Model validation in international HVPG cohort
In the HVPG cohort, the ROC of the novel CSPH risk model was shown in
Figure 2. The AUC were 0.91 (95% CI 0.86-0.95), 0.80 (95% CI
0.73-0.87), and 0.83 (95% CI 0.77-0.89) for the CSPH risk model,
ANTICIPATE model, and Baveno VII criteria, respectively (Figure 2, Table
2). In addition, the AUC of CSPH risk model for assessing CSPH is 0.86
(0.78-0.94) and 0.96 (0.91-1.00) in viral cohort and non-viral cohort,
respectively (sFigure 4). According to the NPV and PPV
>90%, the cutoff value of >0 (high-risk) and
< -0.68 (low-risk) were used to rule in and rule out CSPH,
respectively. Notably, the novel model cut off value of >0
(high-risk) with a higher PPV of 0.906 and a specificity of 0.918 rules
in 42.3% of patients with high-risk CSPH, which is higher than the
19.2% of Baveno VII criteria (Table 2). Similar performances were also
observed for ruling out of patients with low-risk (Table 2). Notably,
the novel model narrows down the grey zone to 22.5%, which is
significantly lower than 50.3%, using Baveno VII criteria (Table 2).
Model validation of decompensation incidences in the
international follow-up cohort
Overall, there were 248 (22.5%), 241 (21.9%), and 613 (55.6%)
cirrhotic patients were categorized into low-risk, medium-risk and
high-risk CSPH groups based on the novel model. Over a median follow-up
of 39.0 (25.2-55.2) months, the 3-year cumulative incidences of
decompensation among the follow-up cohort was substantially higher in
the high-risk CSPH group (15.8%) as compared to the low-risk CSPH
(1.7%) or medium-risk CSPH group (2.5%) without NSBBs treatment
(p<0.001) (Figure 3). Moreover, the 3-year and 5-year AUCs of
the CSPH risk model were higher than that of ANTICIPATE model and Baveno
VII criteria (sFigure 5).