A case-control study identifies people who already have an outcome and people who do not, then looks backwards at what differed.
A case-control study identifies people who already have an outcome and people who do not, then looks backwards at what differed. The direction is the opposite of a cohort study and it makes rare outcomes studiable.
To study something affecting one person in ten thousand prospectively, a cohort would need enormous numbers and years. Starting from the cases reaches the same question with hundreds of participants, which is often the difference between a study existing and not.
Controls must come from the same population that produced the cases, differing only in outcome. Where they are drawn from a convenient different source, the comparison measures the difference between populations instead.
People with an outcome search their history for explanations more thoroughly than people without one. Both report honestly and the two groups remember differently, which inflates associations in a direction that cannot be corrected after the fact.
Where exposure can be taken from records written before the outcome occurred, recall bias disappears. Studies using prescription records or occupational logs are substantially stronger than those using interviews about the past.
Because the ratio of cases to controls is chosen by the researcher, the design cannot say how common an outcome is. It gives relative measures, and converting those into absolute risk requires information from elsewhere.
Selecting cases and controls from inside an existing cohort keeps the efficiency while using exposure data collected before anyone had an outcome. This is the strongest common form of the design.
Where the controls came from, how exposure was ascertained, and whether anyone assessing exposure knew which group a participant was in.
Selecting controls at the moment each case occurs, rather than at the end, keeps the comparison aligned in time and avoids several biases at once. It is more work and it is the version that produces estimates closest to what a cohort would have found.
Where a study draws controls from two different sources and reaches the same answer, the result is substantially more robust, because the main weakness of the design is control selection. Studies doing this are uncommon and worth more weight when found.
The claims on this page about working backwards from outcome to exposure, and why that is sometimes the only option can be checked against sources outside this site: