Online Program

Return to main conference page
Tuesday, January 7
Tue, Jan 7, 9:00 AM - 10:45 AM
Porthole
Patient-Centered Outcomes

WITHDRAWN - Using Bayesian Rasch analysis to develop a new breast cancer-specific preference instrument (307826)

Murray Dale Krahn, Toronto Health Economics and Technology Assessment (THETA) Collaborative 
Nicholas Mitsakakis, Biostatistics Research Unit, University Health Network 
*Teresa Tsui, Toronto Health Economics and Technology Assessment (THETA) Collaborative 

Keywords: Bayesian Rasch analysis, preference instrument, item reduction, estimating item parameters, breast cancer

Patient preferences in breast cancer (BC), inform clinical and policy decisions, but are unsatisfactorily derived from generic preference instruments directly or mapped from a BC-specific psychometric instrument. No BC-specific preference instrument exists. Factor analysis and Rasch analysis are used to reduce items to create new preference instruments. We analyzed data from 409 patient responses to the 75 item BC psychometric instrument EORTC QLQ C30 & BR45, forming the core of a BC-specific preference instrument. First, confirmatory factor analysis found 16 factors. Next, on items from each factor (attribute), we fit Bayesian Rasch generalized partial credit models to estimate item parameters. We performed item analysis by visualizing item category responses and information curves. We chose the most informative item per factor with high Fisher information, and distinct category responses spanning the range of the latent construct. Our new application of Bayesian Rasch analysis, combined with qualitative item selection methods, derived the core of a BC preference instrument. Bayesian methods estimate model parameters with greater precision when item responses are sparse or skewed.