The article discusses a method called the minmax combination method to evaluate the performance of discrimination models with two or three categories. This method aims to estimate various metrics like AUC, partial AUC, VUS, and partial VUS by maximizing sensitivity and specificity using the combination of maximum and minimum values of variables. The researchers define and explain these metrics using cutoff probability density functions and apply them to different normal distributions to calculate specific metrics like two-way partial AUC and three-way partial VUS. They also discuss the practical utility of these metrics using real-world data.