The Field Guide · No. 46
Cohort vs. case-control studies: forward from exposure, or back from disease
A cohort study starts with people grouped by an exposure, such as smoking, and counts who develops a disease; a case-control study starts with people who already have the disease and compares their past exposures with similar people who do not.
Updated
Both designs are observational: nobody assigns anyone to smoke, eat a diet or take a drug. The difference is where the study starts. A cohort study starts with the exposure. It takes people who are alike in many ways but differ by one characteristic, the National Cancer Institute's example being female nurses who smoke and those who do not, and compares how often an outcome such as lung cancer turns up in each group. A case-control study starts with the outcome. It gathers people who have the disease (the cases) and a similar group who do not (the controls), then looks back through their histories for the exposures that differ.
That starting point decides what each design is good at. Because a case-control study picks its people by outcome, it suits rare diseases and diseases that take decades to appear: nobody has to follow tens of thousands of people to collect a few hundred cases. It is also quicker and cheaper, and it can check many possible exposures against one disease. A cohort study picks people by exposure, so it suits rare exposures, can track several outcomes at once, and records the order of events, exposure first and disease later.
Each has its own weak spot. A case-control study usually learns about past exposure by asking, and people who are ill may remember their history differently from people who are well, which is recall bias. Finding a fair control group is hard. And because the researchers fixed how many cases and controls to take, the study cannot say how often the disease occurs in exposed and unexposed people, which is why its results come as odds ratios rather than risks. A prospective cohort can take years, cost a lot and lose people along the way; one review of the method gives a rule of thumb that losses should not exceed 20 percent of the sample.
When a headline says people with a disease were more likely to have done something, that is usually a case-control study, and the first question is how the exposure was measured. When it says people who did something were later more likely to get a disease, that is a cohort study, and the question is what else differed between the groups. Neither design can rule out confounding the way a randomized trial can. Smoking and lung cancer show how far the two can get when they point the same way.
Side by side
The table sets the two designs against each other on the questions that decide how to read them.
| Question | Cohort study | Case-control study |
|---|---|---|
| Where it starts | With an exposure, such as smoking or not smoking | With an outcome: people who have the disease and similar people who do not |
| Direction in time | Forward from exposure to outcome, or through records that already exist | Backward, from the disease to past exposures |
| Best suited to | Rare exposures, and several outcomes from one exposure | Rare diseases, and diseases with a long delay before they appear |
| What it can estimate | How often the disease occurs in each group, so incidence and relative risk | Odds ratios; not the rate of disease in exposed and unexposed people |
| Typical weak spots | Cost, years of follow-up, people lost along the way | Recall bias, and choosing a fair control group |
Neither design wins outright. For questions about whether a treatment helps, the Oxford Centre for Evidence-Based Medicine ranks a non-randomized controlled cohort study one level above case-control studies. For a rare disease, a case-control study may be the only design that can collect enough cases to say anything.
Prospective and retrospective cohorts
The National Cancer Institute separates two kinds of cohort. A prospective cohort study follows groups over time and compares them for an outcome. A retrospective cohort study, also called a historic cohort study, compares the medical records of such groups instead, so the follow-up has already happened by the time anyone looks. Song and Chung call the retrospective version timely and inexpensive, and list its costs: less control over what was recorded, and exposure to recall and information bias.
The word "retrospective" causes confusion. The same dictionary lists "retrospective study" as another name for a case-control study, while a retrospective cohort still groups people by exposure. When a paper says retrospective, check which design it means.
Smoking: one question, both designs
In 1950 Richard Doll and Austin Bradford Hill reported a case-control study from twenty hospitals in the London region. Hospital almoners interviewed 649 men and 60 women with lung cancer, along with a control group of non-cancer patients of the same sex and age. Among the men with lung cancer, 0.3 percent were non-smokers. Among the male controls, 4.2 percent were. For women the figures were 31.7 and 53.3 percent.
The next step ran forward in time. From 1951, Doll and colleagues recorded the smoking habits of 34,439 male British doctors and monitored their causes of death for 50 years, a prospective cohort. Men born between 1900 and 1930 who smoked only cigarettes and kept smoking died on average about 10 years younger than lifelong non-smokers. Among men born around 1920, smoking from early adulthood tripled death rates at each age; stopping at 50 halved that hazard, and stopping at 30 avoided almost all of it.
The case-control study found the link fast, from patients already in hospital. The cohort showed the order of events, smoking first and death later, and measured the years lost.
What to remember
- A cohort study starts with an exposure and follows people to see who develops a disease; a case-control study starts with the disease and looks back at exposures.
- Case-control studies suit rare diseases and long delays; cohort studies suit rare exposures and can measure how often a disease occurs.
- Case-control results come as odds ratios and are prone to recall bias; cohorts cost time and lose participants, and neither design rules out confounding.
From the record
Case-control studies are well suited to investigate rare outcomes or outcomes with a long latency period because subjects are selected from the outset by their outcome status.
Asked often
What is the main difference between a cohort study and a case-control study?
The starting point. A cohort study begins with people who differ by an exposure, such as smokers and non-smokers, and follows them to compare outcomes. A case-control study begins with people who have a disease and similar people who do not, and looks back at what each group was exposed to.
Why do case-control studies report odds ratios instead of relative risk?
Because the researchers decide how many cases and controls to include, the study cannot measure how often the disease occurs among exposed and unexposed people. It compares the odds of exposure between the two groups instead, and the National Cancer Institute notes that odds ratios are most often used in case-control studies.
Is a retrospective study the same as a case-control study?
Not always. The National Cancer Institute's dictionary lists retrospective study as another name for a case-control study, but a retrospective cohort study is a different design: it compares existing medical records of groups defined by an exposure. Check which one a paper means.
Which is stronger, a cohort study or a case-control study?
For questions about whether a treatment helps, the Oxford Centre for Evidence-Based Medicine ranks a non-randomized controlled cohort study a level above case-control studies, because a cohort records exposure before outcome. For a rare disease, though, a case-control study may be the only practical way to gather enough cases.
Read the news better, every morning.
We send you the day's real progress in science, medicine and beyond, told plainly and sourced. Free, forever.
Free for good. One click to leave.