Multilevel Propensity Method to Reduce Bias in Comparative Effectiveness Research

Weill Medical Coll Of Cornell Univ, New York NY

Investigators

Linked publications & trials

Abstract

DESCRIPTION (provided by applicant): Cluster randomized clinical trials (RCTs) and observational studies are each vulnerable to treatment group imbalance at baseline. This problem is seen in designs such as these that do not use patient-level randomization. Imbalance, in turn, compromises treatment group comparisons. Objective: The objective of this application is to develop and evaluate a statistical method that will reduce bias introduced by non-equivalent comparison groups in longitudinal comparative effectiveness research. Method: The innovative method, multilevel propensity score matching, will be evaluated in simulation studies and applied to comparative effectiveness research in mood disorders. This approach is in direct continuity with our work that developed and evaluated novel treatment effectiveness analyses for non- randomized studies. There are four aims of the project: Aim 1: Evaluation of Multilevel Matching in Longitudinal Cluster RCTs: Conduct a simulation study to evaluate bias reduction and signal detection with multilevel propensity score matching in longitudinal cluster RCTs where the number of clusters is small (<20). Aim 2: Application to Archival Cluster RCT Data: Use multilevel propensity score matching to reanalyze data from two completed cluster RCTs for geriatric depression, the PROSPECT and TRIAD studies. PROSPECT examined interventions to reduce suicidality and depression. TRIAD evaluated interventions for nurses to improve depression care. Aim 3: Evaluation of Multilevel Matching in Longitudinal Observational Studies: Conduct a simulation study to evaluate bias reduction and signal detection with multilevel propensity score matching in longitudinal, observational studies where the number of observations is not sufficient for quintile stratification. Aim 4: Application to Archival Longitudinal Observational Data: Use multilevel propensity score matching to examine efficacy and safety of antidepressants and mood stabilizer medications in the NIMH Collaborative Depression Study (CDS), a 31 year observational study of mood disorders. Summary: This project is designed to conduct a comprehensive evaluation of a promising statistical methodology. It is anticipated that this method will reduce bias and enhance signal detection in two designs that can contribute to comparative effectiveness research that will more accurately identify the optimal treatment for patients. PUBLIC HEALTH RELEVANCE: Reducing the burden of mental illness at the public health level requires identification of safe and effective treatments. To address this challenge, comparative effectiveness research must provide empirical evidence that guides the selection of treatment. The objective of this application is to develop an innovative statistical method that will reduce bias and, therefore, more accurately identify the optimal treatment for patients and better inform clinicians on best practices.