On the Proportional Likelihood Ratio Model for Sparse Correlated Data

• Activity Number: 277 - SPEED: Biometrics and Environmental Statistics Part 1
• Type: Contributed
• Date/Time: Tuesday, August 9, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Jiayi Tong* and Yang Ning and Md Nazmul Islam and Annie Qu and Yong Chen
• Companies: University of Pennsylvania and Cornell University and UnitedHealth Group and UC Irvine and University of Pennsylvania
• Keywords: Composite likelihood; Correlated data; Proportional likelihood ratio model; Sparse data

Abstract

Sparse data are often encountered in many biomedical studies, such as in matched case-control design and familial aggregation analysis. In this paper, we extend the semiparametric proportional likelihood ratio model in Luo and Tsai (2012) to sparse independent data incorporating stratum-specific baseline density functions. In this case, the maximum likelihood estimator is inconsistent as it involves estimating many stratum-specific density functions. To circumvent this problem, we construct weighted pseudolikelihood by a conditioning procedure which eliminates the stratum-specific density functions. We further extend the model and inferential procedure to sparse correlated data. The optimal weights in the pseudolikelihood to retain the maximum statistical efficiency are derived for both sparse independent data and sparse correlated data. The performance of the proposed method is evaluated through simulation studies and a real data example of claims data from the UnitedHealth Group (UHG) Clinical Discovery Database.