1 Aliment Pharmacol Ther 2002; 16: 361±372. Flumazenil vs. placebo in hepatic encephalopathy in patients with cirrhosis: a meta-analysis C. GOULENOK*, B. BERNARD*, J. F. CADRANEL, D. THABUT*, V. DI MARTINO*, P. OPOLON* & T. POYNARD* *Service d'heâpato-gastroenteârologie, HoÃpital PitieÂ SalpeÃtrieÁre, Paris, France; UniteÂ d'heâpatologie, HoÃpital Laennec, Creil, France Accepted for publication 11 October 2001 SUMMARY Background: Randomized controlled trials testing umazenil in hepatic encephalopathy have shown con- icting results. Aim: To compare umazenil and placebo in hepatic encephalopathy in patients with cirrhosis. Methods: An overview of randomized controlled trials comparing umazenil and placebo in hepatic encephalopathy in patients with cirrhosis was performed. For each end-point, heterogeneity and treatment ef cacy were assessed by Peto and Der Simonian methods. As most trials were crossover in nature, a sensitivity analysis was performed including the two treatment periods. Results: Six double-blind randomized controlled trials, including 641 patients (326 treated with umazenil and 315 with placebo), were identi ed. The treatment duration ranged from 5 min to 3 days. Heterogeneity tests between control groups were not signi cant. The mean percentages of patients with clinical improvement ( ve trials) were 27% in treated groups and 3% in placebo groups. This difference was signi cant by both methods (Peto: odds ratio ˆ 6.15; 95% con dence interval, 4.0± 9.5; P < 0.001; Der Simonian: mean rate difference, 29%; 95% con dence interval, 17±41; P < 0.001). The mean percentages of patients with electroencephalographic improvement were 19% in treated groups and 2% in placebo groups. This difference was signi cant only with the Peto method (odds ratio ˆ 5.8; 95% con dence interval, 3.4±9.7; P < 0.001). The sensitivity analysis showed similar results. Conclusions: This meta-analysis shows that umazenil induces clinical and electroencephalographic improvement of hepatic encephalopathy in patients with cirrhosis. INTRODUCTION Hepatic encephalopathy is a complex behavioural problem complicating cirrhosis and acute liver failure. Clinical manifestations range from subtle abnormalities to coma. The prevalence of this syndrome is not well known in cirrhotic patients, although the description of subclinical encephalopathy has received increasing attention in recent years. With the inclusion of Correspondence to: Dr B. Bernard, Service d'heâpato-gastroenteârologie, HoÃpital PitieÂ SalpeÃtrieÁre, Paris Cedex 13, France. subclinical encephalopathy, hepatic encephalopathy may be present in 50±70% of patients with cirrhosis. 1 Moreover, hepatic encephalopathy is a severe complication of cirrhosis and is responsible for a substantial proportion of deaths in cirrhotic patients. It is often triggered by gastrointestinal bleeding, bacterial infections, drug intake, such as sedatives (especially benzodiazepines), electrolyte disturbances, constipation, renal failure and spontaneous or surgical portocaval shunts. The accumulation of unmetabolized ammonia has been considered for some time as the main factor involved in the pathogenesis of hepatic encephalopathy, but additional mechanisms may be associated. The Ó 2002 Blackwell Science Ltd 361
2 362 C. GOULENOK et al. inhibition of c-aminobutyric acid (GABA)±benzodiazepine receptors is probably one of these additional factors. One hypothesis proposes the inhibition of this GABA receptor complex by benzodiazepine-like ligands, which have a high af nity for these binding sites. 2 Recently, Basile and Jones 3 reviewed the main theories of hepatic encephalopathy, and showed that the ammonia hypothesis and GABA hypothesis were not mutually exclusive. Flumazenil is a competitive benzodiazepine receptor antagonist. This drug has a high af nity for these receptors and can rapidly reverse the hypnotic effect of benzodiazepines. 4 In hepatic encephalopathy, the inhibitory effect of GABA±benzodiazepine on the central nervous system is antagonized by umazenil. Flumazenil stops the binding of GABA to its receptors by benzodiazepine agonists. In cirrhotic patients, uncontrolled studies have shown a bene cial effect of umazenil in hepatic encephalopathy. 5±9 Since the publication of these uncontrolled studies, several prospective randomized controlled trials have tested the clinical and electroencephalographic ef cacy of umazenil in hepatic encephalopathy. 10±15 However, the results are controversial and the number of patients included is often small, except for the study of Barbaro et al. 15 There is as yet no de nitive evidence of a favourable effect of umazenil in the treatment of hepatic encephalopathy. To investigate this issue, a meta-analysis was conducted to assess the ef cacy of umazenil in hepatic encephalopathy in patients with cirrhosis. METHODS End points Meta-analysis was performed according to a predetermined protocol based on the recommendations of Sacks et al. 16 Two events were chosen as end-points to estimate the clinical ef cacy: clinical improvement and electroencephalographic improvement. Survival was chosen before completion of the meta-analysis; however, none of the selected trials have studied the effect of umazenil on survival. Literature search MEDLINE and a manual search were combined as we have previously demonstrated that a MEDLINE search alone is not sensitive enough. 17 General reviews, references of published randomized controlled trials, letters to pharmacological companies and Current Contents were also used. Inclusion criteria To be included in the meta-analysis, a randomized controlled trial had to ful l the following criteria: published as an article or abstract; randomized and prospective; included patients with cirrhosis and encephalopathy; and assessed the ef cacy of umazenil in hepatic encephalopathy. Crossover studies were also included. Exclusion criteria Uncontrolled trials and controlled trials without randomization or with insuf cient data were excluded. The decision to include or exclude randomized controlled trials was made before performing the meta-analysis. Range of patient characteristics, liver disease and treatment The following items were provided for patients: number of patients excluded after randomization or lost to follow-up; age; sex; cause of cirrhosis; Child±Pugh score; 18 mean duration of follow-up; and stage of hepatic encephalopathy. For each study, the following items were provided: inclusion criteria for patients; duration and dosage of umazenil; washout period between the two treatment periods; end-points; evaluation criteria; use of placebo; and blindness of the study. Methodological quality The methodological quality was assessed independently for each randomized controlled trial by two observers (C. Goulenok and B. Bernard) using a previously validated questionnaire. 19 Fourteen questions were included in this questionnaire: one about the aim of the study; four concerning the description of the studied population; three concerning the blindness of the study; ve concerning the statistical methods; and one concerning the results. The following scores were given for the rst 13 items: 2, yes; 1, partial; 0, no or unknown. The score ranged from ± 2 to 26.
3 FLUMAZENIL IN HEPATIC ENCEPHALOPATHY 363 Source support This meta-analysis was not supported by any pharmaceutical company, government or other grants. Statistical methods To minimize type 1 errors and to avoid nding a difference where none existed, all analyses were performed according to the intention-to-treat method. Criteria for combinability The following methods were used to assess combinability: comparison of each end-point improvement in control groups by the chi-squared test, and heterogeneity tests using the Der Simonian and Laird method 20 and the Peto method. 21 The Peto method is a variant of the Mantel±Haenszel test. The hypothesis states that there is an overall treatment effect ( xed model). In each study, a quanti cation of the size of the difference observed between the two groups ( umazenil and placebo treatment) was performed using the odds ratio (odds ratio (OR) ˆ exp[(observed ± estimated)/ variance(observed ± estimated)] with an estimate of the 95% con dence interval. An estimation of the overall size of the difference between the two groups was then performed using the con dence interval, a graphic representation, a test of the statistical signi cance of the overall difference and a test of the homogeneity of the results obtained in the individual studies. For the Der Simonian and Laird method, the hypothesis states that the treatment effect is not xed (random model). The estimate is the difference between the percentages (risk of treated ± risk of controls). The Der Simonian and Laird method is less sensitive to heterogeneity and more conservative than the Peto method. These two methods were performed for each analysis and all signi cant results were concordant. For each evaluated end-point, the following strategy was used: (i) heterogeneity was assessed for the results in the control groups; (ii) the ef cacy of umazenil vs. no treatment was assessed by the Peto method; and (iii) an assessment was made by the Der Simonian and Laird method. The inclusion of randomized controlled trials with a crossover design in the meta-analysis raised different problems. The number of patients nearly doubled, and not all patients participated in the second stage of crossover. For crossover randomized controlled trials, we performed the meta-analysis on patients included in the rst stage of treatment, and carried out a sensitivity analysis including the two treatment periods. A signi cance level of 5% was taken as the alpha risk. Each estimate was given with its 95% con dence interval. The comparison of the odds ratios of rates of differences between strata was performed using their 95% con dence intervals. When the percentage of an end-point was 0% or 100% in both groups, relative risks and difference rates were impossible to compute directly by the Peto and Der Simonian methods. Therefore, in these cases, a pseudocount method was used: 0.5 was added to the number of subjects in each group; this method does not change the ratio (equal to unity), and only slightly reduces the con dence interval. RESULTS Identi ed and selected randomized controlled trials A total of eight published randomized controlled trials were identi ed between 1994 and Of these, two were excluded. 22, 23 In the rst, the individual data of 13 included patients were not given. 22 In the second, 23 the similar authors, data and protocols indicated that the study was the same as another investigation. 15 Thus, a total of six randomized controlled trials 10±15 comparing umazenil and placebo in the treatment of hepatic encephalopathy in patients with cirrhosis were selected for the meta-analysis. Design of selected randomized controlled trials and patient characteristics The number of patients included in the meta-analysis was 641 (Table 1). Most patients (533, 83%) had Child±Pugh class C. Cirrhosis was due to alcohol in 39% of patients and post-hepatitic in 59%. The inclusion/exclusion criteria of the patients are detailed in Table 2. All patients had cirrhosis and hepatic encephalopathy. Hepatic encephalopathy ranged from grade I to grade IV (clinical grading), with the majority of grade III or IV (93%). In all but one study, patients who had recently taken benzodiazepines were excluded. In the trial of Cadranel et al., 13 three patients had taken benzodiazepines. In all studies, a biological search for benzodiazepine in plasma was systematically performed, and the response to umazenil
4 364 C. GOULENOK et al. Table 1. Patient characteristics Number of patients Number excluded First period (F/P) First period (F/P) Age (years) Males (%) Child±Pugh score Cause of cirrhosis Study Crossover (F/P) Crossover (F/P) F/P F/P (mean or %) F/P F/P Pomier-Layrargues et al /10 0/0 53/58 91/ /11.3 Alcoholic: 73/50 13/15 0/0 Post-hepatitic: 18/30 Cryptogenic: 9/20 Van der Rijt et al. 11 9/9 1/1 50/53 22/56 Child C: 67/44 Alcoholic: 11/44 18/18 2/2 Child B: ±/33 Post-hepatitic: 22/11 Unknown: 33/22 Other: 67/45 Cadranel et al /8 0/ Child C: 89 Alcoholic: 71 18/12 0/0 Child B: 11 Post-hepatitic: 29 Groeneweg et al /15 0/0 53/52 88/60 Child C: 65/53 ND Child B: 29/40 Child A: 6/7 Gyr et al /11 0/0 57/52 79/55 Child C: 71/55 Alcoholic: 57/27 Child B: 29/36 Post-hepatitic: 36/55 Child A: 0/9 Other: 7/18 Barbaro et al /262 0/0 55/52 68/70 Child C: 86/85 Alcoholic: 40/40 527/527 0/0 Child B: 14/15 Post-hepatitic: 59/59 Cryptogenic: 1/1 F, umazenil; P, placebo.
5 FLUMAZENIL IN HEPATIC ENCEPHALOPATHY 365 Table 2. Inclusion criteria and protocols of randomized controlled trials Study Inclusion criteria Exclusion criteria Treatment protocols Follow-up Pomier-Layrargues Cirrhosis proven by biopsy et al. 10 HE stage IV Intake of bzd, recent heavy alcohol abuse End-stage kidney failure, severe respiratory failure Acidosis (ph < 7.30) Pre-existing neurological disease, heart failure Haemodynamic instability 2 mg of F in 5 min 5.5 h after drug injection Van der Rijt et al. 11 Chronic liver disease histologically con rmed Arterial blood ammonia levels >30lmol/L HE by spectral analysis (all stage) Pre-existing neurological disease Pre-existing endocrinological disease Hypersensitivity to bzd or recent use of bzd 1 mg of F in 10 min 4 h later 0.5 mg followed by 0.25 mg/h during 3 days 72 h after drug injection Cadranel et al. 13 Cirrhosis proven by biopsy Unsteady respiratory and haemodynamic state 1 mg of F in 10 min 72 h HE stage II±IV Active gastrointestinal bleeding during previous 4 h Sodium level < 120 mmol/l Serum creatinine > 150 lmol/l Groeneweg et al. 14 Chronic liver disease HE due to cause other than liver failure Three sequential boluses of F 12 h HE stage I±III Severe gastrointestinal bleeding, acute hepatitis Wilson's disease, haemochromatosis (0.4, 0.8, 1 mg) at 1 min intervals, Severe cerebral atrophy, psychiatric disease then 1 mg/h for 3 h Psychotropic medication Gyr et al. 12 Chronic liver failure HE stage I±III HE resulting from precipitating clinical conditions (bleeding, infection) HE stage IV, metabolic coma, liver tumours Severe cerebral atrophy, psychiatric disease Psychotropic medication (including bzd) Three sequential boluses of F (0.4, 0.8, 1 mg) at 1 min intervals, then 1 mg/h for 3 h 12 h Barbaro et al. 15 Cirrhosis proven by biopsy Intake of bzd in preceding 4 days 1 mg of F in 5 min 3 h HE grade III or Iva Recent heavy alcohol abuse Blood creatinine > twice normal values Severe respiratory failure, acidosis (ph < 7.30) Pre-existing neurological diseases, heart failure Intake of drug for the speci c treatment of HE bzd, benzodiazepine; F, umazenil; HE, hepatic encephalopathy.
6 366 C. GOULENOK et al. could thus be analysed in patients who tested positive or negative for benzodiazepines. The treatment design and duration of follow-up are given in Table 2. All studies were placebo-controlled and double-blind. In three randomized controlled trials, 1mg 13, 15 or 2 mg 10 of umazenil was administered over 5±10 min. In two randomized controlled trials, patients received three sequential boluses of umazenil (0.4, 0.8 and 1 mg) over 3 min, followed by intravenous infusion of umazenil at 1 mg/h for 3 h. 12, 14 In one study, nine patients (50%) received 1 mg of umazenil over 10 min, and the other nine received 0.5 mg of umazenil 4 h later, followed by 0.25 mg/h for 3 days. 11 The evaluation criteria are detailed in Table 3. All studies evaluated clinical improvement and ve evaluated electroencephalographic improvement. 10, 11, 13±15 The clinical manifestations of hepatic encephalopathy were evaluated by different grading systems. In two randomized controlled trials, the Glasgow modi ed coma scale, 24 with a worst score of 8 and a best score of 27, was used. 10, 15 The evaluation was blindly made by two observers, clinical improvement being de ned as a positive change in two of the eight items on the clinical coma pro le within 1 h of administration of the study medication. 10 Two randomized controlled trials used a portosystemic encephalopathy (PSE) grading, 25 with an end-point de ned as a 2-point improvement within 3 h of treatment. 12, 14 The last two randomized controlled trials used clinical grading (0±IV) for the evaluation of improvement in clinical neurological function, as given in Table 4. 11, 13 Clinical improvement was de ned as a decrease in both clinical and electroencephalographic 11, 13 hepatic encephalopathy grade. All electroencephalographic gradings were obtained before and after administration of umazenil or placebo. All electroencephalographic analyses were performed blindly. In three randomized controlled trials, electroencephalographic grading was scored by two different observers 10, 15 or one observer 13 according to Fischer's classi cation. 26 A continuous tracing was recorded 15 min before and 15 min after drug administration, and electroencephalographic improvement was de ned as a decrease in electroencephalographic grade. 10, 13, 15 For one study, electroencephalographic improvement was de ned as a decrease in electroencephalographic score according to the conventional classi cation of encephalopathy, and a variation in the relative power of frequency bands measured by spectral analysis within 15 min. 11 In the last randomized controlled trial, 14 a positive response was de ned as a 1-point improvement in electroencephalographic grading using subscores obtained by visual inspection within 2 h. 27 The methodological quality of each selected randomized controlled trial is shown in Table 5. Meta-analysis The details of the clinical and electroencephalographic improvement for each study are given for the metaanalysis and sensitivity analysis in Table 6. Data Table 3. Improvement criteria Study Clinical improvement and methods EEG improvement and methods Pomier-Layrargues et al. 10 Modi ed Glasgow coma scale Fisher's classi cation Positive change of 2/8 items within 1 h Improvement in EEG tracings within 15 min Van der Rijt et al. 11 Clinical grading EEG conventional grading: Decrease of clinical HE grade within 15 min Improvement of EEG stage within 15 min Spectral analysis: Speeding of EEG rhythms Cadranel et al. 13 Clinical grading Fisher's classi cation Decrease of clinical HE grade within 15 min Improvement in EEG tracings within 10 min Groeneweg et al. 14 Clinical PSE grading EEG grading Two-point improvement within 3 h One-point improvement within 3 h Gyr et al. 12 Clinical PSE grading Not evaluated Two-point improvement within 3 h Barbaro et al. 15 Modi ed Glasgow coma scale Fisher's classi cation Improvement of neurological score Improvement in EEG tracings within 10 min EEG, electroencephalographic; HE, hepatic encephalopathy; PSE, portosystemic encephalopathy.
7 FLUMAZENIL IN HEPATIC ENCEPHALOPATHY 367 Table 4. Details of clinical grading 10, 12 and electroencephalographic grading 12, 14 used in randomized controlled trials Coma grade Mental state EEG grade EEG pro le I Euphoria or depression I Irregular background activity (theta and alpha) Mild confusion Slowness, disorder in sleep rhythm II Drowsiness II Continuous theta activity Inappropriate behaviour Burst of delta waves Accentuation of state I III Stupor III Prevalent delta activity. Polyphasic transients Patient sleeps most of the time but is rousable Sharp and slow wave complexes Incoherent speech, marked confusion IVa Coma IVa Continuous delta activity Coordinated response to painful stimuli Abundant sharp and slow wave complexes EEG reactivity present IVb Coma IVb Slower activity (delta and some polyphasic transients) Hyperextension and pronosupination after EEG reactivity ˆ 0 painful stimuli IVc Coma IVc Discontinuous activity with silent periods No response to painful stimuli V Flat EEG, electroencephalographic. concerning the short-term survival of patients receiving umazenil or placebo were not given. Clinical improvement. Five randomized controlled trials included a total of 317 patients in the treated groups and 306 patients in the placebo groups. 10, 12±15 In one study, results were given for all the patients in a crossover design. 11 Heterogeneity tests between control groups were not signi cant. The mean percentages of patients with clinical improvement were 27% in the treated groups and 3% in the placebo groups (Figure 1). The difference between the two groups was signi cant by both methods (Peto method: OR ˆ 6.15; 95% con dence interval (CI), 4.0±9.5; P < 0.001; Der Simonian method: mean rate difference, 28.6%; 95% CI, 16.7± 40.5; P < 0.001) without signi cant heterogeneity. The sensitivity analysis included the second period of 10, 11, 13, 15 the crossover studies. This analysis was performed with the six randomized controlled trials including a total of 607 treated patients and 598 patients in placebo groups. The mean percentages of patients with clinical improvement were 19% in treated groups and 2% in placebo groups (Figure 2). The difference was signi cant in favour of umazenil with the Peto method (OR, 5.50; 95% CI, 3.8±7.9; P < 0.001) and the Der Simonian method (mean rate difference, 30.9%; 95% CI, 14.3±47.5; P < 0.001). Electroencephalographic improvement. Three randomized controlled trials, 11, 14, 15 including a total of 291 treated patients and 286 in placebo groups, were available. Heterogeneity tests between control groups were not signi cant. The mean percentages of patients with electroencephalographic improvement were 35% in treated groups and 6% in placebo groups. This difference was signi cant with the Peto method (OR ˆ 5.8; 95% CI, 3.4±9.7; P < 0.001) (Figure 3) but not signi cant with the Der Simonian method (mean rate difference, 16.1%; 95% CI, ± 6.8, ; P ˆ 0.17). The sensitivity analysis was performed with ve randomized controlled trials, 10, 11, 13±15 including 593 treated patients and 587 patients in placebo groups. The mean percentages of patients with electroencephalographic improvement were 25% in treated groups and 4% in placebo groups (Figure 4). The difference was signi cant in favour of umazenil with the Peto method (OR, 5.2; 95% CI, 3.7±7.1; P < 0.001) and the Der Simonian method (mean rate difference, 21.8%; 95% CI, 5.2±38.4; P < 0.01). DISCUSSION The mechanisms involved in the development of hepatic encephalopathy are complex. The hypothesis of a poor
8 368 C. GOULENOK et al. Table 5. Questionnaire for assessing methodological quality of randomized controlled trials Pomier- Layrargues et al. 10 Van der Rijt et al. 11 Cadranel et al. 13 Groeneweg et al. 14 Gyr et al. 12 Barbaro et al Is (are) the major end-point(s) clearly stated? Description of the population 2 Are the criteria of inclusion and exclusion clearly stated? 3 Is the number of patients seen and rejected given? 4 Is the number of patients randomized clearly stated for both groups? 5 Are the number of patients who were withdrawn and dropped out and thereasons clearly stated? Blindness of the trial 6 Are the patients blind to treatment? Are the physicians caring for patients blind to treatment? 8 Are the persons assessing major end-points blind to treatment? Statistical method 9 Is the necessary sample size precalculated? Is the method of randomization given and is this randomization blind? 11 Are pre-treatment variables adequately analysed and discussed? 12 Are statistical tests used appropriate? Is handling of withdrawals correct? If `non-signi cant' trial, the smallest sample size per group is: greater or equal to 100 (0), between 30 and 100 () 1), smaller than 30 () 2) 0 ) 2 0 ) 2 ) 1 2 Total score () 2 to 26) For the rst 13 items: 2, yes; 1, partial; 0, no or unknown. hepatic function and portosystemic shunts, responsible for an increase in arterial ammonia, is commonly accepted. 3 The arterial level of ammonia is, however, poorly correlated with the severity of hepatic encephalopathy. 28 Several other hypotheses have been proposed: deposition of manganese in the basal ganglia, 29 de ciency of zinc (responsive to a decrease in urea cycle enzymes) and hyperactivity of the GABA-ergic system. However, as stated by Basile and Jones, 3 the two main theories for hepatic encephalopathy in cirrhotic patients, i.e. the ammonia theory and the GABA-ergic theory, are not mutually exclusive. The rationale for the latter hypothesis is that benzodiazepines have a depressant effect on the central nervous system with a high af nity for GABA receptors. Endogenous benzodiazepine agonists, called benzodiazepine-like substances, have been found in experimental studies, 25, 30 and have been detected in the cerebral tissue, cerebrospinal uid and serum of cirrhotic patients with hepatic encephalopathy. Flumazenil is usually used for the diagnosis of suspected poisoning in emergency situations and for the treatment of benzodiazepine intoxication. 31 This antagonist property has been proposed in the treatment of hepatic encephalopathy. Open studies with umazenil in humans have shown an improvement of hepatic encephalopathy. 5±9 The results of randomized controlled trials are more controversial. However, in the largest recently published study, accounting for 82% of the patients included in this meta-analysis, a positive clinical and electroencephalographic effect of umazenil
9 FLUMAZENIL IN HEPATIC ENCEPHALOPATHY 369 Table 6. Results of meta-analysis and sensitivity analysis Clinical improvement EEG improvement Clinical improvement EEG improvement Meta-analysis Meta-analysis Sensitivity analysis Sensitivity analysis Flumazenil Placebo Flumazenil Placebo Flumazenil Placebo Flumazenil Placebo R/T R/T R/T R/T R/T R/T R/T R/T Pomier-Layrargues et al. 10 5/11 0/10 6/13 0/15 4/13 2/15 Van der Rijt et al. 11 0/9 0/9 6/18 2/18 0/18 0/18 Cadranel et al. 13 6/10 0/8 12/18 0/12 12/18 0/12 Groeneweg et al. 14 3/17 0/15 5/17 2/15 3/17 0/15 5/17 2/15 Gyr et al. 12 5/14 0/11 5/14 0/11 Barbaro et al /265 9/262 96/265 14/262 85/527 13/ /527 22/527 EEG, electroencephalographic; R/T, responders/total patients. was established. 15 In the present meta-analysis, the ef cacy of umazenil was not related to the presence or absence of exogenous benzodiazepines. Thus, umazenil had a positive effect whether or not benzodiazepines were present in sera. Indeed, in ve of the six included studies, a previous intake of psychotropic medication, including benzodiazepines, was an exclusion criterion. 10±12, 14, 15 In addition, screening tests for benzodiazepines were performed in ve of the six trials, and positivity for benzodiazepines in blood or urine samples was an exclusion criterion before data analysis (but not in the study of Cadranel et al. 13 ). In these studies, patients with a positive test for benzodiazepines were included in the intention-to-treat method and excluded from the per protocol analysis. The biochemical test for the detection of benzodiazepines was very sensitive in most studies: 10±13, 15 thin layer chromatography (level of detection, 10±12 ng/ml) was used in two studies, 13, 15 and gas chromatography to detect diazepam or N-desmethyl-diazepine, with a higher level of detection for exogenous benzodiazepines, was used in Figure 1. Mean percentage of patients with clinical improvement (Der Simonian). The trials are listed on the left. For each trial, the small black vertical line shows the difference in ef cacy between the two treatments, with 95% con dence interval. A difference of zero is shown by the long black vertical line. When a trial shows a signi cant difference, the con dence interval does not cross the long black vertical line. If the small black vertical line is on the right side of the gure, the difference is in favour of umazenil. The mean rate difference is shown by the small black vertical line with its con dence interval located on the `total' line. Figure 2. Mean percentage of patients with clinical improvement in the sensitivity analysis (Der Simonian).
10 370 C. GOULENOK et al. Figure 3. Mean percentage of patients with electroencephalographic improvement (Peto). two studies. 10, 15 Twenty-three patients (3.5%) had a positive test; 10±13, 15 of these, 10 were non-responders to umazenil (43%). These results are in agreement with the hypothesis of Barbaro et al. 15 that umazenil may act on benzodiazepine receptor agonist ligands, which are synthesized in situ in the brains of patients with liver failure, independent of exogenous benzodiazepines. Thus, in clinical practice, umazenil can be used Figure 4. Mean percentage of patients with electroencephalographic improvement in the sensitivity analysis (Der Simonian). in cirrhotic patients with hepatic encephalopathy without waiting for the results of benzodiazepines in blood or urine samples. This meta-analysis shows that umazenil signi cantly improves the clinical and electroencephalographic signs of encephalopathy. The methodological quality of each study was high based on the validated questionnaire established by Poynard 19 (mean, 21.6; worst score, ± 2; best score, 26). The clinical homogeneity of the patients included in the selected randomized controlled trials is clear: all patients had cirrhosis and clinical hepatic encephalopathy, and all patients with evident causes of hepatic encephalopathy (gastrointestinal bleeding, hyponatraemia, renal failure, infection) were not included, or were included after treatment of these complications. 13 The sample sizes of the selected randomized controlled trials were different, but there was no heterogeneity between trials and treatment effects. Indeed, heterogeneity tests for clinical and electroencephalographic improvements were not signi cant. The mean percentage of patients with clinical improvement was 27% in treated patients, with a mean rate difference between treated and nontreated patients reaching 29% (P < 0.01). The sensitivity analysis con rmed this result with a mean rate difference of 31% (P < 0.001). For electroencephalographic improvement, the meta-analysis showed a non-signi cant improvement with the Der Simonian method (16%), but a signi cant difference with the Peto method (OR, 5.96; P < 0.001). The sensitivity analysis con rmed this result with a signi cant difference in favour of umazenil with both methods (22%, P < 0.01). No adverse events were mentioned in four randomized controlled trials. 10, 11, 13, 14 In the two remaining trials, adverse events were systematically recorded. 12, 15 In one trial, four patients in the umazenil group had ushing, nausea and irritability, whereas no adverse events were mentioned in the placebo group. 12 No sideeffects were observed in the other trial. 15 Therefore, tolerance to umazenil seems to be acceptable. The ef cacy of umazenil has therefore been clearly shown in the treatment of hepatic encephalopathy. However, data concerning survival are not available. Moreover, hepatic encephalopathy is often precipitated by triggering factors, such as bacterial infection or gastrointestinal bleeding, which require speci c treatment to which umazenil can be added. The consequences of acute hepatic encephalopathy may be