A Semi-Nonparametric Propensity Score Model for Clustered Observational Data

• Activity Number: 457
• Type: Invited
• Date/Time: Wednesday, August 7, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Haibo Zhou*+ and Baiming Zou and Fei Zou
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: Adaptive EM algorithm ; observational data ; semi-nonparametric propensity score

Abstract

Unlike randomized clinical trial data, the treatment allocation in observational data is not random, which could result in confounding problems and lead to biased treatment effect estimate. The propensity score (PS) methods are commonly used to adjust confounding in practice. What further complicates the PS analysis is that samples in some observational data are not independent. Even though a few PS methods based on mixed effects models consider correlated or clustered individuals, they all make a strong normality assumption on the random cluster effects. We relaxed this assumption and developed a robust and efficient semi-nonparametric propensity score (SNP-PS) regression model. Our model makes no specific distribution assumption on the random effects except the distribution function is smooth. A truncated Hermite polynomial along with the normal density is used to approximate the unknown density of the heterogeneity. We developed an adaptive EM algorithm for SNP-PS parameter estimates. We established asymptotic results for the treatment effect estimator. We evaluated the performance of the proposed method by extensive simulations and a breast cancer study.