Valvular calcification in liver cirrhosis: A comparison with non-cirrhotic cardiovascular controls

Figure 4 Discriminative performance, calibration, and receiver operating characteristic analysis of prediction models for aortic valve calcification. ^aThe standard error of the area under the curve (AUC) was calculated under the non-parametric assumption; ^bStatistical significance of the AUC was assessed using an asymptotic test against the null hypothesis of an area equal to 0.5. A: Discriminative performance and calibration of models Ia and IIa for predicting aortic valve calcification (AVC). The receiver operating characteristic (ROC) curve on the left illustrates the discriminative ability of model Ia for identifying individuals with AVC, accompanied by the corresponding AUC statistics and Hosmer-Lemeshow (HL) goodness-of-fit results. The ROC curve on the right displays the performance of model IIa, an extended predictive model that includes additional covariates. AUC values with standard errors, significance levels, and 95% confidence intervals are shown directly beneath each ROC curve. Model calibration is summarized using the HL goodness-of-fit test, with nonsignificant P values indicating good agreement between predicted and observed outcomes; B: ROC curve for the fully adjusted multivariable prediction model (model IIIa) of AVC; C: Calibration plot for the fully adjusted multivariable prediction model (model IIIa) of AVC. The ROC curve shows good discrimination, with an AUC of 0.834 (standard error 0027; 95% confidence interval: 0.780-0.887; P < 0.001). The calibration plot compares observed and predicted probabilities of AVC across deciles of predicted risk, with the diagonal indicating perfect calibration. Model calibration was adequate, as indicated by the nonsignificant Hosmer-Lemeshow goodness-of-fit test (χ² = 5.509, df = 8, P = 0.702). ROC: Receiver operating characteristic.