Assessing the Predictive Performance of Biomarkers in the Presence of Competing Risks Due to Transplantation

• Activity Number: 83 - Recent Methodological Developments and Applications in Statistical and Machine Learning Approaches for Predictive Modeling Using Competing Risk Data
• Type: Topic-Contributed
• Date/Time: Monday, August 9, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Section on Risk Analysis
• Author(s): Cuihong Zhang and Jing Ning and Steven Belle and Robert Squires and Jianwen Cai and Ruosha Li*
• Companies: The University of Texas Health Science Center at Houston and The University of Texas MD Anderson Cancer Center and University of Pittsburgh and University of Pittsburgh and The University of North Carolina at Chapel Hill and The University of Texas Health Science Center at Houston
• Keywords: Area under curve; Dependent Censoring; Predictive discrimination; Joint modeling; Liver failure

Abstract

In medical studies, some therapeutic decisions could lead to dependent censoring as a competing risk event for the survival outcome of interest. This is exemplified by a study of pediatric acute liver failure, where death was subject to dependent censoring due to liver transplantation. Existing methods for assessing the predictive performance of biomarkers often pose the independent censoring assumption and are not applicable in the presence of treatment-induced dependent censoring. In this work, we propose to tackle the dependence between the failure event and dependent censoring event using auxiliary information in multiple longitudinal risk factors. We propose estimators of sensitivity, specificity, and area under curve, to discern the predictive power of biomarkers for the failure event. Point estimation and inferential procedures were developed by adopting the joint modeling framework. The proposed methods performed satisfactorily in extensive simulation studies. We applied them to examine the predictive value of various biomarkers and risk scores for the death outcome in the motivating example.