Abstract
Optimal cut‐points selection plays a fundamental role in multi‐stage classification. The popular Youden index method primarily emphasizes overall diagnostic accuracy while overlooking imbalance among classification rates across disease stages, whereas other methods improve balance at the cost of substantially reduced overall accuracy. Such trade‐offs are undesirable in clinical decision‐making. This article aims to develop a flexible optimal cut‐points selection method that simultaneously preserves overall diagnostic accuracy and balance among stage‐specific classification rates. By introducing a tuning parameter, the proposed method integrates the arithmetic mean and the geometric mean of stage‐specific classification rates into a unified objective function. Estimation procedures for confidence intervals are developed under both parametric and nonparametric settings. Simulation studies are conducted to evaluate the performance of the proposed cut‐points selection method and the confidence interval estimation approaches. The proposed method is applied to a real pancreatic cancer dataset to illustrate its practical utility.

This publication has 44 references indexed in Scilit: