Figure 4.

Decision curves display the relationship between Net Benefit and the risk threshold for treatment. Decision curves apply to situations where a binary decision must be made in light of a patient’s risk for an outcome. For example, on the basis of a patient’s 10-year risk of cardiovascular disease, a clinician must decide whether to prescribe statin treatment. Two default strategies are the “treat all” strategy, represented by the gray line, and the “treat none” strategy, represented by the black line. Decision curves are helpful when investigators have an idea of the appropriate risk threshold for treatment or when different end-users have different thresholds. Decision curves cannot be used to choose a risk threshold. Shown are a decision curve for a model that uses clinical variables to assign risk of the outcome (model 1) and an expanded model that additionally uses a biomarker (model 2). The threshold probability is the minimum risk threshold at which a clinician/patient informed of the costs and benefits of treatment would prefer treatment. Model 2 offers higher net benefit than model 1 for risk thresholds of 5%–30%. Below 3%, models 1 and 2 offer the same net benefit as the “treat all” strategy. Above 30%, the “treat none” strategy offers the highest net benefit. The numbers in the legend are the AUC values for each risk model.