Strategy in Dichotomizing a Continuous Biomarker for Survival Data Analysis

• Activity Number: 321
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Dung-Tsa Chen*+ and Ying-Lin Hsu and Po-Yu Huang
• Companies: Moffitt Cancer Center and National Chung Hsing University, Taiwan and National Chung Hsing University, Taiwan
• Keywords: biomarker ; cox proportional hazards model ; log-rank test ; dichotomization

Abstract

In biomedical research, it is common to see a continuous biomarker dichotomized to show its graphical association with survival outcome. This practice is different from other types of outcome variables, which could be graphically represented easily to demonstrate association of a continuous biomarker with the outcome variable, such as boxplot to displaying distribution of a biomarker between two groups in a binary outcome variable and scatterplot to show correlation between a biomarker and a continuous outcome variable. In contrast, graphical presentation of a biomarker in survival data poses a unique issue because it will require dichotomization of the biomarker. Common practice of dichotomization or multiple grouping includes median-cutoff, one-third cutoff, quartile cutoff, or optimal cutoff based on the best split. Then log-rank test with the Kaplan-Meier method will be used to generate survival plot with a p value to evaluate its clinical association. However, the results are not always consistent with the ones based on Cox proportional hazards model using the continuous biomarker measurement. Here we propose a strategy to determine appropriate cutoff for dichotomization.