Finite Mixture Clustering of Risk Behaviors for an Infectious Disease

• Activity Number: 8 - Machine Learning Methods and Applications: Making an Impact in Biomedical Research
• Type: Invited
• Date/Time: Sunday, July 28, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Joseph Kang*
• Companies: Centers for Disease Control and Prevention (CDC)
• Keywords: Finite mixutre clustering; Latent class analysis; Estimating equation; EM algorithm; Infectious disease; NHANES

Abstract

Health behaviors associated with sexually transmitted diseases often occur in a correlated and multi-dimensional pattern. Such a pattern is complex and hence requires clustering methods. Finite mixture clustering is a versatile data mining method in the sense that any distributional forms can be taken. Among finite mixture clustering methods, the latent class analysis (LCA) has been effectively used in health science. Despite the popularity of the LCA, however, it remains challenging to associate the cluster membership variable with other variables due to the uncertainty of clusters identified by the LCA. This presentation will discuss a statistical approach regarding how to effectively associate LCA indicators with other covariates. Our approach is based on the expected estimating equation (EEE) framework. Viewing the cluster indicator as a missing variable, the EEE will take the mathematical expectation of the cluster indicator in estimation procedures. Our analysis investigates the association of sexual risk behavior clusters with herpes simplex 2 (HSV-2), which is a common sexually transmitted disease (STD).