Analyzing and Interpreting Data

A well-planned and carefully executed analysis is essential for any epidemiologic study, even one conducted during the frenzy of a field investigation. An analysis plan, or at least table shells, should be drafted to guide the analysis. To assess the relationship between an exposure and a health outcome, measures of association should be used that are appropriate for the study design—risk ratios for cohort studies, odds ratios for case–control studies, and prevalence ratios for cross-sectional studies. Measures of public health impact can be calculated to reflect the contribution of an exposure, either harmful or beneficial, on occurrence of the outcome among a population. Although tests of statistical significance address the role of chance in an apparent exposure–outcome association, they largely have been replaced by confidence intervals that reflect the range of values of the association that are consistent with the study data. When two or more exposures seem to be associated with the outcome, or when confounding might be present, stratified analysis and logistic regression can be used to clarify the contributions of each exposure. Before accepting that an apparent association is real, consider whether chance, bias, or investigator error might account for the finding. The strength of the evidence, as well as epidemiologic judgment, should guide public health decision-making and action.