Introduction
Acute-on-chronic liver failure (ACLF) is a serious clinical condition associated with high short-term mortality.1–3 It has been estimated that approximately 50% of ACLF patients have no identifiable triggers.1 In China, hepatitis B virus (HBV) infection has been recognized as the most common etiology for ACLF, which is mainly attributable to a relatively high incidence of HBV infection in the country.3,4 Patients with ACLF caused by HBV infection or acute-on-chronic hepatitis B liver failure (ACLF-HBV) are characterized by the sudden onset and rapid progression of ACLF and high risk of short-term death, which the treatment off ACLF-HBV challenging.5,6 Currently, liver transplantation (LT) is the only curative treatment of ACLF and ACLF-HBV patients who have failed medical treatment.6,7 Obviously, early and accurate prediction of prognosis is critical in navigating optimal treatment decisions that lead to improved survival.8
The Model for End-Stage Liver Disease (MELD) score, a well-validated standard scoring system, is used to prioritize organ allocation to patients diagnosed with end-stage liver disease in need of LT and as a prognostic tool for ACLF.9 In addition to MELD, a number of prognostic scoring systems and models, including our proposed nomogram, have been developed to predict the prognosis of ACLF patients. However, the existing prognostic models are not satisfactory and, lack predictive accuracy.
Advances in ultrasonography such as transient elastography (TE) and two-dimensional shear-wave elastography (2D-SWE) allow noninvasive liver stiffness measurement (LSM).10–16 We recently evaluated the accuracy of 2D-SWE to predict the prognosis of patients with ACLF-HBV. Our data, together with those of previous studies demonstrated that 2D-SWE was better and more suitable than TE for LSM of patients with ACLF.17–24 Subsequently, we established a prognostic model (Jin’s model) based on 2D-SWE values in combination with MELD for the prediction of the prognosis of patients with ACLF-HBV.20 Jin’s model (previously the MELD-SWE score) was shown superior to MELD alone in predicting the prognosis of ACLF-HBV patients.20 However, until now, a prognostic model combining 2D-SWE with other independent prognostic factors obtained prospectively from ACLF-HBV patients has not been reported. Intrigued by the prognostic value of 2D-SWE and based on our previous findings, this prospective study was aimed to develop and validate a novel nomogram by integrating 2D-SWE values with other independent prognostic factors. The findings of this study are expected to provide a better model for predicting the prognosis of patients with ACLF.
Methods
Patient enrollment
Patients were consecutively screened for their eligibility to participate in this study at the Third Affiliated Hospital of Sun Yat-Sen University (Guangzhou, Guangdong, China) between November 2013 and August 2016. The patient enrollment process is shown in Supplementary Figure 1. The inclusion criteria were: (1) ACLF caused by HBV infection; (2) the diagnosis of ACLF was made following the criteria the Asian Pacific Association for the Study of the Liver.8 The exclusion criteria were: (1) ACLF with causes other than HBV infection, such as hepatitis C virus (HCV) infection, alcoholic liver disease, drug-induced hepatitis, autoimmune liver disease, or Wilson’s disease; and (2) lost to follow-up. A total of 449 patients with ACLF caused by HBV infection were recruited and were randomly allocated in a 2:1 ratio to a training cohort (n=315) and a test cohort (n=134). A group of 102 ACLF patients were prospectively enrolled in a validation cohort at the Third Affiliated Hospital of Sun Yat-Sen University between September 2019 and October 2021.
Prior to enrollment, written informed consent was obtained from each patient. The study was reviewed and approved by the Human Research Ethic Committee of the Third Affiliated Hospital of Sun Yat-Sen University [no. (2020) 02-253-01]. The study protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki.
Treatment, follow-up, and prognosis
The enrolled patients received standard treatment as previously described,3 follow-up for at least 90 days, or until death or liver transplantation, starting at the time of diagnosis with ACLF. The prognosis of each patient was categorized as either a survivor, with 3 months of liver transplant-free survival or a nonsurvivor. Survival was defined as the resolution of ACLF and cessation of intensive treatment such as artificial liver support. Nonsurvival was defined as patient death or need of liver transplantation because of lack of response to intensive treatment within the 90 days of follow-up.
Measurement of liver stiffness with 2D-SWE
2D-SWE was performed on each patient after hospital admission as previously reported.20 In brief, liver stiffness was measured by ultrasonography (US) using and Aixplorer system (SuperSonic Imagine, Aix-en-Provence, France) with an SC6-1 convex probe (1–6 MHz). A 4 cm × 3 cm 2D-SWE rectangular elasticity box was placed 1–2 cm under the liver capsule in a parenchyma area free of large vessels. Two examiners with experience in performing over 500 2D-SWEs, performed the procedures and were blinded to the patients’ clinical and laboratory data. The median of five independent kPa values was used in the analysis.
Conventional US
Conventional B-mode US was performed after completion of 2D-SWE. Ten parameters were measured: liver surface, liver parenchyma, thickness of the right lobe, biliary sludge, thickness of the gallbladder wall, presence of gallbladder enlargement, spleen index (length, thickness), ascites index (maximum depth of ascites), diameter of the portal vein, and re-opening of the paraumbilical vein, as previously reported.20,25 Gallbladder enlargement was defined as a length >90 mm and/or an anterior-posterior diameter >40 mm. Biliary sludge was defined as presence of low-level echoes without posterior acoustic shadowing and a shift with position changes. The ascites index was calculated as the maximum depth of ascites. The spleen index was recorded as the product of the longitudinal and transverse diameters of the spleen.
Laboratory tests
Blood samples were collected from the patients, and laboratory testing was performed to determine alanine aminotransferase, aspartate aminotransferase, gamma-glutamyl transpeptidase, and alkaline phosphatase levels; total bilirubin, albumin, total cholesterol, serum creatinine, serum sodium, prothrombin time, prothrombin activity, international normalized ratio, white blood cell count, platelet count, and hemoglobin. Patients were also tested for markers of HBV infection, including hepatitis B e antigen, hepatitis B surface antigen, and HBV DNA viral load.
Development of a new nomogram to predict prognosis in the training cohort
2D-SWE values, conventional B-mode US, laboratory test results, and demographic and clinicopathological characteristics of the ACLF patients in the training cohort were included in univariate regression analysis to identify factors significantly associated with prognosis. Multivariate regression analysis of factors with potential prognostic value was then used in to identify independent prognostic factors. Events included in the multivariate analysis included deaths from ACLF, and underwent LT within 3 months of ACLF. Patients who survived longer than 3 months were recorded as nonevents. 2D-SWE values and the factors independently associated with the prognosis were then employed to construct the novel prognostic nomogram.
Performance evaluation and validation of the nomogram for predicting prognosis
The performance of the new model was evaluated and compared with existing prognostic prediction scoring models, including MELD, MELD combined with sodium (MELD-Na), and the previously reported Jin’s model.20 The MELD score R = 3.8 ln [bilirubin (mg/dL)] + 11.2 ln (INR) + 9.6 ln [creatinine (mg/dL)] + 6.4 (cause of cirrhosis: cholestatic or alcoholic 0, other 1). The MELD-Na score = MELD + 1.59 × [135 − Na (mmol/L)],Na = serum sodium >135 mmol/L was calculated as 135, and serum sodium <120 mmol/L was calculated as 120. Jin’s model score = MELD − SWE = 1.3 × MELD + 0.3 × 2D-SWE (kPa) (MELD-SWE score).
Decision curves of the models were plotted, and decision-curve analysis (DCA) was performed to evaluate the clinical utility and net benefit. Calibration curves were created to evaluate the agreement between the predicted and observed probability. For assessment of predictive accuracy, receiver operating curves (ROCs) were generated to calculate standard indices.
Statistical analysis
Statistical analysis was performed with IBM SPSS Statistics 25 for Windows (IBM Corp., Armonk, NY, USA), R (version 3.6.2), and MedCalc (version 19.8) statistical software. Continuous data were reported as means±standard deviation or medians and interquartile range, as appropriate. Student’s t-test or the Mann-Whitney U test was used to compare differences between groups. Categorical data were reported as numbers and percentages (%), and chi-square test or Fisher’s exact test was used to compare difference between groups. Survival probability curves were plotted by the Kaplan-Meier method and between-differences were compared using log-rank tests. Univariate and multivariate Cox proportional hazards analyses were performed to evaluate the association of each variable and clinical outcome. ROC curve characteristics and the area under the ROC curve (AUC) were used to evaluate the prediction of prognosis. DeLong’s test was used to compare differences of the AUC values of prognostic models. P-values <0.05 was considered statistically significant.
Results
Demographic and clinicopathologic characteristics of the patients
During the study period, 449 consecutive patients with ACLF caused by HBV infection were randomly assigned to training (n=315) and test (n=134) cohorts. Baseline demographic characteristics, clinicopathologic features, and laboratory testing data are summarized in Table 1. The demographic characteristics were similar, and there were no significant differences in age and sex between the training and test cohorts. Conventional US parameters and 2D-SWE values in the training and test cohorts were comparable.
Table 1Demographic characteristics, clinicopathologic features, and laboratory tests of ACLF patients in the training and test cohorts
| Training cohort (n=315) | Test cohort (n=134) | p-value |
|---|
| Sex | | | 0.824 |
| Male (n) | 275 | 118 | |
| Female (n) | 40 | 16 | |
| Age (years) | 44 | 44 | 0.781 |
| Prognosis | | | 0.084 |
| Nonsurvivor (n) | 208 | 77 | |
| Survivor (n) | 107 | 57 | |
| BMI (kg/m2) | 22.77 | 22.06 | 0.411 |
| Liver parenchyma (n) | | | 0.355 |
| Uniform | 180 | 72 | |
| Less uniform | 62 | 23 | |
| Uneven | 73 | 39 | |
| Liver capsule (n) | | | 0.528 |
| Smooth | 67 | 34 | |
| Under smooth | 87 | 35 | |
| Rough | 161 | 65 | |
| Thickness of right liver lobe (mm) | 103 | 102 | 0.167 |
| Gallbladder wall thickness (mm) | 6 | 6 | 0.080 |
| Thickening of gallbladder wall (n) | | | 0.832 |
| No | 9 | 19 | |
| Yes | 296 | 125 | |
| Gallbladder enlargement (n) | | | 0.961 |
| No | 98 | 42 | |
| Yes | 217 | 92 | |
| Gallstones (n) | | | 0.294 |
| No | 115 | 42 | |
| Yes | 200 | 92 | |
| Pleural effusion (n) | | | 0.413 |
| No | 302 | 131 | |
| Yes | 13 | 3 | |
| Ascites maximum depth (mm) | 40 | 43 | 0.589 |
| Portal vein diameter (mm) | 12 | 11 | 0.419 |
| Umbilical vein re-open (n) | | | 0.546 |
| No | 136 | 62 | |
| Yes | 179 | 72 | |
| Enlarged spleen (n) | | | 0.201 |
| No | 98 | 50 | |
| Yes | 217 | 84 | |
| Splenic portal vein (mm) | 7 | 7 | 0.617 |
| AST (U/L) | 133 | 137.5 | 0.719 |
| ALT (U/L) | 134 | 137.5 | 0.779 |
| ALB (g/L) | 33.3 | 33.2 | 0.911 |
| TB (µmol/L) | 390.8 | 423.4 | 0.082 |
| GGT (U/L) | 70 | 69 | 0.958 |
| ALP (U/L) | 135 | 135.5 | 0.659 |
| CHE (U/L) | 3,681 | 3,706.5 | 0.556 |
| Na (mmol/L) | 137 | 137 | 0.911 |
| Cr (µmol /L) | 70.6 | 72.5 | 0.556 |
| AFP (ng/ml) | 50.6 | 42.2 | 0.230 |
| WBC (109/L) | 6.55 | 6.73 | 0.238 |
| Neu (%) | 65.8 | 67.3 | 0.512 |
| RBC (109/L) | 3.66 | 3.63 | 0.959 |
| HGB (g/L) | 113 | 114.5 | 0.537 |
| PLT (109/L) | 98 | 98.5 | 0.826 |
| PT (s) | 25.1 | 26.2 | 0.055 |
| PTA (%) | 34 | 32 | 0.063 |
| INR | 2.28 | 2.41 | 0.060 |
| Hepatorenal syndrome | 18 | 8 | 0.915 |
| Follow-up time (days) | 80.0 | 76.5 | 0.393 |
| SWE (kPa) | 38.5 | 38.75 | 0.615 |
The prognosis of each patient was classified as ‘survivor’ or ‘nonsurvivor’ on the basis of the 90-day follow-up, and no significant difference was observed between the training and test cohorts.
Univariate and multivariate analysis of factors associated with the prognosis of ACLF patients in the training cohort
Univariate analysis was performed to identify factors significantly associated with prognosis. As shown in Table 2, variables including age, conventional US parameters (i.e., liver capsule, thickness of right liver lobe, gallbladder wall thickness, and gallstones), 2D-SWE measurements, and some laboratory tests were significantly associated with patient prognosis in in the training cohort. The results of multivariate logistic regression analysis [odds ratios (ORs) with 95% confidence intervals (CIs) are shown in Table 3. The variables independently associated with the prognosis of patients with ACLF caused by HBV infection were age (OR: 1.04, 95% CI 1.01–1.08)), TB OR: 1.01, 95% CI: 1.00–1.01), Neu (OR: 139.55, 95% CI: 6.41– 3,037.71), INR (OR: 2.74, 95% CI: 1.78–4.21), and SWE (OR: 1.05, 95% CI: 1.02–1.08), as shown in Table 3.
Table 2Univariate analysis of prognostic factors in the training cohort
| Variable | Odds ratio | 95% Confidence Interval (CI) | p-value |
|---|
| Sex | | | |
| Male | | | 1 |
| Female | 1.19 | (0.60–2.37) | 0.614 |
| Age* | 1.05 | (1.03–1.07) | <0.001 |
| BMI | 0.94 | (0.87–1.01) | 0.091 |
| Liver parenchyma | | | |
| Uniform | | | 1 |
| Less uniform | 1.24 | (0.68–2.26) | 0.482 |
| Uneven | 1.30 | (0.73–2.29) | 0.375 |
| Liver capsule* | | | |
| Smooth | | | 1 |
| Under smooth | 2.42 | (1.33–4.41) | 0.004 |
| Rough | 1.87 | (0.91–3.83) | 0.087 |
| Thickness of right liver lobe* | 0.95 | (0.93–0.98) | <0.001 |
| Gallbladder wall thickness* | 1.10 | (1.01–1.18) | 0.022 |
| Thickening of gallbladder wall | | | |
| No | | | 1 |
| Yes | 2.00 | (0.65–6.19) | 0.228 |
| Gallbladder enlargement | | | |
| No | | | 1 |
| Yes | 0.98 | (0.59–1.62) | 0.941 |
| Gallstones* | | | |
| No | | | 1 |
| Yes | 1.66 | (1.01–2.74) | 0.047 |
| Pleural effusion | | | |
| No | | | 1 |
| Yes | 1.41 | (0.44–4.55) | 0.568 |
| Ascites index | 1.00 | (1.00–1.01) | 0.250 |
| Portal vein diameter | 1.00 | (0.83–1.21) | 0.988 |
| Umbilical vein reopen | | | |
| No | | | 1 |
| Yes | 1.14 | (0.71–1.82) | 0.598 |
| Enlarged spleen | | | |
| No | | | 1 |
| Yes | 0.89 | (0.54–1.47) | 0.660 |
| Splenic portal vein | 1.01 | (0.86–1.19) | 0.873 |
| AST | 1.00 | (1.00–1.00) | 0.985 |
| ALT | 1.00 | (1.00–1.00) | 0.101 |
| ALB* | 0.94 | (0.89–0.99) | 0.011 |
| TB* | 1.01 | (1.00–1.01) | <0.001 |
| GGT* | 1.00 | (0.99–1.00) | 0.015 |
| ALP | 0.99 | (0.99–1.00) | 0.070 |
| CHE | 1.00 | (1.00–1.00) | 0.554 |
| Na* | 0.92 | (0.87–0.98) | 0.004 |
| Cr* | 1.01 | (1.00–1.02) | 0.017 |
| AFP* | 1.00 | (1.00–1.00) | 0.016 |
| WBC* | 1.08 | (1.01–1.15) | 0.019 |
| Neu* | 355.01 | (35.34– 3,566.28) | <0.001 |
| RBC | 0.92 | (0.69–1.21) | 0.540 |
| HGB | 0.99 | (0.98–1.00) | 0.066 |
| PLT* | 1.00 | (0.99–1.00) | 0.018 |
| PT* | 1.13 | (1.09–1.18) | <0.001 |
| PTA* | 0.92 | (0.89–0.94) | <0.001 |
| INR* | 2.82 | (2.03–3.91) | <0.001 |
| Follow-up time | 0.98 | (0.98–0.99) | <0.001 |
| SWE* | 1.03 | (1.01–1.06) | 0.005 |
Table 3Multivariate analysis of independent prognostic factors in the training cohort
| Variables | Odds Ratio | 95% Confidence Interval (CI) | p-value |
|---|
| Age* | 1.04 | (1.01–1.08) | 0.007 |
| BMI | 0.93 | (0.84–1.04) | 0.226 |
| Liver capsule | | | |
| Smooth | | | 1 |
| Under smooth | 1.84 | (0.84–4.05) | 0.129 |
| Rough | 0.86 | (0.32–2.35) | 0.776 |
| Thickness of right liver lobe | 0.99 | (0.95–1.02) | 0.465 |
| Thickening of gallbladder wall | 1.03 | (0.93–1.13) | 0.623 |
| TB* | 1.01 | (1.00–1.01) | <0.001 |
| Na | 1.04 | (0.96–1.13) | 0.294 |
| Neu* | 139.55 | (6.41– 3,037.71) | 0.002 |
| INR* | 2.74 | (1.78–4.21) | <0.001 |
| SWE* | 1.05 | (1.02–1.08) | 0.003 |
| Gallstones | | | |
| No | | | 1 |
| Yes | 1.76 | (0.91–3.42) | 0.094 |
| ALB | 0.94 | (0.87–1.02) | 0.122 |
| GGT | 1 | (0.99–1.01) | 0.848 |
| ALP | 1 | (0.99–1.01) | 0.576 |
| Cr | 1 | (0.99–1.01) | 0.757 |
| AFP | 1 | (1.00–1.00) | 0.326 |
| HGB | 1 | (0.99–1.02) | 0.606 |
| PLT | 1 | (0.99–1.00) | 0.104 |
Construction of the novel nomogram integrating 2D-SWE values and other independent prognostic factors
The five independent prognostic factors, age, TB, Neu, INR, and SWE identified by multivariate logistic regression analysis, were used to construct a novel nomogram for the prediction of prognosis in the training cohort (Fig. 1). The value of each independent prognostic factor was given a score on a 100-point scale axis, and a total score were calculated by adding up the individual score. The probability to predict the prognosis of patients with ACLF was estimated.
Performance evaluation of the novel nomogram by decision-curve analysis
DCA was conducted to evaluate the clinical utility and net benefit of the new nomogram, which were further compared with those of MELD, MELD-Na, and Jin’s model. As shown in Figure 2, the results indicate that the newly developed nomogram had an overall good net benefit within the wide, practical range of threshold probabilities, and impacted patient outcomes.
Performance evaluation and validation of the novel nomogram by ROC curve analysis
The predictive accuracy of the proposed nomogram was evaluated and compared with existing models, including MELD, MELD-Na, and Jin’s model. The AUC, sensitivity, and specificity values (Table 4) were calculated using the ROC curves in Figure 3. In the new nomogram, ROCs yielded AUCs of 0.849 in the training cohort and 0.861 in the prospective validation cohort, which were significantly greater than the AUCs of 0.758 for MELD, 0.750 for MELD-Na, and 0.777 for Jin’s model in the training cohort (all p <0.05, Fig. 3). Table 4 shows that the AUC values for the four models in the test cohort were similar to the values in the training cohort, with the highest AUCs in the newly developed prognosis model. The result along with a high sensitivity and specificity demonstrates the improved discrimination ability of the proposed nomogram to better predict the prognosis of ACLF than MELD, MELD-Na, and Jin’s model.
Table 4Performance comparison of the newly developed model with the existing models
| Training cohort
| Test cohort
|
|---|
| SWE M | MELD | MELD-Na | Jin’s | SWE M | MELD | MELD-Na | Jin’s |
|---|
| Sensitivity (%) | 77.6 | 67.3 | 72.0 | 75.7 | 66.7 | 66.7 | 82.5 | 75.4 |
| Specificity (%) | 77.9 | 75.0 | 67.3 | 70.7 | 75.3 | 70.1 | 32.4 | 57.1 |
| AUC (95% CI) | 0.849 (0.776–0.779) | 0.758 (0.673–0.750) | 0.750 (0.673–0.720) | 0.777 (0.707–0.757) | 0.804 (0.667–0.818) | 0.744 (0.667–0.727) | 0.777 (0.701–0.825) | 0.737 (0.614–0.792) |
Test of the novel nomogram with the calibration curve analysis
Calibration curves were plotted to assess the agreement between the predicted probability and actually observed probability in the training cohort and test cohorts, which were the compared with those of MELD, MELD-Na, and Jin’s model. As shown in Figure 4, there was good agreement between prediction by the newly established nomogram and actual observation in the training and test cohorts. The data validate the performance accuracy of the newly developed prognostic nomogram.
Discussion
To the best of our knowledge, this is the first prospective study to integrate 2D-SWE values with other independent prognostic factors for the prediction of patient prognosis with ACLF-HBV. The novel findings can be summarized as: (1) Univariate and multivariate regression analyses revealed that five independent prognostic factors, including age, Neu, INR, and 2D-SWE, significantly affect the prognosis of patients with ACLF-HBV. (2) The novel prognostic nomogram was successfully constructed using 2D-SWE values in combination with the other independent prognostic factors to predict the prognosis of ACLF-HBV patients. (3) ROC curve analysis found that the AUCs of the newly constructed model in both the training and the prospective validation cohorts were significantly greater than those of the existing MELD, MELD-Na, and Jin’s prognostic models. (4) DCA supported the superior overall net benefit and clinical utility of the newly developed nomogram in predicting the prognosis of ACLF-HBV patients. (5) The novel prognostic nomogram was well-calibrated according to the good agreement between the predicted and observed probabilities in the calibration curve analysis. The major findings of this study further suggest that the new nomogram has enhanced prognostic accuracy, and, therefore appears promising as an improved prognostic tool to be used in the clinical practice.
As in the 2D-SWE-based noninvasive nomogram for predicting the prognosis of ACLF patients, older age, high TB, and INR have been reported to adversely affect the prognosis of ACLF patients in previous studies.25,26 Shi et al reported that multivariate time-to-death analysis found older age, high INR, and high levels of serum TB to be independent predictors for early death of ACLF patients, within 28 days.26 In addition, Zheng et al found that a number of variables, including age, TB, serum sodium were significantly correlated with the prognosis of patients with acute-on-chronic hepatitis B liver failure (ACHBLF).25 The previous studies support our findings that these prognostic factors were independently associated with the poor prognosis of ACLF.
Interestingly, this study found that a high percentage of peripheral neutrophils was associated with death or liver transplantation in patients with HBV-related ACLF. Generally, increased neutrophils were believed to be related to bacterial infection, severe systemic inflammation, increased susceptibility to sepsis, or even organ dysfunction. Bacterial infection is an important or even fatal complication of ACLF, and has been related to increased grades of systemic inflammation, worse clinical course, and increased 90-day mortality.27 The underlying mechanisms of the correlations are not clear, but the interaction of inflammatory mediator and neutrophils in liver failure is a possible explanation. Inflammatory mediators such as IL1, IL8, and TNFα are known promote neutrophil extravasation of liver, trigger neutrophil activation, and in turn kill hepatocytes by inducing oxidative stress and expressing the Fas ligand.28
Recent advances in the use of ultrasound examination for LSM has led the World Health Organization (WHO)4 and other professional associations10,12 to recommend the use of TE to assess liver fibrosis and cirrhosis, and point out that liver stiffness may be impacted by necrosis or inflammation. Liver necrosis and inflammation may vary with the stage of ACLF-HBV. Studies have found that LSM and MELD had comparable efficacy to predict liver failure,11,14–16 and have shown that dynamic monitoring may be beneficial. 2D-SWE is more suitable for obtaining LSM in ACLF than TE because it can used in patients with ascites. In addition, 2D-SWE is compatible with ultrasound scanners that can evaluate liver morphological changes and potential tumor detection. Previously, Jin’s model showed that the MELD-SWE score combined with the MELD and SWE values was better than MELD alone in predicting the prognosis of ACLF-HBV patients.20 In this study, a new model incorporating serological and conventional ultrasound indicators highlights the value of 2D-SWE in predicting the prognosis of chronic and acute liver failure.
It is pertinent to mention that the constructed nomogram based on 2D-SWE in combination with other four prognostic factors (age, TB, Neu, and INR) was accurate, feasible, and reliable. Its performance was better than that of existing prognostic scoring models, such as MELD, MELD-Na, and Jin’s model, for ACLF-HBV patients. First, we conducted the ROC curve analysis to obtain the AUC values. As a result, the AUC value of the newly constructed model was greater than the AUC value of MELD, MELD-Na, and Jin’s model. In a previous study, Jalan et al.29 reported AUCs of 0.65 for MELD and 0.67 for MELD-Na to predict the 90-day mortality of ACLF patients, and both are lower than the AUC of the newly developed nomogram, which was more accurate than the existing prognostic models in predicting the prognosis of patients with ACLF-HBV. Second, DCA analysis found the overall net benefit and clinical utility of the newly developed nomogram were comparable to the existing prognostic scoring systems or models, i.e., MELD, MELD-Na, and Jin’s model, to predict the prognosis of ACLF-HBV patients. Thirdly, the new nomogram was well-validated in the test cohort by calibration curve analysis, indicating good agreement between the predicted and observed probabilities in patients with ACLF-HBV. In addition to our previous findings, the results of this study demonstrate that the new nomogram improved the accuracy of ACLF prognosis and is potentially useful in clinical practice. Considering the relatively high incidence of HBV infection in China, clinical challenges in predicting the prognosis of ACLF caused by HBV infection, and the high mortality of patients with ACLF, this nomogram should be a choice for priority stratification to identify individuals in need of liver transplantation.
Despite the strengths mentioned above, the study has potential limitations. For example, the ACLF-HBV patients in the training, test, and prospective validation cohorts were enrolled from a single hospital in Guangdong province in southern China. Future studies should select patients from multiple centers in more diverse geographical areas and among various tiers to yield more representative results. As chronic hepatitis B was identified as the underlying liver disease in these ACLF patients, the newly developed prognostic nomogram may not be applicable to ACLF patients with etiologies other than HBV infection, such as alcohol-related hepatitis, HCV infection, alcohol-related liver diseases, drug-induced hepatitis, autoimmune liver disease, and Wilson’s disease, which were excluded from this study. It should be acknowledged that the baseline characteristics were used to develop this model for the early prediction of ACLF prognosis, and that most of the patients did not have multiple organ failures at the time of hospital admission. In this study, we did not compare the model with those taking multiple organ failures into account such as the chronic liver failure-sequential organ failure assessment, or sequential organ failure assessment. It is necessary to externally validate the findings, including the reproducibility, reliability, and accuracy, at multiple centers nationwide.
In conclusion, a novel prognostic nomogram was successfully developed by integrating 2D-SWE values with other independent prognostic factors. The prognostic nomogram had high prognostic accuracy and overall better performance compared with the existing prognostic models. As such, the novel prognostic nomogram may assist hepatologists/surgeons to navigate optimal treatment and management plans, reduce mortality, and ultimately improve the overall care of patients with ACLF-HBV.