A comparison of methods in the presence of censored cost data under different censoring mechanisms

Abstract

Several approaches have recently been proposed in order to derive an accurate estimate of mean costs given censoring. The aim of this study was to compare methods for estimating mean costs given censoring across different censoring mechanisms and censoring levels. 736 "complete" cases from the CHART study were used to form a "complete" set where the mean cost was known. This "complete" cohort was used to generate simulated data sets. The accuracy of methods was measured by comparing the difference between estimates and the "true" cost. The Uncensored cases method, Cox's PH model and the Weighted method CHU consistently gave better estimates of mean costs across different censoring mechanisms and censoring levels. Estimates of mean costs from all methods deteriorated as the censoring level increased. The Uncensored cases method, Cox's PH model and the Weighted method CHU may be appropriate methods for estimating mean costs given censoring in short-term studies.