Methods
Calibration of a prediction model
Also called: calibration curve, Brier score, calibration slope, ECE
A calibrated model's predicted risks match observed rates: of patients given a 20% risk, about 20% have the event. Calibration matters as much as discrimination for clinical use.
Calibration asks whether predicted probabilities can be taken at face value. It is usually shown as a calibration curve plotting predicted risk against observed event rates, and summarized with the calibration slope (ideally 1), calibration-in-the-large, expected calibration error (ECE) or the integrated calibration index (ICI).
The Brier score, the mean squared difference between predicted probability and actual outcome, combines calibration and discrimination into one number; lower is better.
Poor calibration is common when a model is trained on one population and applied to another, which is why external validation and recalibration matter before any prediction is used to inform care.
Tools that use this
- Giant Aneurysm Strategy StudioPredicted good functional outcome (mRS 0-3) for clip only, bypass and clip, or bypass and parent artery occlusion in giant intracranial aneurysms.
- Aneurysm Recanalization StudioIndividualized risk of major recanalization after endovascular coiling over 12 to 48 months, with packing density what-ifs. C-index 0.83 on 3,530 aneurysms.
- MatLevel-CDSCounty-level map and decision support for ACOG levels of maternal care: drive time to Level III care, maternity care deserts and postpartum complication risk.
Related terms
Educational reference. Not medical advice and not for clinical decision-making.