Decision Curve Analysis
Compare a model's net benefit against treat-all and treat-none across the full range of threshold probabilities, and find where — if anywhere — the model actually wins.
- decision curve analysis
- net benefit
- clinical utility
- threshold probability
Net benefit: model
0.844
Net benefit: treat all
0.871
Net benefit: treat none
0.000
Decision curve: model vs. treat-all vs. treat-none
ModelTreat allTreat none
Where the model wins
The model's net benefit exceeds both treat-all and treat-none for pt in roughly [0.40, 0.89]. Ask whether a clinician's real threshold for this decision falls inside that window.
Net benefit is not a universal utility function
It encodes one particular exchange rate between false positives and false negatives, and ignores test cost, capacity, delay, downstream treatment efficacy, and unequal harms. It only answers whether the model beats guessing or treating everyone, given an assumed risk tolerance pt.