AutoScore-Survival

AutoScore-Survival generates interpretable, integer-valued scoring systems for time-to-event (survival) outcomes by selecting predictors with random survival forests and weighting them using Cox proportional hazards regression for clinical risk stratification.

Key Features:

• Machine Learning Integration: Employs random survival forests for variable selection from right-censored survival data.
• Cox Regression for Score Weighting: Uses Cox proportional hazards regression to assign weights to selected predictors.
• Parsimonious Scoring System: Produces more parsimonious scoring systems compared with penalized likelihood approaches and stepwise selection while maintaining performance.
• Integer-valued Scores: Generates integer-valued scores to support interpretability and application in clinical risk assessment.
• Systematic Methodology: Provides a guideline-driven, reproducible framework for developing time-to-event scores from baseline covariates.

Scientific Applications:

• ICU 90-day survival prediction: Applied to predict 90-day survival outcomes for intensive care unit patients.
• Model comparison and benchmarking: Compared with random survival forests and traditional clinical scores, achieving an integrated area under the curve of 0.782 (95% CI: 0.767–0.794).

Methodology:

Automates score generation from time-to-event data and baseline covariates via dataset splitting, variable selection using random survival forests, and score derivation using Cox proportional hazards regression.