A1274
Title: Semiparametric brier score framework for evaluating prediction accuracy under interval censoring
Authors: Yi-Kuan Tseng - National Central University (Taiwan) [presenting]
Tzu-Ling Wang - National Tsing Hua University (Taiwan)
Abstract: A semiparametric framework is proposed to evaluate prediction accuracy for interval-censored survival data using the Brier score. Brier score estimators are first developed at fixed prediction horizons, accommodating a broad class of survival models, including Cox proportional hazards, accelerated failure time (AFT), and transformation models. To enhance clinical interpretability, the Murphy decomposition is extended to interval-censored settings, partitioning the prediction error into three distinct components: miscalibration, discrimination, and intrinsic uncertainty. Furthermore, the methodology is generalized to time-dependent Brier scores, facilitating dynamic performance assessment. This extension incorporates joint modeling and imputation strategies to address challenges such as measurement errors, sparse inspection schedules, and incomplete covariate histories. The finite-sample performance is validated through extensive simulations and applied to the Taiwan AIDS cohort data to demonstrate its practical utility. Finally, user-friendly software is developed to facilitate the adoption of these robust tools in modern survival analysis.
