Multimarker model for non-invasive assessment of variceal bleeding in chronic hepatitis C cirrhotic patients

Table 17 Diagnostic performance of the variceal risk score for predicting high-grade varices

Parameter2	Apparent performance (95%CI)1	Bootstrap-corrected (1000 samples) (95%CI)1
AUC	0.94 (0.89-0.97)	0.92 (0.86-0.96)
Cut-off value	> 0.45
Sensitivity (%)	91.4 (81.5-96.6)	89.8 (79.5-95.2)
Specificity (%)	88.9 (78.4-95.4)	87.2 (76.8-93.8)
PPV (%)	87.5	85.9
NPV (%)	92.3	90.7

¹The “Apparent performance” column shows the model’s diagnostic accuracy on the original dataset. The “Bootstrap-corrected” column shows performance after internal validation using 1000 bootstrap samples, which corrects for overoptimism and estimates the model’s expected performance on new data. Area under the curve, sensitivity, and specificity are presented with 95% confidence intervals.

²The variceal risk score (VRS) was derived from a binary logistic regression model incorporating free triiodothyronine (FT3), platelet count (PLT), international normalized ratio (INR), and portal vein diameter (PVD): VRS = (-1.15 × FT3) + (-0.018 × PLT) + (1.92 × INR) + (0.28 × PVD). The cutoff (> 0.45) was selected using Youden's index. Bootstrap validation served as an internal sensitivity analysis, confirming the score's stability against sampling variability.

AUC: Area under the curve; PPV: Positive predictive value; NPV: Negative predictive value.