Logistic regression analysis: When the odds ratio does not work - An example using intimate partner violence data

The odds ratio is one of the most common measures used to assess the relati onship between exposure to violence and adverse health outcomes, adjusting for possible confounding factors. A reason for the odds ratio's popularity is that it is relatively easy to calculate from the coefficients of a logis tic regression model. For most etiologic studies of disease, the odds ratio is a suitable estimate of risk because incidence or prevalence of disease is rare (<10%). However; health outcomes studied in violence research are o ften more prevalent (e.g., fatigue, insomnia, stomach pain, and shortness o f breath). In these cases, the odds ratio usually overestimates the strengt h of association, sometimes erroneously tripling the magnitude. Data from a study measuring the health effects of intimate partner violence are used t o illustrate the problem of incorrectly using odds ratios. Methods to calcu late relative risks and prevalence ratios from logistic regression models a re presented.