Predictive Bayesian Analysis of the U.K. Prospective Diabetes Study Data

• Activity Number: 72
• Type: Contributed
• Date/Time: Monday, August 12, 2002 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Stat. Sciences*
• Author(s): Arzu Onar*+ and K.D.S. Young and C. Cull and Rury Holman
• Affiliation(s): University of Miami and University of Miami and University of Surrey and University of Oxford
• Keywords: Hierarchical models ; Inverse Gaussian Distribution ; Gibbs Sampler ; Metropolis Hastings Algoritm ; Degradation Analysis

Abstract

There is a worldwide epidemic of diabetes, with 221 million individuals forecast to be diabetic by the year 2010. Approximately 95% of these will have Type 2 diabetes. The landmark UK Prospective Diabetes Study (UKPDS) is a randomized trial of different antidiabetic therapies, which followed 5102 patients with newly-diagnosed Type 2 diabetes for a median of 10.4 years. The aim of this paper is to model the rate of deterioration of HbA1c (an overall measure of glycaemic exposure) levels and explore the influence of certain covariates on the rate of degradation. The analysis is based on a multi-level Bayesian hierarchical model, where HbA1c is assumed to follow a geometric Brownian motion and, thus, the incremental changes in the log-transformed HbA1c levels are modeled via a Gaussian process with a positive drift, leading to predicted average first-passage times over a pre-determined threshold for individual patients. Such predictions, obtained via an MCMC approach, are useful in managing patient care and for calculating the health economic implications of different treatment strategies, e.g., insurance costs, as well as in aiding physicians in assigning treatments.