General characteristics

From July 2010 to June 2022, 361 patients with HBV–ACLF were screened at Qilu Hospital of Shandong University. Twenty-one patients with HBV–ACLF were excluded for co-infected with HAV, HCV or HEV. Nineteen patients were excluded for alcoholic hepatitis and 12 for autoimmune hepatitis. Twenty-three patients were excluded for hepatocellular carcinoma and 6 for incomplete clinical data. Finally, 280 patients were enrolled, among which 204 patients received additional glucocorticoid treatment (Fig. 1). Then, the patients received additional glucocorticoid treatment were randomly divided into a training cohort and a validation cohort. The training cohort included twice as many patients as the validation cohort [18]. Therefore, the training cohort included 136 patients and the validation cohort included 68 patients.

Fig. 1

Flow diagram depicting the patients’ selection process

In this study, there were 7 patients in the glucocorticoid treatment group and 5 patients in the routine treatment group received liver transplantation. No significant difference was observed between them (P = 0.25). In the glucocorticoid treatment group, 3 patients in the training cohort and 4 patients in the validation cohort received liver transplantation (P = 0.17).

In this study, there were 66 (32.35%) patients in the glucocorticoid treatment group and 31 (40.79%) patients in the routine treatment group received artificial liver support (P = 0.19). Meanwhile, all the patients in this study received antiviral drugs. In the routine treatment group, 1 patient received lamivudine (LAM), 7 patients received adefovir dipivoxil (ADV), 30 patients received entecavir (ETV), 28 patients received tenofovir disoproxil fumarate (TDF) and 10 patients received tenofovir alafenamide fumarate (TAF). In the glucocorticoid treatment group, 4 patients received LAM, 19 patients received ADV, 89 patients received ETV, 65 patients received TDF and 27 patients received TAF. No significant difference was observed of the antiviral treatment strategy between the two groups (P = 0.94).

Table 1 presents the basic characteristics of the participants. There were no significant differences of the demographic, biochemical and clinical parameters between patients received additional glucocorticoid treatment and those not.

Table 1 Basic characteristics of patients with HBV–ACLF received glucocorticoid treatment or not

Baseline characteristics of the patients received glucocorticoid treatment are shown in Table 2. No significant differences of the demographic, biochemical and clinical parameters were observed between patients in the training cohort and validation cohort.

Table 2 Basic characteristics of patients received glucocorticoid treatment in the training and validation cohortThe treatment efficacy of glucocorticoids in HBV–ACLF

After 90-day follow-up, the mortality of patients in the glucocorticoid treatment group was 86/204 (42.16%), which was significantly lower than that of patients in the routine treatment group (45/76, 59.21%, P = 0.01). The mean survival time was 66.39 (SE 2.22, 95% CI 62.04–70.74) days in the patients received additional glucocorticoid treatment and 55.53 (SE 4.07, 95% CI 47.55–63.50) days in those not. Patients received additional glucocorticoid treatment showed significantly better survival that those not (P < 0.01, Fig. 2A).

Fig. 2

Treatment efficacy of glucocorticoids in HBV–ACLF. A Kaplan–Meier graph showing survival probability in patients with HBV–ACLF received additional glucocorticoid treatment or not. B Comparison of log10 [HBV DNA] between the glucocorticoid treatment group and routine treatment group

There were no significant differences in log10 [HBV DNA] levels between the glucocorticoid treatment group (median 4.73, interquartile range 3.38–6.1) and routine treatment group (median 5.07, interquartile range 4.21–6.23) before treatment (P = 0.1, Table 1). Meanwhile, there were also no significant differences of log10 [HBV DNA] levels between the glucocorticoid treatment group and routine treatment group at 30 days (median 3.89, interquartile range 3.0–4.78 vs median 3.94, interquartile range 3.51–4.67, P = 0.38), 60 days (median 3.49, interquartile range 3.0–3.81 vs median 3.51, interquartile range 3.0–3.81, P = 0.32) and 90 days (median 3.0, interquartile range 2.7–3.48 vs median 3.0, interquartile range 2.7–3.08, P = 0.1) after treatment (Fig. 2B).

The Adverse events of glucocorticoid treatment in HBV–ACLF

The incidence of newly onset infection in the glucocorticoid treatment group was 31.37% (64/204), which was higher than that in the routine treatment group (17/76, 22.37%). However, no significant difference was observed between them (P = 0.14).

The incidence of UGIH was 9.31% (19/204) in the glucocorticoid treatment group, whereas 7.89% (6/76) in the routine treatment group (P = 0.71). The incidence of HRS was 7.84% (16/204) in the glucocorticoid treatment group, whereas 10.53% (8/76) in the routine treatment group (P = 0.48). The incidence of electrolyte disturbance was 42.16% (86/204) in the glucocorticoid treatment group, whereas 39.47% (30/76) in the routine treatment group (P = 0.69).

Predictors of 90-day mortality of HBV–ACLF after glucocorticoid treatment

Baseline demographic, clinical and laboratory parameters for the prediction of 90-day mortality in patients received additional glucocorticoid treatment were investigated by univariate analysis with cox proportional hazard regression model. In the training cohort, the univariate analysis identified eight variables including age (P = 0.002), TBIL (P < 0.001), ALB (P = 0.018), INR (P = 0.038), Cr (P = 0.013), HGB (P = 0.025), HE (P < 0.001), SIRS (P < 0.001) (Table 3). Then, these variables were introduced into a multivariate stepwise cox regression. HE (HR 1.940, 95% CI 1.379–2.728, P < 0.001), INR (HR 1.055, 95% CI 1.006–1.107, P = 0.027), TBIL (HR 1.003, 95% CI 1.002–1.005, P < 0.001), age (HR 1.031, 95% CI 1.010–1.052, P = 0.004) and SIRS (HR 3.312, 95% CI 1.855–5.913, P < 0.001) were identified to be independent predictors for 90-day mortality (Table 4).

Table 3 Factors associated with 90-day mortality in univariate analysis in the training cohortTable 4 Independent predictors for 90-day mortality identified after multivariate analysis in the training cohortCalculation of the prognostic index in the training cohort

A prognostic model was calculated by combining the 5 prognostic predictors with the regression coefficients reported in Table 4. It was calculated according to the following formula:

$$} = \, 0. * } + \, 0.0 * } + 0.00 * } + \, 0.0 * } + . * }$$

Diagnostic value of the HITAS score

The HITAS scores of non-survivors were higher than those of survivors in the training cohort (t = − 9.754, P < 0.01) (Fig. 3A), the validation cohort (t = − 6.234, P < 0.01) (Fig. 3B) as well as the entire cohort (t = − 11.404, P < 0.01) (Fig. 3C). The AUC of the HITAS score was 0.88 (standard error [SE] 0.03, 95% confidence interval [CI] 0.82–0.93) in the training cohort (Fig. 3D), 0.87 (SE 0.04, 95% CI 0.77–0.94) (Fig. 3E) in the validation cohort and 0.87 (SE 0.02, 95% CI 0.82–0.92) (Fig. 3F) in the entire cohort.

Fig. 3

HITAS scores between survivors and non-survivors, and the ROC curves of HITAS scores. A HITAS scores of non-survivors were higher than those of survivors in the training cohort. B HITAS scores of non-survivors were higher than those of survivors in the validation cohort. C HITAS scores of non-survivors were higher than those of survivors in the entire cohort. D ROC curves of HITAS scores in the training group (AUC 0.88). E ROC curves of HITAS scores in the validation group (AUC 0.87). F ROC curves of HITAS scores in the entire cohort (AUC 0.87)

In the training cohort, two cutoff points were chosen for HITAS score to predict the 90-day mortality in patients with HBV–ACLF received additional glucocorticoid treatment. A high cutoff point was chosen based on the ROC analysis of this model in the training cohort to provide a specificity of at least 85%. A low cutoff point was chosen to provide a sensitivity of at least 90% [19]. Finally, 2.5 was chosen as a low cutoff point and 3.47 was chosen as a high cutoff point. The diagnostic accuracy of HITAS score is presented in Table 5. In the training cohort, the cutoff of 2.5 provided a sensitivity of 95% and a specificity of 56%. The cutoff of 3.47 provided a sensitivity of 69% and a specificity of 91%, respectively. In the validation cohort, the cutoff point of 2.5 provided a sensitivity of 81% and a specificity of 76%. The cutoff point of 3.47 provided a sensitivity of 68% and a specificity of 84%, respectively.

Table 5 Diagnostic accuracy of the HITAS score in predicting the 90-day mortality of HBV–ACLF patients received glucocorticoid treatment.

Using the cutoff points of 2.5 and 3.47, the HITAS score identified patients that have low (HITAS score ≤ 2.5), intermediate (2.5 < HITAS score ≤ 3.47), and high (HITAS score > 3.47) risks of death (Table 6). In the training cohort, only 3 of the 48 patients (6.25%) in the low-risk group died within 90 days. 14 of the 43 patients (32.56%) in the intermediate-risk group died. 38 of the 45 patients (84.44%) in the high-risk group died. In the validation cohort, only 6 of the 34 patients (17.65%) in the low-risk group died within 90 days. 4 of the 7 patients (57.14%) in the intermediate-risk group died. 21 of the 27 patients (77.78%) in the high-risk group died. In the entire cohort, only 9 of the 82 patients (10.98%) in the low-risk group died within 90 days. 18 of the 50 patients (36.00%) in the intermediate-risk group died. 59 of the 72 patients (81.94%) in the high-risk group died.

Table 6 Stratification of the risks of death by HITAS score in patients with HBV–ACLF received glucocorticoid treatment

In the training cohort (Fig. 4A), the AUC of HITAS score (AUC 0.88 SE 0.03, 95% CI 0.82–0.93) was significantly higher than MELD score (AUC 0.79 SE 0.04, 95% CI 0.71–0.85; P = 0.02) and CTP score (AUC 0.75 SE 0.04, 95% CI 0.67–0.82; P < 0.01). There was no significant difference between AUC of MELD score and CTP score (P = 0.50). In the validation cohort (Fig. 4B), the AUC of HITAS score (AUC 0.87 SE 0.04, 95% CI 0.77–0.94) was significantly higher than that of MELD score (AUC 0.75 SE 0.06, 95% CI 0.63–0.85; P = 0.04) and CTP score (AUC 0.72 SE 0.07, 95% CI 0.59–0.82; P = 0.02). No significant difference was found between AUC of MELD score and CTP score (P = 0.63). In the entire cohort (Fig. 4C), the AUC of HITAS score (AUC 0.87 SE 0.02, 95% CI 0.82–0.92) was significantly higher than that of MELD score (AUC 0.78 SE 0.03, 95% CI 0.71–0.83; P < 0.01) and CTP score (AUC 0.74 SE 0.06, 95% CI 0.68–0.80; P < 0.01). There was no significant difference between AUC of MELD score and CTP score (P = 0.40).

Fig. 4

Receiver operating characteristic (ROC) curves and Kaplan–Meier graphs in patients with HBV–ACLF received glucocorticoid treatment. A Comparison among ROC curves of HITAS score, MELD score and CTP score in the training cohort. B Comparison among ROC curves of HITAS score, MELD score and CTP score in the validation cohort. C Comparison among ROC curves of HITAS score, MELD score and CTP score in the entire cohort. D Kaplan–Meier graphs showing survival probability in patients within the low-risk group, intermediate-risk group and high-risk group in the training cohort. E Kaplan–Meier graphs showing survival probability in patients within the low-risk group, intermediate-risk group and high-risk group in the validation cohort. F Kaplan–Meier graphs showing survival probability in patients within the low-risk group, intermediate-risk group and high-risk group in the entire cohort

After 90-day follow-up, the mortality was 55/136 (40.44%) in the training cohort, 31/68 (45.59%) in the training cohort and 86/204 (42.16%) in the entire cohort. The mean survival time was 67.5 (SE 2.69, 95% CI 62.23–72.77) days in the training cohort, 64.16 (SE 3.92, 95% CI 56.49–71.84) days in the validation cohort and 66.39 (SE 2.22, 95% CI 62.04–70.74) days in the entire cohort. Then, we used Kaplan–Meier survival analyses to compare the survival probability in patients within the low-risk group, intermediate-risk group and high-risk group. As shown in Fig. 4D, E, F, the cutoff points of 2.5 and 3.47 successfully identified patients with low, intermediate, and high risks of death after 90-day follow-up in the training cohort, validation cohort and entire cohort (P < 0.01, respectively).

Clinical application of the model

Algorithm for the application of HITAS score to predict 90-day mortality of patients with HBV–ACLF received glucocorticoid treatment is presented in Fig. 5. The HITAS score is an early prediction model for the prognosis of HBV–ACLF, which might be used to identify HBV–ACLF patients with favorable responses to glucocorticoid treatment.

Fig. 5

Algorithm for application of HITAS score to predict 90-day mortality in patients with HBV–ACLF received glucocorticoid therapy