Heterogeneous latent variable modeling of multivariate outcomes

Thomas M. Menino Convention & Exhibition Center 

In biomedical studies, multiple outcomes are often measured on the same individuals, providing complementary views of health status. Jointly analyzing these outcomes and understanding their heterogeneity can provide more robust assessment of disease progression and help guide customized treatment decisions. In this work, we propose to model the heterogeneity in multivariate outcomes by formulating a latent individual frailty and linking it to observed covariates via quantile regression while accounting for outcome-specific variations. We develop an efficient estimation procedure based on the conditional score principle. Our modeling and estimation framework can flexibly accommodate continuous, binary, and categorical outcomes. The proposed method demonstrates favorable asymptotic properties and strong finite-sample performance in numerical studies.

Multivariate outcome

Latent variables

Quantile regression

Conditional score.

Heterogeneity 

Survey Research Methods Section