Establishing a predictive nomogram for 21‑day transplant-free survival in drug-induced liver failure

Abstract

Background

The high prevalence of drug-induced liver failure (DILF) have drawn great attention from clinicians.

Aim

To further delineate the clinical features of DILF and develop an easily applicable nomogram, based on readily-discernable clinical data, to predict transplant-free survival (TFS) at different time points.

Methods

202 DILF patients were enrolled between January 2016 and December 2022, and were followed up from DILF diagnosis to death, liver transplantation, or 91 days afterward, whichever came first. The primary endpoint, though, was 21-day TFS. Clinical data was collected from all patients, and independent risk factors associated with death/liver transplantation was identified using both uni- and multi-variate Cox regression analyses.

Results

Independent risk factors incorporated into the predictive nomogram are neutrophils (HR = 1.148, 95% CI = 1.048–1.257), prothrombin time (HR = 1.048, 95% CI = 1.017–1.080), albumin (HR = 0.880, 95% CI = 0.823–0.941), acute kidney injury (HR = 2.487, 95% CI = 1.134–5.452), and hepatic encephalopathy (HR = 3.378, 95% CI = 1.744–6.543). The resulting nomogram was highly predictive, with an area under the curve of 0.947 for 21-day TFS.

Conclusions

Compared to existing models, such as the Model for End-Stage Liver Disease score, the predictive nomogram is more accurate, only requires easily-measurable clinical and laboratory metrics, as well as being able to directly calculate TFS at various time points.

1. Introduction

Drug-induced liver injury (DILI) refers to toxic liver damage caused by a variety of medicinal products, such as non-prescribed and prescribed natural remedies, chemical drugs, biological agents, nutraceuticals, as well as their excipients and metabolites. Over time, DILI could eventually progress to drug-induced liver failure (DILF), which is characterized by increasing jaundice, dysfunctional coagulation, hepatic encephalopathy (HE), and ascites. Furthermore, DILF represents a significant compromise in hepatic function, with respect to synthesis, detoxification, metabolism, immunity, and biotransformation, thereby manifesting as severe hepatic impairment [1]. In fact, DILF serves as the predominant form of acute liver failure (ALF) in Western countries, comprising 10–52% of all cases [2]; it is most commonly caused by acetaminophen poisoning. Moreover, DILI has become the second most prevalent cause of liver disease hospitalization in China, at a rate of 23.8 per 100 000 individuals, as documented by a nationwide retrospective study of 25 927 DILI patients, admitted to 308 hospitals, between 2012 and 2014 [3]. Within this cohort, 280 individuals (1.08%) developed DILF. Additionally, another Chinese clinical study of 21 789 patients found that DILI was responsible for 50% of ALF cases [4]. However, there is still a paucity of data regarding DILF prevalence in Asian countries, particularly in China, even with significant improvements in our comprehension of DILF pathogenesis, complications, and prognoses. In this study, we aimed to further delineate the clinical features of DILF, as well as develop an easily applicable nomogram to predict transplant-free survival (TFS) at different time points. Our nomogram was able to predict TFS at 21, 60, and 90 days, and was more accurate than currently-existing analytical methods, such as Model for End-Stage Liver Disease (MELD) score, thereby potentially serving as a highly-accurate novel diagnostic strategy for determining DILF prognoses.

2. Materials and methods

2.1. Study design, participants, and data collection

Data was retrospectively collected from 202 DILF patients, of which 109 received artificial liver therapy, who were admitted to the Department of Infection of the First Affiliated Hospital of Nanchang University between January 2016 and December 2022. All included indicators were measured at the time of the patient’s first presentation to hospital. The following inclusion criteria was applied: (1) Both the diagnostic criteria for DILI, within the Guidelines for the Diagnosis and Treatment of Pharmacological Liver Injury (2017), and for liver failure, within the Guidelines for the Diagnosis and Treatment of Liver Failure (2018), were followed. (2) Roussel Uclaf Causality Assessment Method (RUCAM) score ≥6. (3) Complete clinical data was available. (4) Age ≥18 years. Patients were excluded from the study based on the following criteria: Combination of DILF with (1) Acute viral hepatitis infection, or other viruses, such as cytomegalovirus and Epstein-Barr, (2) Other chronic liver diseases, such as autoimmune, genetic metabolic, etc., (3) Other serious physical/mental diseases, including cardiac, pulmonary, renal, vascular, neurological, metabolic, immunodeficiency, or neoplasms. Additionally, DILF patients with (4) Recent use of immunosuppressive agents (ex. hormone inhibitors); (5) Pregnant or lactating, (6) RUCAM < 6, (7) Liver injury prior to drug use, or unclear relationship present between livery injury and drug usage, (8) Incomplete clinical information.

Clinical characteristics were summarized from all patients, ranging from gender, age, hospitalization length, previous medication use, presence of pre-existing diseases, MELD scores, DILI type, signs, symptoms, treatments, complications, regressions, as well as routine blood test results in terms of biochemistry, coagulation, and cytokines. MELD scores were calculated using the following equation: [3.78 × ln(bilirubin in mg/dL)+11.2 × ln(INR)+9.57 × ln(creatinine in mg/dL)+6.43] [5]. Additionally, the presence of acute kidney injury (AKI) was diagnosed if an increase in serum creatinine ≥0.3 mg/dL occurred within 48 h, or by ≥50% within the preceding 7 days [6, 7]. Afterwards, risk factors associated with death or liver transplantation were identified by both uni- and multivariate Cox regression analyses, then included in the nomogram to create a predictive model. For RUCAM scores, they fall into 5 different categorizations, based on the likelihood of DILI: Highly probable (>8), Very probable (6–8), Probable (3–5), Unlikely (1–2), and Able to be ruled out (0); DILI is diagnosed based on RUCAM ≥ 6. As for ALF, it was defined by the following criteria: (1) Evidence of coagulopathy, based on international normalized ratio (INR)≥1.5, (2) HE, of any grade being present, defined as neuropsychiatric abnormalities during the course of liver disease, including cognitive, affective/emotional, behavioral, and bioregulatory domain involvement, based on the West Haven Criteria [8]. (3) Acute illness onset at <26 weeks after hepatic insult, and (4) No preexisting cirrhosis. Patient follow-up took place from the date of diagnosis to 91 days after, death, or liver transplantation, whichever came first. This is due to the fact that the onset time for acute liver injury is <90 days, while for chronic liver injury, it is >90 days.

Furthermore, DILI was classified into 3 different clinical types, based on serum enzyme characteristics outlined in the Guidelines for the Diagnosis and Treatment of DILI [9, 10]: (1) Hepatocellular injury, in which serum alanine aminotransferase (ALT) was ≥3 times the upper limit of normal (ULN), and the R ratio, which is (measured ALT/ALT ULN)÷(measured alkaline phosphatase [ALP]/ALP ULN), being ≥5, (2) Cholestatic injury, in which ALP ≥ 2 × ULN, and R ≤ 2, and 3) Mixed injury type, in which ALT ≥ 3 × ULN, ALP ≥ 2 × ULN, and 2<R < 5.

During follow-up, DILF patients were divided into survival and death/liver transplantation groups, based on prognostic criteria defined in the Guidelines for the Diagnosis and Treatment of Liver Failure (2018). Survival was defined as having significant improvements in clinical symptoms such as fatigue, nausea, abdominal distension, bleeding tendency, jaundice, ascites, etc., as well as for hepatic functional indices, ranging from total bilirubin (TBIL), which has decreased to <5× normal, prothrombin time activity (PTA)>40%, or INR < 1.6. Additionally, HE has disappeared. By contrast, death/liver transplantation, included patients who died or underwent liver transplantation during hospitalization and follow-up.

The protocol was performed in accordance with the guidelines outlined in the Declaration of Helsinki and was approved by the Ethics Committee of The First Affiliated Hospital of Nanchang University. Since the study was a retrospective study, most of the study subjects have died or lost contacts, and all statistics were anonymous, so the Ethics Committee of The First Affiliated Hospital of Nanchang University agreed to waive the need for informed consent.

2.3. Statistical analyses and establishing the predictive nomogram

Data were presented as mean ± standard deviation (SD) if they were continuous variables with normal distribution. For continuous variables that were not normally distributed, they were presented as medians (interquartile range [IQR]). Categorical data were expressed as counts with percentages.

All data analyses were conducted using R software (version 4.2.1; R Foundation for Statistical Computing, Vienna, Austria; http://www.r-project.org/). Non-parametric and parametric continuous variables were compared using, respectively, Mann–Whitney U and Student’s t tests, while categorical variables were analyzed with chi-square test or Fisher’s exact test. TFS between subgroups, stratified based on risk factors, were compared using Kaplan–Meier and log-rank tests.

Both uni- and multi-variate Cox regression analyses were used to identify independent risk factors associated with death/liver transplantation in the internal cohort, in which all risk factors with p < 0.05 under the univariate analysis were included in the multivariate analysis. Hazard ratios (HR) and 95% confidence intervals (CI) were determined for those independent risk factors, and incorporated into the predictive nomogram, via proportionally converting each multivariate regression coefficient on a 0–100-point scale, using the replot package of R. Therefore, different values for independent variables, in the form of points, are added to derive a total point value, which was converted to a corresponding predicted probability for TFS, at different time points.

The predictive performance of the nomogram was then measured using time-dependent receiver operating characteristic curve (ROC), generated by the time ROC package in R, and the resulting calibration curve was plotted using bootstrap sampling to reduce overfitting bias; the closer the calibration curve is to the diagonal line, the higher the predictive accuracy of the nomogram. Area under the ROC curve (AUROC), and concordance index (C-index) were also measured to assess the accuracy and consistency of the predictive nomogram. The net benefit of the nomogram for clinical decision-making was determined using decision curve analysis (DCA), while its discrimination and calibration were assessed by bootstrapping with 1000 replicates[11]. All statistical analyses were two-tailed, and p < 0.05 was considered statistically significant.

3. Results

3.1. Patient baseline characteristics, clinical profiles, and drugs implicated in ALF

Table 1 summarizes the baseline characteristics and clinical profiles for the 202 DILF patients. Mean patient age was 48.8 ± 15.7 years, and 72 (50.7%) were female. The most common etiology behind DILF was previous usage of herbal and dietary supplements (HDS), at 59.2%; other drugs implicated in DILF include ‘Other’ (ex. anti-tuberculosis drugs, antibiotics), at 30.3%, programmed cell death protein 1 (PD1)/programmed death-ligand 1 (PDL1), at 4.9%, and acetaminophen at 5.6%. The most common cause of liver injury was hepatocellular, with R > 5. With respect to complications, infections and HE were the most prevalent.

Table 1.

Baseline clinical features of 202 patients with drug-induced liver failure (DILF).

	Training cohort				Validation cohort n = 60	P **
	Total n = 142	Survival n = 103	Death/Liver transplant n = 39	P *
Female (n [%))	72 (50.7)	53 (51.5)	19 (48.7)	0.771	22 (36.7)	0.890
Age (years)	48.8 ± 15.7	47.6 ± 15.6	51.8 ± 15.9	0.160	52.9 ± 12.6	0.073
Drugs implicated, n (%)				0.710		0.307
Acetaminophen	8 (5.6)	7 (6.8)	1 (2.6)		6 (10.0)
PD1/PDL1	7 (4.9)	5 (4.9)	2 (5.1)		6 (10.0)
Herbal and dietary supplements	84 (59.2)	62 (60.2)	22 (56.4)		34 (56.7)
Other	43 (30.3)	29 (28.2)	14 (35.9)		14 (23.3)
R value, n (%)				0.069		0.258
Hepatocellular	72 (50.7)	58 (56.3)	14 (35.9)		38 (63.3)
Cholestatic	35 (24.6)	24 (23.3)	11 (28.2)		11 (18.3)
Mixed	35 (24.6)	21 (20.4)	14 (35.9)		11 (18.3)
Baseline indices (Interquartile ratio [IQR])
White blood cells,109/L (3.5–9.5)	5.7 (4.6, 8.0)	5.4 (4.5, 6.9)	7.1 (4.7, 9.2)	0.014	5.9 (4.7, 9.6)	0.073
Hemoglobin, g/L (115–150)	124.0 (113.0, 135.0)	124.0 (117.0, 137.0)	116 (102.0, 134.0)	0.031	127.5 (115.3, 136.0)	0.929
Neutrophils,109/L (1.8–6.3)	3.6 (2.6, 5.1)	3.3 (2.4, 4.3)	4.3 (2.8,6.60)	0.004	3.7 (2.8, 5.6)	0.139
Platelets, 109/L (125–350)	195.0 (132.5, 272.25)	210.0 (155.0, 279.0)	134.0 (108.0, 204.0)	0.001	188.5 (95.3, 258.0)	0.775
INR (0.85–1.15)	1.3 (1.0,1.8)	1.2 (1.0, 1.5)	1.9 (1.5, 2.9)	<0.001	1.3 (1.1, 1.9)	0.689
Prothrombin time (9.8–12.1)	14.5 (12.0, 20.3)	13.3 (11.7, 16.1)	21.6 (16.4, 31.2)	<0.001	14.5 (12.2, 21.5)	0.865
Albumin, g/L (40–55)	33.9 (31.1, 37.1)	35.0 (31.9, 38.4)	31.9 (29.1, 33.5)	<0.001	33.8 (29.3, 37.1)	0.716
ALT, U/L (7–40)	374.5 (135.6, 1090.6)	490.7 (149.4, 1180.0)	241.1 (88.0, 663.0)	0.124	488.0 (234.3, 1221.4)	0.676
AST, U/L (13–35)	311.0 (93.0, 762.9)	334.0 (95.5, 765.7)	198.0 (73.0, 716.0)	0.609	326.1 (172.7, 818.9)	0.609
Total bilirubin, μmol/L (0–21)	243.3 (161.7, 325.3)	235.5 (143.1, 310.6)	295.3 (194.4,405.9)	0.005	206.5 (115.5, 356.0)	0.466
Direct bilirubin, μmol/L (0–4)	150.6 (93.6, 220.2)	141.4 (93.2, 200.6)	215.6 (128.0, 270.1)	0.001	144.3 (78.2, 220.0)	0.525
ALP, U/L (50–135)	176.5 (145.8, 223.0)	177.0 (146.2, 234.0)	173.0 (143.0, 209.0)	0.710	179.1 (132.5, 246.2)	0.143
GGT, U/L (7–45)	146.0 (81.8, 272.0)	150.0 (86.0, 280.0)	133.0 (70.0, 216.0)	0.171	151.5 (80.5, 323.6)	0.141
Creatinine, μmol/L (41–81)	61.4 (49.0, 72.3)	59.7 (46.7, 70.7)	65.0 (53.0, 82.7)	0.044	59.3 (47.3, 71.8)	0.743
Alpha–fetoprotein, ng/mL (<7.0)	8.3 (2.8, 33.8)	7.5 (3.0, 26.1)	9.2 (2.2, 84.5)	0.631	7.1 (2.6, 43.5)	0.158
Complications (n, %)
Ascites	11 (7.7)	2 (1.9)	9 (23.1)	<0.001	12 (20.0)	0.016
Infection	82 (57.7)	55 (53.4)	27 (69.2)	0.026	34 (56.7)	0.504
Acute kidney injury	11 (7.7)	0 (0.0)	11 (28.2)	<0.001	6 (10.0)	0.588
Hepatic encephalopathy	19 (13.4)	4 (3.9)	15 (38.5)	<0.001	12 (20.0)	0.285
Disease severity score (IQR)
Model for end–stage liver disease	19.0 (17.0, 24.0)	18.0 (16.0, 22.0)	25.0 (23.0, 31.0)	<0.001	20.0 (16.0, 25.0)	0.968
Treatment (n, %)
N–acetylcysteine	41 (28.9)	33 (32.0)	8 (20.5)	0.292	17 (28.3)	0.541
Glucocorticoid	19 (13.4)	11 (10.7)	8 (20.5)	0.095	10 (16.7)	0.520
Artificial liver support system	74 (52.1)	41 (39.8)	33 (84.6)	<0.001	35 (58.3)	0.443

ALP: alkaline phosphatase; ALT: alanine aminotransferase; AST: aspartate aminotransferase; GGT: gamma-glutamyl transpeptidase; INR: international normalized ratio; PD1: programmed cell death protein 1; PDL1: programmed death-ligand 1.

^*Survival versus Death/Liver transplant subgroups.

^**Training versus Validation datasets.

Patients were random divided into 2 cohorts: 142 in training, and 60 in the validation cohort, and clinical characteristics were largely similar between them; no statistically significant differences were present in terms of gender, age, implicated drugs, R value, white blood cell (WBC) count, hemoglobin (HGB), ALT, aspartate aminotransferase (AST), ALP, gamma-glutamyl transpeptidase (GGT), and alpha-fetoprotein (AFP) measurements.

The training cohort was also sub-divided into 2 groups, survival, and death/LT, in which death/LT patients had significantly lower platelet (PLT) and albumin (ALB) levels, along with higher neutrophil, INR, prothrombin time (PT), TBIL, direct bilirubin (DBIL), and creatinine (CR). All of these were associated with a higher risk of developing complications.

3.2. Comparison of 21-day TFS between different subgroups

Kaplan-Meier analyses of 21-day TFS found that individuals treated with glucocorticoid had significantly lower survival rates, at 79.31%, compared to untreated ones, at 97.11% (p = 0.013, Figure 1A). Lower 21-day TFS was also observed for patients treated with artificial liver support systems (ALSS), at 92.70%, compared to those who were not, at 96.8% (p < 0.0001, Figure 1B). By contrast, patients treated with N-acetylcysteine (NAC) had significantly higher 21-day TFS, at 94.83%, than for untreated individuals, at 91.30% (p = 0.0029, Figure 1C). Additionally, 21-day TFS for DILF individuals, induced by HDS, was higher than for non-HDS-induced, at respectively, 97.46% and 90.48% (p = 0.033, Figure 1D). Furthermore, higher 21-day TFS was present among patients without HE, at 99.42%, compared to HE patients, at 67.74% (p < 0.0001, Figure 1E).

Figure 1.

Kaplan-Meier analysis of 21-day transplant-free survival (TFS) among drug-induced liver failure (DILF) patients, based on (A) glucocorticoid (B) artificial liver support system (ALSS), and (C) N-acetylcysteine (NAC) usage, as well as (D) whether DILF was induced by herbal and dietary supplements (HDS) versus other medications, and (E) whether DILF patients had hepatic encephalopathy (HE). Log-rank test was used to compare cumulative TFS survival across groups. ‘0’ in X-axis indicates time of DILF diagnosis.

3.3. Development and calibration of the predictive nomogram

Uni- and multi-variate Cox regression analyses were conducted to identify statistically significant patient parameters for the predictive nomogram, as shown in Table 2. These variables included neutrophils (HR = 1.148, 95% CI = 1.048–1.257), PT (HR = 1.048, 95% CI = 1.017–1.080), ALB (HR = 0.880, 95% CI = 0.823–0.941), AKI (HR = 2.487, 95% CI = 1.134–5.452), and HE (HR = 3.378, 95% CI = 1.744–6.543), which were all incorporated into the predictive nomogram (Figure 2).

Table 2.

Uni- and multivariate Cox regression analyses for DILF progression (Hazard ratio [HR], 95% confidence interval [CI]).

Parameters	Univariate HR (95% CI)	P	Multivariate HR (95% CI)	P
Female	1.163 (0.684, 1.977)	0.577
Age (years)	1.011 (0.993, 1.030)	0.238
R value
○ Hepatocellular	–	–
○ Cholestatic	1.334 (0.683, 2.606)	0.398
○ Mixed	2.016 (1.092, 3.719)	0.025
Baseline indices
○ White blood cells, 109/L	1.124 (1.047, 1.207)	0.001
○ Hemoglobin, g/L	0.980 (0.965, 0.994)	0.005
○ Neutrophils, 109/L	1.172 (1.084, 1.267)	<0.001	1.148 (1.048, 1.257)	0.003
○ Platelets, 109/L	0.996 (0.993, 0.999)	0.008
○ International Normalized Ratio	2.714 (2.109, 3.493)	<0.001
○ Prothombin time, s	1.092 (1.069, 1.117)	<0.001	1.048 (1.017, 1.080)	0.002
○ Albumin, g/L	0.867 (0.823, 0.914)	<0.001	0.880 (0.823, 0.941)	<0.001
○ Alanine aminotransferase, U/L	1.000 (0.999, 1.000)	0.069
○ Aspartate aminotransferase, U/L	1.000 (0.999, 1.000)	0.339
○ Total bilirubin, μmol/L	1.003 (1.001, 1.005)	<0.001
○ Alkaline phosphatase, U/L	1.000 (0.999, 1.001)	0.785
○ Creatinine, μmol/L	1.008 (1.000, 1.016)	0.052
○ Alpha-fetoprotein, ng/mL	1.001 (1.000, 1.002)	0.111
Complications
○ Ascites	4.913 (2.721, 8.871)	<0.001
○ Infection	1.409 (0.803, 2.470)	0.232
○ Acute kidney injury	6.006 (3.150, 11.451)	<0.001	2.487(1.134,5.452)	0.023
○ Hepatic encephalopathy	8.268 (4.808, 14.215)	<0.001	3.378(1.744,6.543)	<0.001
Treatment
○ N-acetylcysteine	0.363 (0.181, 0.729)	0.004
○ Glucocorticoid	2.204 (1.158, 4.195)	0.016
○ Artificial liver support system	4.038 (2.079, 7.844)	<0.001

Figure 2.

Predictive nomogram for determining 21-, 60-, and 90-day TFS among DILF patients, based on 5 parameters: Neutrophil counts, prothrombin time (PT), albumin levels, presence of acute kidney injury (AKI), and HE.

The AUROC for the nomogram, with respect to predicting 21-day TFS was 0.947, indicating that it had high predictive value. This was further supported by calibration curves between training and validation cohorts, which showed the well calibration by 1000 times bootstrap sampling. For both datasets, there was a close correspondence between 21-day TFS probabilities predicted by the nomogram versus the actual probabilities (Figure 3A-B). Additionally, DCA curves for both datasets showed that the nomogram had great clinical utility, compared to assuming that all patients would have TFS after 21 days, or that none of the patients would (Figure 3C-D).

Figure 3.

Calibration curves of the predictive nomogram for 21-day TFS among (A) training and (B) validation cohorts. Decision curve analyses (DCA) for determining the clinical utility of the nomogram, based on the (C) training and (D) validation cohorts.

3.4. Comparison of the predictive value of the nomogram with respect to 60- and 90-day TFS, as well as MELD scores

Even though the nomogram was the most predictive for 21-day TFS, as demonstrated by AUROC values of 0.947 and 0.993 for, respectively, the training and validation cohorts, it was also highly predictive for 60- and 90-day TFS among both cohorts, with AUROCs of 0.890 (training) and 0.687 (validation) for the former, and 0.888 (training) and 0.881 (validation) for the latter (Figure 4A-B). Most of those values were higher than for the MELD equation, whose AUROC was 0.841 for the training cohort, and 0.795 for the validation cohort (Figure 4A-B). All of these findings thus illustrated that the predictive nomogram was able to identify TFS for DILF patients at 21-, 60-, and 90-days, with the highest accuracy being present for 21-days.

Figure 4.

Measuring the predictive performance for the nomogram, with respect to 21-, 60-, and 90-day TFS, as well as versus Model for End-Stage Liver Disease (MELD) score, via receiver operating characteristic (ROC) curve analyses, for the (A) training and (B) validation cohorts.

4. Discussion

In the present study, we highlighted DILF clinical characteristics and prognoses, and found that the main etiology was HDS. We identified, based on uni- and multivariate Cox regression analyses, that neutrophil counts, PT, ALB, AKI, and HE were independent risk factors for poor DILF prognoses, and established a predictive nomogram based on those factors. This nomogram was found to be highly predictive for 21-day TFS, under both training and validation datasets. Furthermore, it was more accurate for predicting TFS, compared to MELD score, and was able to do so at multiple time points, such as 60 and 90 days. Based on these merits, our nomogram could serve as a potential novel diagnostic approach for predicting DILF prognoses, particularly in terms of TFS.

Our finding of HDS being the main etiology for DILF contrasts with findings from developed countries, where acetaminophen poisoning is the main cause of acute DILF. In fact, 45.7% of recorded DILF cases were caused by acetaminophen poisoning in North America, while in the United Kingdom, 65.4% of DILF/DILI cases in the Liver Intensive Therapy Unit of Kings College Hospital were caused by acetaminophen [12]. On the other hand, acetaminophen overdose is rare or non-existent in most parts of Asia and Africa [13, 14]. This is likely due to other medicinal products being more commonly-used there; in the case of China, Chinese herbal medicine is widely used for treating various diseases, owing to patients wrongly believing that they have no toxic side effects. Therefore, different medicinal therapies must be accounted for when identifying the underlying etiological factors behind DILF, among different demographics and geographical regions.

With respect to low graft-free survival rates, numerous clinical features have been associated, ranging from high ammonia, lactic acid, phosphate, INR for PT, and coagulation factor VIII, as well as low platelet counts, coagulation factor V, and VII [14]. However, we have identified an additional clinical parameter, neutrophils, which are the main human immune cells, and serve as the first barrier against invading pathogen. We found that neutrophil counts are related to DILF prognosis, owing to those counts reflecting, to a certain extent, bodily inflammation. This association between neutrophils and DILF is supported by a previous study reporting that neutrophil consumption by RB6-B18C5, an anti-APGr-β1 antibody, alleviated acetaminophen-induced acute liver injury, thereby confirming that neutrophils play a role in liver injury [15]. Furthermore, another study observed that the combination of NAC and sivelestat significantly lowered liver injury, owing to sivelestat inhibiting neutrophil elastase, a proinflammatory protein secreted by activated neutrophils. Indeed, this combinatorial treatment was more effective than NAC alone, suggesting a potential treatment strategy for acetaminophen-induced liver injury [16]. All of these observations thereby strengthen our finding that lowered neutrophil counts were associated with greater 21-day TFS.

As for PT, previous studies have shown that the thrombin required for plasma coagulation in acetaminophen-induced ALF mice is 10 times higher than those with simple hepatotoxicity [17]. Furthermore, patients with PT that was continually increasing on day 4 post-acetaminophen poisoning, as well as having a peak of ≥180 s, have been identified by Harrison et al. as having survival rates <8%, and in turn requiring liver transplantation [18]. Nevertheless, even though PT and INR increases are characteristic of acute liver injury and ALF, wide-ranging heterogeneity in clotting abnormalities exists [19]. Our findings, though, was in line with Harrison et al. in which prolonged PT was associated with adverse DILF outcomes.

Another factor identified is ALB, in which lower serum levels has been associated with liver failure, likely due to decreased hepatocyte mass, caused either by the disease itself, or multiple therapeutic interventions. Indeed, we found that greater ALB reductions were associated with increased risks for worsening adverse outcomes. On the other hand, increased ALB has been linked to improved immune cell function, as well as lowered advanced liver disease severity and infection risk [20]. Advanced liver disease, such as ALF, is often associated with AKI, such as hepatorenal syndrome, which is a ‘functional’ type of renal failure that often occurs in cirrhosis patients, in a background of marked arterial circulation abnormalities, as well as overactive endogenous vasoactive systems [21]. In fact, a retrospective analysis of >1,600 patients found that AKI is common in ALF patients, where it affects both short- and long-term prognosis, but rarely leads to chronic kidney disease. In line with that analysis, our findings also showed that AKI was associated with lowered 21-day TFS among DILF patients. It is worth noting that individuals with acetaminophen-induced ALF and AKI usually have better prognoses than for other ALF etiologies [22], which, considering the majority of DILF cases in our study was due to HDS, may serve as an underlying basis behind the association between AKI and lowered 21-day TFS.

Currently, predictive scoring models for liver disease include the King’s College Criteria standard, MELD score, Acute Liver Failure Study Group Prognostic Index, and drug-induced acute liver failure-5 [23], of which MELD is one of the most commonly-applied. We thus compared the performance of our predictive nomogram with that of MELD score, and found that the AUROC value was significantly higher for predicting TFS, at 21, 60, and 90 days. This was further validated using calibration curves, for both training and validation datasets. As a result, the predictive nomogram was able to directly calculate TFS likelihood, at multiple time points, using easily-measurable patient metrics.

However, there are a number of limitations, one of which is that it is a retrospective analysis of DILF patients from one hospital. Thus, the findings may be biased, due to the small sample size and the retrospective nature of the study. Additionally, only Chinese patients were investigated, which, considering that the predominant etiologies for DILF differ around the world, means that the findings in this study may not be fully applicable to other patient demographics. Future studies are still needed, particularly prospective ones utilizing data collected from multiple hospitals, to validate the findings from this study.

In summary, in light of increasing DILF incidence, we have constructed a predictive nomogram, incorporating the independent risk factors of neutrophils, PT, ALB, AKI, and HE identified by uni- and multivariate Cox regression analyses. The nomogram had a high predictive value for TFS, particularly after 21 days, but it was also applicable for 60 and 90 days. Furthermore, it was more accurate than MELD score, and had great clinical utility under DCA. Therefore, this predictive nomogram is easy to use, only requires data from readily-available clinical and laboratory metrics, and capable of predicting TFS at multiple time points, potentially serving as a novel diagnostic tool for DILF patients.

Funding Statement

The study was supported by the Natural Science Foundation of Jiangxi Province, China [20212ACB206010] and Health Commission of Jiangxi Province, China (20202097).

Credit authorship statement

M.Y.T. drafted the manuscript and contributed to data acquisition and analysis. Z.L.W. and J.L. contributed to data acquisition. W.T.Z. and J.W.F. contributed to data acquisition and manuscript revision. X.P.W. conceived and supervised the study, acquired funding, and revised the manuscript for important intellectual content. All authors reviewed the manuscript.

Disclosure statement

This result was previously reported in EASL (https://www.journal-of-hepatology.eu/article/S0168-8278(24)00656-1/abstract), but was not officially published. The authors declare no competing interests.

Ethics statement

The protocol was performed in accordance with the guidelines outlined in the Declaration of Helsinki and was approved by the Ethics Committee of The First Affiliated Hospital of Nanchang University. Since the study was a retrospective study, most of the study subjects have died or lost contacts, and all statistics were anonymous, so the Ethics Committee of The First Affiliated Hospital of Nanchang University agreed to waive the need for informed consent.

Data availability statement

The datasets analysed during the current study are available from the corresponding author on reasonable request.