A1297
Title: Nonparametric transformation models for doubly censored survival data: A Bayesian approach
Authors: Shouzheng Chen - The Hong Kong Polytechnic University (China) [presenting]
Chong Zhong - The Hong Kong Polytechnic University (Hong Kong)
Xu Zhang - South China Normal University (China)
Abstract: Doubly censored data are frequently encountered in pharmacological and epidemiological studies, where the failure time can only be observed within a certain range and is otherwise either left or right-censored, and some analysis and inference procedures have been established. Predictions of survival times are made for doubly censored data under a nonparametric transformation model where both the monotone transformation function and the model error are unspecified. The nonparametric transformation model is robust to model misspecification, leading to stable predictions in various practical settings. Bayesian inference is facilitated without identifying the model by constructing weakly informative nonparametric priors for the infinite-dimensional parameters. Considering the left and right censoring times for real-life doubly censored data are commonly fixed, for the transformation function, a pseudo-quantile I-splines prior is proposed, which places interior knots of I-spline functions at average quantiles of synthetic doubly censored data. Such a prior characterizes the major body of the transformation function well and outperforms commonly used I-splines priors with equally spaced knots. Comprehensive simulations and an application to an AIDS clinical trial demonstrate that the proposed method outperforms existing approaches.
