Maximizing power in seroepidemiological studies through the use of the proportional odds model

Epidemiological studies of zoonotic influenza and other infectious diseases often rely upon analysis of levels of antibody titer. In most of these studies, the antibody titer data are dichotomized based on a chosen cut‐point and analyzed with a traditional binary logistic regression. However, cut‐points are often arbitrary, particularly those selected for rare diseases or for infections for which serologic assays are imperfect. Alternatively, the data can be left in the original form, as ordinal levels of antibody titer, and analyzed using the proportional odds model. We show why this approach yields superior power to detect risk factors. Additionally, we illustrate the advantages of using the proportional odds model with the analyses of zoonotic influenza antibody titer data.