Skip to main content
. 2026 Jun 11;13:1830474. doi: 10.3389/fmed.2026.1830474

Figure 3.

Panel A displays a calibration plot comparing predicted probability to observed frequency for five models: Decision Tree, Logistic, NB, SVM, and XGBoost, with a dashed diagonal reference line. Panel B shows a decision curve analysis graph plotting net benefit against threshold probability for Treat All, Treat None, and Logistic strategies, highlighting the highest net benefit for Logistic at lower threshold probabilities.

Model calibration and decision curve analysis in the external validation cohort. (A) Calibration plots showing agreement between predicted probabilities and observed event frequencies for the decision tree, logistic regression, naïve Bayes (NB), support vector machine (SVM), and XGBoost models. The dashed diagonal line represents perfect calibration. (B) Decision curve analysis (DCA) comparing the net benefit of the logistic regression model (blue) with the default strategies of treating all patients (red) and treating none (green) across a range of threshold probabilities.