Statistical analysis of case-control studies

Epidemiol Rev. 1994;16(1):33-50. doi: 10.1093/oxfordjournals.epirev.a036143.


Methods of analysis of results from case-control studies have evolved considerably since the 1950s. These methods have helped to improve the validity of the conclusions drawn from case-control research and have helped to ensure that the available data are utilized to their fullest extent. Logistic regression modeling, in its various forms, has become by far the most frequently applied method for multivariable analysis of case-control studies. As with any type of statistical modeling, the appropriateness of its formulation can be verified only partially through examination of the data themselves, and cautious interpretation has been urged (80-83). In this article, I have concentrated on methods that are extremely well suited to the evaluation of fairly specific etiologic issues, where one or two particular exposures are designated as being of a priori interest. In situations where a large number of associations are examined for possible case-control differences, additional complexities arise. Several authors have argued strongly against statistical adjustment for "multiple comparisons" in such situations (6, 7, 84). However, recent work suggests that, when background information is limited, certain forms of multiple-comparison procedures can be useful, specifically within a decision-analysis framework (85-87). Further methodological work relevant to the analysis of case-control studies is needed in at least two important areas. First, as discussed above, we need additional methods for conducting analyses that take appropriate account of the considerable error to which measurements in case-control studies are subject. Only with such methods available can estimates from case-control studies be confidently employed for elucidating pathogenesis, for developing policy, and for individual decision-making. Second, there has been a renewed effort in recent years to clarify the nature of causal effects and to relate these to the typically calculated epidemiologic parameters (88-92). As this work develops further, it is likely that the analysis of case-control studies will be enriched.

Publication types

  • Review

MeSH terms

  • Bias
  • Case-Control Studies*
  • Data Interpretation, Statistical*
  • Humans