This methodological project proposes to evaluate and apply an innovative 2-staged statistical procedure that reduces the impact of self-selection in treatment effectiveness analyses. It is a dynamic adaptation of the Rosenbaum and Rubin propensity adjustment that accounts for changes in treatment and psychopathology over the course of an illness. It is capable of adjusting for selection bias, incorporating multiple observations per subject, and comparing effectiveness of ordinal doses. In the first stage of analyses, a model of propensity for treatment examines demographic and clinical characteristics that distinguish among subjects who receive various levels of treatment (Aim 1). The propensity adjustment is then applied to examine effectiveness of medication and psychotherapeutic interventions in longitudinal, observational studies of Personality Disorders (PD), Body Dysmorphic Disorder (BDD), and Bipolar Depression (Aim 2). In addition, the methodology will be used to account for what is typically an observational aspect of randomized clinical trials (RCTs): missing data due to dropout. It will be used to adjust for the propensity for dropout in treatment effectiveness analyses of archival data from RCTs for the treatment of geriatric depression and chronic depression (Aim 3). Finally, a series of simulation studies will evaluate the performance of the proposed statistical approaches under various specifications, separately for conditions which mirror data from observational and RCT designs (Aim 4). Specifically, Type I error rates, statistical power, and bias will be estimated for each of the statistical procedures. These results will be compared with the performance of more customary approaches. In summary, this interdisciplinary project will inform methodologists about the proposed data analytic strategy and inform clinicians about effectiveness of treatments for PD, BDD, and bipolar depression.
Showing the most recent 10 out of 36 publications
