corticosteroids improve short term survival in …€¦ · web viewcorticosteroids improve liver...
TRANSCRIPT
CORTICOSTEROIDS IMPROVE SHORT TERM SURVIVAL IN PATIENTS WITH SEVERE ALCOHOLIC HEPATITIS (AH): INDIVIDUAL DATA ANALYSIS OF THE LAST THREE RANDOMIZED PLACEBO CONTROLLED DOUBLE BLIND TRIALS
Corticosteroids and pentoxifylline for patients with severe alcoholic hepatitis: a meta-analysis of individual data on 2111 patients.
Alexandre Louvet1, Mark R Thursz2, Dong Joon Kim3*, Julien Labreuche4*, Stephen Atkinson2, Sandeep Singh Sidhu5, John G O’Grady6, Evangelos Akriviadis7, Emmanouil Sinakos7, Robert L Carithers Jr8, Marie-José Ramond9, Willis C Maddrey10, Timothy R Morgan11, Alain Duhamel4, Philippe Mathurin1
1Service des maladies de l’appareil digestif, Université Lille 2 and Inserm U795, Lille, France; 2Department of Hepatology, Imperial College, Norfolk Place, London, United Kingdom; 3Department of Internal Medicine, Hallym University College of Medicine, Chuncheon, Republic of Korea; 4Unité de Biostatistiques, CHRU de Lille, France; 5Department of Gastroenterology, Dayanand Medical College and Hospital, Ludhiana, Punjab, India; 6Institute of Liver Studies, King’s College Hospital, Denmark Hill, London, United Kingdom; 74th Department of Internal Medicine, Aristotle University of Thessaloniki, Greece; 8Department of Medicine, Seattle, WA, USA; 9Hôpital Beaujon, Clichy, France; 10Department of Medicine, University of Texas, Southwestern Medical Center, Dallas, TX, USA; 11Gastroenterology Section, Veterans Affairs Long Beach Healthcare System, Long Beach, CA, USA.
*DJK and JL contributed equally to this work.
Corresponding author:
Philippe Mathurin
Service Maladies de l’Appareil digestif,
Hôpital Huriez
Rue Polonovski
F- 59037 Lille cedex
Phone 33 3 20 44 55 97
Fax : 33 3 20 44 55 64
E-mail: [email protected]
Word count: 5436 (abstract, text, tables, legend of figures, references)
E-mail addresses:
john.o'[email protected]
Author contribution:
Study concept and design: AL, MRT, JL, AD, PM
Acquisition of data: AL, MRT, DJK, SRA, SSS, JGOG, EA, ES, RLC, MJR, WCM, PM
Analysis and interpretation of data: AL, JL, AD, PM
Drafting of the manuscript: AL, JL, AD, PM
Critical revision of the manuscript for important intellectual content: AL, MRT, DJK, JL, SRA, SSS, JGOG, EA, ES, RLC, MJR, WCM, TRM, AD, PM
Statistical analysis: JL, AD
No funding source
No conflict of interests exist
ABSTRACT
This meta-analysis combined individual patient data from eleven randomized controlled trials comparing corticosteroids, pentoxifylline or their combination in severe alcoholic hepatitis. Aims: The aims of this meta-analysis were to compare allocated treatment in terms of: a/ 28-day mortality; b/ 6-month mortality and c/ response to treatment using the Lille model. Methods: 2111 patients were included in 4 meta-analyses: corticosteroids vs. placebo or control, corticosteroids vs. pentoxifylline, corticosteroids+pentoxifylline vs. corticosteroids+placebo/control, pentoxifylline vs. placebo. Results: Primary outcome (28-day mortality): corticosteroids significantly decreased mortality compared to controls (HR, 0.64; 95%CI, 0.48-0.86) or to pentoxifylline (HR, 0.64; 95%CI, 0.43-0.95). In multiple imputation and complete case analyses, the effect of corticosteroids compared to controls remained significant. For corticosteroids versus pentoxifylline, the corticosteroid effect remained significant in the complete case analysis (HR, 0.66, p=0.04) but not in multiple-imputation analysis (HR, 0.71, p=0.08). There was no difference in 28-day mortality between the combination of corticosteroids and pentoxifylline versus corticosteroids alone or between pentoxifylline versus control. Secondary outcomes: there were no significant differences in 6-month mortality whatever the comparison between treatments. Corticosteroids were significantly associated with increased response to therapy by comparison to controls (RR, 1.24; 95%CI, 1.10-1.41) or to pentoxifylline (RR, 1.43; 95%CI, 1.20-1.68). No difference in terms of response to therapy was seen between combination of corticosteroids and pentoxifylline versus corticosteroids alone or pentoxifylline versus controls. Conclusion: corticosteroids improve 28-day survival and response to treatment but the beneficial effect to survival disappears over 6 months. This loss of efficacy over time emphasizes the need for new therapeutic strategies to improve medium term outcome.
Key words: alcoholic hepatitis; prednisolone; survival; meta-analysis
INTRODUCTION
Patients with severe alcoholic hepatitis (AH) constitute a subgroup in which development of new strategies or new molecules are required to improve short- (28-day) or medium-term (3-6-month) mortality. AASLD in 2010
4-6
whereas 2 others questioned their efficacy in AH regardless of disease severity
ADDIN EN.CITE
3
. Five meta-analysis of the literature evaluating the therapeutic benefit of corticosteroids have yielded inconsistent results, 3 concluded that there was a survival benefit in patients with severe AH
ADDIN EN.CITE 2
have recommended treatment with pentoxifylline and corticosteroids for severe AH to reduce 28-day mortality. Only one randomized controlled trial (RCT) comparing pentoxifylline to placebo was available at the time of these recommendations1
and EASL in 20127, 8. Alternatively, the last meta-analysis using individual data from five randomized controlled trials observed that corticosteroids significantly improve 28-day survival in patients with severe alcoholic hepatitis
Within the last decade, 7 head-to-head RCTs
A recent Bayesian network meta-analysis reported that pentoxifylline and corticosteroids (alone or in combination with pentoxifylline or N-acetylcysteine) can reduce short-term mortality whereas no treatment decreases the risk of medium-term mortality
To improve the level of evidence and increase the relevance of the statistical analysis, we performed a new meta-analysis using individual patient data (MIPD) from 11 RCTs testing corticosteroids (CS), pentoxifylline (PTX) or their combination in patients with severe alcoholic hepatitis (Maddrey discriminant function (DF) ≥32). As the primary objective was to assess the short-term (28-day) efficacy of CS, PTX or their combination, only RCTs testing these strategies were considered.
The present MIPD integrates the clinical and biological data of each randomized patient when available. This allows us to evaluate the independent survival impact of each treatment modality after adjustment on classical prognostic factors and to quantify their effect in terms of biological improvement upon therapy. This approach is critical since it has been established that early biological improvement is an important predictor of short-term and medium-term mortality. An early identification of response to therapy can be assessed by the Lille model
The main aims of our MIPD from 2111 patients originating from 11 RCTs published were to compare patients receiving either corticosteroids, or pentoxifylline or their combination to those who did not receive them after adjusting for baseline prognostic factors in terms of: a/ 28-day survival; b ) 6-month survival; c/ response to treatment using the Lille model;
PATIENTS AND METHODS
Patients
The search strategy was performed by two investigators (PM, AL) with input from study investigators using Medline data base and manual searches with keyword capabilities, for RCTs of pharmacological therapy for severe AH. For RCTs testing corticosteroids, the search strategy started after the previously published MIPD integrating 5 RCTs ADDIN EN.CITE 9, 19-23. Studies were selected if they were randomized, published as full articles in English, had specific data on patients with DF (32 or encephalopathy, and if those RCTs compared corticosteroids versus pentoxifylline, one of them versus placebo or therapy without any effect on 28-day mortality, or a combination of these two agents to either one of them. We combined data from the previously published MIPD integrating 5 RCTs
9
and data from 6 RCTs whose principal investigators agreed to participate HYPERLINK \l "_ENREF_9" \o "Mathurin, 2011 #6"
ADDIN EN.CITE 3, 12-16. The previous MIPD had combined five RCTs comparing corticosteroids to placebo or therapy without any effect on 28-day mortality. The other six RCTs compared pentoxifylline plus corticosteroids versus corticosteroids alone (n=2) ADDIN EN.CITE 12, 15, pentoxifylline versus placebo (n=1)
3
, pentoxifylline versus corticosteroids (n=2) HYPERLINK \l "_ENREF_3" \o "Akriviadis, 2000 #61"
ADDIN EN.CITE 13, 14 and pentoxifylline, corticosteroids or their combination in a 2x2 factorial design (n=1)24
; 2) 3 RCTs for which the authors were contacted but did not consent to participate
16
with one group receiving placebo alone, a second group receiving prednisolone alone, a third group receiving pentoxifylline alone, and a fourth group receiving pentoxifylline and prednisolone. Four RCTs testing these treatment modalities were not included for the following reasons: 1) one RCT was published only in an abstract form HYPERLINK \l "_ENREF_16" \o "Thursz, 2015 #8"
ADDIN EN.CITE 10, 11, 25, among which one was published in Spanish25
.
The data managers of the previous MIPD integrating 5 RCTs and of the 6 other trials were asked to provide the following information: study center, age, sex, presence of encephalopathy, presence of ascites, survival data until the end of the treatment period and if available until 6-month follow-up, serum bilirubin level, prothrombin time, serum albumin level, serum AST, serum creatinine, white blood cell count, evolution of bilirubin after 7 days of treatment and type of treatment, Maddrey’s DF [defined as: 4.6 x (patient prothrombin time-control prothrombin time, in seconds) + serum bilirubin level, in milligrams per deciliter]26
. Only patients with severe alcoholic hepatitis were included using the definition used in previous MIPD and RCTs (i. e. DF ( 32 and/or encephalopathy).
Statistical analysis
All analyses were conducted according to the intention-to-treat principle by including all randomized patients, irrespective of whether they subsequently received the intended treatment. Main baseline characteristics were described in each trial and in each meta-analysis according to treatment group. Continuous variables were expressed as means (standard deviation) or as medians (interquartile range) in cases of a non-normal distribution. Categorical variables were expressed as frequencies (percentages). Survival curves were estimated using the Kaplan-Meier method. For each meta-analysis, absolute standardized differences between treatment groups were calculated and values>10% were considered to be a meaningful imbalance. The risk of publication bias cannot be statistically assessed because of the number of trials included per meta-analysis.
In each meta-analysis, the treatment effect on the primary outcome (overall survival at 28 days defined as the period from the first day of assigned treatment to 28 days) was estimated using a Cox proportional hazards regression model stratified by trial. The proportional hazards assumption was assessed by plotting the scaled Schoenfeld residuals of treatment effect against the rank of survival time27, 28. Heterogeneity across trials was examined by the Cochran’s Q test and quantified by calculating the I2 statistics29
; Cochran’s Q test with a p value <0.10 or I2>50% were interpreted as meaningful heterogeneity. Hazard ratio (HRs) and 95% confidence intervals (CI) were derived from Cox regression models as relative effect size measures. In case of significant difference, we also calculated the absolute risk difference and the number needed to treat (NNT) from the marginal probabilities using the method described by Austin30
; 95%CI were estimated using bootstrap method (2000 bootstrap samples). We tested the interaction between baseline discriminant function and treatment effect sizes by including the corresponding interaction term in Cox regression models. A pre-specified secondary analysis was conducted adjusting for age, baseline discriminant function, creatinine (after log-transformation), and albumin as covariates into the Cox’s regression model. To avoid case deletion in primary adjusted analysis due to missing data on baseline covariates, missing data were imputed by multiple imputations (m=10) using regression switching approach (chained equations using all baselines characteristics, death status and the survival times after a logarithmic transformation) with predictive mean matching method for continuous variables, logistic regression model for binary variables, and ordinal logistic regression for ordered categorical variables3231
. Imputation involves filling in the missing values with plausible values and performing the statistical analyses on this completed data set. The imputation process must be repeated several times and the results of analyses performed on each completed data set are then combined according to rules proposed by Rubin. We also reported complete-cases adjusted analysis as sensitivity analysis. Treatment effects on 6-month mortality were analyzed as a pre-specified secondary endpoint using the same statistical methods as for primary outcome. We have also provided the treatment effects on 3-month mortality as an unplanned analysis.
The response to the assigned treatment assessed with the Lille model, a score ranging from 0 to 1, was a secondary end point. Response to therapy defined using the Lille model were analyzed as a binary outcome (Lille model <0.45) in which patients can be classified as satisfactory responders to the allocated, as an ordinal outcome (≤0.16; 0.16 to <0.56; ≥0.56) in which patients can be also classified as complete responders (Lille score ≤ 0.16), partial responders (Lille score between 0.16 and 0.56), and null responders (Lille score > 0.56)
9 and as a continuous outcome. We assessed the prognostic value of response therapy (binary and 3-level ordinal variables) on 28-day mortality using a trial-stratified Cox’s regression model; this analysis was done in overall and in each treated group separately.
We used a Poisson regression model with robust error variance model33
to estimate the treatment effect on binary outcome, an ordinal logistic regression model to estimate the treatment effect on ordinal outcome and a linear regression model to estimate the treatment effect on continuous outcome (after logit transformation of Lille model to satisfy the normality assumption). From these models, we respectively calculated the risk ratio, common odds ratio (OR) and standardized difference (Cohen’ d) with their 95%CI as effect size measures. All models were adjusted for trials without and with pre-specified confounding factors. Similar to primary outcome, to avoid case deletion in adjusted analysis due to missing data on baseline covariates, missing data were imputed by multiple imputations using regression switching approach; a sensitivity adjusted analysis were performed on complete cases. The statistical plan was registered on clinicaltrials.gov (NCT02796469) on June 10th 2016, before the analysis.
Statistical testing was done at the two-tailed ( level of 0.05. Data were analyzed using SAS software (version 9.4, SAS Institute Inc., Cary, NC, USA).
Role of the funding source
This study was not supported by any company or grants. The cost was borne by the authors’ institutions.
RESULTS
We obtained data for 2111 participants from 11 trials; 6 trials involving 956 participants comparing corticosteroids versus control, 2 trials involving 666 participants comparing corticosteroids versus pentoxifylline, 3 trials involving 886 participants comparing the combination of corticosteroids and pentoxifylline to corticosteroids alone, and 3 trials involving 694 participants comparing pentoxifylline versus control (Supplemental Figure 1). Individual data was not available for one trial included in the meta-analysis comparing corticosteroids versus control19
; only mortality information could be extracted from the published manuscript using the method developed by Fine et al. and others34, 35. The main baseline characteristics and outcomes (28-day and 6-month mortality, response therapy assessed by Lille model) for each included trial are available in Supplemental Table 1. The characteristics of the patients from the 4 excluded studies (see Patients for more details) are provided in Supplemental Table 2. Overall, 1972 (93.4%) of 2111 participants had complete information for primary endpoint and pre-specified confounding factors. Baseline characteristics and Kaplan-Meier survival curves according to treatment groups per meta-analysis are shown in Table 1 and Figure 1.
28-day mortality (Primary outcome)
Overall, 383 of 2111 patients died within the 28 days following randomization (mortality at 28 days using the Kaplan-Meier method: 18.5%). As shown in Figure 2, corticosteroid use was significantly associated with a decrease in mortality rate by comparison to controls (HR, 0.64; 95%CI, 0.48 to 0.86) or to pentoxifylline (HR, 0.64; 95%CI, 0.43 to 0.95) with no evidence of heterogeneity between trials (I2 statistic=31% and 4%, respectively). The estimated risk difference in favor of corticosteroids was 7.8% (95%CI, 2.6 to 12.6, NNT=13) for comparison with controls and 6.6% (95%CI, 1.1 to 12.2, NNT=15) for comparison to pentoxifylline. After excluding the trial without available individual data, the effect size of corticosteroids versus controls was not modified (HR, 0.63; 95%CI, 0.46 to 0.85). In multiple imputation and complete case analyses adjusted for pre-specified confounding factors, the effect of corticosteroids against control remained significant (Table 2). For corticosteroids against pentoxifylline, the corticosteroid effect remained significant in the complete case adjusted analysis (HR, 0.66; 95%CI, 0.43 to 0.98, p=0.039) but not in multiple-imputation adjusted analysis (HR, 0.71; 95%CI, 0.48 to 1.05; p=0.083).
There was no difference in 28-day mortality in meta-analysis comparing the combination of corticosteroids and pentoxifylline versus corticosteroids alone (HR, 0.89; 95%CI, 0.62 to 1.28) or in the meta-analysis comparing pentoxifylline versus controls (HR, 0.96; 95%CI, 0.68 to 1.34). Similar findings were found after the pre-specified adjustment (Table 2). However, heterogeneity was apparent in pentoxifylline versus control meta-analysis (Figure 2, I2=55%, p=0.11). Similar results were found using random-effects models (Supplemental Table 3).
We tested whether the assigned treatment influenced 28-day mortality according to baseline disease severity. Thus, we tested the interaction between Maddrey’s discriminant function (treated as a continuous variable) and treatment group on the main outcome (28-day mortality) in each meta-analysis. We did not find any significant interaction in any of the meta-analyses (p=0.71 in first meta-analysis, p=0.09 in second meta-analysis, p=0.33 in third meta-analysis and p=0.49 in fourth meta-analysis). In addition, we also assessed this interaction treating Maddrey’s discriminant function as a binary variable using the 75th percentiles as a cut-off value for the most severely ill patients and confirmed that there was no interaction between disease severity and treatment efficacy (see Supplemental Table 4).
6-month mortality
Of the 11 included trials, two trials involving 66 and 50 participants, respectively, with no follow-up information after 30 days, were excluded from the 6-month mortality analysis; one was included in the first meta-analysis comparing corticosteroids versus controls
Response therapy assessed by Lille model
One of the 11 included trials without individual data involving 71 participants was excluded from the response-to-therapy analysis; this trial was included in the first meta-analysis comparing corticosteroids versus control. Overall, 923 (58.3%) of 1581 patients with available Lille model score were classified as responders to therapy (score <0.45).
On overall patients, regardless of treatment, patients classified as responders had a higher 28-day survival than non-responders (HR, 0.21; 95%CI, 0.16 to 0.28). This difference between responders and non-responders was observed in each treated group; the HR (95% CI) for responders was 0.16 (0.10 to 0.28) in the corticosteroids group, 0.26 (0.13 to 0.49) in the pentoxifylline group, 0.23 (0.11 to 0.48) in the corticosteroids+pentoxifylline group, and 0.24 (0.15 to 0.39) in the control group. Using the classification of response in three groups (complete, partial and null) on overall patients, 28-day survival was significantly different; using null response category as reference, the HR (95% CI) was 0.14 (0.09 to 0.22) for complete response and 0.31 (0.23 to 0.41) for partial response). This difference between the three response groups was also observed in each allocated treatment group (data not shown).
As shown in Figure 4, corticosteroids were significantly associated with increased incidence of the response therapy by comparison to control (67.3% vs. 54.0%; RR, 1.24; 95%CI, 1.10 to 1.41) or to pentoxifylline (67.3% vs. 47.2%; RR, 1.43; 95%CI, 1.20 to 1.68). This difference was not observed in meta-analysis comparing the combination of corticosteroids and pentoxifylline versus corticosteroids alone (60.4% vs. 64.8%; RR, 0.94; 95%CI, 0.83 to 1.06) or in meta-analysis comparing pentoxifylline versus control (50.2% vs. 51.5%, RR, 0.98; 95%CI, 0.82 to 1.16). When response to therapy was analyzed as a continuous or an ordinal outcome (Supplemental Figures 4 and 5), similar results were found. There was no evidence of heterogeneity across trials whatever the meta-analysis, except in the fourth meta-analysis (pentoxifylline versus controls) treating response-to-therapy as a continuous outcome (Supplemental Figure 4, I2=64%, p=0.065). Multivariate analysis adjusted for pre-specified confounding factors yielded similar results (Supplemental Tables 6-7-8). Similar results were found using random-effects models (Supplemental Table 9). Because of the large amount of missing data, we did not analyze bilirubin at 28 days (data only available in 748/2111 patients, 35.4%).
DISCUSSION
The present MIPD performed using individual data of 11 RCTs showed that patients treated with corticosteroids had higher 28-day survival than patients treated with either placebo/control or pentoxifylline. Pentoxifylline alone or its combination with corticosteroids did not improve short-term (28-day) survival or evolution of liver injury. No difference was observed in 3-month or 6-month survival regardless of the allocated treatment. When considering the large sample size of this MIPD, the present results add substantial evidence to the amount data for therapeutic management of patients with severe AH. Differences in treatment efficacy may vary in patients with clinical or histological alcoholic hepatitis, although this hypothesis cannot be tested in the present MIPD. Although a survival benefit was observed only during the 28-day therapeutic period, such an impact provides a strong argument supporting the use of prednisolone for severe alcoholic hepatitis and should be considered by scientific societies when drawing their therapeutic guidelines.
This MIPD also confirms the prognostic value of the Lille model in a large cohort from several centers regardless of the allocated therapy. This emphasizes early improvement as a key phenomenon driving short-term mortality. The Lille model can be a suitable end-point in the studies designed to test new treatments in severe alcoholic hepatitis. However, the accuracy of this score as well as that of the MELD score must be evaluated further with newer therapeutic agents.
Corticosteroids improve liver function, response to therapy, and 28-day survival as compared to the other tested options. The lack of statistical interaction between treatment response and baseline disease severity does not support not treating the most severely ill patient (i.e. with the highest Maddrey’s DF) with corticosteroids. However, severe renal failure was an exclusion criterion in most trials. Thus, the benefit of corticosteroids has not been shown in patients with severe renal failure. We feel that the present MIPD ends the controversy surrounding corticosteroid treatment and hope that the hesitation about their use in severe alcoholic hepatitis will decrease over time. However, corticosteroids cannot be viewed as an ideal treatment because they do not improve medium- (3-6-month) or long-term (>6-month) mortality.
Future studies design testing drugs to improve liver injury in AH should assess survival in relation to duration of drug exposure in order to avoid the potential bias of analysis of outcome far after the cessation of treatment. We suggest that future studies evaluate 3 or 6 months of drug exposure and consider 3- or 6-month survival as end-points to test the impact of molecules targeting liver injury. We feel that the optimal time point should be 3 months, as alcohol relapse starts around 2-3 months and starts to impact survival between 3 and 6 months36
.
In the present MIPD, pentoxifylline neither improved short-term survival and nor modified response to therapy. MIPD have a truth survival advantage over than meta-analysis of the literature37
. The lack of survival impact observed here with pentoxifylline or its combination may appear in contrast to the last meta-analysis of the literature concluding to a potential effect of pentoxifylline on 28-day mortality
We analyzed the impact of allocated therapy on medium-term mortality (6-month mortality) and did not observe any differences in survival whatever the allocated therapy. This raises several concerns on the medium or long-term impact of molecules in alcoholic hepatitis. The proportional hazard assumption for the effect of corticosteroids versus controls or pentoxifylline is violated, which may be related to the important discordance of duration of drug exposure (28 days) and length of period of evaluation (6 months). All RCTs were not designed for medium term and none for long-term. Analysis of medium- and long-term data to determine efficacy of a drug given for a short-term period of time (28 days) in patients with severe alcoholic hepatitis is debatable. Indeed, a recent prospective study with sequential assessment of clinical, biological parameters and alcohol consumption showed that short-term and medium-term outcome (before 6 months) is mainly related to the severity of liver injury at baseline and early improvement in hepatic function whereas long-term outcome (after 6 months) is mainly influenced by alcohol consumption and must be analyzed in relation to early improvement in liver injury36
. Such caution in interpretation of long-term data has been also highlighted in non-hepatic chronic diseases. As an example, a randomized controlled trial showed that beneficial effect of tight control of blood pressure disappeared during the long term when no strategy to maintain such control was made
In summary, the present MIPD show that corticosteroids improve 28-day survival and response to treatment but this protective HR disappears over 6 months. Such loss of efficacy over the time highlights the urgent need for clinicians to focus on new therapeutic strategies aiming to improve medium-term outcome.
REFERENCES
1.O'Shea RS, Dasarathy S, McCullough AJ. Alcoholic liver disease. Hepatology 2010;51:307-28.
2.EASL clinical practical guidelines: management of alcoholic liver disease. J Hepatol 2012;57:399-420.
3.Akriviadis E, Botla R, Briggs W, et al. Pentoxifylline improves short-term survival in severe acute alcoholic hepatitis: a double-blind, placebo-controlled trial. Gastroenterology 2000;119:1637-48.
4.Daures JP, Peray P, Bories P, et al. [Corticoid therapy in the treatment of acute alcoholic hepatitis. Results of a meta-analysis]. Gastroenterol Clin Biol 1991;15:223-8.
5.Imperiale TF, McCullough AJ. Do corticosteroids reduce mortality from alcoholic hepatitis? A meta-analysis of the randomized trials. Ann Intern Med 1990;113:299-307.
6.Reynolds TB, Benhamou JP, Blake J, et al. Treatment of acute alcoholic hepatitis. Gastroenterol Int 1989;2:208-16.
7.Christensen E, Gluud C. Glucocorticoids are ineffective in alcoholic hepatitis: a meta-analysis adjusting for confounding variables. Gut 1995;37:113-8.
8.Rambaldi A, Saconato HH, Christensen E, et al. Systematic review: glucocorticosteroids for alcoholic hepatitis--a Cochrane Hepato-Biliary Group systematic review with meta-analyses and trial sequential analyses of randomized clinical trials. Aliment Pharmacol Ther 2008;27:1167-78.
9.Mathurin P, O'Grady J, Carithers RL, et al. Corticosteroids improve short-term survival in patients with severe alcoholic hepatitis: meta-analysis of individual patient data. Gut 2011;60:255-60.
10.De B, Mandal S, Sau D, et al. Pentoxifylline Plus Prednisolone versus Pentoxifylline Only for Severe Alcoholic Hepatitis: A Randomized Controlled Clinical Trial. Ann Med Health Sci Res 2014;4:810-6.
11.De BK, Gangopadhyay S, Dutta D, et al. Pentoxifylline versus prednisolone for severe alcoholic hepatitis: a randomized controlled trial. World J Gastroenterol 2009;15:1613-9.
12.Mathurin P, Louvet A, Duhamel A, et al. Prednisolone with vs without pentoxifylline and survival of patients with severe alcoholic hepatitis: a randomized clinical trial. JAMA 2013;310:1033-41.
13.Park SH, Kim DJ, Kim YS, et al. Pentoxifylline vs. corticosteroid to treat severe alcoholic hepatitis: a randomised, non-inferiority, open trial. J Hepatol 2014;61:792-8.
14.Sidhu SS, Goyal O, Singla M, et al. Pentoxifylline in severe alcoholic hepatitis: a prospective, randomised trial. J Assoc Physicians India 2012;60:20-2.
15.Sidhu SS, Goyal O, Singla P, et al. Corticosteroid plus pentoxifylline is not better than corticosteroid alone for improving survival in severe alcoholic hepatitis (COPE trial). Dig Dis Sci 2012;57:1664-71.
16.Thursz MR, Richardson P, Allison M, et al. Prednisolone or pentoxifylline for alcoholic hepatitis. N Engl J Med 2015;372:1619-28.
17.Singh S, Murad MH, Chandar AK, et al. Comparative Effectiveness of Pharmacological Interventions for Severe Alcoholic Hepatitis: A Systematic Review and Network Meta-analysis. Gastroenterology 2015;149:958-70 e12.
18.Louvet A, Naveau S, Abdelnour M, et al. The Lille model: a new tool for therapeutic strategy in patients with severe alcoholic hepatitis treated with steroids. Hepatology 2007;45:1348-54.
19.Cabre E, Rodriguez-Iglesias P, Caballeria J, et al. Short- and long-term outcome of severe alcohol-induced hepatitis treated with steroids or enteral nutrition: a multicenter randomized trial. Hepatology 2000;32:36-42.
20.Carithers RL, Jr., Herlong HF, Diehl AM, et al. Methylprednisolone therapy in patients with severe alcoholic hepatitis. A randomized multicenter trial. Ann Intern Med 1989;110:685-90.
21.Mendenhall CL, Anderson S, Garcia-Pont P, et al. Short-term and long-term survival in patients with alcoholic hepatitis treated with oxandrolone and prednisolone. N Engl J Med 1984;311:1464-70.
22.Phillips M, Curtis H, Portmann B, et al. Antioxidants versus corticosteroids in the treatment of severe alcoholic hepatitis--a randomised clinical trial. J Hepatol 2006;44:784-90.
23.Ramond MJ, Poynard T, Rueff B, et al. A randomized trial of prednisolone in patients with severe alcoholic hepatitis. N Engl J Med 1992;326:507-12.
24.Paladugu H, Sawant P, Dalvi L, et al. Role of pentoxifylline in treatment of severe acute alcoholic hepatitis – a randomized controlled trial. J Gastroenterol Hepatol 2006;21:A459.
25.Garrido Garica JR, Sanchez Hernandez G, Melchor Lopez A, et al. Pentoxifilina versus esteroide en la sobrevivencia a corto plazo en hepatitis aguda alcohólica severa. Med Int Mex 2012;28:227-33.
26.Maddrey WC, Boitnott JK, Bedine MS, et al. Corticosteroid therapy of alcoholic hepatitis. Gastroenterology 1978;75:193-9.
27.Schoenfeld D. Partial residuals for the proportional hazards regression model. Biometrika 1982;69:239-41.
28.Therneau T-M, Grambsch, P.-M. Modeling Survival Data: Extending the Cox Model. New-York, 2000.
29.Higgins JP, Thompson SG, Deeks JJ, et al. Measuring inconsistency in meta-analyses. BMJ 2003;327:557-60.
30.Austin PC. Absolute risk reductions and numbers needed to treat can be obtained from adjusted survival models for time-to-event outcomes. J Clin Epidemiol 2010;63:46-55.
31.van Buuren S, Groothuis-Oudshoorn K. mice: Multivariate Imputation by Chained Equations in R. J Stat Soft 2011;45:1-67.
32.Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York, 1987.
33.Zou G. A modified poisson regression approach to prospective studies with binary data. Am J Epidemiol 2004;159:702-6.
34.Fine HA, Dear KB, Loeffler JS, et al. Meta-analysis of radiation therapy with and without adjuvant chemotherapy for malignant gliomas in adults. Cancer 1993;71:2585-97.
35.Salerno F, Camma C, Enea M, et al. Transjugular intrahepatic portosystemic shunt for refractory ascites: a meta-analysis of individual patient data. Gastroenterology 2007;133:825-34.
36.Louvet A, Labreuche J, Artru F, et al. Main drivers of outcome differ between short and long-term in severe alcoholic hepatitis: A prospective study. Hepatology 2017.
37.Poynard T, Munteanu M, Ratziu V, et al. Truth survival in clinical research: an evidence-based requiem? Ann Intern Med 2002;136:888-95.
38.Holman RR, Paul SK, Bethel MA, et al. Long-term follow-up after tight control of blood pressure in type 2 diabetes. N Engl J Med 2008;359:1565-76.
Table 1. Main baseline characteristics according to treatment groups in each meta-analysis
Meta-analysis n°1
Meta-analysis n°2
Meta-analysis n°3
Meta-analysis n°4
Corticosteroids
Control
Corticosteroids
Pentoxifylline
Corticosteroids + Pentoxifylline
Corticosteroids
Pentoxifylline
Control
Number of patients
490
466
333
333
441
445
345
349
Age, yr, mean (SD)
48.7 (10.7)
48.4 (10.1)
49.0 (10.2)
48.4 (10.1)
48.9 (9.8)
49.5 (10.1)
47.1 (10.2)
47.4 (10.6)
Men, n(%)
295/454 (65.0)
275/431 (63.8)
217/333 (65.2)
214/333 (64.3)
300/441 (68.0)
291/445 (65.4)
223/345 (64.6)
227/349 (65.0)
Ascites, n(%)
122/163 (74.8)
115/148 (77.7)
-
-
113/167 (67.7)
132/170 (77.6)
36/49 (73.5)
37/52 (71.2)
Encephalopathy, n(%)
134/452 (29.6)
142/429 (33.1)
64/273 (23.4)
56/270 (20.7)
90/439 (20.5)
97/442 (21.9)
60/319 (18.8)
79/323 (24.5)
Bilirubin at day 0, µmol/l
279
(165 to 436)
290
(181 to 435)
271
(168 to 415)
257
(173 to 406)
282
(144 to 422)
265
(159 to 408)
277
(187 to 408)
305
(189 to 424)
Prothrombin time, s
19.0
(16.7 to 22.0)
19.0
(16.8 to 22.7)
20.2
(17.3 to 23.7)
21.0
(17.8 to 24.7)
20.2
(17.8 to 24.0)
20.0
(17.7 to 23.2)
20.5
(17.1 to 24.1)
19.6
(17.1 to 23.0)
Albumin, g/l, mean (SD)
25.5 (5.7)
25.5 (5.8)
25.6 (6.0)
25.2 (5.4)
25.3 (5.9)
25.3 (6.0)
25.5 ± 6.2
25.8 ± 6.4
Serum creatinine, µmol/l
76
(59 to 111)
71
(58 to 100)
65
(53 to 89)
64
(53 to 83)
66
(53 to 88)
64
(53 to 87)
69
(55 to 91)
70
(56 to 90)
AST, IU/I
100
(72 to 141)
96
(74 to 136)
-
-
110
(83 to 160)
116
(86 to 157)
143
(109 to 193)
126
(84 to 198)
White blood cells, no/mm3
990
(700 to 1430)
1020
(690 to 1460)
920
(660 to 1240)
870
(590 to 1260)
930
(650 to 1340)
980
(700 to 1430)
950
(620 to 1360)
960
(640 to 1325)
Maddrey DF
50
(40 to 67)
52
(41 to 68)
55
(43 to 73)
57
(44 to 78)
56
(43 to 73)
54
(42 to 74)
55
(43 to 75)
54
(43 to 70)
Values are no/total no. (%) or median (IQR) unless otherwise indicated.
All absolute standardized differences (ASD) were <10% except for prothrombin time in meta-analysis n°2 (ASD=15%) and n°4 (ASD=11%), ascites in meta-analysis n°3 (ASD=23%), encephalopathy in meta-analysis n°4 (ASD=14%) and ASAT in meta-analysis n°4 (ASD=11%).
Abbreviations: AST=aspartate aminotransferase; DF=discriminant function; IQR=interquartile range; SD=standard deviation.
Table 2. Adjusted treatment effect sizes on 28-day mortality
Trials/Patients
Multiple imputation analysis
Complete case analysis1
Meta-analysis
(numbers)
HR (95%CI)
P
HR (95%CI)
P
Corticosteroids vs. control 2
5* / 885
0.53 (0.39 to 0.73)
<0.0001
0.50 (0.36 to 0.69)
<0.0001
Corticosteroids vs. pentoxifylline
2 / 666
0.71 (0.48 to 1.05)
0.083
0.66 (0.43 to 0.98)
0.039
Corticosteroids+pentoxifylline vs. corticosteroids
3 / 886
0.90 (0.62 to 1.29)
0.56
0.94 (0.64 to 1.37)
0.74
Pentoxifylline vs. control
3 / 694
0.90 (0.63 to 1.27)
0.54
0.91 (0.64 to 1.30)
0.60
1 pre-specified as sensitivity analysis. Complete-case analysis was performed in 848 patients for comparison between corticosteroids vs. control, 639 for comparison between corticosteroids vs. pentoxifylline, 850 for comparison between corticosteroids+pentoxifylline vs. corticosteroids, and 663 for comparison between pentoxifylline vs. control.
2 excluding the study by Cabré et al.19
Hazard ratios (HRs) and their 95% confidence intervals (CIs) were calculated using a Cox’s regression model stratified by trial and adjusted for pre-specified baseline confounding factors (age, discriminant function, creatinine and albumin).
Table 3. Adjusted treatment effect sizes on 6-month mortality
Trials / Patients
Multiple imputation analysis
Complete case analysis1
Meta-analysis
(numbers)
HR (95%CI)
P
HR (95%CI)
P
Corticosteroids vs. control 2
4 / 819
0.87 (0.69 to 1.10)
0.26
0.88 (0.69 to 1.11)
0.26
Corticosteroids vs. pentoxifylline
2 / 666
1.00 (0.76 to 1.31)
1.00
1.00 (0.76 to 1.31)
0.98
Corticosteroids+pentoxifylline vs. corticosteroids
3 / 886
0.85 (0.67 to 1.07)
0.16
0.85 (0.67 to 1.08)
0.18
Pentoxifylline vs. control
2 / 644
0.97 (0.74 to 1.27)
0.81
1.01 (0.76 to 1.34)
0.95
1 pre-specified as sensitivity analysis. Complete-case analysis was performed in 782 patients for comparison between corticosteroids vs. control, 639 for comparison between corticosteroids vs. pentoxifylline, 850 for comparison between corticosteroids+pentoxifylline vs. pentoxifylline, and 613 for comparison between pentoxifylline vs. control.
2 excluding the study by Cabré et al.19
Hazard ratios (HRs) and their 95% confidence intervals (CIs) were calculated using a Cox’s regression model stratified by trial and adjusted for pre-specified baseline confounding factors (age, discriminant function, creatinine and albumin).