A Bayesian Approach to Estimate the Transition Rates Between Disease Stages Using Hidden Markov Models

• Activity Number: 475
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Lola Luo*+ and Dylan Small and Jason Roy
• Companies: University of Pennsylvania and The Wharton School and University of Pennsylvania
• Keywords: Hidden Markov Model ; Bayesian ; healthcare databases ; Chronic Kidney Disease ; transition rates

Abstract

Chronic Kidney Disease (CKD) is a world-wide public health problem. It has been classified into five stages of severity, with stage 1 being the mildest and stage 5 being severe illness. Estimating the transition rates between stages is important for public health planning and understanding the effectiveness of treatments. One way to obtain data on CKD is to use large healthcare databases. Some of the advantages of using healthcare databases are affordability and the large amount of information. There are also unique challenges that arise when such data are used. For example, the large amount of data may make it difficult to estimate the models, the data may be "dirty" from measurement or data entry errors, the number of observations and durations between observations may vary drastically between and within subjects. For estimating transition rates between stages for CKD, we consider a Hidden Markov Model (HMM). We propose a multi-state HMM with a Bayesian approach that will take into account some of the above-mentioned challenges. We evaluated the methods using simulated data and applied the methods to electronic health record data from Geisinger Health System (GHS) in PA.