Statistical analysis
In baseline characteristics, continuous variables with normal distribution were expressed as mean ± standard deviation (SD) or median if not normally distributed which were using the t-test, and categorical variables were analyzed using chi-square test. All tests were two-sided; P<0.05 was considered significant.
Based on the Cox proportional-hazards model, we combined with the nonlinear regression model which contains a variety of machine learning to obtain a nonlinear survival model. In the non-linear part of building RSI and CLI, we used a variety of machine learning methods to compare, such as Gradient Boosting, Support Vector Machine, Random Forest, K-Nearest Neighbor, and Neural Network. The best cut-off for dividing patients into low-risk and high-risk groups was picked using X-tile plots based on the association with patients’ survival time. The Kaplan-Meier curve analysis and the log-rank test was used to estimate the cumulative survival curves of recurrence during the follow-up period. The prognostic effect of the risk score model was observed by using the receiver operating characteristic (ROC) curve with area under curve (AUC) value and Harrell’s concordance index (C-index). All analyses were conducted using R (version 4.2.2).