Decision Curve Analysis (DCA)

Decision curve analysis evaluates the clinical usefulness of a prediction model by plotting its net benefit across a range of threshold probabilities and comparing it with treating all or no patients.

의사결정곡선분석은 예측모형의 순편익을 여러 임계확률에 걸쳐 그려 모두 치료하거나 아무도 치료하지 않는 전략과 비교함으로써 모형의 임상적 유용성을 평가합니다.

How do I show whether a prediction model would actually improve clinical decisions?

예측모형이 실제로 임상적 의사결정을 개선하는지 어떻게 보여주나요?

Decision curve analysis assesses whether using a risk prediction model to guide treatment does more good than harm by calculating net benefit over a range of threshold probabilities at which a clinician would act. It complements discrimination and calibration measures and fits when comparing a model, or several models, with the default strategies of treating everyone or no one.

When to use Decision Curve Analysis (DCA)

• Binary outcome with predicted risks from one or more models
• Deciding whether a model or marker should be used to guide treatment
• Comparing competing prediction models on clinical consequences rather than accuracy alone

When to use something else

• Only discrimination is of interest: use ROC curve analysis and the AUC
• Checking agreement between predicted and observed risks: use a calibration plot

Assumptions and requirements

• Binary outcome with predicted probabilities for each patient
• Reasonably calibrated predicted probabilities
• Range of threshold probabilities that is clinically meaningful
• Validation data, ideally separate from model development data

What the analysis reports

• Decision curve of net benefit by threshold probability
• Treat-all and treat-none reference curves
• Range of thresholds where the model adds net benefit
• Net reduction in interventions per 100 patients

How to report it

Decision curve analysis showed that the model provided greater net benefit than the treat-all and treat-none strategies across threshold probabilities from [value] to [value].

Related methods

• ROC Curve Analysis
• Calibration Plot
• Net Reclassification Improvement

Run Decision Curve Analysis (DCA) on Quorant