Decision curve analysis. The decision curves display the net benefit of using a risk prediction model for decision, weighing the expected benefit of a possible intervention or treatment for patients correctly identified as high risk with the costs or harms associated with a false-positive decision. The relative weights of benefits and costs (cost:benefit ratio) are also reflected by the decision threshold on the probability scale. The dashed blue lines correspond to using the clinical model only and are identical on the 4 panels. The decision curves represent the standardized net benefit of using a model (dashed blue or plain red lines) relative to a default strategy of changing management for no patient (“None,” black line), or all patients (“All,” gray line). The horizontal line (“None”) represents the default strategy of changing management for no patients (net benefit is 0 at all thresholds because there is no change in patient management). The line for changing management for all patients (black line) (ie, considering all patients at “positive”) starts at 1 at a threshold of 0 because false positives are given no weight relative to true positives. Consequently, falsely tagging an individual as positive has no negative consequence. When the risk threshold (corresponding to a given cost:benefit ratio) increases, false positives are given more weight comparatively to true positives, so that changing management for all patients leads to a decreasing net benefit. The dashed blue or plain red lines, corresponding to using models to predict risk and changing management for those predicted at higher risk, allows us to assess the evidence in favor of using the risk-prediction model over the “None” strategy. The thin lines present bootstrap 95% confidence intervals.