Regression Analysis of Multiple-Source Longitudinal Outcomes: A "Stirling County" Depression Study

Epidemiologic studies of psychiatric disorders have increasingly relied on multiple sources of information to improve the validity of diagnoses and repeated assessments over time to provide a longitudinal perspective. In this paper, the authors present a general multivariate logistic regression method for the simultaneous analysis of discrete outcomes that exhibit such features. This approach permits risk factor and agreement analyses within a unified framework and appropriately uses data from subjects who may be missing some outcomes. The authors use this approach to analyze data from a "Stirling County" study of depression. During a 3- to 4-year period in the early 1990s, 631 subjects were assessed in two separate interviews, on each occasion with two diagnostic schedules (the DePression and AnXiety schedule (DPAX) and the Diagnostic Interview Schedule (DIS)). The female:male ratio of depression was found to be different for the DPAX and the DIS (0.8 and 1.6, respectively). Education was inversely associated with depression, while the effects of time, the subject's age, and the interviewer's sex were essentially null. With respect to the outcomes' association, agreement between the DPAX and the DIS was low. In addition, stability of the DPAX over time was significantly higher than that of the DIS. No covariates were found to affect significantly the association between outcomes.