Imputation Method Based on Sliding Window for Right-Censored Data

Abstract

Censored data arise in almost all important statistical analyses. For example, in patient-based studies, biostatistics data often subject to right censoring due to the detection limits, or to incomplete data. In the context of regression analysis, improper handling of these problems may lead to biased parameter estimates. Recently, imputation techniques are popularly used to impute censoring observations and the data are then analyzed through techniques that can handle censoring. In this sense, we introduce a new imputation strategy called sliding window method based on predictive model imputation (SWPM). In the present study, to assess the success of the proposed imputation method, the classical predictive model (PM) is used as a benchmark method. Hence, we compared two imputation methods for evaluating the right-censored data. The focus here is to assess and analyze through simulation and real data studies the performances of our imputation techniques based on different censoring levels and sample sizes.