Supplementary Material S2. XGBoost Hyperparameter Configuration

Model: XGBoost Gradient Boosting Classifier
Python package: xgboost version 3.2.0

┌─────────────────────────────────────┬──────────────┬──────────────────────────────────────────┐
│ Hyperparameter                      │ Value        │ Rationale                                │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ n_estimators                        │ 200          │ Sufficient for convergence               │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ max_depth                           │ 4            │ Moderate depth to prevent overfitting    │
│                                     │              │ given limited EPV (~8.2)                 │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ learning_rate                       │ 0.05         │ Conservative rate for stable training    │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ subsample                           │ 0.8          │ Row subsampling for regularization       │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ colsample_bytree                    │ 0.8          │ Column subsampling for regularization    │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ scale_pos_weight                    │ 4132/304 │ Class imbalance correction             │
│                                     │ ≈ 13.6  │ (CS minority vs VB majority)             │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ eval_metric                         │ logloss      │ Standard probabilistic loss              │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ random_state                        │ 42           │ Reproducibility seed                     │
├─────────────────────────────────────┼──────────────┼──────────────────────────────────────────┤
│ n_jobs                              │ 4            │ Parallel processing                      │
└─────────────────────────────────────┴──────────────┴──────────────────────────────────────────┘

Hyperparameter selection rationale:
Hyperparameters were selected based on published recommendations for clinical
prediction models with limited events per variable (EPV=8.2), prioritizing
regularization to prevent overfitting. The chosen configuration (max_depth=4,
subsample=0.8, colsample_bytree=0.8) performed adequately on the temporal
validation set, and no further grid search was deemed necessary given the
primary objective of demonstrating generalizability across time rather than
maximizing AUC.