What is a Cohort Study vs a Case-Control Study?

A cohort study and a case-control study are the two main ways to study a cause that nobody can assign to people, such as smoking. A cohort study starts with the suspected cause: it records who has the habit, then follows everyone for years and counts who gets the disease. A case-control study starts with the disease: it takes people who already have it, finds similar people who do not, and looks back at who had the habit. Both are observational studies: the researchers change nothing. The forward design gives real rates, at a price in time: the British doctors study recorded the smoking of 34,439 doctors in 1951 and followed them for 40 years to find that about half of regular smokers were killed by the habit. The backward design is fast and works for rare diseases, but it can only compare how common the habit is among the ill and the well, never how often the habit leads to the disease. Cohort starts with the habit and waits. Case-control starts with the disease and asks.

A cohort study follows people forward from a suspected cause to see who develops an outcome. A case-control study works backward from people who have the outcome to see what they were exposed to. Both are observational: nobody assigns the cause, and the researchers only record. The founding pair of studies concerned smoking. In 1948 and 1949, two researchers interviewed 649 men with lung cancer in London hospitals. For each, they picked a control: a patient of the same sex, within the same five-year age band, in the same hospital, with some other illness. That is a case-control study. Only 2 of the 649 patients had never smoked, against 27 of the 649 controls. The study could say smoking was far commoner among the ill. It could not say how often smokers got cancer, because the researchers had chosen how many patients and how many controls to include. The number such a study produces is an odds ratio, not a rate, and the smoking histories came from memory. So in 1951 the same researchers wrote to British doctors, recorded the smoking habits of the 34,439 who replied, and followed them for 40 years. That is a cohort study. Death rates in continuing smokers were about three times those of lifelong non-smokers at ages 45 to 64, and about half of regular smokers were killed by the habit. The cohort bought real rates and no reliance on memory, at the price of those 40 years. One word trips readers: the controls here are people without the disease, not a group given a dummy treatment as in an experiment's control group. And since neither design assigns anything, both can be fooled by a confounder, a hidden factor tied to the habit and to the disease. So ask of any health headline which way the study was read. Forward from the cause gives a rate. Backward from the disease gives a comparison.

A cohort study follows people forward from a suspected cause and a case-control study looks back from the outcome, and a delivery company can check its vans both ways. Forward: at New Year the depot notes which 100 of its 200 vans were serviced on schedule and which 100 were not, then counts breakdowns through December. Six of the serviced vans break down and 18 of the unserviced, so 6 in 100 against 18 in 100: a rate, after a year of waiting. Backward: in December the depot pulls the files of the 24 vans that broke down and of 24 that did not, and checks who was serviced. Six of the broken vans had been serviced against 18 of the fine ones. The same answer in an afternoon, but no rate, because the depot chose how many broken and fine vans to pull. The broken vans are the cases and the fine ones the controls, and if breakdowns were rare, the file pull would be the only way to find enough. Where the match ends: a depot has every van's file, while a case-control study depends on what people remember doing.