Two cohorts were enrolled in this study, and the flow chart was in Figure 1. In the first cohort, we enrolled HBV-ACLF patients who presented with a new episode of acute hepatic insult manifesting as acute jaundice, coagulopathy, ascites, upper gastrointestinal bleeding, and hepatic encephalopathy (HE) between 1 January 2018 and 1 October 2018. Patients were divided into survivor and non-survivor groups according to their survival status of 30 days after admission. Patients undergoing liver transplantation were excluded. In the second cohort, a cross-sectional investigation was performed using a cohort of 142 subjects including healthy controls (HCs), patients with CHB, liver cirrhosis (LC), liver cirrhosis-acute decompensation (LC-AD) and ACLF. All HCs, CHB and LC subjects were recruited from the outpatient clinic. Among the 38 enrolled ACLF patients, 21 survived patients were placed into ACLF-S group while the remaining 17 were placed into ACLF-D group. Patients with ACLF received a standard treatment, including nucleoside analogs for HBV DNA-positive patients; diammonium glycyrrhizinate and Ademetionine for protection of liver function; sodium restriction, diuretics and paracentesis combined with albumin infusion for ascites; lactulose and L-ornithine aspartate for HE; prophylactic antibiotics for bacterial infections and renal replacement therapy for hepato-renal syndrome. Eleven non-survivors (11/38) and 35 survivors (35/124) received plasma exchange (PE) combined with hemofiltration (HF) treatment. All subjects were enrolled or recruited in the First Affiliated Hospital of Zhejiang University. This study was conducted in compliance with the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the First Affiliated Hospital of Zhejiang University. Written informed consent was obtained from all patients or their legal representatives.

ACLF was diagnosed according to the Asian Pacific Association for the Study of the Liver (APASL) criteria published in 2014 (3). According to the APASL criteria, ACLF was defined as an acute hepatic insult manifesting as jaundice (serum levels of bilirubin ≥ 5 mg/dl) and coagulopathy (international normalized ratio [INR] ≥ 1.5 or prothrombin activity <40%) within 4 weeks in a patient with previously diagnosed or undiagnosed CLD/cirrhosis. Cirrhosis patients with a history of AD who met the above criteria were also diagnosed as ACLF. The Model for End-stage Liver Disease (MELD), the chronic liver failure-sequential organ failure assessment (CLIF-SOFA), CLIF-C ACLF (18) scores were calculated to evaluate the severity of patients. The diagnosis of LC was based on previous liver biopsy, endoscopic signs, radiologic imaging, clinical symptoms and laboratory parameters. HBV reactivation was defined as a ≥ 2 log increase in the HBV DNA level from previously stable baseline level or a level ≥ 100 IU/mL in patients in whom HBV DNA had been undetectable, or ≥ 20,000 IU/mL in those negative for HBV DNA at baseline (19). The exclusion criteria included: age <18 or >80 years, liver malignancy or any other type of cancer, history of liver transplantation or bone marrow transplantation, and any severe disease in other systems.

Data were collected at enrollment, and included vital signs, physical examination results, histories, complications and precipitating events. The laboratory measurements collected included white blood cell (WBC) count, neutrophils, red blood cell (RBC) count, platelet (PLT) count, C-reactive protein (CRP), alpha-fetoprotein (AFP), ferritin, total protein (TP), alanine aminotransferase (ALT), aspartate aminotransferase (AST), serum levels of albumin (ALB), cholinesterase, gamma-glutamyl transpeptidase (GGT), triglycerides, cholesterol, total bilirubin (TB) level, serum levels of sodium, potassium (K), creatinine (Cr), blood urea nitrogen (BUN), international normalized ratio (INR), pulse oximetry, HBV infection biomarker levels, and HBV-DNA levels. Image measurements, including ultrasound, CT, and MRI, were also collected. Follow-up information was collected from medical records and follow-up phone calls at 30 days.

Blood samples were obtained on the day following hospitalization and were centrifuged at 3,500 rpm for 10 min to separate the serum, which were stored at −80°C. GH was measured using an enzyme-linked immunosorbent assay (ELISA) with an ELISA detection kit (ab190811, Abcam). IGF-1, IGF-2, IGF receptor-1, and IGF receptor-2 levels in prospective cohort were detected using an immunoassay (QAH-IGF-1, RayBiotech). IGF-1 level in cross-sectional cohort was detected using an ELISA detection kit (ab211651, Abcam).

Statistical analyses were performed using Statistical Package for the Social Sciences (SPSS, v.22.0; SPSS, Inc., Chicago, IL, USA) and MEDCALC (MedCalc Software, Belgium). Continuous data are expressed as means ± standard deviations or medians with interquartile ranges (p25, p75), while categorical data are expressed as numbers (percentages). All tests were two-tailed, and P < 0.05 were considered indicative of statistical significance. Student's t-tests or non-parametric Mann–Whitney U-tests, as appropriate, were used to compare continuous data. Categorical data and ordered categorical data were compared using chi-square tests and Spearman rank correlation tests, respectively. The area under the receiver operating characteristic curve (AUROC) of the various prognostic scoring systems was compared by Z-tests using Delong's method.

A total of 124 HBV-ACLF patients were enrolled in the prospective cohort. As shown in Table 1, the mean patient age was 46.61 ± 12.71 years, and 19 (15.3%) patients were female. Among the whole cohort, 69 (55.6%) patients were cirrhotic, and 25 (20.2%) had acute decompensation (AD) history. The MELD, CLIF-SOFA, CLIF-C ACLF scores were 23.09 ± 5.61, 8.15 ± 1.95, and 42.01 ± 9.79, respectively.

Based on 30-day survival, the patients were divided into non-survivor (n = 38) and survivor groups (n = 86). The mean age of survivor and non-survivor groups was 44.58 ± 12.15 and 51.21 ± 12.9, while 12.8% in survivor group were female and 21.1% in non-survivor group. Compared to the survivor group, non-survivors were older (P = 0.009) and had lower levels of cholesterol (P = 0.003) and albumin (P = 0.035). Conversely, neutrophil counts (P = 0.005), INR (P < 0.0001), TB (P = 0.001) and creatine (P = 0.005) levels, and MELD (P < 0.0001), CLIF-SOFA (P < 0.0001), and CLIF-C ACLF (P < 0.0001) scores were higher in the non-survivor group.

In the prospective cohort of HBV-ACLF patients, the median GH levels in non-survivors were 1106.55 pg/ml, which were significantly lower (P < 0.0001) than in survivors (2930.55 pg/ml) (Figure 2A). All patients were divided into four groups according the number of organ failures; patients with ≥3 organ failures had significantly lower GH levels compared to 1 (P < 0.001), and 2 (P = 0.045) organ failure groups (Figure 2B). Furthermore, patients with a MELD score > 25 had lower GH levels than those with a score between 20 and 25 (P = 0.009) (Figure 2C). However, there were no significant difference in GH levels between the MELD > 25 group and the MELD <20 group. Comparisons of GH levels between patients with and without certain kinds of organ failure are shown in Supplementary Figure 1. Interestingly, patients with encephalopathy had lower GH levels than those without. We divided patients into no HE, HE I~II, HE III~IV groups (Figure 2D), and then found that GH level in no HE group was significantly higher than HE III~IV group (P = 0.002), and showed a higher trend than HE I~II group (P = 0.056).

The relationships between GH levels and other indicators are shown in Supplementary Table 1. GH levels were negatively related to neutrophil counts, CRP, and BUN, but positively related to ALB, triglycerides, cholesterol, and alkaline phosphatase. The serum levels of IGF-1, IGF-2, IGF-1R, and IGF-2R were measured, as shown in Supplementary Table 2, but there were no significant differences between survivors and non-survivors.

The characteristics of HCs and patients in the cross-sectional cohort are listed in Table 2. There were no differences in mean age or sex among groups. The median GH levels in each group were 427.66 pg/ml (HCs), 125.91 pg/ml (CHB), 362.94 pg/ml (LC), 1246.05 pg/ml (LC-AD), and 1299.8 pg/ml (ACLF). Comparisons of GH levels among the different groups are shown in Figure 3A. Compared to HC (P < 0.001), CHB (P < 0.001), and LC (P < 0.001), the ACLF group had significantly higher GH levels. Similarly, patients in the LC-AD group had higher GH levels than those in the HC (P < 0.001), CHB (P < 0.001), and LC (P < 0.001) groups. In ACLF group, ACLF-S group had higher GH level than ACLF-D group, which was shown in Supplementary Figure 2. A detailed comparison of GH levels among groups is listed in Supplementary Table 3. Correlation analysis revealed a positive correlation between GH levels and disease severity (r = 0.462, P < 0.001).

Distribution of IGF-1 among these groups was different from GH. As shown in Figure 3B, IGF-1 levels in HC group were higher than LC (P < 0.001), LC-AD (P < 0.001) and ACLF (P < 0.001) group. Interestingly, IGF-1 levels in LC-AD group were significantly lower than in LC group (P < 0.001), but GH levels revealed much higher (P < 0.001). Consistent with the result in prospective cohort, there was no difference of IGF-1 levels between ACLF-S and ACLF-D groups. A detailed comparison of IGF-1 levels among groups is listed in Supplementary Table 4. Correlation analysis revealed a negative correlation between IGF-1 levels and disease severity (r = −0.641, P < 0.001).

Univariate analyses showed that age, HE, GH, WBC, neutrophil count, ALB, TB, Cr, BUN, cholesterol, and INR were significantly associated with 30-day outcomes of HBV-ACLF patients in the prospective cohort. Then, multivariate analyses revealed that HE, GH, WBC, TB, BUN, and INR were independently associated with prognosis at day 30 (Table 3).

Next, we analyzed the prognostic value of GH. The AUC of GH for predicting 30-day outcomes was 0.793 (Figure 4A). Then we divided patients into low-GH and high-GH groups according to the cut-off value (2,001 pg/mL). The 30-day survival rate in low-GH group was significantly lower than in the high-GH group (45.5 vs. 88.4%, P < 0.001) (Figure 4B).

Then we analyzed the prognostic efficacy among different prognostic models. The AUC of Child-Pugh, MELD, CLIF-SOFA and CLIF-C ACLF was 0.758, 0.882, 0.862, and 0.838. A new prognostic model was built based on MELD score by logistic regression analyses, namely MELD-GH, combining GH with MELD to predict 30-day outcome. The formula used to calculate MELD-GH is: 0.413^*MELD-1.606^*Ln (GH). The sensitivity and specificity of MELD-GH was 83.72% and 92.11%. The AUC of MELD-GH was 0.937, which was significantly superior to the Child-Pugh (P < 0.0001), MELD (P = 0.0185), CLIF-SOFA (P = 0.034), and CLIF-C ACLF (P = 0.0227) scores for predicting 30-day outcomes (Figure 4C and Table 4). When MELD-GH < −5, the mortality rate only reached 3.5%, and when MELD-GH > 0, the mortality rate increased above 90% (Figure 4D). The detailed comparison of each two prognostic models were listed in Supplementary Table 5.