Baseline characteristics
A total of 1377 cACLD patients were enrolled in the study from China, Japan, Southern Korea, Egypt, Singapore, and India between January 2009 and August 2021. The training cohort consisted of 586 patients who were consecutively included from China and Japan; the validation cohort consisted of 532 patients consecutively included from Singapore, South Korea, and Egypt; and the cross-sectional HVPG cohort comprised 259 patients from India and China. The mean ages were 55.1 (± 12.9), 53.2 (± 10.7), and 53.0 (± 10.4) years in the training, validation, and HVPG cohorts, respectively. The medium follow-up time was 38.2 (26.4-45.3) and 44.6 (25.7-68.1) months in training and validation cohorts, respectively. The mean HVPG was 12.0 (3.5) mmHg in HVPG cohort. Baseline characteristics of three cohorts were summarized (Table 1, Supplementary Table 1).
Table 1
Baseline characteristics in training and validation cohorts
Parameters
|
Training cohort (n = 586)
|
Validation cohort (n = 532)
|
Age, years
|
55.1 (12.9)
|
53.2 (10.7)
|
Male
|
327 (55.8)
|
400 (75.2)
|
Creatinine, µmol/L
|
64.4 (40.6)
|
95.3 (91.0)
|
INR
|
1.1 (0.1)
|
1.1 (0.1)
|
LSM, kPa
|
20.2 (13.9)
|
19.0 (13.0)
|
Albumin, g/L
|
41.1 (5.6)
|
41.3 (4.2)
|
Total bilirubin, µmol/L
|
20.0 (12.2)
|
17.9 (14.9)
|
ALT, U/L
|
57.3 (60.1)
|
59.4 (48.8)
|
AST, U/L
|
57.2 (58.2)
|
55.4 (35.6)
|
PLT, × 109/L
|
126.7 (52.2)
|
157.8 (71.0)
|
Varices
|
264 (45.1)
|
130 (24.4)
|
MELD
|
8.7 (2.7)
|
8.6 (2.9)
|
Etiology
|
|
|
Viral hepatitis
|
402 (68.6)
|
434 (81.6)
|
NASH
|
66 (11.2)
|
39 (7.3)
|
Alcohol
|
41 (7.0)
|
23 (4.3)
|
AIH
|
12 (2.0)
|
4 (0.7)
|
PBC
|
11(1.8)
|
5 (0.9)
|
Other
|
54 (9.2)
|
27 (5.1)
|
Follow-up, months
|
38.2 (26.4-45.3)
|
44.6 (25.7-68.1)
|
Data are presented as the means (standard deviations), median (IQR), or n (%). Abbreviations: AIH, Autoimmune Hepatitis; ALT, alanine aminotransferase; AST, aspartate aminotransferase; HVPG, hepatic venous pressure gradient; INR, international normalized ratio; IQR, interquartile range; LSM, liver stiffness measurement; NASH, non-alcoholic steatohepatitis; MELD, model of end-stage liver disease; PBC Primary Biliary Cholangitis; PLT, platelet count.
|
Univariate and multivariate Cox proportional hazard regression analyses
In the training cohort, 74.6% (437/586) of cACLD patients had unfavorable Baveno VI criteria (LSM > 20 kPa or PLT < 150 × 109/L). We performed Cox proportional hazard regression analyses to investigate the independent factors of developing into hepatic decompensation. In the univariate analyses, 13 variables were involved for the analysis (Table 2), and age, international normalized ratio (INR), MELD, LSM, albumin, total bilirubin, ALT, PLT, and varices presented the statistical significances (p < 0.05). In the multivariate analyses with forward likelihood ratio method, LSM, PLT, albumin, ALT, and varices were eventually presented the significant differences (p < 0.05), which indicated that these 5 variables were independent factors for hepatic decompensation.
Table 2
Risk factors for hepatic decompensation in training cohort
Parameters
|
Univariate analyses
|
Multivariate analyses
|
|
HR
|
p value
|
HR
|
p value
|
Age
|
1.028 (1.009 - 1.048)
|
0.005
|
-
|
-
|
BMI
|
0.996 (0.953 - 1.040)
|
0.847
|
-
|
-
|
Creatinine
|
0.998 (0.987 - 1.009)
|
0.715
|
-
|
-
|
INR
|
67.239 (15.348 - 294.578)
|
< 0.001
|
-
|
-
|
MELD
|
1.150 (1.081 - 1.224)
|
< 0.001
|
-
|
-
|
Male
|
1.087 (0.708 - 1.669)
|
0.704
|
-
|
-
|
LSM
|
1.046 (1.035 - 1.057)
|
< 0.001
|
1.036 (1.023 - 1.050)
|
< 0.001
|
Albumin
|
0.878 (0.843 - 0.914)
|
< 0.001
|
0.934 (0.893 - 0.976)
|
0.003
|
Total bilirubin
|
1.025 (1.013 - 1.038)
|
< 0.001
|
-
|
-
|
ALT
|
0.987 (0.979 - 0.995)
|
0.001
|
0.984 (0.976 - 0.992)
|
< 0.001
|
AST
|
0.995 (0.990 - 1.001)
|
0.083
|
-
|
-
|
PLT
|
0.985 (0.979 - 0.993)
|
< 0.001
|
0.987 (0.981 - 0.994)
|
< 0.001
|
Varices
|
4.161 (2.444 - 7.083)
|
< 0.001
|
1.918 (1.904 - 3.363)
|
0.023
|
Data are presented with 95% confidence interval. Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; INR, international normalized ratio; LSM, liver stiffness measurement; MELD, model of end-stage liver disease; PLT, platelet count.
|
Establishment and performances of CHESS criteria and ABC model in training cohort
These independent factors for the prediction of hepatic decompensation were included to develop the CHESS criteria as following formula: (0.036*LSM [kPa]) + (-0.013*PLT [109/L]) + (-0.068*Albumin [g/L])) + (-0.016*ALT [U/L]) + (0.651*Varices [present: 1, absent: 0]). This model showed the lower information criterion (AIC 827 and BIC 840, Supplementary Table 2), which indicated that the model was parsimonious and not over fitted. Two cutoff values were determined from the CHESS criteria according to the break-point of regression models estimation method, i.e., -4.4 and -3.1 (Figure 1). The < -4.4, -4.4 to -3.1 and > -3.1 indicated the low risk, medium risk, and high risk of developing into hepatic decompensation, with a 3 year-tAUC of 0.851 (0.800-0.901), (Supplementary Table 3). In training cohort, the risks of hepatic decompensation in low risk, medium risk, and high risk groups were 1.4%, 8.7% and 47.6%, respectively.
Taking the cACLD patients with Grade 0 as the reference, Grade 2 and Grade 3 groups presented significant higher rates of cumulative incidences of hepatic decompensation, excepting Grade 1 (Grade 0 vs Grade 1, HR = 3.1 [0.3-29.2], p = 0.311, Grade 0 vs Grade 2, HR=11.3 [1.5-86.8], p = 0.019; Grade 0 vs Grade 3, HR= 74.5 [10.3-536.6] p < 0.001; Figure 2A). Pairwise comparisons in four grades also showed significant difference in rates of hepatic decompensation (all p adjust < 0.05), except for comparison of Grade 0 and Grade 1 (p adjust = 0.294). Meanwhile, when taking the cACLD patients who met the Baveno VI criteria (Grade 0) and Grade 1 together as the reference (low risk), Grade 2 (medium risk) and Grade 3 (high risk) groups still presented significant higher rates of cumulative incidences of decompensation (low risk vs medium risk, HR = 5.6 [1.8-17.2], p = 0.003; low risk vs high risk, HR = 36.8 [13.4-101.0], p < 0.001; Figure 2B).
Additionally, ALBI-FIB-4, ALBI, LSM > 20 kPa, MELD, and PLT < 150 × 109/L had the 3-year tAUCs of 0.754 (0.688-0.820), 0.702 (0.629-0.775), 0.687 (0.626-0.745), 0.645 (0.570-0.720), 0.632 (0.594-0.671), respectively. Comparing the ABC model with above-mentioned 5 models, all the differences were statistically significant (Supplementary Table 3, Figure 3A).
Performances of the ABC model in validation cohort
In validation cohort, when taking the cACLD patients with Grade 0 as the reference, Grade 2 and Grade 3 groups presented significant higher rates of cumulative incidences of hepatic decompensation, excepting Grade 1 (Grade 0 vs Grade 1, HR = 1.1 [0.2-7.7], p = 0.935; Grade 0 vs Grade 2, HR = 7.8 [1.7-36.7], p < 0.009; Grade 0 vs Grade 3, HR = 27.9 [6.4-121.3], p < 0.001; Figure 2C). Meanwhile, when taking the cACLD patients who met the Baveno VI criteria (Grade 0) and Grade 1 together as the reference (low risk), Grade 2 (medium risk) and Grade 3 (high risk) groups still presented significant higher rates of cumulative incidences of hepatic decompensation (low risk vs medium risk, HR = 7.5 [2.3-24.9], p = 0.001; low risk vs high risk, HR = 26.8 [9.0-80.2], p < 0.001; Figure 2D).
Additionally, the 3-year tAUCs of ABC model for the prediction of hepatic decompensation was 0.843 (0.742-0.943). Comparing with ALBI-FIB-4, ALBI, LSM > 20 kPa, MELD, and PLT < 150 × 109/L that had the 3-year tAUCs of 0.720 (0.586-0.854), 0.710 (0.578-0.842), 0.718 (0.618-0.819), 0.537 (0.385-0.688), 0.665 (0.576-0.753), respectively. The tAUCs of ABC model was significantly higher than all the above-mentioned 5 models (Supplementary Table 3, Figure 3B).
Performance of the ABC model to predict CSPH in HVPG cohort
In HVPG cohort, the ABC model can be used to diagnose the CSPH with an AUC of 0.818 (Figure 4A). The AUC of ABC model was higher than above-mentioned 5 models to diagnose CSPH (Figure 4A). The ideal cut-off of ABC model to diagnose CSPH was -4.1. Notably, the positive predictive value of high risk group to diagnose CSPH was 0.930. Accordingly, the median values of HVPG in low risk, medium risk, and high risk groups was 10.0 (IQR: 4.0), 12.0 (IQR: 4.0) and 13.0 (IQR: 4.0) mmHg, respectively (Figure 4B).