Evaluates whether predictive models maintain calibration across health systems.
Computes Greenwood-Nam-D'Agostino multi-site calibration curves, site-specific Brier scores,
calibration slope/intercept, expected calibration error (ECE), and decile prediction drift.
Usage
consortium_calibration(y_true, y_prob, site_labels, n_bins = 10L)
Arguments
- y_true
Binary vector of true outcome values (0 or 1).
- y_prob
Numeric vector of predicted event probabilities (between 0 and 1).
- site_labels
Vector of site names or identifiers.
- n_bins
Number of calibration risk quantile bins (default 10).
Value
A consortium_calibration S3 object containing $site_metrics, $calibration_curves, and $decile_drift.
Details
Evaluates cross-site predictive model transportability, multi-site calibration drift,
Greenwood-Nam-D'Agostino tests, and clinical net benefit via Decision Curve Analysis.
Multi-Site Calibration and Decile Drift
