Collection of data documenting risk factors: safeguards in conducting case-control studies

When data are analyzed, an assumption is made that they have been properly collected. This assumption is often incorrect. A review of many case-control and cohort studies shows that the standards used to avoid bias are often overlooked. Research methods in a well-designed study should: (1) predetermine the patient selection method; (2) specify the causal or protective agent at the outset; (3) provide for unbiased data collection; (4) avoid differences in patient recall of past events; (5) avoid constrained selection of cases and controls; (6) use similar diagnostic tests; (7) use the same case finding methods; (8) use the same demographic criteria; (9) use populations with similar clinical risk factors; (10) assure that the patient is not inadvertently treated with the study agent before the diagnosis is made; (11) assure that patients exposed to the study agent have an equal chance of inclusion in the groups, whether or not they are ill [1]. A prospective, randomized, double-blind study design does not guarantee that the foregoing research methods have been rigorously invoked. Even after careful adherence to correct research methods, a small sample size can lead to incorrect acceptance of the null hypothesis that the two groups are similar (type II error). All these points are illustrated with examples from the medical literature in infectious diseases.