Prm259 - The Challenges of Estimating Multi-State Model Transitions in Rare Diseases: Informing an Economic Decision Model for Duchenne Muscular Dystrophy

Abstract

n¼210). k-fold cross-validation was applied, with median survival and restricted mean survival (RMS) as the outcomes. The mean prediction errors quantifying the SE were calculated. k¼5 was used in the base case. RESULTS: When the normal k-fold cross-validation was conducted where the data were randomly split, the prediction error was "unexpectedly" increasing with the number of prognostic variables included. This "unexpectedly" increasing trend was removed when we increased the sample size (n¼2100) or used k¼2. When the data were non-randomly split so that baseline characteristics were different between the split sets, the prediction error was consistently decreasing with the number of prognostic variable included.