Bayesian Design and Data Analysis for an Ordinal Staging Score in Patients with Familial Adenomatous Polyposis

• Activity Number: 410
• Type: Topic Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Sijin Wen*+ and Jeffrey S. Morris and Patrick M. Lynch
• Companies: West Virginia University and MD Anderson Cancer Center and MD Anderson Cancer Center
• Keywords: Bayesian Design ; Ordinal data ; Concordance ; Intra-class Correlation Coefficient ; Multiple rater model ; MCMC

Abstract

In a clinical study for patients with Familial Adenomatous Polyposis, it is a high priority to evaluate whether a range of severities of colorectal polyposis could be sub-grouped in such a way that the experts could agree on severity when a panel of endoscopists reviews a series of videos and assigns each video an ordinal staging score. We used a Bayesian multiple rater model to assess concordance of Insight Polyposis Staging System (IPSS) Score with ordinal 5-point scale categories across multiple raters. Markov chain Monte Carlo methods were utilized to estimate the posterior distributions of parameters. This method allowed us to estimate the rater variation and overall variation, and therefore to obtain a model-based intra-class correlation coefficient which measured the homogeneity within raters relative to the total variation. In order to obtain a reasonable statistical power in the trial design, we estimated the number of raters and the number of videos based on the simulation studies. The Bayesian approach used to evaluate the concordance of severities of colorectal polyposis showed that the IPSS staging system achieved acceptable agreement as to stage assignment.