OptSurvCutR: Validated Cut-point Selection for Survival Analysis

The stratification of subjects based on continuous predictors is a common yet challenging task in time-to-event analysis, particularly when relationships are non-linear and require multiple thresholds. Arbitrary cut-point selection inflates Type I error rates and produces biased effect estimates. This manuscript presents OptSurvCutR (Optimal Survival Cut-Points in R), addressing this challenge through three key logical processes: (1) data-driven determination of the optimal number of cut-points using information criteria; (2) simultaneous identification of multiple thresholds via a genetic algorithm; and (3) integrated bootstrap validation to assess cut-point stability and control false discovery risk. First, the find_cutpoint_number() function determines the optimal model complexity by comparing metrics such as AIC, AICc, or BIC, while optionally adjusting for covariates. Second, find_cutpoint() identifies the precise threshold locations by optimising survival metrics (e.g., log-rank statistic, hazard ratio) using either an exhaustive systematic search or an efficient genetic algorithm. Finally, validate_cutpoint() assesses the robustness of identified thresholds by generating 95% confidence intervals through bootstrap resampling. We demonstrate the packages complete workflow using two case studies: a plant science example modelling the non-monotonic effect of temperature on rapeseed germination, and a clinical bioinformatics analysis stratifying colorectal cancer patients by a microbial biomarker after adjusting for clinical covariates. These examples illustrate how OptSurvCutR uncovers complex survival patterns often missed by traditional dichotomisation. The package provides a transparent, extensible framework for validated cut-point selection, making it broadly applicable for researchers working with time-to-event data. Source code and documentation are freely available at https://github.com/paytonyau/OptSurvCutR.