Fig. 2

Calibration and decision curves of the ASAP model and the updated models predicting HCC risk. Calibration curves of the predicted probabilities versus the observed probabilities of (a) the original ASAP model reported by Yang et al [8]; and (b) recalibration-in-the-large developed based on the method described by Vergouwe et al [15] in the validation cohort (n = 1012). A nonparametric calibration curve with the 95% confidence limits (CL) (red slide line with dashed lines) was created with the Loess algorithm. Observed HCC occurrence (green triangles) with the 95% CL was plotted against the average predicted probability in each group. The blue straight diagonal line serves as a reference for perfect calibration. The brown bar chart at the bottom of the figure presents the distribution of the predicted probabilities of the cases with outcomes (above the line) and those without outcomes (below the line) (“1” vs. “0”). (c) Decision curves showing the net benefit correlated with the utility of the original ASAP model reported by Yang et al [8] and the updated models (recalibration-in-the-large, recalibration, and model revision) derived using the methods described by Vergouwe et al [15]. Notably, the coefficients of the models are listed in Table 2. The logit (P) calculation formula of the ASAP model was {− 7.57711770 + 0.04666357[age] - 0.57611693[sex]+ 0.42243533[log(AFP)] + 1.10518910[log(PIVKA-II)]}; the formula of recalibration-in-the-large was {-10.82011770 + 0.04666357[age] − 0.57611693[sex] + 0.42243533[log(AFP)] + 1.10518910[log(PIVKA-II)]}; the formula of recalibration was {− 12.60892 + 0.0556229754[age] − 0.6867313806[sex] + 0.5035429134[log(AFP)] + 1.3173854072[log(PIVKA-II)]}; and the formula of the model revision was {-12.95521 + 0.06178[age] − 0.7449[sex]+ 0.55643[log(AFP)] + 1.28268[log(PIVKA-II)]}. Abbreviations: AFP, alpha-fetoprotein; CL, confidence limits; HCC, hepatocellular carcinoma; Loess: locally weighted linear regression; PIVKA-II, prothrombin induced by vitamin K absence-II