2016). Ideally, both groups are similar except for the exposure status. Additionally, the cohort can be divided based on exposure categories at study entry.
For example, an investigator could recruit people living with HIV (PLWH) who smoke and do not smoke (never smoked) from the same community and follow them over five years to determine the relationship between smoking status and HIV and the incidence of heart disease and stroke in this population. Alternatively, at study entry, the smokers could be categorized based on the smoking pack-years (less than five pack-years or greater than five pack-years) to determine whether heart disease and stroke are associated with the amount and duration of smoking.
The prospective cohort studies are also referred to as longitudinal studies. It is used to answer a specific question(s) in a selected area. Investigators recruit a sample of participants and follow them over time, from the present to the future. At pre-determined time-points, characteristics are measured (using interviews, questionnaires, biological assays, physiologic measures) to understand the relationship between the cohort and study outcome. See figure 1 .
Prospective and Retrospective Cohort Designs
During the recruitment phase, the investigator must identify potential participants who plan to move and difficult to reach during the study’s follow-up phase. The eligibility criteria should reflect this consideration. The investigator should collect contact information from the enrolled participants, telephone, email address, mailing address, and at least two friends or family members the investigator can contact if they move or die during the follow-up phase (Hulley, 2013). Additionally, the study protocol should schedule periodic contact with the participants, such as telephone calls to provide assessment results, study newsletter, or study incentives (gift cards) to keep the participants engaged.
In continuing with the HIV study example, study participants are recruited from local New York City HIV primary care clinics. The study plans to evaluate participants annually for ten years to determine heart disease and stroke incidence. PLWH are eligible to join if they smoke cigarettes with well-controlled HIV (undetectable viral load). At study entry, individual exposures for smoking are determined (smoking pack-years), medical history and cardiovascular health are evaluated. Participants identified at baseline to have heart disease or a history of stroke are excluded from the study. Participants are categorized into two groups based on smoking exposure, less than five pack-years or greater than five pack-years for this study. The independent variables ((predictor variables) (smoking pack-years, blood pressure, weight, waist circumference, lipid levels), and the dependent variable ((outcome), history of heart disease, and stroke) are assessed annually. The longitudinal design allows investigators to compare changes over time (Fitzmaurice, 2008) and determine if the level of exposure (smoking pack-years) and other variables are associated with the outcome (incidence of heart disease and stroke).
A primary strength of the prospective cohort design is that it allows investigators to determine the number of new cases (incidence) occurring over time. From our example, the incidence of new-onset heart disease and stroke among the study participants. Additionally, measuring the predictor variables before the onset of the outcome (heart disease and stroke) strengthens the ability to assess the sequence of events and infer the causal basis of an association between the predictor variables and the outcome (Hulley, 2013).
A limitation of using this design is that it requires a large sample size. Alexander and colleagues (2015) recommend at least 100 participants. Additionally, the cost of conducting the study may be costly in terms of participant recruitment, the number of staff to conduct the research, and the collection, storage, and analysis of the outcome measurements. Moreover, some conditions (i.e., breast cancer, chronic obstructive disease), despite being relatively common, could occur at low rates in any given evaluation period and not provide meaningful results. Therefore, participants need to be followed for a longer duration, thus increasing cost and the possibility of participants withdrawing from the study or losing them during follow-ups (Hulley, 2013).
Retrospective cohort studies are also called historical cohort studies. The term historical is fitting since data analysis occurs in the present time, but the participants’ baseline measurements and follow-ups happened in the past (Hulley, 2013). This type of study is feasible if an investigator has access to a dataset that fits the research question. The dataset must also have adequate measurements about the predictor variables. See figure 1 .
Generally, the participants for a retrospective cohort design are generated for other purposes, such as electronic medical records or an administrative database like medicare (Hulley, 2013). This design’s primary goal is to review past data (predictor variables) to examine events or outcomes. Institutional review board approval is required for this design even though actual patient interactions do not occur. For example, to ascertain the incidence of heart disease and stroke among PLWH who smoke, electronic medical records of 500 HIV patients from a local HIV primary clinic are examined over ten years, 2010–2020. For this illustration, HIV patients are categorized by their smoking exposure status: smoking less than five pack-years or greater than five pack-years. The outcome of interest is the incidence of heart disease and stroke.
A strength of the retrospective cohort design is the immediate ability to analyze the outcome since it is already assembled with collected measurements and the participants’ follow-ups. This type of design is also inexpensive to conduct. A primary limitation of this study is that the available dataset may be incomplete, inaccurate, or measurements undertaken that do not match the research question (Hulley, 2013). In other words, the investigator(s) do not have control over the data collection methods and procedures.
During the scheduled evaluation periods, investigators count the incidence or the number of participants who develop the outcome of interest (i.e., heart disease and stroke). The methods to measure incidence are risks and rates (Alexander, 2015). Both terms can provide additional information about the exposure of interest (smoking, nonsmoking) by calculating the risk ratio and rate ratio (Alexander, 2015).
The term risk is also known as cumulative incidence. It is defined as the number of participants who develop the outcome of interest divided by the total population (participants from the cohort) at risk (Alexander, 2015). For instance, investigators conduct a study to evaluate the association between smoking and heart disease and stroke among PLWH who attend an HIV primary clinic in lower Manhattan. The investigators follow a total of 1000 PLWH for ten years. Among the 1000 PLWH, 500 were smokers, and 500 were nonsmokers. Participants were evaluated annually. A total of 125 heart disease cases and stroke were diagnosed in the smoking group, while 25 heart disease cases and stroke were diagnosed in the non-smoking group. All the cases of heart disease and stroke were diagnosed at the fifth year follow-up. (See Table 1 for calculations).
| Disease Heart Disease/Stroke | No Disease No Heart Disease/Stroke | Total | Total Person-Time (years) | |
|---|---|---|---|---|
| Exposed Smoker | 125 a | 375 b | 500 (a + b) | (125×5) + (375×10) = 4375 (a × 5 + ) + (b × 10 $ ) |
| Unexposed Nonsmoker | 25 c | 475 d | 500 (c + d) | (25×5) + (475×10) = 4875 (c × 5 + ) + (d × 10 $ ) |
| Total | 150 (a + c) | 859 (b + d) | 1000 (a + b + c + d) | 9250 [(a × 5 + ) + (b × 10 $ )] + [(c × 5 + ) + (d × 10 $ )] |
