Identifying Optimal Cut-Points for Continuous Predictors to Discriminate Disease Outcomes

Variables are often dichotomized for decision making in clinical practice and appropriate management of patients requires optimizing a cut-point to discriminate disease risk. If true cut-points for one or more variables exist, the challenge is identifying them. We examine dichotomization methods to identify which methods recover a true cut-point and present evidence that maximizing odds ratio, Youden's statistic, Gini Index, chi-square statistic, relative risk and kappa statistic theoretically recover a cut-point. Simulations evaluating these statistics for recovery of a cut-point indicate that the chi-square statistic and Gini Index have the smallest bias and variability. There are limited methods for simultaneously optimizing cut-points for more than 1 variable. We propose a method for jointly dichotomizing two or more variables and conduct simulations to compare joint and marginal dichotomization for the ability to recover the cut-points. Our results show that cut-points selected jointly exhibit smaller error and similar bias relative to marginal selection.
Dr. Wolf is an Assistant Professor of Biostatistics in the Department of Public Health Sciences at the Medical University of South Carolina (MUSC). She has a PhD in biostatistics from MUSC, a Master’s degree in environmental chemistry from UNC Wilmington, and a Bachelor’s degree in chemistry and anthropology from Rice University. Her statistics research interests focus on developing statistical methods for biomarker discovery and disease prediction modeling. Her translational interest focus on development of prediction models and diagnostic tools for rheumatic diseases and on examining the impact of environmental contaminants in the food chain of human populations.

[@haoduong6565]

Dr. Wolf, thanks for the great talk! Can you please share example using Python if available?