Case-Control Study Design
- Introduced by
- Jerome L. Schlesselman, Brian MacMahon, Thomas Pugh
- Year
- 1950s-1970s
- Type
- Research Design
- Family
- observational design
What does it do?
A case-control study identifies individuals with a disease or outcome (cases) and a comparison group without the outcome (controls), then measures prior exposure retrospectively.
Developed in the 1950s–1970s by epidemiologists like Schlesselman and MacMahon, case-control studies are especially efficient for rare diseases, as they sample cases enriched for the outcome, avoiding the need for enormous cohorts. They are a mainstay of clinical epidemiology, observational research, and outbreak investigations.
The intuition behind the method
Case-control studies work backward: start with people who have the disease, then ask 'What were you exposed to before you got sick?' Compare to controls who don't have the disease. If cases are much more likely to report a past exposure than controls, the exposure is likely a risk factor. The key statistic is the Odds Ratio (OR), which can be calculated from the case-control sample alone. The main challenges are recall bias (cases may remember exposures differently than controls) and selection bias (how you choose controls affects your results). For rare diseases, case-control is far more efficient than cohort: you enroll 100 cases and 100 controls rather than 100,000 disease-free people followed for years.
History
Case-control methodology emerged in the 1950s as epidemiologists sought to study rare diseases efficiently. Jerome Schlesselman's monograph (1982) formalized the statistical foundation. Case-control designs proved decisive in establishing causal links: the classic case-control study of thalidomide and birth defects (1960s) and smoking and lung cancer (Doll & Hill) exemplify the design's power to detect exposures quickly. Nested case-control studies (selecting controls from within a cohort) and matched case-control designs evolved to increase precision. Today, case-control studies remain fundamental in epidemiology, particularly for investigating outbreaks, occupational hazards, and rare adverse events.
When to use it
Use case-control studies when: (1) the disease or outcome is rare (e.g., rare cancers, birth defects), making cohort studies prohibitively expensive, (2) a long latency period exists between exposure and outcome (years to decades), (3) studying multiple exposures for a single outcome, (4) conducting outbreak or epidemic investigations (comparing exposed vs unexposed), (5) generating hypotheses from existing data (medical records, disease registries), (6) resources are limited (case-control is cheaper and faster than cohort), (7) studying past exposures where follow-up is impossible.
Limitations and considerations
- Recall bias: cases, having experienced the outcome, may recall past exposures differently than controls. If cases exaggerate or minimize exposure reporting, OR is biased.
- Selection bias: how controls are chosen critically affects OR. Controls must come from the same source population that generated cases; using hospitalized controls when cases are community-derived introduces bias.
- Temporal ambiguity: retrospective exposure measurement can obscure causality. Did exposure precede outcome, or did outcome influence exposure recall?
- Cannot calculate incidence: case-control samples are outcome-enriched, so you cannot directly estimate disease incidence or absolute risk. OR is the only estimable association measure.
- Exposure misclassification: retrospective exposure assessment (especially for distant exposures) is error-prone, biasing OR toward the null.
Source conditions do not replace independent scientific review or validation by an analysis engine.
Strengths
- Efficiency for rare outcomes: outcome-enriched sampling makes case-control far more efficient than cohort studies for rare diseases. A study of rare cancer might enroll 300 cases and 300 controls instead of 100,000 people followed for years.
- Speed and cost: retrospective data collection is faster and cheaper than prospective follow-up.
- Multiple exposures: a single case-control study can examine many exposures simultaneously, identifying new risk factors.
- Established disease registries: many diseases have registries (cancer, birth defects), providing ready access to well-characterized cases.
- Latency: case-control naturally handles long induction-latency periods between exposure and outcome diagnosis.
Applications
- Outbreak investigation: during a foodborne illness outbreak, identify cases (ill persons) and controls (those exposed but not ill), compare foods consumed, and identify the source.
- Birth defects epidemiology: enroll infants with congenital anomalies as cases, select population-based controls, ask mothers about medication use, infections, or exposures during pregnancy.
- Occupational health: workers with mesothelioma (rare outcome) are cases; controls are workers without asbestos-related disease; retrospectively assess asbestos exposure from employment records.
- Drug adverse effects: patients hospitalized with liver failure (cases) are compared to hospitalized controls (without liver failure); medication history is reviewed to identify hepatotoxic drugs.
- Cancer etiology: newly diagnosed cancer patients (cases) are frequency-matched to cancer-free controls by age/sex; dietary, smoking, and reproductive histories are collected to identify risk factors.
How it works
How it works
Define cases: individuals with the outcome of interest (e.g., diabetes, myocardial infarction, congenital anomaly). Cases can be newly diagnosed (incident) or all persons with the disease in a defined period (prevalent). Define cases clearly: diagnostic criteria, date of diagnosis, disease status confirmed by medical records. Select controls from the source population that generated the cases—controls represent exposure distribution in the population that produced the cases. Controls must be disease-free (at least for the specific outcome). Matching: controls can be matched 1:1 or 1:k to cases on age, sex, race, or other confounders; matching improves power but requires statistical adjustment (McNemar for matched pairs). Alternatively, use unmatched controls and adjust for confounders in regression. Measure prior exposure from medical records, interviews, questionnaires, or biobanked specimens. For historical exposure (distant past), use medical records or occupational histories; for recent exposure, interviews are feasible.
Formula
Identify cases (with outcome) and controls (without outcome); measure prior exposure retrospectively from medical records, interviews, or surveys; match controls to cases on age, sex, or other variables if needed.
Construct a 2×2 contingency table: rows = exposure status (yes/no), columns = case/control status. Compute Odds Ratio: OR = (a × d) / (b × c), where a = exposed cases, b = unexposed cases, c = exposed controls, d = unexposed controls. Calculate 95% CI using log scale. For matched data (n matched pairs), use McNemar's test: χ² = ((b − c)²) / (b + c), which tests whether discordant pairs are balanced. For unmatched data, use logistic regression to adjust for confounders and compute adjusted OR. Report adjusted OR with CI and p-value. Check for interaction between exposure and confounders.
Formula
Tabulate 2×2 table (exposure × outcome); compute Odds Ratio (OR) = (a×d)/(b×c); apply McNemar's test for matched data or logistic regression for unmatched data with confounders.
OR = 1: no association between exposure and outcome. OR > 1: exposure associated with increased odds of having the outcome (risk factor). OR < 1: exposure associated with decreased odds (protective factor). For example, OR = 3 means the odds of the exposure among cases are 3 times the odds among controls. When the disease is rare (prevalence <10%), OR approximates Relative Risk. When disease is common, OR overstates the effect. Confidence intervals excluding 1.0 suggest statistical significance. Causality requires more than statistical association: consider strength (OR >2–3), consistency across studies, dose-response, temporal plausibility, and biological mechanism. Assess alternative explanations: could selection bias, recall bias, or unmeasured confounding explain the observed association?
Formula
Interpret OR: OR=1 no association, OR>1 increased odds of exposure in cases, OR<1 decreased odds. In rare outcomes, OR approximates Relative Risk. Consider recall bias, selection bias, and unmeasured confounding.
Common pitfalls
- Poor control selection: controls selected from hospitalized patients (when cases are community cases) or controls with a related disease incur selection bias. Controls should be disease-free and representative of the population source.
- Ignoring matching in analysis: if cases and controls are matched on age/sex, standard logistic regression is inappropriate; use conditional logistic regression or McNemar's test for matched analysis.
- Differential recall bias: cases may recall exposures more thoroughly or distort them more than controls. Minimize by using objective records (e.g., medical charts, employment records) instead of recall surveys.
- Post-hoc selection of controls: adjusting control selection after seeing the data (e.g., selecting additional controls to balance a particular covariate) introduces bias. Define control criteria a priori.
- Prevalence-incidence bias: if prevalent cases (all disease cases in population) rather than incident cases (newly diagnosed) are enrolled, survivors of severe disease may be enriched, distorting OR.
Frequently asked questions
When is Odds Ratio a good approximation for Relative Risk?
The Odds Ratio (OR) approximates the Relative Risk (RR) when the disease is rare in the population (typically <10% prevalence). At low prevalence, the odds of disease ≈ risk of disease, so OR ≈ RR. When prevalence is >10%, OR overstates the effect relative to RR. For example, if RR = 2, the OR might be 2.5 or higher at higher prevalence. Always clarify that case-control studies yield OR, not RR. If clinical interpretation requires RR, you must convert OR using the baseline risk (prevalence of disease in the source population).
What is a nested case-control study, and why use it?
A nested case-control study selects cases and controls from within a prospective cohort. All case members are identified from the cohort as outcomes occur; controls are randomly sampled from non-cases in the cohort, often matched to cases on age and sex. Nested case-control combines cohort and case-control strengths: temporal sequence and incidence data are available (as in cohorts), yet cost is lower because only case and control subsets are measured for expensive exposures (e.g., biomarkers, genetic testing). Odds Ratio from nested case-control approximates cohort Relative Risk, even if disease is not rare.
How do I choose between matched and unmatched controls?
Matched controls (1:1 matched to cases on age, sex, etc.) increase precision and reduce confounding by matched variables. However, matching is statistically 'expensive': you must use conditional logistic regression, and you cannot estimate matched variables' independent effects. Unmatched controls allow adjustment for many confounders in logistic regression and may be more cost-effective. Matching is most useful for strong confounders (age, sex, calendar year) and when controls are hard to find. If potential confounders are numerous, consider unmatched design with regression adjustment.
How do I distinguish between selection bias and information bias in case-control studies?
Selection bias arises from how you select cases and controls. If cases and controls come from different source populations (e.g., hospitalized cases vs. community controls), the OR is biased. Information bias (misclassification) arises from errors in measuring exposure or outcome. Recall bias is a type of information bias: cases with a serious illness may recall past exposures differently than controls. Both bias OR, but in different ways. To minimize selection bias, ensure cases and controls are from the same source population. To minimize information bias, use objective records (medical charts, biomarkers) instead of recall.
Related methods
Sources
Schlesselman, J. J. (1982). Case-Control Studies: Design, Conduct, Analysis. Oxford University Press.
ISBN: 978-0195027815Copied.Rothman, K. J., Lash, T. L., & Greenland, S. (2008). Modern Epidemiology (3rd ed.). Lippincott Williams & Wilkins.
ISBN: 978-0781755657Copied.Greenland, S., & Thomas, D. C. (1990). On the need for the rare disease assumption in case-control studies. American Journal of Epidemiology, 132(2), 374–375.
Open publication referenceCopied.