survival-models

What problem does it solve? Right-censored time-to-event outcomes such as disease progression, conversion, relapse, or mortality cannot be modeled as ordinary regression, and naive approaches produce biased predictions. This Skill trains censor-aware prognosis models from subject-level features while preserving proper handling of censored observations. ## Core Features & Use Cases - Multiple Survival Estimators: Supports Cox proportional hazards, Random Survival Forest, DeepSurv, and XGBoost Cox survival through a single CLI. - Censor-Aware Evaluation: Uses Harrell's concordance index with grouped cross-validation folds for multi-site cohorts, exporting risk scores, fold assignments, and metrics. - Use Case: Given a CSV of neuroimaging ROI features with follow-up days and a progression indicator, train a DeepSurv model with 5-fold grouped cross-validation to produce relative risk scores for each subject. ## Quick Start Train a Cox proportional hazards model on my prognosis.csv file using followup_days as the duration column and progressed as the event column with 5-fold cross-validation.