Pattern Discovery of Health Curves with an Ordered Probit Model and Functional PCA

• Activity Number: 551
• Type: Contributed
• Date/Time: Wednesday, August 3, 2016 : 10:30 AM to 12:20 PM
• Sponsor: Royal Statistical Society
• Author(s): Shijia Wang* and Liangliang Wang and Jason Sutherland
• Companies: Simon Fraser University and Simon Fraser University and University of British Columbia School of Population and Public Health
• Keywords: The ordered probit model ; functional principle component analysis ; Bayesian smoothing

Abstract

We target at discovering the patterns of patients' health status based on their daily healthcare locations/types and other health-related information. In the proposed ordered probit model, the response variable is ordinal healthcare locations/types, and one of the covariates is a continuous smooth function of time for the health wellness, called `health curve'. We use the Markov Chain Monte Carlo method to estimate the parameters and the health curves. Then we apply functional principal component analysis to the estimated individual health curves to discover common health patterns. Our methods are demonstrated through the application on a stroke dataset and simulation studies. Whilst this paper focuses on the method's application to a healthcare problem, the proposed model and its implementation has the potential to be applied to many application domains in which the response variable is ordinal and there is a hidden functional covariate.