How is a cohort study defined?

Question

Basic question here, because I don't even see it mentioned in e.g. wikipedia. Does the definition of a cohort study requires strictly that groups are exposed/not exposed to something (e.g. smoking, alcohol)? It would be a restrictive definition if so. What about if workers are paid less/paid more as 'exposure' & 'tenure' (short/long) as outcome/event.

Answer 1

A cohort study is an observational study wherein each study participant is observed/measured on the dependent variable at two or more points in time. Any explanatory variable(s) may or may not also be observed/measured at each time of observation/measurement. The observation of the dependent variable across time allows measurement of it's rate of change (over time) in individual participants, and estimation of average rate of change in the target population. The analytic question of interest in a cohort study is does estimated rate of change differ across values of, or rate of change of, an explanatory variable?

Some points on nomenclature

• The dependent variable is typically referred to as the outcome in epidemiology.

• The explanatory variable is typically referred to as the exposure in epidemiology.

• A defined cohort means that the participants share some value in the dependent variable, the explanatory variable, or both. For example, an exposure defined cohort might be one where all participants have no exposure when first observed/measured; some of them will become exposed at different levels across the study. An outcome defined cohort might be one where all participants do not have the outcome when first observed/measured. Cohort studies may be defined on both dependent variable and explanatory variable.

• When the dependent variable takes only two values (0 or 1), and the cohort is defined on the dependent variable by all participants having 0 at the start of the study, the change rate is called an incidence rate.

• Observational study means that exposure and changes in exposure are not randomly assigned by the researchers in the sense that they are in randomized control trials and other experimental designs.

Answer 2

Here is an example of a cohort in the context of a website-based business (but it generalizes to many other kinds).

Imagine you sell a product online. You care about how many people convert (i.e. subscribe) to your website every day.

Say in month 1, 100 people visited your website and 50 people subscribed.

Then in month 2, 150 people visited your website and 100 people subscribed.

Then in month 3, 125 people visited your website and 50 people subscribed.

Great. Your overall conversion rate is $\frac{50+100+50}{100+150+125} = 53\%$.

But that treats people who signed up in the first month the same as people who signed up in the third month. Does that make sense to you? Maybe people in the first month have since canceled their subscription. Maybe between months 1 and months 3 you've changed your website and that impacts your conversion rate. Wouldn't you like to know that?

To answer these kinds of question, we split up our users into cohorts. One natural way (in the online business) is to split by time of visit to the site.

In this case, we'd have 3 cohorts - month 1 (conversion rate 50%) month 2 (66%) and month 3 (40%).

If I track these groups of users separately, I can get a better sense of how changes in my product affect my users. For example, maybe people in month 1 all signed up because they saw a great advertisement. Then in month 2 I removed that ad, and my month 2 cohort had a poor conversion rate.

Cohort analysis plays a key role in understanding your business.

Answer 3

Let's first look at the term cohort. Originally it has meant a group of people that were born during a specific period, in specific place and identified by period of birth. This definition made it useful to determine things like death rates as people in the cohort aged over time. Cohort is understood much more broadly though, to mean any group of people that are defined by a specific time period and place.

A cohort study is a type of longitudinal study, examining a cohort retrospectively or prospectively. It's defining features start by identifying a cohort of interest (usually lots of people), over a long period of time (usually years).

A retrospective study design would examine a cohort from the past, for example, all of the female employees that worked at Chemical Company X during the 1970s. A prospective study would examine a cohort defined in present time and followed in the future, such as everyone born in Canada this year. At the time of cohort definition, the cohort is assessed for exposure to to a risk factor (e.g., working with a mutagenic chemical) and as they are followed in time they are monitored for an outcome of interest (e.g., developing cancer). A comparison can then be made between the sub-groups that were exposed and non-exposed (e.g., cancer incidence rates).