This methodological project proposes a novel statistical strategy. The strategy integrates two components that are fundamental to longitudinal analysis of treatment effectiveness in an observational study. First, treatment status and level of psychopathology are dynamic processes, in that they change over the course of an illness. Second, there are clinical and demographic characteristics, which define, in part, the propensity of an individual to be treated. Standard data analytic strategies fail to capture the complex nature of treatment effectiveness over extended follow-up. Rosenbaum and Rubin (1983) have shown that the propensity scoring method can be used for causal inference from observational data. This proposal extends their approach to a dynamic model for analysis of longitudinal treatment effectiveness data. The procedure that is proposed here will adapt a mixed-model approach to propensity score methodology (Aim 1). It will be used to examine antidepressant treatment effectiveness in subjects who were initially identified with affective disorders and have been followed-up over 15 years as part of the NIMH Collaborative Study of the Psychobiology of Depression. This dynamic data analytic approach provides a framework for incorporating multiple observations per subject, over the repeated episodes and recoveries that typically comprise a chronic illness, into an evaluation of treatment effectiveness (Aim 2). Furthermore, incorporating the propensity for treatment in the analyses reduces the bias that is inherent in an observational study of effectiveness. The methodology that is proposed here will also be applied to archival randomized clinical trial (RCT) data sets. The propensity for study completion and the propensity for missingness will be accounted for in the evaluation of treatment effectiveness in RCTs (Aim 3). Its use could preclude the need for both endpoint and completer analyses. The performance of the proposed methodology will be evaluated and compared to standard procedures in a Simulation Study (Aim 4).
Showing the most recent 10 out of 36 publications
