Glucocorticoid therapy in acute liver failure: Survival outcomes, predictors, and a response prediction model

Abstract

BACKGROUND

Glucocorticoids (GC) are a potential therapy for acute liver failure (ALF). However, their clinical efficacy remains controversial, with significant interpatient heterogeneity.

AIM

To assess the impact of GC therapy on 28-day survival of patients with ALF and identify early treatment-response factors.

METHODS

In this single-centre retrospective cohort study, 179 patients with ALF from the past 12 years were included: 84 received GC treatment, and 95 served as non-GC controls. The primary outcome was 28-day survival. GC-treated patients were further stratified into responders and nonresponders to analyse the determinants of efficacy. Survival distributions were compared using Kaplan-Meier curves with the log-rank test. Independent predictors of GC response were identified through multivariate logistic regression. Statistical significance was set at P < 0.05.

RESULTS

The 28-day survival rate was significantly greater in the GC group than in the control group (58.3% vs 30.5%, P < 0.001). An early increase in prothrombin activity (PTA ≥ 4.5% by day 3), along with baseline model for end-stage liver disease (MELD) score < 28.5 and blood ammonia concentration < 135.5 μg/dL, was independently associated with GC response. A model combining these factors predicted GC responsiveness with an accuracy of 95.2%.

CONCLUSION

GC therapy improves 28-day ALF survival. An early increase in the PTA, combined with baseline MELD score and blood ammonia level, effectively identifies patients who are most likely to benefit.

Key Words: Acute liver failure; Glucocorticoids; Prediction model; Survival; Efficacy

Core Tip: This retrospective cohort study of 179 acute liver failure patients demonstrates that glucocorticoid (GC) therapy significantly improves 28-day survival (58.3% vs 30.5%, P < 0.001). An early increase in prothrombin activity ≥ 4.5% by day 3, alongside baseline model for end-stage liver disease score < 28.5 and blood ammonia < 135.5 μg/dL, forms a highly accurate predictive model for GC response (95.2% accuracy). This model provides a practical tool to select patients most likely to benefit from GC treatment.

INTRODUCTION

Acute liver failure (ALF) is a severe syndrome characterized by abrupt onset and rapid progression, manifesting as sudden hepatic injury, coagulopathy [international normalized ratio (INR) ≥ 1.5], and hepatic encephalopathy (HE) (≥ grade II) in patients without preexisting liver disease and is associated with extremely high mortality. The aetiology of ALF is diverse, and its pathogenesis is complex, with excessive inflammatory responses and immune activation considered key drivers of extensive hepatocyte necrosis[1]. Although artificial liver support systems and liver transplantation (LT) have improved outcomes for some patients, the overall global mortality of ALF remains approximately 30%[2]. In China, severe donor shortages critically limit access to transplantation. ALF constitutes only 1.73% of all liver transplant cases[3]. These realities highlight the urgent need for effective medical therapies.

Glucocorticoids (GCs) are a focal point in ALF management because of their potent immunomodulatory and antiinflammatory effects. Preclinical evidence has demonstrated that GCs mitigate immunemediated liver injury through multiple pathways, including by suppressing cytotoxic responses, modulating immune signalling, and inducing inflammatory cell apoptosis[4]. Clinically, early high-dose GC administration has been associated with reduced hepatocyte destruction and improved survival[5]. However, their utility is constrained by heterogeneous treatment responses and an increased risk of infections, underscoring the need for a highly individualized therapeutic strategy. Moreover, determining the optimal timing for LT presents a dilemma: Intervening too early may expose patients who could recover with medical therapy alone to unnecessary surgical risks, whereas delaying evaluation beyond a critical window (e.g., > 5 days) may preclude transplantation because of disease progression[6]. Therefore, early identification of patients with ALF who are most likely to benefit from GC therapy is critical for improving patient outcomes.

Controversy persists regarding the efficacy of GC therapy in ALF. This study aimed to determine whether GC treatment improves transplantfree survival in patients with ALF and to identify factors predictive of treatment response in GCtreated patients, ultimately informing early clinical identification of potential beneficiaries.

MATERIALS AND METHODS

Study population and data collection

This single-center retrospective cohort study was conducted at Beijing YouAn Hospital, Capital Medical University. Hospitalized patients aged 18 years to 84 years who were diagnosed with ALF between September 2010 and December 2021 were consecutively enrolled. The manuscript was prepared according to the STROBE Statement. The diagnosis of ALF followed the Chinese Guidelines for the Diagnosis and Treatment of Liver Failure (2018 edition)[7], defined by acute onset with grade II or higher HE plus at least two of the following: (1) Severe gastrointestinal symptoms; (2) Serum total bilirubin (TBIL) ≥ 10 × upper limit of normal or a daily increase ≥ 17.1 μmol/L; (3) Prothrombin activity (PTA) ≤ 40% or INR ≥ 1.5; and (4) Imaging evidence of progressive liver shrinkage. Patients with concurrent shock, baseline infection, malignancy, or contraindications to GC therapy were excluded. The 28-day survival status was recorded.

Patients assigned to the GC treatment group received intravenous methylprednisolone (1 mg/kg/day), with the initiation and duration of therapy determined by the treating clinicians on the basis of the patient’s clinical course. The primary endpoint was 28-day transplant-free survival, which was determined through medical records and telephone follow-up. For patients who died during hospitalization, the date of death was confirmed via case record retrieval. Patients who were discharged were followed up by phone to determine their condition on day 28. The 28-day period was calculated from the date of admission. Demographic and clinical data, including laboratory parameters (complete blood count, liver and kidney function, and coagulation profile) at baseline and on day 3 of treatment, were extracted from the hospital’s electronic medical records. The model for end-stage liver disease (MELD) score was calculated as 11.2 × ln (INR) + 9.57 × ln (creatinine) + 3.78 × ln (TBIL) + 6.43. The MELD-Na score was calculated as MELD - [0.025 × MELD × (140 - Na)] + 140.

Statistical analysis

Statistical analyses were performed using R software (version 4.1.0), SPSS 23.0 and EasyR (https://www.easyr.cc Solutions, Inc., Shanghai). Continuous variables were compared between two independent groups using Student’s t test or the Mann-Whitney U test, as appropriate. Factors influencing GC response were analysed by logistic regression. The optimal cut-off points for continuous predictive factors were determined by maximizing the Youden index. Patients were then stratified on the basis of these cut-off points for early predictive indicators. Survival outcomes across subgroups were evaluated with Kaplan-Meier curves and compared using the log-rank test. To assess longitudinal changes, we employed generalized estimating equations. Preliminary assessment using histograms and Q-Q plots indicated that TBIL, INR, and PTA were approximately normally distributed on the original scale in our sample (Shapiro-Wilk test P > 0.05). Therefore, analyses were performed on the original scale to maintain clinical interpretability. For all three outcomes, we specified a Gaussian distribution with identity link. We evaluated three candidate correlation structures [independent, exchangeable, AR (1)] using the Quasi-likelihood under Independence Criterion; the exchangeable structure was retained as optimal. Robust sandwich variance estimators were used to ensure valid inference. Model diagnostics included plots of standardized residuals vs fitted values and Q-Q plots, which supported the adequacy of the Gaussian assumption. The primary focus was the group-by-time interaction effect. Two-sided P < 0.05 was considered statistically significant.

Ethics

The study was conducted in accordance with the ethical principles of the Declaration of Helsinki. The protocol was submitted for review, for which approval was obtained from the Ethics Committee of Beijing YouAn Hospital, Capital Medical University (No. LL-2024-139-K).

RESULTS

Corticosteroid therapy improves survival in patients with ALF

A total of 179 patients with ALF were included in the analysis. The most common aetiology was druginduced liver injury (42.46%), followed by viral hepatitis (36.88%). Among drug-induced cases, the three leading causes were traditional Chinese medicine (42.11%), antipyreticanalgesic drugs (21.05%), and antituberculosis agents (13.16%), as detailed in Supplementary Figure 1.

Among the total cohort, 84 patients received GC therapy, while 95 did not. Baseline characteristics of the two groups are presented in Supplementary Table 1. The median age differed significantly between the groups. After adjusting for age, the 28-day survival rate remained significantly greater in the GC-treated group than in the non-GC group (58.33% vs 30.53%, P < 0.001), as shown in Figure 1.

Figure 1 Survival comparison between glucocorticoid-treated and untreated patients with acute liver failure.

Factors influencing treatment response to GC therapy in patients with ALF

Among the total cohort, 84 patients received GC therapy. The timing and duration of treatment are summarized in Figure 2A. Most patients (94%) initiated GC within 1-3 days of disease onset, with the treatment course typically lasting less than one week, as shown in Figure 2A. On the basis of 28-day survival, these 84 patients were stratified into GC responders (survivors, n = 49) and GC non-responders (non-survivors, n = 35). Comparisons between groups revealed that compared with responders, non-responders had significantly worse baseline coagulation, higher blood ammonia levels, and higher MELD and MELD-Na scores, as detailed in Table 1.

Figure 2 Corticosteroid use and biochemical changes in patients with acute liver failure. A: Timing and duration of hormone use in patients with acute liver failure; B: Block diagram showing the temporal trends of serum total bilirubin, international normalized ratio, and prothrombin activity response and nonresponse to corticosteroids in patients with acute liver failure over a period of 3 days. TBIL: Total bilirubin; INR: International normalized; PTA: Prothrombin activity.

Table 1 Comparison of responder and nonresponder groups to corticosteroid therapy for acute liver failure, n (%)/median (25^th-75^th percentiles).

Parameters	Overall (n = 84)	Responders group (n = 49)	Nonresponders group (n = 35)	P value
Age	38.00 (26.75-50.00)	37.00 (28.00-49.00)	41.00 (26.00-56.00)	0.57
Gender (male)	43 (51.2)	24 (49.0)	19 (54.3)	0.63
ALT (U/L)	1869.00 (720.78-3601.50)	1379.00 (506.50-3347.00)	2287.00 (810.50-4062.50)	0.12
AST (U/L)	880.00 (289.50-2225.50)	812.00 (216.00-1680.00)	1342.00 (460.00-3178.40)	0.06
TBIL (μmol/L)	255.85 (172.18-346.75)	249.40 (175.90-327.20)	280.20 (170.55-392.95)	0.55
DBIL (μmol/L)	149.40 (90.55-217.50)	137.60 (93.30-212.40)	163.50 (89.10-219.35)	0.73
Albumin (g/L)	32.35 (28.80-34.90)	32.70 (29.40-35.40)	31.60 (27.85-33.30)	0.04
UREA (mmol/L)	4.14 (2.13-6.61)	4.40 (2.13-6.90)	3.74 (2.34-5.70)	0.50
Cr (μmol/L)	62.50 (48.50-108.14)	62.00 (47.00-108.14)	68.00 (49.50-99.50)	0.51
NH3 (ug/dL)	148.00 (110.00-192.25)	133.00 (93.00-172.01)	171.00 (141.50-259.50)	0.00
Na+ (mmol/L)	136.95 (134.65-139.65)	136.70 (134.70-139.10)	137.20 (134.45-139.95)	0.75
K+ (mmol/L)	3.79 (3.58-4.30)	3.73 (3.53-4.22)	3.84 (3.59-4.36)	0.30
PT (second)	42.35 (28.20-70.58)	30.00 (21.60-56.30)	52.20 (40.65, 88.70)	< 0.001
PTA (%)	17.00 (10.75-30.85)	25.00 (13.00-38.80)	15.00 (7.80-18.00)	< 0.001
INR	3.76 (2.43-5.73)	2.88 (1.89-4.98)	4.55 (3.58-7.22)	0.00
FIB (g/L)	1.04 (0.77-1.55)	1.10 (0.80-1.57)	0.98 (0.73-1.37)	0.45
WBC (109/L)	9.57 (7.29-11.80)	9.40 (7.05-11.52)	10.15 (7.60-13.82)	0.35
NEUT (109/L)	6.20 (5.16-9.22)	5.91 (5.06-8.54)	7.52 (5.49, 11.60)	0.12
HGB (g/L)	125.00 (111.00-142.00)	124.00 (113.00-142.00)	126.00 (111.00-145.00)	0.88
PLT (109/L)	150.55 (68.51)	153.04 (65.77)	147.06 (72.99)	0.70
AST/ALT	0.62 (0.33-1.06)	0.60 (0.31-1.03)	0.67 (0.34-1.12)	0.66
NLR	4.84 (2.85-7.96)	4.07 (2.64-7.16)	5.94 (3.88-10.60)	0.04
MEDL	26.50 (20.75-33.00)	23.00 (18.00-28.00)	32.00 (27.00-37.00)	< 0.001
MELD-Na	28.50 (21.75-34.00)	23.00 (19.00-30.00)	34.00 (29.00-37.50)	< 0.001

ALT: Alanine transferase; AST: Aspartate transferase; TBIL: Total bilirubin; DBIL: Direct bilirubin; Cr: Creatinine; PT: Prothrombin time; PTA: Prothrombin activity; INR: International normalized ratio; NH3: Blood ammonia; FIB: Fibrinogen; WBC: White blood cell; NEUT: Neutrophil; NH3: Blood ammonia; HGB: Haemoglobin; PLT: Platelet; NLR: Neutrophil-lymphocyte ratio; MELD: Model for end-stage liver disease.

In an effort to identify early indicators predictive of the response to GC therapy, longitudinal changes in TBIL, INR, and PTA were analyzed in GC responders and nonresponders at three time points, namely, baseline, day 3 of GC treatment, and 3 days after completion, as shown in Figure 2B. Responders exhibited downwards trajectories in TBIL and INR alongside an upwards trajectory in PTA. We have supplemented this section in the main text. For the indicator of TBIL, there was no significant difference between the two groups at baseline (P = 0.4969). However, after 3 days of hormone treatment, there was a significant difference between the two groups (P = 0.0263), but after hormone therapy, there was still a significant difference between the two groups (P = 0.0011).

While decreasing trends were also observed for MELD and MELD-Na scores, the differences between groups did not reach statistical significance. Subsequent univariate and multivariate analyses revealed baseline PTA, ΔPTA, and blood ammonia level to be independent factors influencing the response to GC treatment (P < 0.05), as detailed in Table 2.

Table 2 Univariate and multivariate analyses of the efficacy of corticosteroid therapy in patients with acute liver failure.

Characteristic	Univariate analysis		Multivariate analysis
	OR (95%CI)	P value	OR (95%CI)	P value
Age	0.84 (0.54-1.29)	0.42	-	-
Gender	1.24 (0.52-2.95)	0.63	-	-
WBC	0.88 (0.57-1.35)	0.55	-	-
NEUT	0.82 (0.53-1.27)	0.37	-	-
HGB	1.20 (0.78-1.85)	0.41	-	-
NLR	0.63 (0.39-1.01)	0.05	-	-
PLT	1.09 (0.70-1.69)	0.69	-	-
ALT	0.68 (0.43-1.08)	0.10	-	-
AST	0.51 (0.24-1.08)	0.08	-	-
AST/ALT	0.96 (0.62-1.48)	0.85	-	-
TBIL	0.86 (0.56-1.33)	0.50	-	-
UREA	0.77 (0.42-1.43)	0.42	-	-
Cr	0.88 (0.57-1.36)	0.58	-	-
INR	0.38 (0.16-0.92)	0.03	1.28 (0.60-2.70)	0.53
PT	0.39 (0.20-0.77)	0.01	0.83 (0.39-1.77)	0.63
PTA	3.24 (1.77-5.94)	< 0.001	4.21 (1.09-16.34)	0.04
ΔPTA	4.96 (2.39-10.33)	< 0.001	7.78 (2.79-21.68)	< 0.001
Blood ammonia	0.46 (0.27-0.79)	0.01	0.42 (0.19-0.96)	0.04

WBC: White blood cell; NEUT: Neutrophil; HGB: Haemoglobin; NLR: Neutrophil-lymphocyte ratio; PLT: Platelet; ALT: Alanine transferase; AST: Aspartate transferase; TBIL: Total bilirubin; UREA: Blood urea; Cr: Creatinine; INR: International normalized ratio; PT: Prothrombin Time; PTA: Prothrombin activity; ΔPTA: Prothrombin activity on day 3-baseline prothrombin activity; NH3: Blood ammonia; OR: Odds ratio.

A predictive model for GC treatment response in patients with ALF

The optimal cut-off for ΔPTA was 4.5%, demonstrating the highest predictive performance, with an area under the receiver operating characteristic (AUROC) of 0.83 (sensitivity 85.71%, specificity 65.7%). This was followed by the baseline MELD score (cut-off of 28.5; AUROC of 0.78; sensitivity of 79.59%, and specificity of 71.43%), baseline PTA (cut-off of 22.85%; AUROC of 0.76; sensitivity of 55.1%, and specificity of 88.57%), and blood ammonia concentration (cut-off of 135.5 μg/dL; AUROC of 0.69; sensitivity of 53.06%, and specificity of 80.0%), as detailed in Figure 3 and Supplementary Table 2. In contrast, ΔMELD and ΔMELD-Na had poor predictive capacity [area under the curve (AUC): 0.59] but the difference was not statistically significant Kaplan-Meier survival analysis based on these cut-off values revealed significant differences in 28-day survival between risk-stratified groups (P < 0.01), confirming their discriminatory value, as shown in Figure 4.

Figure 3 Receiver operating characteristic curves for predicting performance of various indicators. AUC: Area under the curve; MELD: Model for end-stage liver disease; PTA: Prothrombin activity.

Figure 4 Kaplan-Meier curves showing cumulative survival in the acute liver failure cohort stratified by delta prothrombin activity, model for end-stage liver disease, model for end-stage liver disease-Na, prothrombin activity, blood ammonia, delta model for end-stage liver disease, and delta model for end-stage liver disease-Na. Significance was estimated using the log-rank test. PTA: Prothrombin activity; MELD: Model for end-stage liver disease.

Predictive accuracy was further improved by combining ΔPTA, MELD, and blood ammonia. In two-indicator models, patients meeting both favourable cut-offs had a GC response rate of 85%, whereas those meeting neither cut-off were 5.88%, as shown in Supplementary Table 3 and Figure 5. The three-indicator model provided even finer stratification: Patients who fulfilled all three criteria (ΔPTA ≥ 4.5%, MELD < 28.5, and blood ammonia concentration < 135.5 μg/dL) had a response rate of 95.24%, compared with only 7.14% for those who did not meet all three criteria, as shown in Supplementary Table 3. After bootstrap internal validation, the optimism-corrected AUC was 0.826, with a corrected sensitivity of 69.1% and specificity of 85.8%. The apparent performance (AUC = 0.977) was likely over-optimistic due to cutoff optimization within the same dataset, a known limitation of such approaches. The validated performance (AUC = 0.826) still represents good discrimination and supports the clinical utility of this triple-indicator model. Therefore, this triple-indicator model more precisely identifies patients with a low likelihood of benefit, offering greater clinical utility.

Figure 5 Kaplan-Meier curves showing cumulative survival in the acute liver failure cohort stratified by the combination of delta prothrombin activity, model for end-stage liver disease, and blood ammonia. Significance was estimated using the log-rank test. PTA: Prothrombin activity; MELD: Model for end-stage liver disease; OR: Odds ratio.

DISCUSSION

Although ALF has a low global incidence, it is characterized by rapid disease progression, extremely high early mortality, and an unpredictable clinical prognosis. Current management strategies-including hepatoprotective agents, plasma exchange, haemodiafiltration, and LT-aim to improve survival. However, these approaches (with the exception of LT) yield limited therapeutic benefits. Prior to the widespread adoption of LT, ALF was associated with a mortality rate of 80%-85%[8]. In recent years, advances in intensive care management have substantially improved ALF survival[2]. Even so, approximately half of patients still die without access to emergency LT, underscoring the critical role of LT in ALF management. However, LT has notable limitations: It is a costly intervention requiring major surgery and lifelong immunosuppression. Perioperative care for patients with ALF is particularly challenging, and their survival rates remain consistently lower than those of patients undergoing elective LT. Compounding these issues, organ donor shortages further restrict access to LT for the majority of patients with ALF[6].

Recent research has revealed that sustained and extensive hepatic inflammation is among the core pathological mechanisms driving the development and progression of ALF[9], providing a theoretical rationale for the use of GCs to suppress excessive immune-mediated inflammation. A prior study demonstrated that early GC administration can inhibit immune response-induced hepatocellular injury and microvascular dysfunction, thereby preserving liver regenerative capacity[10]. However, the clinical value of GCs in the management of patients with ALF remains controversial. Multiple randomized controlled trials (RCTs) conducted in the 20^th century failed to confirm that GCs improve the survival of patients with ALF[11-13]. Similarly, earlier RCTs and observational studies from the United States and Europe also reported no significant survival benefit with GC use[11,13]. In contrast, our study revealed that GCs significantly improved the 28-day survival rate of patients with ALF, which aligns with findings from a Japanese study[5]. Notably, our study included a relatively larger sample size, whereas the United States and European RCTs referenced previously had a smaller sample (e.g., one key RCT enrolled only 40 patients[12]).

In our study, the main aetiologies of ALF were drug-induced (primarily herbal medicine) and hepatitis B virus infection. GC therapy was usually initiated after admission following the exclusion of contraindications, at a dosage of 0.5-1.0 mg/kg of body weight per day for an average duration of 3-7 days. In contrast, previous landmark studies (such as the 1993 EASL trial) primarily enrolled patients with viral hepatitis or ALF of indeterminate cause, and their GC regimens (timing, dosage, and duration) were not specified in detail. The low survival rates reported in those studies also suggest that their cohorts may have had more advanced disease. These factors, along with potential racial/ethnic differences, may partly explain the discrepancy between our findings and earlier results.

In recent years, several Japanese studies have consistently reported that early, high-dose, short-course GC therapy (e.g., 1000 mg/day methylprednisolone for 3 consecutive days) markedly increased survival of patients with ALF[5,14]. An analysis of our study population characteristics revealed that GC treatment was initiated within 1-3 days of hospital admission, corresponding to the early stage of ALF-which is consistent with the timing reported in the study by Yasui et al[15].

The immunosuppressive effects of GC inevitably raise concerns regarding infection risk. However, evidence linking GC therapy to a higher incidence of infections in ALF remains inconclusive. For instance, Yasui et al[15] reported that the use of GCs did not significantly increase infection rates. Notably, in patients who developed infections, the interval between GC initiation and LT was prolonged, with a reported median time to infection onset of 15 days[15]. In our cohort study, compared to the non-GC group, patients receiving GC therapy had a significantly lower incidence of gastrointestinal bleeding (8.3% vs 21.1%, P = 0.018) and respiratory tract infection (25.0% vs 38.9%, P = 0.047). This finding is counterintuitive, given the conventional understanding that corticosteroids increase the risk of bleeding and infection. Several factors may explain this observation: Although baseline characteristics were generally balanced between the groups (Supplementary Table 1), patients who received GC were younger. Additionally, the GC group may have experienced faster recovery and shorter hospitalization. Differential mortality rates could also affect the observed complication rates. Furthermore, unmeasured confounding factors may still influence the outcomes. We emphasize that no causal inferences can be drawn from these associations, and the findings should be interpreted with caution.

This shortcourse approach aligns with the concept proposed by Hisanaga et al[16], who suggested that GCs could serve as a transitional or adjuvant therapy to sustain patients awaiting LT or spontaneous recovery, provided that treatment duration is minimized. These observations collectively suggest that the efficacy and safety of GC therapy in patients with ALF may critically depend on timely initiation and precise control of treatment duration. A “shortcourse, rapidreview” strategy appears to be a relatively safe and practical option. In summary, while the conflicting results in previous studies may be attributed to factors such as limited sample size, delayed GC initiation, or prolonged treatment, our findings highlight the importance of optimized treatment timing and duration in maximizing therapeutic benefit while minimizing risks.

The higher DAMA rate in the GC group represents an important source of potential bias, as DAMA is a complex phenomenon influenced by multiple factors and should not be equated with disease severity. While the worst-case analysis yielded a P value of 0.059, the consistent direction of benefit across all analyses-with absolute survival differences ranging from 14.1% to 49.1%-suggests that the observed treatment effect is unlikely to be entirely artifactual. The finding under the most extreme assumption did not reach statistical significance, but when considered alongside the highly significant results in more plausible scenarios, a true treatment effect cannot be ruled out.

Determining the optimal timing for LT in patients with ALF poses a significant clinical challenge. ALF is a highly dynamic condition, and early outcome prediction is difficult because of rapid clinical evolution. Excessive delay in decision-making may preclude LT as the disease progresses, leading to preventable mortality. Conversely, excessively early transplantation can result in patients who might otherwise recover from medical therapy, such as unnecessary major surgery, lifelong immunosuppression, and substantial resource expenditure, and may deprive other suitable candidates for a graft. The clinical window for LT in patients with ALF is narrow and is typically limited to 2-5 days after admission. Missing this window can result in loss of the survival opportunity afforded by transplantation[6]. Therefore, early application of reliable prognostic models is critical for optimizing outcomes. The MELD score and HE grade are key indicators of liver failure severity and prognosis. A MELD score ≥ 35 or grade 4 HE generally predicts end-stage disease with a poor prognosis, largely independent of therapeutic intervention. Consistent with this, the team led by Zhao et al[17] reported extremely low short-term survival in patients with MELD ≥ 35 and HE grade 4, irrespective of steroid use. Notably, among patients with MELD scores of 25-35 and no HE or HE grades 1-3, survival was significantly greater in those receiving GC than in those who did not[17].

Our findings also underscore the significant prognostic value of the MELD score and blood ammonia level in determining the response to GC therapy in patients with ALF. However, ALF is a highly dynamic condition, and prognostic variables measured at admission may evolve, leading to divergent clinical trajectories. Continuous monitoring of key predictors could therefore provide valuable insights for tracking disease progression. For instance, in acetaminopheninduced ALF, patients whose prothrombin time continued to deteriorate between 3 days and 4 days post-overdose had higher mortality than those whose prothrombin time improved. Similarly, rising serum alpha-fetoprotein levels within the first 3 days of admission have been associated with better survival, and arterial blood ammonia levels correlate closely with progressive cerebral oedema in patients with ALF[18,19]. These observations collectively suggest that dynamic changes in predictive variables are more informative than baseline values alone. In support of this concept, the Kumar et al[18] developed an early dynamic model for ALF based on the evolution of multiple indicators over the first three hospital days. This model demonstrated excellent discriminative ability, with an AUROC of 0.91 in the derivation cohort and one of 0.92 in the validation cohort[18]. In line with this approach, we analysed the utility of the change in PTA from baseline to day 3 of GC therapy (ΔPTA), combined with baseline MELD score and blood ammonia concentration. Our model-defined by ΔPTA ≥ 4.5%, baseline MELD < 28.5, and blood ammonia concentration < 135.5 μg/dL-can effectively guide the decision to continue GC treatment or prioritize LT. Compared with static indicators such as HE grade or ALT level alone, it had superior early discriminative performance[17]. This finding aligns with the report by Kuwano et al[19] that dynamic improvement in PTA predicts a favourable prognosis. Our study further refines this by specifying the optimal evaluation time point and critical thresholds, thereby enhancing clinical operability.

In conclusion, the dynamic stratification model established in this study enables the early identification (by day 3) of GC responders. This approach may help avoid unnecessary prolonged GC therapy and its associated risks (e.g., infection) in nonresponders while simultaneously minimizing either futile LT or missed transplant opportunities due to delayed assessment. We propose that a “short-course, rapid-assessment” GC strategy coupled with an effective early evaluation protocol could be a key determinant of successful GC therapy in patients with ALF.

In summary, the results of this study support the role of early, short-term GC therapy in improving survival for a subset of patients with ALF. We further propose, for the first time, a dynamic decisionmaking model centred on day-3 ΔPTA. This model provides a practical framework for individualized immunomodulatory management in patients with ALF and holds significant potential for clinical translation.

Although our three-parameter model demonstrated excellent predictive ability (AUC of 0.977), its clinical implementation requires consideration of several practical factors. First, rapid blood ammonia testing is crucial for timely decision-making but may not be available in all medical institutions. Second, continuous PTA monitoring to calculate ΔPTA requires close clinical observation and laboratory support, potentially increasing resource demands. Third, the model requires simultaneous availability of all three parameters within a short timeframe (e.g., 3-7 days after GC initiation), which may be challenging in emergency settings. To address these limitations, future studies should explore simplified scoring systems, point-of-care testing technologies, and clinical decision support tools. Implementation research across diverse health care settings will help determine the broader applicability of the model. The apparent predictive performance (AUC = 0.977) was likely over optimistic due to the derivation of predictor cutoff within the same dataset. After bootstrap internal validation with 1000 samples, the optimistic corrected AUC was 0.826, which still represents good predictive performance. However, external validation in independent cohorts is necessary to confirm the model’s generality. The graphical was shown in Supplementary Figure 2. The mortality rate of patients with ALF is high, and the use of GC can significantly reduce the mortality rate of patients. A new 3-day diagnostic model was established for patients with ALF receiving GC therapy. On the third day of treatment, the responders consistently showed ΔPTA ≥ 4.5% and a simultaneous decrease in bilirubin/INR. The composite predictive factors of ΔPTA ≥ 4.5%, baseline MELD < 28.5, and blood ammonia < 135.5 μg/dL achieved a sensitivity of 95.24% in identifying treatment responders, making early clinical decisions between continuing steroid therapy or prioritizing transplant evaluation possible.

This study had several limitations. First, the single-centre retrospective design may have introduced selection bias. Second, the absence of external validation limits the generalizability of our predictive model. Third, residual confounding from unmeasured factors, including care setting variations and discharge against medical advice, cannot be excluded. Finally, given the lack of international consensus on the use of GCs in ALF, our results require validation in prospective multicentre studies.

CONCLUSION

GC treatment significantly increased the 28-day survival rate of patients with ALF without increasing the incidence of adverse events. ΔPTA, baseline MELD score, and baseline blood ammonia level are independent early predictors of GC treatment response. A multivariable prediction model incorporating these three indicators can accurately identify patients who are likely to benefit from GC therapy while also effectively screening highrisk nonresponders, who may then be prioritized for timely LT assessment.

ACKNOWLEDGEMENTS

We express our gratitude to our colleagues who provided us with assistance.