Methods for estimating the optimal number and location of cut points in multivariate survival analysis: a statistical solution to the controversial effect of BMI

In clinical practice, researchers usually categorize continuous variables for risk assessment. Many algorithms have been developed to find one optimal cut point to group variables into two halves; however, there is often need to determine the optimal number of cut points and their locations at the same time. In this paper we proposed a new AIC criterion, where the AIC values were corrected with cross-validation and Monte Carlo method, to select the optimal number of cut points. In addition, the cross-validation and Monte Carlo methods were used to correct the p value and relative risk. To provide the biomedical researchers with an easy tool, we developed an R function that utilized the genetic algorithm to find the location of the optimal cut points. Furthermore, we conducted simulation experiments to study the performance of our proposed method. In the end we applied our method to study the effect of body mass index on cervical cancer survival, which had inconsistent reports in the literature.