Noninvasive models for predicting poor prognosis of chronic HBV infection patients precipitating acute HEV infection

Hepatitis E virus (HEV) infection contributes to a considerable proportion of acute-on-chronic liver failure (ACLF) in patients with chronic hepatitis B virus (HBV) infection. This study aimed to predict the prognosis of chronic HBV infection patients precipitating acute HEV infection. A total of 193 patients were enrolled in this study. The performances of three chronic liver disease prognostic models (CTP score, MELD score, and CLIF-C ADs) were analyzed for predicting the development of ACLF following HEV superimposing chronic HBV infection. Subsequently, the performances of five ACLF prognostic assessment models (CTP score, MELD score, CLIF-C ACLFs, CLIF-C OFs, and COSSH-ACLFs) were analyzed for predicting the outcome of those ACLF patients. Of 193 chronic HBV infection patients precipitating acute HEV infection, 13 patients were diagnosed ACLF on admission, 54 patients developed to ACLF after admission, and 126 patients had non-ACLF during the stay in hospital. For predicting the development of ACLF, CTP score yielded a significantly higher AUROC compared with MELD score and CLIF-C ADs (0.92, 0.88, and 0.86, respectively; all p < 0.05). For predicting the poor prognosis of ACLF patients, the COSSH-ACLFs yielded a significantly higher AUROC compared with CLIF-C ACLFs, CLIF-C OFs, MELD score, and CTP score (0.89, 0.83, 0.81, 0.67, and 0.58, respectively; all p < 0.05). In conclusion, the stepwise application of CTP score and COSSH-ACLFs can predict the prognosis of chronic HBV infection patients precipitating acute HEV infection.

Diagnostic criteria. Chronic HBV infection was diagnosed using serum HBsAg and/or HBV DNA positivity for more than six months 11 . Acute HEV infection was diagnosed using two consecutive positive serum anti-HEV immunoglobulin IgM test results, and seroconverted to serum anti-HEV immunoglobulin IgG positivity during follow-up. ACLF was diagnosed according to the consensus recommendations of the Asian Pacific Association for the Study of the Liver (APASL) 2014 for ACLF 12 : jaundice (total bilirubin ≥ 5 mg/dL) and coagulopathy (INR ≥ 1.5) complicated within 4 weeks by ascites and/ or hepatic encephalopathy in a patient with or without previously diagnosed chronic liver disease 12 . The poor prognosis of ACLF was defined as death or liver transplantation within 28 days after admission. clinical laboratory measurements. Serum anti-HEV immunoglobulin IgM and IgG were tested using enzyme-linked immunosorbent assay (ELISA) (MP Biomedicals, Singapore) according to the manufacturers' instructions. Liver functions were tested using fully-auto-biochemistry-analysis instruments (7600 Series; Hitachi, Japan). Serum HBV markers were detected using ELISA methods (ARCHITECT i2000 SR; Abbott, Germany). HBV viral load quantification was detected using real-time PCR (ABI 7500; Applied Biosystems Inc, United States). Routine blood tests were performed using automated blood cell analyzers (XT-2000i; Sysmex, Japan). choice of prognostic assessment models. Until 2013, Child-Turcotte-Pugh (CTP) score and model for end-stage liver disease (MELD) score were the only two available methods to assess prognosis of ACLF patients. According to the chronic liver failure (CLIF) Consortium of EASL, the CLIF acute decompensation score (CLIF-C ADs) can be used to predict prognosis of hospitalised cirrhotic acute decompensation patients without ACLF 13 . The CLIF-SOFA score was a modified version of SOFA score, proposed by the EASL CLIF Consortium, which can be used to predict the prognosis of ACLF 14 . The CLIF-C OFs is a simplified version of CLIF SOFA score, and has the same performance with CLIF-SOFA score for predicting the prognosis of ACLF 14,15 . The CLIF-Consortium ACLF score (CLIF-C ACLFs) was proposed for predicting the prognosis of ACLF based on two large prospective studies performed by the EASL CLIF Consortium 14,15 . Wu et al. developed a prognostic score for HBV-related ACLF, named the Chinese Group on the Study of Severe Hepatitis B (COSSH)-ACLF score (COSSH-ACLFs) 16 . Compared with the EASL-ACLFs, the COSSH-ACLFs identified approximately 20% more ACLF patients 16 . calculation of prognostic assessment models.

Statistical analysis.
All statistical analysis in this study was performed using statistical software SPSS 15.0 (SPSS Inc. USA) and MedCalc 16.1 (MedCalc Software, Belgium). The Kolmogorov-Smirnov test was used to check the normality of data. Data was showed as follows: normal distribution data as mean ± SD, non-normal distribution continuous data as median (IQR), and categorical data as number (percentage). The data between two groups were compared with Chi-squared-test (for categorical data), Mann-Whitney-test (for non-normal continuous data), and t-test (for normal data), respectively. Pearson correlation coefficient was used to perform correlation analysis between two variables. The areas under Receiver Operator Characteristic (ROC) curves (AUROCs) were calculated to evaluate the performances of the prognostic assessment models. The AUROCs were compared using the Delong test 17 . The optimal cut-offs were obtained by maximizing Youden index (sensitivity + specificity − 1). All significance tests were two-tailed, and p ≥ 0.05 was considered no significant difference between two groups/methods.  comparison between AcLf group and non-AcLf group. Of 193 chronic HBV infection patients precipitating acute HEV infection, 13 patients were diagnosed ACLF on admission, 54 patients developed to ACLF within 28 days after admission, and 126 patients had non-ACLF during the stay in hospital. The comparison between ACLF group and non-ACLF group was showed in Table 1. Patients with ACLF had significantly higher age (mean, 54 vs 47 years, p < 0.001), total bilirubin (median, 16.5 vs 2.19 mg/dl, p < 0.001), white blood cell count (median, 6.1 vs 5.1 × 10 9 cells/L, p = 0.007), INR (median, 1.85 vs 1.08, p < 0.001), CTP score (median, 11 vs 6, p < 0.001), MELD score (median, 21 vs 7, p < 0.001), and CLIF-C ADs (median, 49 vs 37, p < 0.001); but significantly lower ALT (median, 310 vs 434, p < 0.001), AST (median, 176 vs 249, p = 0.021), GGT (median, 94 vs 135, p = 0.015), and platelet count levels (mean, 99 vs 141 × 10 9 cells/L, p < 0.001) compared with patients without ACLF. correlation between prognostic models and the outcome of patients. Based on the fact that only the patients developed ACLF during hospitalization after enrollment could be counted for prediction. First, the correlation analysis was performed between 54 patients developed ACLF within 28 days after admission and 126 patients who had non-ACLF during the stay in hospital. The results showed that the development of ACLF after admission significantly correlated with CTP score (r = 0.76, p < 0.001), MELD score (r = 0.64, p < 0.001), and CLIF-C ADs (r = 0.64, p < 0.001) at the baseline (Table 2). Next, the correlation analysis was performed between 30 ACLF patients with poor prognosis and 37 ACLF patients with favorable outcome. The results showed that the poor prognosis of ACLF significantly correlated with COSSH-ACLFs (r = 0.72, p < 0.001), CLIF-C ACLFs (r = 0.66, p < 0.001), CLIF-C OFs (r = 0.64, p < 0.001), MELD score (r = 0.52, p < 0.001), and CTP score (r = 0.43, p < 0.001) ( Table 2).
AUROCs comparison of prognostic models. ROC curve analysis was performed to predict the development of ACLF (a) and poor prognosis (b) of ACLF patients (Fig. 2). Pairwise comparison of AUROCs was presented in Table 3. For predicting the development of ACLF, CTP score had a significantly higher AUROC than MELD score and CLIF-C ADs (0.92, 0.88, and 0.86 for CTP score, MELD score, and CLIF-C ADs, respectively; all p < 0.005).
In order to evaluate the ability of prognostic models in predicting the outcome of ACLF patients, we divided the 67 ACLF patients into two groups: favorable outcome group (37 patients) and poor prognosis (liver transplantation/death) group (30 patients). For predicting the poor prognosis of ACLF patients, COSSH-ACLFs had a significantly better diagnostic performance than CLIF-C ACLFs, CLIF-C OFs, MELD score, and CTP score (AUROC of 0.89, 0.83, 0.81, 0.67, and 0.58 for COSSH-ACLFs, CLIF-C ACLFs, CLIF-C OFs, MELD score, and CTP score, respectively; all p < 0.05) ( Table 4).

Discussion
In China, the rate of HBsAg positivity is 7.18% in 2006, and 6.0% in 2016 18   For predicting the development of ACLF, CTP score had a significantly higher AUROC than MELD score and CLIF-C ADs (0.92, 0.88, and 0.86 for CTP score, MELD score, and CLIF-C ADs, respectively; all p < 0.005). For predicting the poor prognosis of ACLF patients, COSSH-ACLFs had a significantly better diagnostic performance than CLIF-C ACLFs, CLIF-C OFs, MELD score, and CTP score (AUROC of 0.89, 0.83, 0.81, 0.67, and 0.58 for COSSH-ACLFs, CLIF-C ACLFs, CLIF-C OFs, MELD score, and CTP score, respectively; all p < 0.05).  Table 3. AUROCs of prognostic models for predicting the development of ACLF. AUROC, area under the receiver operating characteristic curve; ACLF, acute-on-chronic liver failure; CTP score, Child-Turcotte-Pugh score; MELD score, model for end-stage liver disease score; CLIF-C ADs, Chronic Liver Failure Consortium acute decompensation score. (2020) 10:2753 | https://doi.org/10.1038/s41598-020-59670-4 www.nature.com/scientificreports www.nature.com/scientificreports/ For chronic HBV infection patients precipitating acute HEV infection, the first question need to be resolved is which patients would develop ACLF. In this study, three prognostic assessment models (CTP score, MELD score, and CLIF-C ADs) were compared for predicting the development of ACLF in chronic HBV infection patients precipitating acute HEV infection. We found that CTP score, MELD score, and CLIF-C ADs were higher in patients developed ACLF, compared with patients had non-ACLF. For predicting the development ACLF, the CTP score yielded a significantly higher performance compared with MELD score and CLIF-C ADs. For those patients, the second question need to be resolved is which patients would have poor prognosis after the development of ACLF. In our study, five prognostic assessment models were analyzed for predicting the poor prognosis of those ACLF patients. We found that COSSH-ACLFs yielded a significantly more diagnostic performance compared with CLIF-C ACLFs, CLIF-C OFs, MELD score, and CTP score for predicting the poor outcome of ACLF patients following HEV superimposing chronic HBV infection.
Therefore, we proposed stepwise application of CTP score and COSSH-ACLFs to predict the outcome of chronic HBV infection patients precipitating acute HEV infection. The CTP score is the best model for predicting the development of ACLF in chronic HBV infection patients precipitating acute HEV infection. The COSSH-ACLFs is the best model for predicting the poor outcome of ACLF patients. The CTP-COSSH-ACLFs algorithm, which sequentially combines CTP-score and COSSH-ACLFs, can discriminate patients at low risk of developing ACLF with excellent prognosis from those at high risk of developing ACLF with impaired prognosis and need specialized care.
Cirrhosis is considered prerequisite for ALCF in Europe and America; however, in Asia, it is considered that ACLF can develop in patients without cirrhosis, including chronic HBV infection patients 12 . In this study, of 67 ACLF patients, 42 (62.7%) had cirrhosis background, 25 (37.3%) had no cirrhosis background. The EASL and AASLD ACLF definitions were proposed and validated only in patients with cirrhosis from Europe and North America, where alcoholic liver disease is the major aetiology 23 . However, in the Asia-Pacific regions including China, the major aetiology of ACLF is HBV infection. Therefore, we defined ACLF according to the ACLF consensus recommendations of the APASL 2014 12 .
Of course, several limitations in this study should be noticed. First, this study is a retrospective single-center study. The results in this study need to be further validated in a large sample, multi-center, and perspective study. Second, this study was performed in a tertiary hospital with a high percentage of seriously ill patients than in the general population. The spectrum bias may appear when extrapolating the results of this study to general  Table 4. AUROCs of prognostic models for predicting the poor prognosis of ACLF. Poor prognosis, death or liver transplantation; ACLF, acute-on-chronic liver failure; COSSH-ACLFs, Chinese Group on the Study of Severe Hepatitis B ACLF score; CLIF-C ACLFs, Chronic Liver Failure Consortium ACLF score; CLIF-C OFs, Chronic Liver Failure Consortium organ failure score; MELD score, model for end-stage liver disease score; CTP score, Child-Turcotte-Pugh score.  Table 5. Diagnostic thresholds of prognostic models. ACLF, acute-on-chronic liver failure; the optimal cut-off points were determined by maximizing Youden index; Se, sensitivity; Sp, specificity; PPV, positive predictive value; NPV, negative predictive value; +LR, positive likelihood ratio; −LR, negative likelihood ratio. (2020) 10:2753 | https://doi.org/10.1038/s41598-020-59670-4 www.nature.com/scientificreports www.nature.com/scientificreports/ population 24,25 . Third, HEV genotyping were not performed in this study. Although we did not detect the HEV genotypes of patients in this cohort, we had reason to believe that the genotypes of almost patients in this study were genotype 4 HEV. Over the past twenty years, HEV genotype 1 has been replaced by HEV genotype 4 as the most common genotype (over 80% of HEV infection patients) in China 26 .

Cut-offs Se (%) Sp (%) PPV (%) NPV (%) +LR −LR
In conclusion, the stepwise application of CTP score and COSSH-ACLFs can predict the outcome of chronic HBV infection patients precipitating acute HEV infection. The CTP-COSSH-ACLFs algorithm provides a method that contributed to regulate the large flow of patients between primary health care institutions and tertiary hospitals: patients had no ACLF tendency can be treatment in primary health care institutions, whereas those had ACLF tendency will need to be redirected to a tertiary hospital for specialized management.