|
Activity Number:
|
416
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 6, 2003 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
Section on Bayesian Stat. Sciences
|
| Abstract - #301513 |
|
Title:
|
Bayesian Degradation Models Applied to the United Kingdom Prospective Diabetes Study
|
|
Author(s):
|
Arzu Onar*+ and Karen D. S. Young and Rury Holman and Carole A. Cull and Richard Stevens
|
|
Companies:
|
University of Miami and University of Surrey and Diabetes Trials Unit and Diabetes Trials Unit and Diabetes Trials Unit
|
|
Address:
|
417J Jenkins Bldg., Coral Gables, FL, 33124,
|
|
Keywords:
|
geometric Brownian motion ; MCMC ; cross-validation ; first passage times ; prediction
|
|
Abstract:
|
Type 2 diabetes is a worldwide epidemic fuelled by increasing obesity and reduced levels of exercise and is forecast to affect 210 million people worldwide by the year 2010. The models proposed here aim to predict the length of time a treatment will keep patients' blood sugar levels under a pre-defined and clinically meaningful threshold. This research was based on the results of the landmark United Kingdom Prospective Diabetes Study, which followed 5102 patients with newly diagnosed Type 2 diabetes for a median of 10.4 (range 6 to 20) years. An earlier version of this work led to models which successfully differentiated among treatments in terms of how long they stayed effective on average given a patient's profile based on age, gender, race and body mass index as well as other clinical covariates, yet they are unable to make predictions for individual patients with any useful level of precision due to excessive variability. The work proposed here presents a modification which allows more precise predictions at the patient level by incorporating a patient level frailty parameter in order to account for unexplained variability which includes disease severity at diagnosis.
|
