Determinants of long-term outcome in severe alcoholic hepatitis

Authors


Correspondence to:

Dr S. Verma, Senior Lecturer Medicine, Honorary Consultant Hepatology, Brighton and Sussex Medical School, Falmer, Brighton BN1 9PX, UK.

E-mail: s.verma@bsms.ac.uk

Summary

Background

Although short-term outcome in severe alcoholic hepatitis (SAH) is well described, its long-term course remains uncharacterised.

Aim

To assess determinants of long-term outcome in SAH.

Methods

Data were recorded from a cohort with SAH (admission Discriminant Function (DF) ≥32). Kaplan–Meier (KM) and Cox proportional hazards survival analyses were performed to determine predictors of outcome.

Results

One hundred and nine patients were included; 63.3% male, aged 49.6 ± 9.4 years with median follow-up of 40.7 months (95% CI 37.2–44.3). Median DF was 58, 86.2% had cirrhosis and 65.1% received corticosteroids and/or pentoxifylline. Overall mortality was 57.8%, 96.8% of deaths being liver-related and 65.1% occurring after the index hospitalisation. Estimated 5-year survival was 31.8%. Hepatorenal syndrome was the only baseline factor independently associated with mortality (HR 3.78, 95% CI 1.98–7.19, P < 0.0001), although it predicted short-term, rather than long-term outcome (median survival 0.52 months, 95% CI 0.43–0.61). Of the 87 patients (79.8%) who survived index hospitalisation, 65.1% experienced recidivism. Abstinence at last follow-up remained the only independent predictor of survival in multivariate analysis (HR 0.370, 95% CI 0.168–0.818, P = 0.014). Five-year survival was higher in abstainers (75.3%) compared with relapsed and continued drinkers (26.8% and 21.0%, respectively, P = 0.005). However, the survival benefit from abstinence only became statistically significant at 18 months postdischarge (HR 2.714, 95% CI 0.995–7.404, P = 0.051).

Conclusions

Estimated 5-year survival after index hospitalisation with SAH is 31.8% with alcohol relapse occurring in two-thirds of patients. Abstinence remains the only independent predictor of long-term survival. Novel strategies to improve abstinence after admission with SAH are urgently needed.

Introduction

Rates of alcohol-related liver disease (ALD) have risen by almost 50% in the UK over the last 7 years and increased alcohol consumption underlies the substantial rise in liver-related mortality over the past two decades.[1] Severe alcoholic hepatitis (SAH), defined by an admission Discriminant Function (DF) ≥32 or spontaneous hepatic encephalopathy (HE),[2] is arguably the most fulminant and life-threatening form of ALD.

The short-term outcome of SAH is well described and mortality, despite abstinence, is high (up to 30% at 1 month).[3] This can be improved in selected individuals with medical therapy, principally corticosteroids (CS) and/or pentoxifylline (PTX), although recent meta analyses and systematic reviews have given conflicting results.[4-6] Data regarding the long-term course of SAH beyond 1 year are sparse, although 1-year survival has been associated with CS therapy (abrogated by 2 years), peripheral leucocytosis and neutrophil infiltrate at liver biopsy.[7] Five-year survival in alcoholic hepatitis has previously been reported; however, these data are four decades old with most patients not having severe disease or cirrhosis.[8]

Absolute alcohol cessation is widely accepted to be the sine qua non for recovery in SAH.[3, 9, 10] While this may seem self-evident, the prevalence and patterns of recidivism after SAH are poorly described and data regarding its impact upon survival are conflicting.[11, 12] We therefore sought to describe outcome following index hospitalisation with SAH and to identify factors associated with long-term mortality, both at baseline and after discharge from hospital.

Patients and methods

Our institute is a large teaching hospital in South East England with a referral base of ~300 000 patients. In our region, alcohol-related hospital admissions have doubled over the last 7 years (855/100 000 in 2003/04 to 1986.5/100 000 in 2010/11) and our primary care trust (PCT) ranks in the top quartile of PCTs for alcohol-specific mortality with death from chronic liver disease (CLD) being higher than the national average.[13]

The study period was from January 2006 until August 2011, the starting time point chosen due to the availability of electronic pathology records. Those with a primary coded diagnosis of ALD (all ‘K70’ ICD-10 codes) were identified from the hospital's database of in-patient admissions. Medical and electronic records of consecutive patients admitted with a coded diagnosis of ALD were then retrospectively reviewed to identify those with SAH as their index presentation of liver disease.

SAH was defined by:

  1. Presence of jaundice (bilirubin >80 μmol/L) and coagulopathy (INR ≥1.5) with admission Discriminant Function (DF) ≥32.[2]
  2. Long-standing alcohol excess (>50 g/day in women and >70 g/day in men until at least 6 weeks prior to admission).

plus any one of the following additional criteria:

  1. AST:ALT ratio >2.
  2. Enlarged, tender liver with peripheral blood leucocytosis.
  3. Presence of hepatic encephalopathy at presentation.
  4. Corroborative liver biopsy, where available.

Patients with any other form of coexistent chronic liver disease (viral hepatitis, iron overload, biliary or autoimmune disease) or history of jaundice ≥3 months were excluded.

In the context of SAH management, the need for histological confirmation is controversial and the frequency of liver biopsy varies according to local facilities and expertise.[14, 15] A minority of our cohort (n = 12, 11.0%) had biopsy-proven AH, reflective of clinical practice in most UK centres.[15] A clinical diagnosis of SAH proves correct in 96% of cases, provided a threshold bilirubin ≥80 μmol/L is applied,[16] a criterion fulfilled by all our patients. Our short-term outcomes (see below) were comparable to studies mandating histological confirmation of alcoholic hepatitis.[11]

Data collection

A retrospective review of clinical records was performed and the following data obtained: patient demographics, nature, duration and quantity of alcohol consumption, presence of cirrhosis, ascites, infections, hepatorenal syndrome (HRS), hepatic encephalopathy (HE) and portal hypertension-related gastrointestinal (GI) haemorrhage. The following prognostic scores were calculated at admission: Child–Pugh Score (CPS), DF,[2] Glasgow Alcoholic Hepatitis Score (GAHS),[17] Model for End-stage Liver Disease (MELD)[18] and MELD-sodium (MELD-Na).[19]

At our institute, medical therapy for SAH is prescribed at the discretion of the treating hepatologist, although it usually comprises corticosteroids (CS) (prednisolone 30 mg or 40 mg daily) and/or pentoxifylline (PTX) 1200 mg daily for 4 weeks. HRS was treated with terlipressin and albumin. All patients received dietetic input and were offered referral to alcohol liaison counsellors.

Study definitions

Spontaneous bacterial peritonitis (SBP) and HRS: as previously described.[20, 21]

Cirrhosis: any one of the following: corroborative histology, nodular liver margin or splenomegaly on imaging.

Recidivism: defined as any amount of regular alcohol use after index hospitalisation with SAH. This was largely determined from self-reported alcohol consumption in hospital records, including documentation from Emergency Department and out-patient clinic visits, subsequent in-patient admissions and correspondence from primary care physicians. Patients were classified into four groups according to drinking behaviour following index hospitalisation:

  1. Consistent abstainers: continued abstinence following discharge sustained until last follow-up.
  2. Continued drinkers: evidence of early return (within 3 months of discharge) to regular alcohol use of any quantity, ongoing at last follow-up.
  3. Relapsers: those who abstained for at least 3 months after discharge, although subsequently relapsed, with ongoing alcohol use at last follow-up.
  4. Eventual abstainers: those who initially resumed drinking, although had been abstinent for ≥3 months at last follow-up.

In view of the small number of patients in group 4, for purposes of data analysis, these cases were included in group 1.

Overall survival was calculated from the date of hospitalisation with SAH and censored at death or last hospital follow-up, determined as of July 2012. Cause of death was obtained from hospital, Coroner and primary care records. Deaths arising from hepatocellular carcinoma (HCC), liver failure and portal hypertension-related GI bleeding were deemed to constitute liver-related mortality (LRM). In those surviving the index hospitalisation, the number of subsequent admissions and need for liver transplantation (LT) were recorded. Those undergoing LT were censored alive at the date of transplantation.

Statistical analysis

Data are presented as mean ± standard deviation, median (interquartile range) or number (%) and all reported P values are two-tailed. The Mann–Whitney U- and Student's t-tests were used to compare nonparametric and parametric continuous variables, respectively, and categorical data were compared using the χ2 test. Kaplan–Meier (KM) survival tables and curves were generated and factors compared using the log rank test. Cox proportional hazards survival analysis was performed to identify factors associated with long-term mortality. Parameters with P value <0.10 in univariate analysis were selected for inclusion in a multivariable Cox regression model to determine independent predictors of survival. Survival analyses were performed in the whole cohort and then repeated to include only those who survived index hospitalisation. Receiver operator characteristic (ROC) curves were generated to evaluate the use of prognostic scores in predicting outcome beyond 12 months and compared using area under the curve (AUROC) analysis. In light of the anticipated high short-term mortality, the median follow-up time was calculated using the reverse KM method.[22] Statistical analyses were undertaken using spss version 19.0 (SPSS Inc., Chicago, IL, USA) and figures produced using Prism Version 5 (GraphPad, San Diego, CA, USA).

This study was classified as service evaluation by both our institute's Research and Development Department and the National Research Ethics Service, and consequently formal ethics approval was not deemed necessary.

Results

Patient characteristics

During the study period, there were 151 953 acute admissions of whom 1167 (0.77%) had a primary coded diagnosis of liver disease (all K70-77 codes). Of the latter, there were 630 (54%) with a coded diagnosis of ALD (all K70 codes). One hundred and seventy-four patients with suspected SAH as index presentation of liver disease were identified. Sixty-five cases were excluded due to: coexistent viral hepatitis (n = 8), decompensated alcohol-related cirrhosis rather than SAH (n = 22), alcohol-related fatty liver (n = 4), incomplete clinical data (n = 29) and diagnostic uncertainty (n = 2).One hundred and nine patients were therefore found eligible for the study. This did include seven patients (6.4% of the whole cohort), who were admitted during the study period with a coded diagnosis of ALD, although their index presentation with SAH predated January 2006. As all these seven patients had complete follow-up data after their first presentation with SAH, it was, in our opinion, reasonable to include them in the study. Their inclusion did not alter our study results and helped to ensure a robust sample size.

Table 1 shows demographic, clinical and laboratory data at entry for the whole cohort, stratified according to survival at last follow-up. This was a young and predominantly male cohort with severe disease; admission DF was ≥40 in 81.7% and admission MELD ≥21 in 77.1%. Of the 12 who underwent liver biopsy, cirrhosis was present in 11 (91.7%). An additional 83 patients were considered to have cirrhosis based on radiological criteria. Therefore, based on histological and imaging criteria, 94 (86.2%) had cirrhosis, consistent with an earlier study where cirrhosis was confirmed on biopsy in ~90% with SAH.[7] Of those with cirrhosis, 83.0% were Child–Pugh class C at accession. Median length of hospital stay was 12.5 [14] days and estimated median follow-up time from hospitalisation was 40.7 months (95% CI 37.2–44.3). Of those surviving index hospitalisation, loss to follow-up within 6 months occurred in four patients (4.6%) and a cumulative 10 cases (11.5%) by 12 months.

Table 1. Patient demographics and baseline parameters in 109 patients at index hospitalisation with severe alcoholic hepatitis, stratified according to survival at last follow-up
Alive (n = 46)Dead (n = 63) P valuea
  1. Data presented as mean ± standard deviation, median (IQR) or number (%).

  2. Normal values: bilirubin 0–21 μmol/L, INR 0.8–1.2, albumin 35–52 g/L, ALP 40–129 IU/L, ALT 0–41 IU/L, AST 0–40 IU/L, γGT 10–71 IU/L, urea 1.7–8.3 mmol/L, creatinine 62–106 μmol/L.

  3. a

    Comparisons between those alive and dead at last follow-up.

Age (years)48.7 ± 9.850.3 ± 9.10.360
Male gender28 (60.9%)41 (65.1%)0.652
Smoker24 (52.2%)27 (42.9%)0.336
Duration of alcohol use (years)15 (12)10 (12)0.595
Alcohol consumption (units/day)18.5 ± 11.716.8 ± 7.80.375
Spirits use28 (60.9%)32 (50.8%)0.296
Diabetes mellitus4 (8.5%)6 (9.7%)1.000
Parameters at hospitalisation
Bilirubin (μmol/L)228 (98)216 (194)0.593
INR1.8 (0.5)1.7 (0.6)0.609
Albumin (g/L)29 (6)29 (6)0.655
ALP (IU/L)215 (148)202 (155)0.908
ALT (IU/L)41 (31)46 (28)0.404
AST (IU/L)145 (83)140 (82)0.387
γGT (IU/L)413 (524)513 (774)0.782
Urea (mmol/L)3.2 (3.3)3.4 (3.2)0.459
Creatinine (μmol/L)51 (40)56 (36)0.731
Leucocyte count (109/L)9.6 (7.0)11.1 (6.9)0.356
Discriminant function59 (33)55 (33)0.585
Glasgow alcohol Hepatitis score8 (2)8.0 (3)0.955
MELD score23 (5)23 (8)0.652
MELD-Na score27 (7)26 (8)0.801
Underlying cirrhosis41 (89.1%)53 (84.1%)0.454
Child–Pugh score11 (3)11 (2)0.583
Complications at/during hospitalisation
Ascites40 (87.0%)55 (87.3%)0.958
Hepatic encephalopathy13 (28.3%)29 (46.0%)0.060
Infection18 (39.1%)34 (54.0%)0.126
Spontaneous bacterial peritonitis3 (6.5%)6 (9.5%)0.731
Hepatorenal syndrome3 (6.5%)17 (27.0%)0.006
Gastrointestinal haemorrhage5 (10.9%)4 (6.3%)0.397
Received corticosteroids and/or pentoxifylline32 (69.6%)39 (61.9%)0.407
Corticosteroids and pentoxifylline19 (41.3%)20 (31.7%)0.304
Corticosteroid monotherapy5 (10.9%)3 (4.8%)0.227
Pentoxifylline monotherapy8 (17.4%)16 (25.4%)0.319

The prevalence of various complications including infections, HRS and HE is shown in Table 1. A total of 64 infections occurred in 52 patients (prevalence 47.7%), which included 9 episodes of SBP (14.1% of all infections) (see Table S1). Overall, 33 infections (51.6%) were culture-positive (54.5% Gram-negative organisms and 6.1% fungal). As regards the SBP episodes, organisms cultured included E. coli (n = 5), Enterococci (n = 1) and Streptococcus mitis (n = 1), the remainder being culture-negative (n = 2).

Outcomes

During the study period, there were 63 deaths, overall mortality being 57.8%. Twenty-two patients died during the index hospital admission (in-hospital mortality 20.2%). The cause of death was liver failure in all, with median survival of 15 days (95% CI 7–23). Factors associated with mortality during index hospitalisation were the presence of HE, infection and HRS, and admission bilirubin, urea, creatinine, ALT, leucocyte count, MELD score and MELD-Na (P ≤ 0.05) (see Table S2). A further 41 of the 87 patients (47.1%) who survived index hospitalisation died during follow-up with a median survival of 30.2 months (95% CI 20.8–39.6). The cause of death was liver-related in 39 cases [liver failure (n = 29), portal hypertension-related GI haemorrhage (n = 9) and HCC (n = 1)]. Therefore, of the 63 total deaths, 61 (96.8%) were liver-related.

A total of seven patients (6.4% overall and 8% of those surviving index hospitalisation) were considered for LT. One was listed, but died of variceal haemorrhage prior to transplantation and in three, LT evaluation was terminated because of recidivism. Three patients (2.7% of the overall cohort and 3.4% of those surviving hospitalisation) were eventually transplanted after a median of 61.6 (range 18.4–87.4) months. Three of the surviving patients with HRS (Table 1) were not considered for LT due to recidivism. Another six patients who were abstinent died within 3 months of hospitalisation and were therefore also not considered for LT. Overall median survival for the entire cohort was 22.4 months (95% CI 11.1–33.7), the estimated 5-year survival being 31.8% (Figure 1).

Figure 1.

Kaplan–Meier survival curve showing survival in 109 patients following index hospitalisation with severe alcoholic hepatitis.

Factors predicting outcome in the whole cohort

Table 2 shows the results of the univariate survival analysis for 20 continuous variables and 19 categorical variables measured at baseline. Parameters associated with mortality were serum urea and creatinine, peripheral blood leucocyte count and the presence of infection, HE and HRS (P < 0.10 for all). There was significant collinearity between leucocyte count and the presence of infection, and multicollinearity between urea, creatinine and the presence of HRS. Therefore, the following variables were entered into the multivariate Cox regression analysis: presence of HE, HRS and infection. Only the presence of HRS remained independently associated with mortality in the multivariable model (HR 3.842, 95% CI 2.018–7.312, P < 0.0001) (Table 3). However, HRS was fatal within 3 months of accession in 85.0% of cases, with median survival of 0.52 months (95% CI 0.43–0.61). Of the three patients who survived HRS, all received standard medical therapy for HRS and, in addition, two were treated and responded to medical therapy for SAH.

Table 2. Univariate analysis of demographic and baseline parameters associated with death in patients with severe alcoholic hepatitis (n = 109) [bold = P < 0.10]
VariableHazard ratio95% CI P value
LowerUpper
Age1.0090.9841.0360.482
Male gender1.3360.7952.2440.274
Smoker0.9540.5781.5720.852
Duration of alcohol use0.9890.9601.0190.483
Alcohol consumption0.9980.9711.0250.871
Spirits use0.7540.4601.2360.263
BMI0.9910.9321.0540.769
Diabetes mellitus0.9160.3942.1290.916
Bilirubin1.0010.9991.0030.269
INR1.0090.7031.4480.962
Albumin0.9530.9001.0100.104
ALP1.0010.9981.0030.664
ALT1.0010.9951.0060.813
γGT1.0001.0001.0010.108
Urea 1.036 1.000 1.073 0.053
Creatinine 1.006 1.002 1.010 0.003
Sodium0.9710.9361.0070.118
Leucocyte count 1.035 0.994 1.079 0.097
CRP1.0050.9951.0140.333
Discriminant Function1.0010.9951.0070.785
Glasgow Alcohol Hepatitis Score1.0500.8691.2690.611
Glasgow Alcohol Hepatitis Score ≥91.1340.6911.8630.618
MELD1.0270.9791.0770.275
MELD ≥210.7810.4231.4440.431
MELD-Na1.0300.9801.0830.242
MELD-Na ≥281.1830.7041.9880.526
Child–Pugh score1.0820.9321.2570.299
Ascites1.1780.5602.4770.666
Gastrointestinal haemorrhage0.6840.2481.8870.464
Hepatic encephalopathy 2.013 1.224 3.310 0.006
Hepatorenal syndrome 4.661 2.632 8.253 <0.0001
Infection 1.726 1.049 2.841 0.032
Medical treatment0.9060.5451.5080.705
Corticosteroids & pentoxifylline0.9460.5561.6110.839
Corticosteroid monotherapy0.4180.1311.3410.143
Pentoxifylline monotherapy1.3090.7412.3130.354
β-blocker use0.8630.4461.9670.863
Proton pump inhibitor use0.8260.4851.4040.480
Paracetamol use1.2100.7322.0000.458
Table 3. Multivariable analysis of parameters associated with death in patients with severe alcoholic hepatitis (n = 109)
VariableBS.E.WalddfP valueHazard ratio95% CI
LowerUpper
Hepatic encephalopathy0.2190.2990.53710.4641.2450.6932.238
Hepatorenal syndrome1.3460.32816.8021<0.00013.8422.0187.312
Infection0.2490.2750.81910.3651.2830.7482.201

AUROC analysis for clinical prognostic scores

In area under the curve (AUROC) analysis, no baseline clinical prognostic score proved discriminatory in predicting outcome at 12 (Figure 2a), 24 or 48 months (Figure 2b). In a further analysis, none of the baseline clinical prognostic scores was significantly associated with overall survival, either as continuous variables or using published cut-offs to define severe disease (GAHS ≥9, MELD ≥21 and MELD-Na ≥28 [23-25] (data not shown).

Figure 2.

AUROC analysis of baseline prognostic scores at index hospitalisation with severe alcoholic hepatitis in predicting survival at 12 (a) and 48 months (b).

Effect of medical therapy

Medical treatment (taken as a whole or stratified by drug) was not associated with a survival benefit in either the Cox analysis (Table 2) or in a separate binary logistic regression of survival to 12, 24 or 48 months (data not shown). In-patient mortality was not significantly different between treated and untreated groups (21.1% vs. 18.4%, P = 0.737). The prevalence of HRS did not differ according to PTX treatment (22.2% treated vs. 13.0% untreated, P = 0.317). Of the 38 patients not treated medically with CS and/or PTX (Table 1), infection and gastrointestinal haemorrhage were present in 14 (36.8%) and 5 (13.1%) respectively.

Factors predicting outcome in those surviving index hospitalisation

Eighty-seven patients (79.8%) survived the index hospitalisation with SAH and all except one had available data regarding alcohol use after discharge. Median survival in this cohort was 31.6 months (95% CI 20.8–39.6). At last follow-up, 37 patients (43.0%) were abstinent, of whom 30 were consistent abstainers and seven eventual abstainers. Forty-nine (57.0%) were drinking alcohol at last follow-up, of whom 33 were continued drinkers and 16 relapsed after initial abstinence. Therefore, 56 (65.1%) suffered an alcohol relapse at some stage, although at last follow-up, 49 (57.0%) were still drinking alcohol. In those with recidivism, mean alcohol consumption was lower during relapse compared with accession, although the differences were not statistically significant (12.4 ± 8.9 vs. 17.3 ± 12.9 units/day, P = 0.088). Ten patients had two or more discrete episodes of SAH, and have been reported separately.[26]

Table 4 shows the univariate survival analysis in those surviving index hospitalisation. Parameters associated with mortality (at P < 0.10 level of significance) were paracetamol use and abstinence status at last hospital follow-up. There was no significant difference in mean daily in-patient paracetamol dose between survivors and those who died (0.51 ± 0.36 g/day vs. 0.96 ± 1.61 g/day, P = 0.233). In multivariate analysis (Table 5), only alcohol abstinence remained statistically significantly associated with outcome (HR for death 0.370 (95% CI 0.168–0.818), P = 0.014). Median survival in continued or relapsed drinkers was 24.1 (95% CI 16.9–31.3) months, significantly lower than those who remained or eventually achieved abstinence, where survival plateaued at 75.3% after 19.6 months (P = 0.003) (Figure 3a). When those with recidivism were stratified into continued and relapsed drinkers, 5-year survival was significantly higher in abstainers (75.3%) compared with relapsed and continued drinkers (26.8% and 21.0%, respectively, P = 0.005) (Figure 3b). In a separate binary logistic regression, there was no statistically significant difference in survival at 12 months according to abstinence (HR 1.714, 95% CI 0.612–4.802, P = 0.305). By 18 months, there was a trend to increased survival (HR 2.714, 95% CI 0.995–7.404, P = 0.051), although this only assumed statistical significance at 24 and 48 months (HR 3.207, 95% CI 1.224–8.400, P = 0.018 and HR 6.243, 95% CI 2.356–16.546, P < 0.001 respectively).

Table 4. Univariate analysis of demographic and baseline parameters associated with death in those surviving index hospitalisation with severe alcoholic hepatitis (n = 87) [bold = P < 0.10]
VariableHazard ratio95% CI P value
LowerUpper
Age1.0110.9791.0430.516
Male gender1.1870.6332.2260.592
Smoker0.9460.5081.7640.862
Duration of alcohol use0.9920.9561.0290.669
Alcohol consumption0.9790.9391.0210.329
Spirits use0.6760.3661.2490.211
BMI0.9960.9251.0720.907
Diabetes mellitus0.8610.3062.4200.777
Bilirubin0.9990.9961.0010.329
INR1.0270.6631.5910.905
Albumin0.9660.8991.0380.347
ALP1.0010.9971.0040.734
ALT0.9980.9901.0050.553
γGT1.0000.9991.0010.930
Urea0.9350.8461.0340.192
Creatinine0.9930.9821.0030.185
Sodium0.9870.9381.0380.599
Leucocyte count0.9930.9351.0540.819
CRP0.9910.9771.0040.182
Discriminant function1.0000.9911.0080.943
Glasgow Alcohol Hepatitis Score0.9160.7161.1720.487
Glasgow Alcohol Hepatitis Score ≥90.7920.4191.4970.473
MELD0.9560.8871.0310.243
MELD >210.5850.2851.2030.145
MELD-Na0.9620.9001.0280.251
MELD-Na >280.5460.2521.1830.125
Child–Pugh score1.0100.8401.2140.916
Ascites0.7140.3291.5530.396
Gastrointestinal haemorrhage0.7420.2292.4080.619
Hepatic encephalopathy0.7210.3441.5110.386
Hepatorenal syndrome0.0460.00019.4080.317
Infection0.8200.4291.5670.548
Medical treatment0.8490.4561.5820.607
Corticosteroids and pentoxifylline0.7320.3591.4940.391
Corticosteroid monotherapy0.5780.1771.8840.363
Pentoxifylline monotherapy1.4800.7392.9640.269
β-blocker use1.1250.4732.6770.790
Proton pump inhibitor use0.9000.4591.7650.759
Paracetamol use 2.088 1.064 4.099 0.032
Alcohol relapse at any time1.7350.7983.7690.164
Abstinence at last follow-up 0.329 0.151 0.716 0.005
Table 5. Multivariable analysis of parameters associated with death in those surviving index hospitalisation with severe alcoholic hepatitis (n = 87)
VariableBS.E.WalddfP valueHazard ratio95% CI
LowerUpper
Paracetamol use0.5030.3522.03510.1541.6530.8293.298
Abstinence at last follow-up−0.9930.4046.03710.0140.3700.1680.818
Figure 3.

Kaplan–Meier survival curves showing survival, depending on presence or absence of abstinence (a) and patterns of recidivism (b) at last follow-up.

Readmission to hospital occurred in 68 (78.2%) of those surviving the index episode, in whom there were a further 348 hospitalisations. The median number of readmissions per patient was significantly higher in relapsed or continued drinkers compared with those who consistently abstained (5 vs. 1, P = 0.001). The specific indication for each readmission was not recorded, although the vast majority appeared to be liver-related.

Amongst the 46 patients alive at last follow-up, clinical prognostic scores at this point were significantly better in abstainers than continued or relapsed drinkers (median MELD 9.4 vs. 13.5, P = 0.008, median CPS 6 vs. 8, P = 0.031).

Discussion

This is one of the first published studies to describe the long-term course in patients with SAH and to comprehensively examine predictors of long-term outcome. Several issues are noteworthy. Firstly, of those surviving index hospitalisation, about half died during follow-up, the estimated 5-year survival for the whole cohort being 31.8%. In fact, 65.1% of all deaths occurred after hospital discharge and almost all were liver-related. Secondly, the prevalence of recidivism was high (overall 65.1%, and at last follow-up 57.0%), which contributed both to the high overall mortality and rates of hospital readmission. Although HRS was an independent predictor of mortality, it proved fatal within 3 months of accession in 85.0%, consistent with a recent French study, where survival in a similar cohort was 16.7%.[27] Similarly, none of the current prognostic scores calculated at accession (DF, GAHS, MELD and MELD-Na) was discriminatory in predicting survival beyond 12 months. Abstinence at last follow-up remained the only independent predictor of long-term outcome (HR 0.370, 95% CI 0.168–0.818, P = 0.014), with 5-year survival significantly higher in abstainers (75.3%) compared with relapsed and continued drinkers (26.8%, and 21.0% respectively, P = 0.005).

Lifelong sobriety has been considered vital for recovery from SAH.[3, 9, 10] Although this may seem intuitive, published data regarding the prevalence and patterns of recidivism after SAH and its impact on survival are conflicting. This is most likely reflective of inadequate follow-up, as well as inclusion of patients with the entire ALD spectrum.[12, 28, 29] Recidivism occurred in almost two-thirds of our cohort, much higher than previously reported in patients with SAH (28–38% at 6 and 12 months respectively).[11, 12] Besides the aforementioned factors, additional reasons for this difference could include cultural variations. Although per capita alcohol consumption has stabilised or fallen in most Western European countries, the UK bucks this trend having amongst the highest rates of alcohol use in the European Union.[30] To our knowledge, only one other published study has assessed the impact of recidivism on SAH, finding no survival advantage from abstinence at either 3- or 12-month follow-up.[12] This is not surprising in view of our data suggesting that the survival benefit from sobriety in SAH may not be apparent within the first year. In addition, the lower prevalence of both recidivism and severe disease in the Spanish study[12] may be another explanation for the discordant results.

Our findings corroborate recent data suggesting that abstinence has a positive impact, even in advanced forms of ALD[28, 29, 31] (although these studies focused largely on alcohol-related cirrhosis rather than SAH). In the French study, where only 18% had SAH, improvement in CPS occurred within 3 months of abstinence, with a survival benefit apparent by 8 months.[28] However, our data suggest that with more fulminant forms of ALD, such as SAH, improvement in survival due to abstinence can be delayed to 18 months (Figure 3a,b). This is likely to reflect the more protracted nature of this illness. Furthermore, contrary to previous authors,[8, 28] we found no significant difference in the survival of continued vs. relapsed drinkers (P = 0.182), where outcome was equally poor. Transient abstinence is therefore of little benefit in SAH and patients should be given a clear and consistent message regarding the need for immediate and sustained sobriety.

The time lag between abstinence and improved survival provides a window of opportunity for medical therapy to prevent early deaths. Our patients were treated with various drug regimens, including combination therapy (CS and PTX), despite lack of corroborative evidence from recent trials.[6] However, these data only became available towards the end of our study period and prescribing patterns at our institution have subsequently been altered. Only 43% of our cohort received CS, consistent with an earlier trial where <40% were considered eligible for medical treatment.[32] This highlights the ongoing controversy regarding optimal medical treatment of SAH and the unmet need for safer drugs. We did not observe medical therapy to predict survival at 12 months,[7, 33] or at even later time points, although we accept that this study was neither specifically designed nor powered to assess this.

For many hepatologists, the fact that abstinence determines long-term survival after SAH will be unsurprising. However, recognition of this fact is paramount, not least because recidivism is one of the few potentially modifiable prognostic factors. Our data support interventions to improve abstinence in SAH during the first year after hospitalisation. Unfortunately, trials investigating the use of anticraving drugs are currently lacking because of a tendency to exclude those with liver disease due to hepatotoxicity concerns. However, a recent study of the γ-amino butyric acid B receptor (GABAB) agonist baclofen in a cohort with ALD (a third with CPS C disease) found a significant reduction in recidivism compared with placebo (71% vs. 29%, OR 6.3, 95% CI 2.4–16.1, P = 0.0001) with an excellent safety profile.[34]

The 6-month abstinence rule has traditionally precluded LT in SAH.[35] Only 2.7% of our cohort subsequently underwent LT as definitive treatment, consistent with the low rate of transplantation in ALD as a whole.[28] The small number of patients transplanted and high mortality in our study suggest that maybe transplantation should have been more aggressively pursued. Such a strategy has recently been advocated in highly selected individuals with treatment-refractory SAH as their index presentation of ALD.[33] Six of our abstinent patients who died within 3 months of hospital discharge could therefore potentially have been considered for a LT. While this remains an emotive issue, the high recidivism rates, which we have observed (including the ~20% potential for relapse after initial abstinence), beg the question as to whether it is possible during the acute illness to confidently identify those likely to remain abstinent after LT. Unfortunately, due to the small numbers, our data do not help identify the subgroup likely to benefit from early transplantation nor permit us to determine a time point at which sustained abstinence can be predicted.

In those surviving hospitalisation, we identified in-patient paracetamol use to be associated with mortality in univariate (HR 2.088, 95% CI 1.064–4.099, P = 0.032), although not in a subsequent multivariate analysis. The presence of other stronger predictors of early mortality, such as HRS and infections, may explain why this association was not seen in the whole cohort. Paracetamol-induced liver injury is recognised to occur at therapeutic doses in chronic alcohol users due to glutathione depletion and cytochrome P-450 isoenzyme CYP2E1 induction.[36, 37] A recent study reported that use of 4 g/day of paracetamol was safe in chronic alcohol abusers; however, most did not appear to have severe ALD.[38] Indications for paracetamol prescribing were not recorded in the current study and although it is possible that paracetamol is a surrogate marker for concomitant illness, we identified no statistically significant difference in peripheral leucocyte count, C-reactive protein or incidence of infection according to paracetamol use.

Limitations inherent in our study methodology include the retrospective collection of data and reliance upon self-reported alcohol use to determine recidivism. Nonetheless, we attempted to corroborate abstinence by cross-referencing documentation from multiple sources (including out-patient clinics, Emergency Department visits and correspondence with primary care) and using objective evidence where possible, including Emergency Department blood alcohol testing. Our categorisation of recidivism is also likely to be a crude representation of complex behaviour. By combining consistent and eventual abstainers (Groups 1 and 4) for analysis, we may risk underestimating the benefit of abstinence, through inclusion of those who abstain at a late stage with terminal disease. However, the two groups had similar MELD (11.1 vs. 9.3, P = 0.308) and CPS (7 vs. 5.5, P = 0.157) at last follow-up, making this unlikely. Finally, during the study period, we were unable to assess if abstinence was associated with regression of cirrhosis, as has been reported in CLD of other aetiologies.[39] Nonetheless, it is now our practice to perform transient elastography in all individuals with ALD on a yearly basis.

In conclusion, after index presentation with SAH only ~30% survive long-term, with almost all deaths (96.8%) being liver-related. Two-thirds of the total deaths occur after initial hospitalisation, undoubtedly due to the high prevalence of recidivism (65.1%). Alcohol use at the last follow-up remains the only independent predictor of long-term outcome, 5-year survival being approximately three-fold higher in abstainers than continued drinkers. Our data are worrying, especially in light of the rising tide of ALD in the UK, and underscore the urgent need for innovative strategies to improve abstinence following hospitalisation with SAH. The continued focus on short-term survival as the goal for therapy ignores the significant burden of late-stage mortality in SAH as demonstrated by this study.

Authorship

Guarantor of the article: Dr S. Verma.

Author contributions: JP data collection and analysis, study design and analysis and writing initial draft; SG statistical support; MH study design and critical revisions; SV study concept and design, data collection and critical revisions. All authors contributed equally to and approved the final draft of the manuscript.

Acknowledgements

Declaration of personal interests: SV has served as an advisory board member for Janssen, has received research funding from Janssen and Gilead, and received travel grants from BMS and Gilead.

Declaration of funding interests: None.

Ancillary