Bayesian hierarchical change point model with variable selections in hidden Markov model setting

Principal Investigator

Name
Hong Li

Degrees
PhD

Institution
medical university of south carolina

Position Title
associate professor

Email
liho@musc.edu

About this CDAS Project

Study
PLCO (Learn more about this study)

Project ID
PLCO-759

Initial CDAS Request Approval
Apr 5, 2021

Title
Bayesian hierarchical change point model with variable selections in hidden Markov model setting

Summary
We will develop a Bayesian hierarchical change point model with variable selections in hidden Markov chain setting to identify patients in different risk groups (indolent vs. aggressive disease, disease vs. non-diseased or different stage of the disease). We plan to apply the developed approach on PLCO data. Modeling longitudinal PSA (or CA125) trajectories uses a change point model. Patients in different hidden states may have different change points, different slopes of the trajectories, or different risk factors to predict the trajectory development. Risk factors in the change point model and disease transition model could be time-dependent or time-independent. The developed model will also be able to predict personalized risk of disease development in the next few years, and identify contributing risk factors to patients in different hidden disease stage.

Aims

1. Develop Bayesian hierarchical change point model with variable selection in hidden Markov model setting. The developed model allows each patient to have a unique random intercept, random slope before the change point, random change point time, and random slope after the change point. The difference in slope before and after a change point is constrained. Risk factors in the change point model and disease transition model could be time-dependent or time-independent. The developed model will also be able to predict personalized risk of disease development in the next few years, and identify contributing risk factors to patients in different hidden disease stage.
2. Apply the proposed approach on PLCO data. The proposed model may provide a useful tool for clinicians to identify high risk patients which may provide an opportunity to treat those patients earlier and increase their chance of survival.

Collaborators

Brian Neelon, Medical university of south carolina