CutpointsOEHR

CutpointsOEHR implements an R package that identifies two cut-points using the optimal equal-HR (Hazard Ratio) method to discretize continuous predictors with asymmetric U-shaped relationships to log relative hazards and selects cut-points that minimize the Akaike Information Criterion (AIC) of Cox proportional hazards models.

Key Features:

• Novel Discretization Approach: Implements the optimal equal-HR method for finding cut-points when the relationship between a continuous predictor and log relative hazards is U-shaped and potentially asymmetric.
• Equal Hazard Ratios: Determines two cut-points that produce equal log relative hazard values for balanced risk categorization.
• Minimization of AIC: Optimizes cut-points to minimize the Akaike Information Criterion (AIC) of Cox proportional hazards models, improving model fit.
• Comprehensive Simulation Study: Performance was evaluated via Monte Carlo simulations varying censoring proportions, baseline hazard functions, and asymmetry levels.
• Comparative Advantage: Demonstrated superior discrimination and overall performance compared with median split, upper and lower quantiles, and minimum p-value approaches in asymmetric U-shaped scenarios.
• Real-World Application: Applied to a small cell lung cancer dataset to obtain clinically meaningful cut-points with predictive performance in Cox regression models.

Scientific Applications:

• Clinical and Epidemiological Research: Categorizes continuous variables for survival analysis in studies with right-censored outcomes and U-shaped predictor–hazard relationships.
• Cox Model Optimization: Enhances interpretability and predictive performance of Cox proportional hazards models by providing data-driven discretization for continuous predictors.
• Oncology and Risk Stratification: Supports patient classification and risk assessment in oncology studies, exemplified by application to small cell lung cancer data.

Methodology:

Identifying two cut-points that equalize log relative hazard values, optimizing those cut-points to minimize the Cox model AIC, and validating performance via Monte Carlo simulations varying censoring proportions, baseline hazard functions, and asymmetry levels.