The Cohort Study

In the previous parts, every study we saw was experimental, meaning the researcher decided who received which treatment. An RCT used chance and randomization to decide, and weaker studies were run with no control group or with a historical control group. From this part on we enter the other family of studies, observational studies. In this family the researcher gives no treatment to anyone and builds no group. People have received the treatment, or have been exposed to whatever, that happened to them in real life, and the researcher only records and compares them. The first and most important member of this family is the cohort study.

How a cohort is built

In a cohort study, the researcher finds people who are already known to have been exposed or not exposed to a factor (exposure), and groups them on that basis. A cohort is a group of people who share a characteristic and are followed together over time; the word is used both for the whole study population and for each of the groups within it, for example the cohort of smokers and the cohort of non-smokers. That factor can be a treatment, a drug, a vaccine, or something like smoking, which the researcher cannot give to anyone and can only observe who already has it. After grouping, all the groups are followed over time, sometimes for years or even decades, to see in which group the outcome of interest occurs more often. So a cohort is a longitudinal study, and in its classic form a prospective one, because people are followed from a certain point onward and data are collected along the way.

A famous example is the British Doctors Study. About forty thousand doctors were placed in four cohorts according to whether they were non-smokers or light, moderate or heavy smokers, and were followed for forty years, from the 1950s to the 1990s. This study became one of the main foundations for proving the link between smoking and lung cancer, and showed that the more one smokes, the greater the risk. The value of this example is that it shows a cohort that is well designed and well run can, despite being observational, reach a conclusion that becomes the foundation of an important scientific belief.

Why researchers turn to a cohort instead of an RCT

A cohort has several practical advantages over an RCT. It is usually cheaper and simpler to run. It is also more ethically defensible, because the researcher neither withholds a beneficial treatment from anyone nor gives anyone a harmful factor. No one can randomly make a group of people smoke to study the effect of smoking; for a question like that, a cohort is the only reasonable way. That is why, when the question is about the risk of a factor or the course of a disease, a cohort is usually considered the most suitable design.

Where it falls short

The main weakness of a cohort is exactly what the RCT was built to solve. Because real life built the groups, not chance, there is no guarantee that the groups are completely alike apart from the factor of interest, so that there is no bias. Smokers may also differ from non-smokers in social class, occupation, diet or occupational exposures, and each of these differences can explain part of the result. The researcher can measure the differences they know about and account for them in the analysis, but can do nothing about the differences they do not know about. Randomization solved both kinds at once; a cohort solves only the first kind, and only partially.

The second weakness concerns rare diseases. If the outcome of interest is uncommon, then to see enough of it we must follow either a very large number of people or for a very long time, and this is sometimes practically impossible.

The single-group cohort

There is another form of cohort that has only one group. Instead of comparing exposed and unexposed, the researcher selects a group of people who all entered the study from a shared starting point. This group is called an inception cohort. For example, all of them had a positive screening test, or all were diagnosed at an early stage of a disease. Then this single group is assessed repeatedly at regular intervals to see in what fraction of them the disease develops, or over what time span various outcomes occur.

This study has no comparison group, and that is not a flaw, because its question is not a question of effect. Here no one asks what effect a factor has; the question is what this disease does over time. To answer that, following one group is enough. This is the same qualification we made in the first part: comparing two groups is the rule for questions about treatment effect, not the rule for all studies.

Summary

A cohort groups people by exposure or non-exposure and follows them over time to see in which group the outcome occurs more often. For questions where an RCT is not possible or not ethical, it is the strongest tool available, but because chance did not build the groups, there always remains the possibility that the difference between the groups is due to something other than the factor of interest, and that the effect we see is not due only to the factor we have in mind.

In a cohort the direction runs from factor to outcome: first we know who smokes, then we wait years to see who gets sick. For rare diseases this wait is sometimes not possible. The next part is about a study that runs the direction in reverse: it starts from patients and goes backward to see who was exposed.