Chronic hepatitis B (CHB) virus infection is one of the most common causes of liver disease, affecting more than 400 million people worldwide. Nucleos(t)ide analog (NA) therapy has dramatically improved the outcomes of CHB in the last 15 years.1 However, it is estimated that 600,000 deaths per year are still reported in patients with CHB.2 Liver-related mortality in CHB can be classified into three categories: (i) complications of liver cirrhosis or hepatocellular carcinoma (HCC), (ii) gradual hepatic decompensation with liver failure, and (iii) acute decompensation, otherwise known as acute-on-chronic liver failure (ACLF).3

ACLF is defined as an acute hepatic insult manifesting as jaundice and coagulopathy, complicated within 4 weeks by ascites and/or encephalopathy in a patient with previously diagnosed or undiagnosed chronic liver disease. The clinical and immunological features of ACLF are differentiated from those of acute liver failure.4 HBV flares occur in 40-50% of hepatitis B envelope antigen (HBeAg)-positive CHB patients and in 15-30% of HBeAg-negative CHB patients. Up to 8% of these patients develop hepatitis B related acute-on-chronic liver failure (ACLF-HBV).5 The prognosis of ACLF-HBV has been reported to be very poor, with 3-month mortality rates of over 50% without liver transplantation.3 Although liver transplantation has been recognized as the only definitive therapy, some patients can survive with medical treatment alone due to development of potent oral antiviral agents.

Identification of prognostic factors for ACLF-HBV patients is critical, because emergent liver transplantation (LT) is not readily available in many cases because of organ shortage. Previously proposed factors associated with adverse outcomes of ACLF-HBV include pre-existing cirrhosis, prolonged prothrombin time (PT), elevated bilirubin, low albumin level, low platelet count and age.67 Most of these are baseline characteristics estimated at initial admission. Because of the rapid progressive course of ACLF, both the initial static features and the early dynamic changes in clinical features after admission can be helpful to predict the outcome. However, there is a shortage of studies evaluating the effectiveness of changing clinical events after hospitalization in combination with baseline characteristics.

Therefore, the aims of this study were to identify the prognostic factors including both the baseline upon admission and the clinical features during hospitalization in ACLF-HBV patients.

For acute-on-chronic liver failure (ACLF) in CHB patients, earlier diagnosis and identification of poor prognosis predictors are necessary to distinguish the patients who require transplantation from those who will survive with only medical care. The objectives of this study are to identify the risk factors for unfavorable outcomes in hepatitis B related acute-on-chronic liver failure (ACLF-HBV) and to observe and compare the changing clinical manifestations of different outcome groups. Our data demonstrates that both baseline clinical features and changing clinical presentations during hospitalization could suggest unfavorable outcomes and the need for earlier LT consideration.

Although ACLF is believed to be reversible, the reversibility depends on the severity and nature of the acute insulting cause and the extent of the underlying chronic liver disease.4 One of the causes of ACLF in CHB is acute exacerbation of CHB, a disease presentation distinct from other types of chronic liver diseases due to its sudden virologic surge followed by biochemical flares. Its pathophysiology and management are also very different from acute liver failure by other causes.3 More than 50% of ACLF-HBV patients will die without LT. LT is the only definitive therapy for ACLF-HBV patients who do not improve with only medical treatment. Therefore, prognostic models to predict spontaneous recovery from ACLF-HBV should be established to select appropriate candidates for LT and to prevent avoidable LT.

Previous studies proposed some predictive scoring models for ACLF such as Child-Turcotte-Pugh (CTP), Acute Physiology and Chronic Health Evaluation (APACHE), Sequential Organ Failure Assessment (SOFA) and Model of End-stage Liver Disease (MELD) scores. However, these studies did not consider the different etiologies of ACLF.414 Furthermore, some regression models using various factors believed to be related to adverse outcomes were proposed to be superior to both the MELD score and the CTP score.1516 However, these studies estimated only the predictive value of initial clinical features. In addition, unfortunately, none of these models demonstrated consistently reliable accuracy in predicting the outcome of ACLF-HBV. Therefore, we investigated the baseline characteristics of ACLF-HBV patients upon admission as well as the changing clinical presentations during hospitalization.

Previous studies identified several baseline factors that were positively associated with mortality in multivariable analysis: hepatic encephalopathy, age, the MELD score, bilirubin level, prolonged prothrombin time, and HBV DNA level.7 The effect of nucleos(t)ide analog therapy such as ETV and LAM is still controversial.171819 Our results are almost comparable to those of the above studies. The significantly different baseline characteristics between FG and UFG were age, ascites (≥grade 2), hepatorenal syndrome, serum bilirubin level, prothrombin time, pre-existing cirrhosis of Child-Pugh class B or C and MELD score. However, the type of nucleos(t)ide analog therapy was not related to the outcome. Furthermore, initial HBV DNA level did not show significant distinction. These data imply that ACLF-HBV is a distinguishable disease entity from acute hepatitis B or chronic hepatitis B for which the key point of treatment is viral suppression.

In addition to the baseline characteristics of ACLF-HBV, our study suggested that dynamic changing clinical features could be more useful for predicting the prognosis. By multivariable analysis, the aggravation of hepatic encephalopathy to severe grade, not the presence of initial severe hepatic encephalopathy, is an independent poor prognosis predictor for ACLF-HBV. Although we could not find a significant difference in infection rate in FG and UFG, SIRS and infection developed at a significant rate in the Death group compared to the LT group according to subgroup analysis of UFG. The hepatic dysfunction in UFG may play a role in the increased susceptibility to infection.20 In spite of a similar degree of initial hepatic dysfunction in the LT and Death groups, the additional organs affected by infection were important in determining the possible condition for LT. Based on this result, if signs of infection appeared during hospitalization in ACLF-HBV patients, early consideration of LT as well as active control of the infection could be helpful to reduce the 3-month mortality rate.

Recent studies demonstrated extremely high short-term mortality (≥90%) in ACLF patients with MELD scores ≥30 under conventional medications.1421 However, variable cut-off values for the initial MELD score ranging from 25 to 30 have been proposed.7 Our multivariable analysis results demonstrated that an initial MELD score over 28 was a significant risk factor. However, initial MELD scores of FG (17-33) showed wide overlapped distribution with those of UFG (14-53). Therefore, there is insufficient evidence to decide on the necessity for LT using only the initial MELD score. Starting from the 7th day of hospitalization, the MELD scores increased by more than 3 in UFG and decreased by more than 1.9 in FG. These results suggest that serial measurement of MELD scores could be helpful for predicting outcomes and selecting a management plan. Although there were statistical limitations in deducing the absolute scale of the delta MELD score or the timing due to the small study population, we were able to confirm the likelihood of an unfavorable outcome with an increase of more than 3 in the MELD score on the 7th day of hospitalization. A decrease in the MELD score of more than 1.9 on the 7th day of hospitalization was found to be related to a favorable outcome. Lee WC et al.21 measured the MELD scores of 51 non-cirrhotic ACLF-HBV patients and presented estimated weekly changes in the MELD scores. The respective increase and decrease in MELD scores for patients who spontaneously recovered and those who required LT was analogous to our data. However, we measured the serial MELD scores of most cirrhotic ACLF-HBV patients at a shorter interval due to the early dynamicity of ACLF and compared the analytic predicted value of the serial MELD scores.

Our study had some limitations. First, this was designed as a retrospective study. Thus, the therapeutic modality of ACLF-HBV was determined according to the physician's decision, not with a planned protocol. This might lead to a bias in the prognosis group allocation and analysis. Also, there were missing data such as serial MELD scores and the type, duration of the antiviral agent used. Second, the number of patients was too small to fully analyze the effective prognosis predictors for two or three subgroups. Because this study was based on single center data, data collection from multi centers could provide more valuable results. Most enrolled patients were referred from other hospitals at different times in both FG and UFG. It is possible that the patients who were referred earlier may have had more opportunities to consider LT for survival. Although the exact onset timing of acute hepatic decompensation before transfer was not available from medical records, the mean actual hospitalization duration till death or LT was shorter in the Death group (20 days) than in the LT group (45 days).

In conclusion, it is important to assess the dynamic progression of clinical features (aggravation of hepatic encephalopathy (≥grade III), development of infection or SIRS) earlier as well as the initial clinical presentation (initial high MELD score ≥28, presence of moderate or severe ascites) to predict the prognosis of ACLF-HBV. Early consideration of LT is necessary for the patients with these risk factors. Furthermore, serial measurement of MELD scores to determine delta MELD score can be useful to predict unfavorable prognosis.