Bias in Case-Control Studies

Types of Bias

1. Recall Bias (Memory Bias)

• Cases (people with the disease) tend to recall past exposures more vividly and accurately than controls (healthy people), simply because they have been thinking about what may have caused their condition.
• Example: A patient who suffered a myocardial infarction is more likely to remember dietary habits, smoking, and stress events than a healthy control.
• This leads to differential misclassification of exposure - cases over-report exposures, inflating the odds ratio.
• In teratogenesis studies, mothers of children with birth defects recall medication use more often than mothers of normal children, artificially inflating the association. - Creasy & Resnik's Maternal-Fetal Medicine

2. Selection Bias

• Cases and controls may not be representative of their respective populations in the general community.
• There may be systematic differences in characteristics between those selected and those not selected.
• Particularly problematic when cases and controls are recruited from hospitals (see Berkson's bias below).
• Example: In a skin cancer study, if cases are recruited from a specialist clinic but controls come from a general practice, the two groups may differ in sun-exposure behavior independent of the disease. - Fitzpatrick's Dermatology

3. Berkson's Bias (Berkesonian Bias)

• A special subtype of selection bias, named after Dr. Joseph Berkson.
• Arises because people with different diseases have different rates of hospital admission. If both cases and controls are recruited from hospitals, the controls may be over-represented with conditions that also share the exposure of interest.
• Example: A hospital-based case-control study of lung cancer using cardiac patients as controls - cardiac patients are also more likely to be smokers, making it appear that smoking is less strongly associated with lung cancer than it actually is.
• Park's Textbook of Preventive and Social Medicine

4. Interviewer Bias (Observer Bias)

• Occurs when the interviewer knows who is a case and who is a control (i.e., is not blinded).
• The interviewer may probe cases more thoroughly for positive exposure histories, or unconsciously ask leading questions.
• A practical check: compare the average time taken to interview cases vs. controls - if cases take noticeably longer, interviewer bias may be operating.
• Park's Textbook of Preventive and Social Medicine

5. Bias Due to Confounding

• A confounding variable is associated with both the exposure and the disease, and distorts the apparent association between them.
• Example: In a case-control study of coffee and heart disease, smoking is a confounder if smokers drink more coffee and are also at higher risk of heart disease.
• Confounding is not a bias introduced by error in measurement - it is a real third-variable relationship that can be controlled.

6. Information Bias (Measurement Bias)

• A broader category that includes any systematic error in measuring exposure or outcome.
• Recall bias is a subtype.
• Can also include misclassification of disease status (diagnosis error) or exposure status.
• Creasy & Resnik's Maternal-Fetal Medicine

7. Collider Bias

• A less intuitive but important bias. Occurs when an exposure and an outcome both independently cause a third variable (the "collider"), and the study inappropriately controls for that collider.
• Example: The "obesity paradox" - in cardiac patients, obesity appears protective against mortality. This is because controlling for cardiovascular disease (a collider caused by both obesity and mortality-related factors) induces a distorted association.
• Creasy & Resnik's Maternal-Fetal Medicine

Summary Table