Combining Biomarkers for Risk Prediction Using Approximated Rank Correlation Statistic with Censored Survival Data

• Activity Number: 160 - SPEED: Biometrics
• Type: Contributed
• Date/Time: Monday, July 31, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Eisuke Inoue*
• Companies: Medical Informatics, St. Marianna University School of Medicine
• Keywords: risk prediction ; classification ; maximum rank correlation estimator ; sigmoid function approximation ; survival data

Abstract

Statistical tools for risk prediction and disease classification are useful and are widely used in daily clinical practice. Multiple biomarkers are used for such tools, and studies on how to combine biomarkers have been conducted. For data with survival outcome, Cai & Cheng (2008) proposed a method based on the rank correlation statistic as objective function. In this method, only a few biomarkers can be handled since the objective function incorporates a non-smooth indicator function which is difficult to optimize. Given that most of the tools in the biomedical area are formed by three or more biomarkers, it can be regarded as a limitation of this method. In order to overcome this, we consider to approximate the indicator function with the smooth sigmoid function so that the Newton method can be applied, as with the research of Ma & Huang (2007) on binary endpoint. Applying this method to real data yields almost the same results as the original method. Furthermore, the proposed method is applicable to the case with three or more biomarkers.