Alimentary Pharmacology and Therapeutics
Systematic review with meta-analysis: the haemodynamic effects of carvedilol compared with propranolol for portal hypertension in cirrhosis E. Sinagra†,§,1, G. Perricone†,‡,1, M. D’Amico¶, F. Tine* & G. D’Amico*
*Gastroenterology Unit, A.O. Ospedali Riuniti Villa Sofia Cervello – P.O. Vincenzo Cervello, Palermo, Italy. † Internal Medicine Unit, A.O. Ospedali Riuniti Villa Sofia Cervello – P.O. Vincenzo Cervello, Palermo, Italy. ‡ Hepatology and Gastroenterology Unit, A.O. Ospedale Niguarda Ca’ Granda, Piazza Ospedale Maggiore, Milano, Italy. § Endoscopy and Gastroenterology Unit, Fondazione Istituto San Raffaele Giglio, Cefal u, Italy. ¶ Radiology Department, Universita di Palermo, Palermo, Italy.
Corresponding author: Dr G. D’Amico, Gastroenterology Unit, A.O. Ospedali Riuniti Villa Sofia Cervello – P.O. Vincenzo Cervello, via Trabucco 180, 90146 Palermo, Italy. E-mail: [email protected] These authors equally shared the first authorship. 1
Publication data Submitted 23 November 2013 First decision 9 December 2013 Resubmitted 3 January 2014 Accepted 6 January 2014 EV Pub Online 26 January 2014 As part of AP&T’s peer-review process, a technical check of this meta-analysis was performed by Mr M. Siddiqui.
SUMMARY Background Propranolol is recommended for prophylaxis of variceal bleeding in cirrhosis. Carvedilol is a nonselective beta-blocker with a mild anti-alfa-1-adrenergic activity. Several studies have compared carvedilol and propranolol, yielding inconsistent results. Aim To perform a systematic review and meta-analysis of the randomised clinical trials comparing carvedilol with propranolol for hepatic vein pressure gradient reduction. Methods Studies were searched on the MEDLINE, EMBASE and Cochrane library databases up to November 2013. The weighted mean difference in percent hepatic vein pressure gradient reduction and the relative risk of failure to achieve a hemodynamic response (reduction ≥20% of baseline or to ≤12 mmHg) with each drug were used as measures of treatment efficacy. Results Five studies (175 patients) were included. Indication to treatment was primary prophylaxis of variceal bleeding in 76% of patients. There were overall three acute (60–90 min after drug administration) and three long-term (after 7–90 days of therapy) comparisons. The summary mean weighted difference in % of reduction in hepatic vein pressure gradient was: acute 7.70 (CI 12.40, 3.00), long-term 6.81 (CI 11.35, 2.26), overall 7.24 (CI 10.50, 3.97), favouring carvedilol. The summary relative risk of failure to achieve a hemodynamic response with carvedilol was 0.66 (CI 0.44, 1.00). Adverse events were nonsignificantly more frequent and serious with carvedilol. However, quality of trials was mostly unsatisfactory. Conclusions Carvedilol reduces portal hypertension significantly more than propranolol. However, available data do not allow a satisfactory comparison of adverse events. These results suggest a potential for a cautious clinical use. Aliment Pharmacol Ther 2014; 39: 557–568
ª 2014 John Wiley & Sons Ltd doi:10.1111/apt.12634
557
E. Sinagra et al. INTRODUCTION Portal hypertension is an almost unavoidable complication of cirrhosis1 and variceal haemorrhage is its most dramatic consequence, bearing a mortality of 10–15% per episode.2 The hepatic venous pressure gradient (HVPG) is significantly associated with the risk of variceal bleeding, liver decompensation and hepatocellular carcinoma (HCC).3–5 A reduction in HVPG by 20% of baseline results in a significant reduction in the bleeding risk and a reduction to less than 12 mmHg virtually abolishes the risk.3, 6 Propranolol, a nonselective betablocker (NSBB), significantly reduces HVPG and is currently recommended for the prevention of primary and secondary variceal bleeding. However, it achieves the target HVPG reduction only in approximately one-third of patients, possibly because the increase in porto-collateral resistance impairs the reduction in portal pressure.7–9 This effect may be reduced by adding a vasodilator such as isosorbide mononitrate, which in fact rescues most of nonresponders to propranolol, although at the expense of increasing side effects.10 Carvedilol is a potent NSBB with a mild anti-alfa–1– adrenergic activity, which confers to the drug a potential for a higher portal pressure reduction compared with other NSBB. Several studies have compared the acute or long-term hemodynamic response of carvedilol and propranolol yielding inconsistent results.11 The aim of the present study was to assess the hemodynamic effects of carvedilol compared with propranolol for portal hypertension. With this aim, we performed a systematic review of randomised controlled trials (RCTs) comparing the two drugs for their effect on hepatic vein pressure gradient in cirrhosis. METHODS Study selection and participants All the studies that met the following inclusion criteria were included: (i) the study was an RCT comparing carvedilol with propranolol for acute or long-term effect on HVPG in cirrhotic patients; (ii) reduction in HVPG under treatment was a measure of the treatment effect; (iii) participants in the included studies were cirrhotic patients with clinically significant portal hypertension as indicated by HVPG >10 mmHg and/or presence of oesophageal varices, according to the Baveno III criteria.12 Randomised trials were included irrespective of blinding, publication status or language. Trials using quasi-randomisation, based on case record number, date of birth, day of the week or similar, were excluded. 558
Compared treatments Oral carvedilol was compared with either oral or i.v. propranolol for the acute hemodynamic effect, or oral carvedilol compared with oral propranolol for the long-term effect. The assessment of the hemodynamic effect was considered ‘acute’ when performed during the same hepatic veins catheterisation. When a second catheterisation was required (after days or months of treatment), it was classified as assessment of long-term treatment effect. Treatment outcome The primary measure of the treatment effect was the difference in % HVPG reduction achieved with each of the two compared drugs. The proportion of patients failing to achieve a hemodynamic response to the treatment was considered a secondary outcome measure. Hemodynamic responders were considered those patients achieving a HVPG reduction ≥20% of baseline value or in whom HVPG was lowered to ≤12 mmHg.3, 13, 14 The number and severity of adverse events were also assessed as well as the difference in mean arterial blood pressure reduction (MAP) with each drug as an indicator of systemic treatment effect. Search strategy Medline, Embase, Cochrane Controlled Trial Registry and The Cochrane Library were searched up to November 2013 to retrieve pertinent studies. Databases were searched by the following search terms: carvedilol, propranolol, portal hypertension, randomised trial. A manual search of the reference lists of pertinent studies, articles, reviews, editorials from the English language medical literature was also performed. Proceedings of the most relevant international congresses were handsearched. Data collection and analysis Decisions on which trials to include were taken independently by two of the authors (ES and GP), who were unblinded with regard to the names of the authors, investigators, institution, source, and results of the identified trials. Disagreements were resolved by discussion with all the authors. Excluded trials were identified with the reason for exclusion. Data were extracted by two independent authors (ES and GP) and discrepancies were solved by discussion. The following data were extracted from each trial: (i) trial characteristics: time interval between trial drug administration and hemodynamic effect assessment; definition of the outcome of Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Systematic review with meta-analysis: carvedilol vs. propranolol for portal hypertension interest; (ii) patients’ characteristics: number of patients, mean age, sex, aetiology of cirrhosis, Child-Pugh score, proportion of patients with previous variceal bleeding; (iii) outcome: mean HVPG change under trial treatment; number of patients achieving a hemodynamic response; adverse events observed during the treatment period.
Assessment of study quality Two of the authors (GP and ES) independently assessed bias risk of the trials, without masking the trial names. For this purpose, the risk of overestimation of intervention effects in randomised trials due to inadequate methodological quality was assessed using the following domains.15, 16 Generation of the allocation sequence was defined adequate if sequences were generated by table of random numbers or computer-generated random numbers or similar; unclear if it was not described; inadequate if sequences could be related to prognosis. Allocation concealment was defined adequate when a centralised randomisation or sequentially numbered, sealed, opaque envelopes were used; unclear when studies did not report any concealment approach; inadequate when the following modalities were used: alternation, the use of case record numbers, dates of birth or day of the week, and any procedure that is entirely transparent before allocation, such as an open list of random numbers. Blinding was considered adequate when the study was double-blind or single-blind (HVPG measurements were blindly evaluated); unclear when no information was reported; inadequate when an open design was used. Handling of dropouts and withdrawals was considered adequate when numbers of patients withdrawn or lost and reasons were described; inadequate otherwise. Intention-to-treat analysis was considered adequate when all randomised patients were included in the analyses and inadequate otherwise. Following the above definitions, an included trial was judged as achieving inadequate or unclear control of bias when the methodological quality was considered as inadequate or, respectively, unclear in at least 1 domain. If the quality was judged adequate in all the domains, then the control of bias was considered adequate. Statistical methods Trials and patient characteristics are reported as number of observations, proportions or means standard deviations. Where standard error of mean was reported in the original article, the standard deviation was derived from the standard error. Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Analyses were performed by STATA 10 (College Station, TX 77845 USA, ©1984–2009) by a fixed-effect model if no statistically significant heterogeneity was found and by a random-effects model in case of significant heterogeneity (P < 0.10). The treatment effect was calculated as the weighted mean difference (WMD) between the % HVPG reduction from baseline achieved with propranolol and the % reduction achieved with carvedilol. Data are expressed as % HVPG reduction with propranolol minus % HVPG reduction with carvedilol. Therefore, negative values indicate a higher HVPG reduction with carvedilol. The pooled WMD is reported as the summary statistic of treatment effect together with 95% confidence interval (CI). The pooled relative risk (RR) of failure to achieve a hemodynamic response with either treatment was also assessed and reported with 95% CI. Inter-trial heterogeneity was statistically assessed by a chi-square test and was expressed using I2 values.17 If significant heterogeneity was found, potential reasons for heterogeneity were explored and combinability of trials was reassessed accordingly. Analyses were performed using the intention-to-treat principle including all patients randomised as they were reported in each single study. Trials were first combined independently of whether they assessed acute or long-term hemodynamic effect of the study treatment. However, separate analyses according to the timing of assessment were also performed. To explore the robustness of estimates, the following sensitivity analyses were performed: (i) random-effects model; (ii) standardised mean difference (SMD) instead of WMD to overcome potential measurement differences across trials; (iii) excluding trials with acute assessment; (iv) excluding trials with long-term assessment; (v) excluding trials with inadequate control of bias. The weighted mean difference in arterial blood pressure with the two drugs was also assessed as a measure of a major treatment side effect.
Study protocol The full study protocol is available by contacting the corresponding author of this article. RESULTS Trial retrieval Overall, 39 potentially eligible references were retrieved in the literature search. Thirty references were excluded by reading the title and abstract, because they did not 559
E. Sinagra et al. 38 references identified through database serching
One additional reference identified by handsearch
39 references screened 30 excluded
9 references assessed for eligibility 7 references to 5 pertinent trials
2 RCTs exluded because did not include hemodynamic assessment or β-blocker control group: - one compared carvedilol with band ligation of esophageal varices for primary prophylaxis of variceal bleeding - the other compared carvedilol with nadolol plus isosorbide mononitrate for secondary prophylaxis of variceal bleeding
5 RCTs included In quantitative synthesis (meta-analysis)
refer to randomised trials. The remaining 9 references were fully assessed for inclusion in the study (Figure 1). Among these, two referred to full articles and were excluded for the following reasons: one was a randomised controlled trial of carvedilol vs. variceal band ligation for the prevention of the first variceal bleed;18 the other was a randomised controlled trial of carvedilol vs. nadolol plus isosorbide mononitrate for the prevention of variceal rebleeding, which did not include hemodynamic assessments.19 Seven references20–26 to five pertinent studies were eventually found: two of these referred to the preliminary report20 or to a subgroup analysis25 of a fully published study26 and were therefore excluded. The remaining five references21–24, 26 referred to randomised controlled trials, which fulfilled our inclusion criteria and were included in the quantitative meta-analysis.
Description of included studies The characteristics of the five included studies are summarised in Tables 1 and 2. In two studies, the acute hemodynamic effect was assessed 60 and 90 min after trial drug administration21, 24 respectively. In two studies, the treatment effect was assessed after a mean of 11 weeks22 and 92.7 days.26 In one study,23 both acute and long-term effects were assessed at 90 min and 1 week after initial trial drug administration, respectively. Overall, therefore, there were three acute and three long-term evaluations available for meta-analysis. In the acute comparisons, all treated patients were given a tablet of carvedilol 25 mg, while control patients were given propranolol 0.15 mg/kg i.v. followed by a continuous infusion of 0.2 mg/kg/h in one study,21 40 mg p.o.24 or 80 mg p.o.23 In one acute study,24 560
Figure 1 | Flow diagram of the selection process of randomised controlled trials for inclusion in the metaanalysis. RCT, randomised controlled trials.
propranolol was associated with isosorbide mononitrate 20 mg in the control group. In the evaluation of the long-term hemodynamic response, treated patients received a mean dose of 31 mg of oral carvedilol and controls received a mean of 73 mg of oral propranolol in one study;22 a fixed oral dose of 12.5 mg of carvedilol was compared with a fixed oral dose of 80 mg of propranolol in another;23 and a mean dose of 14 mg of carvedilol and 122 mg of propranolol in the third.26 A second HVPG measurement was obtained in all the included patients in three studies.21, 24, 26 In one study,22 discontinuation of treatment caused missing second HVPG measurement in two patients in the carvedilol group (heart failure one and pericardial effusion the other) and in three in the propranolol group (portal systemic encephalopathy 1 and fatigue 2). In the other study,23 discontinuation of treatment hampered a second HVPG measurement in one patient in the carvedilol group because of severe hypotension and in two patients in the propranolol group because of bleeding. Control of bias was adequate in one trial, unclear in three and inadequate in one trial (Table 2). In the four trials with inadequate or unclear control of bias, this was mostly due to the generation of the allocation sequence or to allocation concealment. The characteristics of the 175 patients included in the five studies are summarised in Table 3. Males were 74%. The number of patients with ascites was reported only in two studies22, 23 and was 44 of 87 (51%); patients with a previous bleeding episode were 42 of 153 (27%; information not reported in one study24). Mean age was 55 years and aetiology was mostly viral or alcoholic in all the Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Systematic review with meta-analysis: carvedilol vs. propranolol for portal hypertension Table 1 | Characteristics of included trials
Author, year
Participants
N of randomised patients C/P
Interventions C; P* C: Carvedilol 25 mg p.o. P: Propranolol (0.15 mg/kg i.v. followed by a continuous infusion of 0.2 mg/kg/h) C: Carvedilol 31 4 mg/day p.o.(range, 12.5–50)* P: Propranolol 73 10 mg/day p.o. (range, 10–160)* Acute: C: Carvedilol 25 mg p.o.; P: Propranolol 80 mg p.o. Maintenance: C: carvedilol 6.25 mg p.o. twice daily P: propranolol 40 mg p.o. twice daily C: Carvedilol 25 mg p.o. P: propranolol 40 mg p.o + isosorbide 5-mononitrate 20 mg p.o. C: Carvedilol 14 7 mg* P: Propranolol 122 64 mg*
Banares, 1999
Cirrhotic patients with oesophageal varices, with or without previous variceal bleeding
14/14 7 more patients were randomised to placebo
Banares,2002
Cirrhotic patients with oesophageal varices, no previous bleeding, and HVPG >12 mmHg
26/25
De, 2002
Cirrhotic patients with oesophageal varices, with or without previous variceal bleeding
18/18
Lin, 2004
Cirrhotic patients with HVPG ≥12 with or without previous variceal bleeding
11/11
Hobolth, 2012
Cirrhotic patients with HVPG≥12 mmHg, independent of varices or previous bleeding
21/17
Time of outcome assessment
N of patients with 1st/2nd HVPG measurement Carvedilol
Propranolol
60 min
14/14
14/14
11 4.1 weeks†
26/24
25/22
Acute: 90 min Long-term: 1 week
Acute: 18/18 Long-term 18/17
Acute: 18/18 Long-term 18/16
90 min
11/11
11/11
92.7 13.6 days†
21/21
17/17
C, Carvedilol group; P, Propranolol group; p.o., orally; i.v., intravenously. * Mean dose standard deviation. † Interval between baseline and second HVPG measurement, mean s.d.
studies but one where only patients with cirrhosis from HBV and/or HCV were included.24 Child-Pugh class C patients were overall 31 of 153 (20%) in four studies and the mean Child-Pugh score was eight in the fifth.24 No appreciable differences in major patient characteristics were observed between the two study groups across the five studies (Table 3).
Outcome evaluation HVPG reduction. Mean baseline HVPG was 18.3 mmHg (median 18, range 16.6–20.4) in propranolol-treated patients and 18.8 mmHg (median 19, range 17.6–19.5) Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
in the carvedilol-treated patients. The mean % HVPG reduction under treatment was 15.6 (median 12.6, range 10.1–23.2) with propranolol and 22.2 (median 19.8, range 18.6–27.7) with carvedilol (crude mean difference 6.61%). The pooled WMD between % HVPG reductions with the two drugs was 7.24 ( 10.50, 3.97; fixed-effect model) indicating a significantly higher HVPG reduction with carvedilol than propranolol. No heterogeneity was found (P = 0.99; I2 = 0.0) (Figure 2). A sensitivity analysis performed by using the standardised mean difference between the % HVPG reductions with the two drugs 561
E. Sinagra et al. Table 2 | Methodological quality assessment of included trials*
Author, year
Generation of the allocation sequence
Allocation concealment
Blinding
Dropouts and withdrawals
Intention-totreat analysis
Overall risk of bias
Banares, 1999 Banares, 2002 De, 2002 Lin, 2004 Hobolth, 2012
Unclear Adequate Unclear Adequate Unclear
Unclear Adequate Inadequate Unclear Adequate
Adequate† Adequate† Adequate‡ Unclear Adequate‡
Adequate Adequate Adequate Adequate Adequate
Adequate Adequate Adequate Adequate Adequate
Unclear Adequate Inadequate Unclear Unclear
* Definition used for rating of each quality item are reported in the method section. † Blinded reading of HVPG measurements. ‡ Double-blind.
Table 3 | Characteristics of participants in the included studies
Banares, 1999 Banares, 2002 De, 2002 Lin, 2004 Hobolth, 2012
Treatment group
Patients N
Age, mean s.d.
C P C P C P C P C P
14 14 26 25 18 18 11 11 21 17
54.6 51.4 57.9 58.4 42.3 47.3 59 61 58.2 56.2
8.8 8.5 7.6 11 11.9 12.9 13.3 13.3 6.8 6.1
Males N
Alcoholic/ viral aetiology N
Child-Pugh class A/B/ C, N
Oesophageal varices large/ small, N
Ascites N
Previous variceal bleeding, N
NR NR 19 15 15 17 8 9 12 12
8/NR 8/NR 6/18 9/16 5/9 10/5 0/11 0/11 18/2 12/1
8/4/2 5/6/3 13/10/3 15/6/4 5/9/4 0/13/5 8.7 3.0† 7.3 2.7† 8/7/6 6/6/4
NR NR 10/16 14/11 NR NR NR NR 4/10 1/11
7* 7* 10 6 12 16 NR NR NR NR
9 9 0 0 7 7 NR NR 5 5
C, Carvedilol; P, Propranolol. * Previous ascites. † Child-Pugh score, mean standard deviation (derived from the standard error reported in the article).
confirmed the principal analysis: pooled SMD 0.50 (CI 0.78, 0.23; fixed-effect model) without significant heterogeneity (P = 0.5; I2 = 0.0). The other sensitivity analyses confirmed that either the acute or the long-term % HVPG reduction is significantly higher with carvedilol than with propranolol (Figure 2) and superiority of carvedilol held true also by excluding the study with inadequate control of bias22 (WMD 7.41, CI 10.79, 4.03; data not shown). The mean % HVPG reduction was ≥20% in 2 of 6 comparisons with propranolol and in 3 of 6 with carvedilol. Overall, a reduction in HVPG of ≥20% or to ≤12 mmHg was achieved with propranolol in 33 of 87 evaluable patients as compared to 57 of 94 with carvedilol (including patients with both acute and chronic assessments; P = 0.003, Fisher’s exact test) (Table 4). 562
The pooled relative risk of failure to achieve a hemodynamic response with carvedilol was 0.66 (CI 0.44–1.00) (Figure 3).
Adverse events. Mean arterial blood pressure (MAP) was reduced more by carvedilol (WMD 10.40; 95% CI: 13.90, 6.90; I2 = 0.0%, P = 0.477) than by propranolol (WMD 6.35; 95% CI 9.86, 2.83; I2 = 0.0%, P = 0.936) (Figure 4). The lack of complete information on the difference in MAP reduction between the two drugs hampered obtaining a summary statistics. However, as the CI of MAP reduction achieved with each drug largely overlapped, it may be concluded that the difference in MAP reduction between the two drugs was not significantly different (Figure 4). On the other hand, even if not statistically significant, this difference was Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Systematic review with meta-analysis: carvedilol vs. propranolol for portal hypertension
Acute assessment Banares (1999)
–7.70 (–13.24, –2.16)
34.70
De (2002)
–4.69 (–23.94, 14.56)
2.88
Lin (2004)
–8.50 (–18.48, 1.48)
10.71
Subtotal (I 2 = 0.0%, P = 0.942)
–7.70 (–12.40, –3.00)
48.29
Banares (2002)
–7.00 (–12.26, –1.74)
38.48
De(2002)
–4.72 (–21.99, 12.55)
3.58
Hobolth (2012)
–6.80 (–17.31, 3.71)
9.66
Subtotal (I 2 = 0.0%, P = 0.970)
–6.81 (–11.35, –2.26)
51.71
–7.24 (–10.50, –3.97)
100.00
Long-term assessment
Heterogeneity between groups: P = 0.789
Overall (I 2 = 0.0%, P = 0.998)
–20 –10 Better carvedilol
0
10 20 Better propranolol
Weighted Mean Difference: fixed effects
Figure 2 | Weighted mean difference (WMD) (fixed-effect model) of the % hepatic vein pressure gradient (HVPG) reduction between propranolol and carvedilol in the subsets of RCTs assessing the acute and, respectively, the longterm treatment effect together with the overall estimation of the whole set of studies. Each study is identified by the name of the first author and year of publication (references in the text). Squares indicate the WMD per each trial and the size of the squares is proportional to the weight of trials. The horizontal bars denote the 95% confidence intervals of WMD. The vertical solid line is the equivalence line, where WMD is equal to 0. Differences are given as values in the propranolol group minus values in the carvedilol group. Therefore, negative values (WMDs on the left of the equivalence line) denote superiority, whereas those on the right denote inferiority of carvedilol. The diamonds represent 95% CI of the WMDs per each subset and in whole set of RCTs. The vertical dashed line represents the pooled WMD of the whole set of trials. The study by De (2002) provided estimation of either acute or long-term treatment effect.
clinically relevant, with at least one symptomatic episode of hypotension reported with carvedilol and a total of 14 patients with orthostatic hypotension out of 65 patients treated with carvedilol, compared to 9/60 with propranolol (P = 0.37) in three of the five studies.22, 23, 26 Other adverse events, even if not fully reported in all the studies, were the following: need for increasing diuretics in 14/47 patients treated with carvedilol and in 6/42 with propranol;22, 26 shortness of breath in 12/47 and 7/42;22, 26 portosystemic encephalopathy in 3/26 and 4/2522 respectively. Overall carvedilol was withdrawn in 6/65 and propranolol in 8/60 treated patients22, 23, 26 because of adverse events. None of these differences was statistically significant. Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
DISCUSSION The present meta-analysis shows that, in the available RCTs comparing propranolol and carvedilol for portal hypertension, carvedilol reduces HVPG significantly more than propranolol. Superiority of carvedilol is consistent across all the trials, regardless of acute or longterm assessment. The summary weighed mean difference between the percent HVPG reduction achieved with propranolol and carvedilol is 7.24 (CI 10.50 to 3.97) in favour of carvedilol. No heterogeneity was found and all the sensitivity analyses confirmed the principal result. Patients in the two treatment groups were fairly comparable across the included studies, regarding demography, aetiology, indicators of disease severity, baseline 563
E. Sinagra et al. Table 4 | Main results of the included studies Carvedilol
Author, year
Baseline HVPG, mmHg mean s.d.*
Acute assessment Banares, 1999 19.5 De, 2002 19.0 Lin, 2004 18.9 Subtotal – Long-term assessment Banares, 2002 19.1 De, 2002 19.0 Hobolth, 2012 17.6 Subtotal – Total
–
4.9 3.8 6.0
Propranolol
% HVPG reduction, mean s.d.*
Hemo-dynamic responders† n/N (%)§
20.4 7.5 27.7 31.5 18.6 11.9 –
9/14 (64) 11/18 (61)‡ NR –
20.4 4.1 16.60 3.9 17.6 4.0 –
–
19 9.8 27.7 31.5 19.3 16.1 –
13/24 (54) 11/17 (65)‡ 13/21 (62) –
20.3 4.2 16.6 4.0 18.4 3.6 –
12 9.4 22.98 20.1 12.5 16.7 –
–
–
–
–
–
–
5.4 3.8 4.2
Baseline HVPG, mean s.d.*
Weighed Mean Difference in % HVPG reduction (CI)¶
% HVPG reduction, mean s.d.*
Hemo-dynamic responders† n/N (%)§
12.7 7.48 22.9 27.4 10.1 11.9
2/14 (14) 9/18 (50)
7.70 4.69 8.50 7.70
5/22 (23) 10/16 (62) 7/17 (41)
7.00 ( 4.72 ( 6.80 ( 6.81 ( 11.35, 7.24 (
NR –
( ( ( (
13.24, 2.16) 23.94, 14.56) 18.48, 1.48) 12.4, 3.00) 12.26, 1.74) 21.99, 12.55) 17.31, 3.71) 2.26) 10.50,
3.97)
* s.d., standard deviation of mean. In the studies by Banares 1999 and 2002 and by Lin 2004, the standard deviation of the mean was calculated from the standard error of the mean reported in the original articles. † Patients achieving a HVPG reduction ≥20% or HVPG ≤12 mmHg. ‡ Hemodynamic response was defined as a reduction in HVPG of ≥20% of baseline in the De’s study. § n, number of responders; N, total number of evaluable patients. ¶ The difference is calculated as % HVPG reduction with propranolol carvedilol.
HVPG and history of previous bleeding. A second measurement of HVPG was available in all the included patients, but three in the De’s study23 and five in the long-term study by Banares.22 Only two of the missing second HVPG measurements were due to bleeding (both in proporanolol-treated patients), while the others were caused by adverse events requiring treatment withdrawal. Control of bias was considered inadequate in only one trial and the beneficial effect of carvedilol over propranolol was confirmed also when excluding this study. On the other hand, control of bias was adequate in only one study: this finding calls for caution when interpreting the present results. The number of patients achieving a reduction in HVPG to ≥20% or to ≤12 mmHg was reported in 4 of the 5 studies and was also markedly higher with carvedilol (57/94 vs. 33/87). However, the reduction in the pooled relative risk of failure to achieve a hemodynamic response failed to reach statistical significance with carvedilol (0.67, CI 0.44–1.01), conceivably because of a type II error. It may also be worth noting that in the only study assessing both acute and long-term effects,23 the number of patients with acute hemodynamic response was almost identical to the number of patients with response after 7 days: respectively 11/18 and 11/17 with 564
% HVPG reduction with carvedilol: negative values favour
carvedilol and 9/18 and 10/16 with propranolol. Although no individual patient data are reported in the article,23 it may be assumed that the acute hemodynamic response tends to be maintained. Superiority of carvedilol over propranolol in reducing HVPG is accredited to be a consequence of the alfa-1 blocking effect reducing intra-hepatic resistance. In this respect it remains unclear the better (although not statistically significant) effect of carvedilol in the trial by Lin,24 comparing carvedilol with propranolol plus isosorbide mononitrate. The authors adduce viral aetiology or the known hyper-responsiveness to alfa-1 blockers in cirrhosis as potential explanations, although equivalence of drug doses with respect to the expected hemodynamic effect should also be accounted for. On the other hand, the only RCT comparing the clinical efficacy of carvedilol with nadolol plus isosorbide mononitrate showed that the two treatments did not differ in the prevention of variceal rebleeding after a median follow-up of 30 months.19 Adverse events were satisfactorily reported in only one study.21 The most serious reported adverse event was orthostatic hypotension (14/65 carvedilol- and 9/60 propranolol-treated patients). Although not statistically significant, the higher incidence of this serious adverse Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Systematic review with meta-analysis: carvedilol vs. propranolol for portal hypertension RR (95% CI)
%Weight
Banares (1999)
0.22 (0.06, 0.85)
8.00
De (2002)
0.82 (0.45, 1.48)
25.10
Subtotal (I 2 = 71.3%, P = 0.062)
0.48 (0.13, 1.85)
33.10
Banares (2002)
0.42 (0.18, 0.99)
16.09
De (2002)
0.97 (0.58, 1.62)
28.55
Hobolth (2012)
0.67 (0.34, 1.29)
22.27
Subtotal (I 2 = 35.1%, P = 0.214)
0.71 (0.44, 1.13)
66.90
Overall (I 2 = 39.9%, P = 0.155)
0.66 (0.44, 1.00)
100.00
Acute assessment
Long-term assessment
.2 .5 1 5 10 Better carvedilol Better propranolol Relative risk: random effects
Figure 3 | Summary relative risk (RR) of treatment nonresponse (reduction in HVPG of ≥20% of baseline or to
Systematic review with meta-analysis: the haemodynamic effects of carvedilol compared with propranolol for portal hypertension in cirrhosis E. Sinagra†,§,1, G. Perricone†,‡,1, M. D’Amico¶, F. Tine* & G. D’Amico*
*Gastroenterology Unit, A.O. Ospedali Riuniti Villa Sofia Cervello – P.O. Vincenzo Cervello, Palermo, Italy. † Internal Medicine Unit, A.O. Ospedali Riuniti Villa Sofia Cervello – P.O. Vincenzo Cervello, Palermo, Italy. ‡ Hepatology and Gastroenterology Unit, A.O. Ospedale Niguarda Ca’ Granda, Piazza Ospedale Maggiore, Milano, Italy. § Endoscopy and Gastroenterology Unit, Fondazione Istituto San Raffaele Giglio, Cefal u, Italy. ¶ Radiology Department, Universita di Palermo, Palermo, Italy.
Corresponding author: Dr G. D’Amico, Gastroenterology Unit, A.O. Ospedali Riuniti Villa Sofia Cervello – P.O. Vincenzo Cervello, via Trabucco 180, 90146 Palermo, Italy. E-mail: [email protected] These authors equally shared the first authorship. 1
Publication data Submitted 23 November 2013 First decision 9 December 2013 Resubmitted 3 January 2014 Accepted 6 January 2014 EV Pub Online 26 January 2014 As part of AP&T’s peer-review process, a technical check of this meta-analysis was performed by Mr M. Siddiqui.
SUMMARY Background Propranolol is recommended for prophylaxis of variceal bleeding in cirrhosis. Carvedilol is a nonselective beta-blocker with a mild anti-alfa-1-adrenergic activity. Several studies have compared carvedilol and propranolol, yielding inconsistent results. Aim To perform a systematic review and meta-analysis of the randomised clinical trials comparing carvedilol with propranolol for hepatic vein pressure gradient reduction. Methods Studies were searched on the MEDLINE, EMBASE and Cochrane library databases up to November 2013. The weighted mean difference in percent hepatic vein pressure gradient reduction and the relative risk of failure to achieve a hemodynamic response (reduction ≥20% of baseline or to ≤12 mmHg) with each drug were used as measures of treatment efficacy. Results Five studies (175 patients) were included. Indication to treatment was primary prophylaxis of variceal bleeding in 76% of patients. There were overall three acute (60–90 min after drug administration) and three long-term (after 7–90 days of therapy) comparisons. The summary mean weighted difference in % of reduction in hepatic vein pressure gradient was: acute 7.70 (CI 12.40, 3.00), long-term 6.81 (CI 11.35, 2.26), overall 7.24 (CI 10.50, 3.97), favouring carvedilol. The summary relative risk of failure to achieve a hemodynamic response with carvedilol was 0.66 (CI 0.44, 1.00). Adverse events were nonsignificantly more frequent and serious with carvedilol. However, quality of trials was mostly unsatisfactory. Conclusions Carvedilol reduces portal hypertension significantly more than propranolol. However, available data do not allow a satisfactory comparison of adverse events. These results suggest a potential for a cautious clinical use. Aliment Pharmacol Ther 2014; 39: 557–568
ª 2014 John Wiley & Sons Ltd doi:10.1111/apt.12634
557
E. Sinagra et al. INTRODUCTION Portal hypertension is an almost unavoidable complication of cirrhosis1 and variceal haemorrhage is its most dramatic consequence, bearing a mortality of 10–15% per episode.2 The hepatic venous pressure gradient (HVPG) is significantly associated with the risk of variceal bleeding, liver decompensation and hepatocellular carcinoma (HCC).3–5 A reduction in HVPG by 20% of baseline results in a significant reduction in the bleeding risk and a reduction to less than 12 mmHg virtually abolishes the risk.3, 6 Propranolol, a nonselective betablocker (NSBB), significantly reduces HVPG and is currently recommended for the prevention of primary and secondary variceal bleeding. However, it achieves the target HVPG reduction only in approximately one-third of patients, possibly because the increase in porto-collateral resistance impairs the reduction in portal pressure.7–9 This effect may be reduced by adding a vasodilator such as isosorbide mononitrate, which in fact rescues most of nonresponders to propranolol, although at the expense of increasing side effects.10 Carvedilol is a potent NSBB with a mild anti-alfa–1– adrenergic activity, which confers to the drug a potential for a higher portal pressure reduction compared with other NSBB. Several studies have compared the acute or long-term hemodynamic response of carvedilol and propranolol yielding inconsistent results.11 The aim of the present study was to assess the hemodynamic effects of carvedilol compared with propranolol for portal hypertension. With this aim, we performed a systematic review of randomised controlled trials (RCTs) comparing the two drugs for their effect on hepatic vein pressure gradient in cirrhosis. METHODS Study selection and participants All the studies that met the following inclusion criteria were included: (i) the study was an RCT comparing carvedilol with propranolol for acute or long-term effect on HVPG in cirrhotic patients; (ii) reduction in HVPG under treatment was a measure of the treatment effect; (iii) participants in the included studies were cirrhotic patients with clinically significant portal hypertension as indicated by HVPG >10 mmHg and/or presence of oesophageal varices, according to the Baveno III criteria.12 Randomised trials were included irrespective of blinding, publication status or language. Trials using quasi-randomisation, based on case record number, date of birth, day of the week or similar, were excluded. 558
Compared treatments Oral carvedilol was compared with either oral or i.v. propranolol for the acute hemodynamic effect, or oral carvedilol compared with oral propranolol for the long-term effect. The assessment of the hemodynamic effect was considered ‘acute’ when performed during the same hepatic veins catheterisation. When a second catheterisation was required (after days or months of treatment), it was classified as assessment of long-term treatment effect. Treatment outcome The primary measure of the treatment effect was the difference in % HVPG reduction achieved with each of the two compared drugs. The proportion of patients failing to achieve a hemodynamic response to the treatment was considered a secondary outcome measure. Hemodynamic responders were considered those patients achieving a HVPG reduction ≥20% of baseline value or in whom HVPG was lowered to ≤12 mmHg.3, 13, 14 The number and severity of adverse events were also assessed as well as the difference in mean arterial blood pressure reduction (MAP) with each drug as an indicator of systemic treatment effect. Search strategy Medline, Embase, Cochrane Controlled Trial Registry and The Cochrane Library were searched up to November 2013 to retrieve pertinent studies. Databases were searched by the following search terms: carvedilol, propranolol, portal hypertension, randomised trial. A manual search of the reference lists of pertinent studies, articles, reviews, editorials from the English language medical literature was also performed. Proceedings of the most relevant international congresses were handsearched. Data collection and analysis Decisions on which trials to include were taken independently by two of the authors (ES and GP), who were unblinded with regard to the names of the authors, investigators, institution, source, and results of the identified trials. Disagreements were resolved by discussion with all the authors. Excluded trials were identified with the reason for exclusion. Data were extracted by two independent authors (ES and GP) and discrepancies were solved by discussion. The following data were extracted from each trial: (i) trial characteristics: time interval between trial drug administration and hemodynamic effect assessment; definition of the outcome of Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Systematic review with meta-analysis: carvedilol vs. propranolol for portal hypertension interest; (ii) patients’ characteristics: number of patients, mean age, sex, aetiology of cirrhosis, Child-Pugh score, proportion of patients with previous variceal bleeding; (iii) outcome: mean HVPG change under trial treatment; number of patients achieving a hemodynamic response; adverse events observed during the treatment period.
Assessment of study quality Two of the authors (GP and ES) independently assessed bias risk of the trials, without masking the trial names. For this purpose, the risk of overestimation of intervention effects in randomised trials due to inadequate methodological quality was assessed using the following domains.15, 16 Generation of the allocation sequence was defined adequate if sequences were generated by table of random numbers or computer-generated random numbers or similar; unclear if it was not described; inadequate if sequences could be related to prognosis. Allocation concealment was defined adequate when a centralised randomisation or sequentially numbered, sealed, opaque envelopes were used; unclear when studies did not report any concealment approach; inadequate when the following modalities were used: alternation, the use of case record numbers, dates of birth or day of the week, and any procedure that is entirely transparent before allocation, such as an open list of random numbers. Blinding was considered adequate when the study was double-blind or single-blind (HVPG measurements were blindly evaluated); unclear when no information was reported; inadequate when an open design was used. Handling of dropouts and withdrawals was considered adequate when numbers of patients withdrawn or lost and reasons were described; inadequate otherwise. Intention-to-treat analysis was considered adequate when all randomised patients were included in the analyses and inadequate otherwise. Following the above definitions, an included trial was judged as achieving inadequate or unclear control of bias when the methodological quality was considered as inadequate or, respectively, unclear in at least 1 domain. If the quality was judged adequate in all the domains, then the control of bias was considered adequate. Statistical methods Trials and patient characteristics are reported as number of observations, proportions or means standard deviations. Where standard error of mean was reported in the original article, the standard deviation was derived from the standard error. Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Analyses were performed by STATA 10 (College Station, TX 77845 USA, ©1984–2009) by a fixed-effect model if no statistically significant heterogeneity was found and by a random-effects model in case of significant heterogeneity (P < 0.10). The treatment effect was calculated as the weighted mean difference (WMD) between the % HVPG reduction from baseline achieved with propranolol and the % reduction achieved with carvedilol. Data are expressed as % HVPG reduction with propranolol minus % HVPG reduction with carvedilol. Therefore, negative values indicate a higher HVPG reduction with carvedilol. The pooled WMD is reported as the summary statistic of treatment effect together with 95% confidence interval (CI). The pooled relative risk (RR) of failure to achieve a hemodynamic response with either treatment was also assessed and reported with 95% CI. Inter-trial heterogeneity was statistically assessed by a chi-square test and was expressed using I2 values.17 If significant heterogeneity was found, potential reasons for heterogeneity were explored and combinability of trials was reassessed accordingly. Analyses were performed using the intention-to-treat principle including all patients randomised as they were reported in each single study. Trials were first combined independently of whether they assessed acute or long-term hemodynamic effect of the study treatment. However, separate analyses according to the timing of assessment were also performed. To explore the robustness of estimates, the following sensitivity analyses were performed: (i) random-effects model; (ii) standardised mean difference (SMD) instead of WMD to overcome potential measurement differences across trials; (iii) excluding trials with acute assessment; (iv) excluding trials with long-term assessment; (v) excluding trials with inadequate control of bias. The weighted mean difference in arterial blood pressure with the two drugs was also assessed as a measure of a major treatment side effect.
Study protocol The full study protocol is available by contacting the corresponding author of this article. RESULTS Trial retrieval Overall, 39 potentially eligible references were retrieved in the literature search. Thirty references were excluded by reading the title and abstract, because they did not 559
E. Sinagra et al. 38 references identified through database serching
One additional reference identified by handsearch
39 references screened 30 excluded
9 references assessed for eligibility 7 references to 5 pertinent trials
2 RCTs exluded because did not include hemodynamic assessment or β-blocker control group: - one compared carvedilol with band ligation of esophageal varices for primary prophylaxis of variceal bleeding - the other compared carvedilol with nadolol plus isosorbide mononitrate for secondary prophylaxis of variceal bleeding
5 RCTs included In quantitative synthesis (meta-analysis)
refer to randomised trials. The remaining 9 references were fully assessed for inclusion in the study (Figure 1). Among these, two referred to full articles and were excluded for the following reasons: one was a randomised controlled trial of carvedilol vs. variceal band ligation for the prevention of the first variceal bleed;18 the other was a randomised controlled trial of carvedilol vs. nadolol plus isosorbide mononitrate for the prevention of variceal rebleeding, which did not include hemodynamic assessments.19 Seven references20–26 to five pertinent studies were eventually found: two of these referred to the preliminary report20 or to a subgroup analysis25 of a fully published study26 and were therefore excluded. The remaining five references21–24, 26 referred to randomised controlled trials, which fulfilled our inclusion criteria and were included in the quantitative meta-analysis.
Description of included studies The characteristics of the five included studies are summarised in Tables 1 and 2. In two studies, the acute hemodynamic effect was assessed 60 and 90 min after trial drug administration21, 24 respectively. In two studies, the treatment effect was assessed after a mean of 11 weeks22 and 92.7 days.26 In one study,23 both acute and long-term effects were assessed at 90 min and 1 week after initial trial drug administration, respectively. Overall, therefore, there were three acute and three long-term evaluations available for meta-analysis. In the acute comparisons, all treated patients were given a tablet of carvedilol 25 mg, while control patients were given propranolol 0.15 mg/kg i.v. followed by a continuous infusion of 0.2 mg/kg/h in one study,21 40 mg p.o.24 or 80 mg p.o.23 In one acute study,24 560
Figure 1 | Flow diagram of the selection process of randomised controlled trials for inclusion in the metaanalysis. RCT, randomised controlled trials.
propranolol was associated with isosorbide mononitrate 20 mg in the control group. In the evaluation of the long-term hemodynamic response, treated patients received a mean dose of 31 mg of oral carvedilol and controls received a mean of 73 mg of oral propranolol in one study;22 a fixed oral dose of 12.5 mg of carvedilol was compared with a fixed oral dose of 80 mg of propranolol in another;23 and a mean dose of 14 mg of carvedilol and 122 mg of propranolol in the third.26 A second HVPG measurement was obtained in all the included patients in three studies.21, 24, 26 In one study,22 discontinuation of treatment caused missing second HVPG measurement in two patients in the carvedilol group (heart failure one and pericardial effusion the other) and in three in the propranolol group (portal systemic encephalopathy 1 and fatigue 2). In the other study,23 discontinuation of treatment hampered a second HVPG measurement in one patient in the carvedilol group because of severe hypotension and in two patients in the propranolol group because of bleeding. Control of bias was adequate in one trial, unclear in three and inadequate in one trial (Table 2). In the four trials with inadequate or unclear control of bias, this was mostly due to the generation of the allocation sequence or to allocation concealment. The characteristics of the 175 patients included in the five studies are summarised in Table 3. Males were 74%. The number of patients with ascites was reported only in two studies22, 23 and was 44 of 87 (51%); patients with a previous bleeding episode were 42 of 153 (27%; information not reported in one study24). Mean age was 55 years and aetiology was mostly viral or alcoholic in all the Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Systematic review with meta-analysis: carvedilol vs. propranolol for portal hypertension Table 1 | Characteristics of included trials
Author, year
Participants
N of randomised patients C/P
Interventions C; P* C: Carvedilol 25 mg p.o. P: Propranolol (0.15 mg/kg i.v. followed by a continuous infusion of 0.2 mg/kg/h) C: Carvedilol 31 4 mg/day p.o.(range, 12.5–50)* P: Propranolol 73 10 mg/day p.o. (range, 10–160)* Acute: C: Carvedilol 25 mg p.o.; P: Propranolol 80 mg p.o. Maintenance: C: carvedilol 6.25 mg p.o. twice daily P: propranolol 40 mg p.o. twice daily C: Carvedilol 25 mg p.o. P: propranolol 40 mg p.o + isosorbide 5-mononitrate 20 mg p.o. C: Carvedilol 14 7 mg* P: Propranolol 122 64 mg*
Banares, 1999
Cirrhotic patients with oesophageal varices, with or without previous variceal bleeding
14/14 7 more patients were randomised to placebo
Banares,2002
Cirrhotic patients with oesophageal varices, no previous bleeding, and HVPG >12 mmHg
26/25
De, 2002
Cirrhotic patients with oesophageal varices, with or without previous variceal bleeding
18/18
Lin, 2004
Cirrhotic patients with HVPG ≥12 with or without previous variceal bleeding
11/11
Hobolth, 2012
Cirrhotic patients with HVPG≥12 mmHg, independent of varices or previous bleeding
21/17
Time of outcome assessment
N of patients with 1st/2nd HVPG measurement Carvedilol
Propranolol
60 min
14/14
14/14
11 4.1 weeks†
26/24
25/22
Acute: 90 min Long-term: 1 week
Acute: 18/18 Long-term 18/17
Acute: 18/18 Long-term 18/16
90 min
11/11
11/11
92.7 13.6 days†
21/21
17/17
C, Carvedilol group; P, Propranolol group; p.o., orally; i.v., intravenously. * Mean dose standard deviation. † Interval between baseline and second HVPG measurement, mean s.d.
studies but one where only patients with cirrhosis from HBV and/or HCV were included.24 Child-Pugh class C patients were overall 31 of 153 (20%) in four studies and the mean Child-Pugh score was eight in the fifth.24 No appreciable differences in major patient characteristics were observed between the two study groups across the five studies (Table 3).
Outcome evaluation HVPG reduction. Mean baseline HVPG was 18.3 mmHg (median 18, range 16.6–20.4) in propranolol-treated patients and 18.8 mmHg (median 19, range 17.6–19.5) Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
in the carvedilol-treated patients. The mean % HVPG reduction under treatment was 15.6 (median 12.6, range 10.1–23.2) with propranolol and 22.2 (median 19.8, range 18.6–27.7) with carvedilol (crude mean difference 6.61%). The pooled WMD between % HVPG reductions with the two drugs was 7.24 ( 10.50, 3.97; fixed-effect model) indicating a significantly higher HVPG reduction with carvedilol than propranolol. No heterogeneity was found (P = 0.99; I2 = 0.0) (Figure 2). A sensitivity analysis performed by using the standardised mean difference between the % HVPG reductions with the two drugs 561
E. Sinagra et al. Table 2 | Methodological quality assessment of included trials*
Author, year
Generation of the allocation sequence
Allocation concealment
Blinding
Dropouts and withdrawals
Intention-totreat analysis
Overall risk of bias
Banares, 1999 Banares, 2002 De, 2002 Lin, 2004 Hobolth, 2012
Unclear Adequate Unclear Adequate Unclear
Unclear Adequate Inadequate Unclear Adequate
Adequate† Adequate† Adequate‡ Unclear Adequate‡
Adequate Adequate Adequate Adequate Adequate
Adequate Adequate Adequate Adequate Adequate
Unclear Adequate Inadequate Unclear Unclear
* Definition used for rating of each quality item are reported in the method section. † Blinded reading of HVPG measurements. ‡ Double-blind.
Table 3 | Characteristics of participants in the included studies
Banares, 1999 Banares, 2002 De, 2002 Lin, 2004 Hobolth, 2012
Treatment group
Patients N
Age, mean s.d.
C P C P C P C P C P
14 14 26 25 18 18 11 11 21 17
54.6 51.4 57.9 58.4 42.3 47.3 59 61 58.2 56.2
8.8 8.5 7.6 11 11.9 12.9 13.3 13.3 6.8 6.1
Males N
Alcoholic/ viral aetiology N
Child-Pugh class A/B/ C, N
Oesophageal varices large/ small, N
Ascites N
Previous variceal bleeding, N
NR NR 19 15 15 17 8 9 12 12
8/NR 8/NR 6/18 9/16 5/9 10/5 0/11 0/11 18/2 12/1
8/4/2 5/6/3 13/10/3 15/6/4 5/9/4 0/13/5 8.7 3.0† 7.3 2.7† 8/7/6 6/6/4
NR NR 10/16 14/11 NR NR NR NR 4/10 1/11
7* 7* 10 6 12 16 NR NR NR NR
9 9 0 0 7 7 NR NR 5 5
C, Carvedilol; P, Propranolol. * Previous ascites. † Child-Pugh score, mean standard deviation (derived from the standard error reported in the article).
confirmed the principal analysis: pooled SMD 0.50 (CI 0.78, 0.23; fixed-effect model) without significant heterogeneity (P = 0.5; I2 = 0.0). The other sensitivity analyses confirmed that either the acute or the long-term % HVPG reduction is significantly higher with carvedilol than with propranolol (Figure 2) and superiority of carvedilol held true also by excluding the study with inadequate control of bias22 (WMD 7.41, CI 10.79, 4.03; data not shown). The mean % HVPG reduction was ≥20% in 2 of 6 comparisons with propranolol and in 3 of 6 with carvedilol. Overall, a reduction in HVPG of ≥20% or to ≤12 mmHg was achieved with propranolol in 33 of 87 evaluable patients as compared to 57 of 94 with carvedilol (including patients with both acute and chronic assessments; P = 0.003, Fisher’s exact test) (Table 4). 562
The pooled relative risk of failure to achieve a hemodynamic response with carvedilol was 0.66 (CI 0.44–1.00) (Figure 3).
Adverse events. Mean arterial blood pressure (MAP) was reduced more by carvedilol (WMD 10.40; 95% CI: 13.90, 6.90; I2 = 0.0%, P = 0.477) than by propranolol (WMD 6.35; 95% CI 9.86, 2.83; I2 = 0.0%, P = 0.936) (Figure 4). The lack of complete information on the difference in MAP reduction between the two drugs hampered obtaining a summary statistics. However, as the CI of MAP reduction achieved with each drug largely overlapped, it may be concluded that the difference in MAP reduction between the two drugs was not significantly different (Figure 4). On the other hand, even if not statistically significant, this difference was Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Systematic review with meta-analysis: carvedilol vs. propranolol for portal hypertension
Acute assessment Banares (1999)
–7.70 (–13.24, –2.16)
34.70
De (2002)
–4.69 (–23.94, 14.56)
2.88
Lin (2004)
–8.50 (–18.48, 1.48)
10.71
Subtotal (I 2 = 0.0%, P = 0.942)
–7.70 (–12.40, –3.00)
48.29
Banares (2002)
–7.00 (–12.26, –1.74)
38.48
De(2002)
–4.72 (–21.99, 12.55)
3.58
Hobolth (2012)
–6.80 (–17.31, 3.71)
9.66
Subtotal (I 2 = 0.0%, P = 0.970)
–6.81 (–11.35, –2.26)
51.71
–7.24 (–10.50, –3.97)
100.00
Long-term assessment
Heterogeneity between groups: P = 0.789
Overall (I 2 = 0.0%, P = 0.998)
–20 –10 Better carvedilol
0
10 20 Better propranolol
Weighted Mean Difference: fixed effects
Figure 2 | Weighted mean difference (WMD) (fixed-effect model) of the % hepatic vein pressure gradient (HVPG) reduction between propranolol and carvedilol in the subsets of RCTs assessing the acute and, respectively, the longterm treatment effect together with the overall estimation of the whole set of studies. Each study is identified by the name of the first author and year of publication (references in the text). Squares indicate the WMD per each trial and the size of the squares is proportional to the weight of trials. The horizontal bars denote the 95% confidence intervals of WMD. The vertical solid line is the equivalence line, where WMD is equal to 0. Differences are given as values in the propranolol group minus values in the carvedilol group. Therefore, negative values (WMDs on the left of the equivalence line) denote superiority, whereas those on the right denote inferiority of carvedilol. The diamonds represent 95% CI of the WMDs per each subset and in whole set of RCTs. The vertical dashed line represents the pooled WMD of the whole set of trials. The study by De (2002) provided estimation of either acute or long-term treatment effect.
clinically relevant, with at least one symptomatic episode of hypotension reported with carvedilol and a total of 14 patients with orthostatic hypotension out of 65 patients treated with carvedilol, compared to 9/60 with propranolol (P = 0.37) in three of the five studies.22, 23, 26 Other adverse events, even if not fully reported in all the studies, were the following: need for increasing diuretics in 14/47 patients treated with carvedilol and in 6/42 with propranol;22, 26 shortness of breath in 12/47 and 7/42;22, 26 portosystemic encephalopathy in 3/26 and 4/2522 respectively. Overall carvedilol was withdrawn in 6/65 and propranolol in 8/60 treated patients22, 23, 26 because of adverse events. None of these differences was statistically significant. Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
DISCUSSION The present meta-analysis shows that, in the available RCTs comparing propranolol and carvedilol for portal hypertension, carvedilol reduces HVPG significantly more than propranolol. Superiority of carvedilol is consistent across all the trials, regardless of acute or longterm assessment. The summary weighed mean difference between the percent HVPG reduction achieved with propranolol and carvedilol is 7.24 (CI 10.50 to 3.97) in favour of carvedilol. No heterogeneity was found and all the sensitivity analyses confirmed the principal result. Patients in the two treatment groups were fairly comparable across the included studies, regarding demography, aetiology, indicators of disease severity, baseline 563
E. Sinagra et al. Table 4 | Main results of the included studies Carvedilol
Author, year
Baseline HVPG, mmHg mean s.d.*
Acute assessment Banares, 1999 19.5 De, 2002 19.0 Lin, 2004 18.9 Subtotal – Long-term assessment Banares, 2002 19.1 De, 2002 19.0 Hobolth, 2012 17.6 Subtotal – Total
–
4.9 3.8 6.0
Propranolol
% HVPG reduction, mean s.d.*
Hemo-dynamic responders† n/N (%)§
20.4 7.5 27.7 31.5 18.6 11.9 –
9/14 (64) 11/18 (61)‡ NR –
20.4 4.1 16.60 3.9 17.6 4.0 –
–
19 9.8 27.7 31.5 19.3 16.1 –
13/24 (54) 11/17 (65)‡ 13/21 (62) –
20.3 4.2 16.6 4.0 18.4 3.6 –
12 9.4 22.98 20.1 12.5 16.7 –
–
–
–
–
–
–
5.4 3.8 4.2
Baseline HVPG, mean s.d.*
Weighed Mean Difference in % HVPG reduction (CI)¶
% HVPG reduction, mean s.d.*
Hemo-dynamic responders† n/N (%)§
12.7 7.48 22.9 27.4 10.1 11.9
2/14 (14) 9/18 (50)
7.70 4.69 8.50 7.70
5/22 (23) 10/16 (62) 7/17 (41)
7.00 ( 4.72 ( 6.80 ( 6.81 ( 11.35, 7.24 (
NR –
( ( ( (
13.24, 2.16) 23.94, 14.56) 18.48, 1.48) 12.4, 3.00) 12.26, 1.74) 21.99, 12.55) 17.31, 3.71) 2.26) 10.50,
3.97)
* s.d., standard deviation of mean. In the studies by Banares 1999 and 2002 and by Lin 2004, the standard deviation of the mean was calculated from the standard error of the mean reported in the original articles. † Patients achieving a HVPG reduction ≥20% or HVPG ≤12 mmHg. ‡ Hemodynamic response was defined as a reduction in HVPG of ≥20% of baseline in the De’s study. § n, number of responders; N, total number of evaluable patients. ¶ The difference is calculated as % HVPG reduction with propranolol carvedilol.
HVPG and history of previous bleeding. A second measurement of HVPG was available in all the included patients, but three in the De’s study23 and five in the long-term study by Banares.22 Only two of the missing second HVPG measurements were due to bleeding (both in proporanolol-treated patients), while the others were caused by adverse events requiring treatment withdrawal. Control of bias was considered inadequate in only one trial and the beneficial effect of carvedilol over propranolol was confirmed also when excluding this study. On the other hand, control of bias was adequate in only one study: this finding calls for caution when interpreting the present results. The number of patients achieving a reduction in HVPG to ≥20% or to ≤12 mmHg was reported in 4 of the 5 studies and was also markedly higher with carvedilol (57/94 vs. 33/87). However, the reduction in the pooled relative risk of failure to achieve a hemodynamic response failed to reach statistical significance with carvedilol (0.67, CI 0.44–1.01), conceivably because of a type II error. It may also be worth noting that in the only study assessing both acute and long-term effects,23 the number of patients with acute hemodynamic response was almost identical to the number of patients with response after 7 days: respectively 11/18 and 11/17 with 564
% HVPG reduction with carvedilol: negative values favour
carvedilol and 9/18 and 10/16 with propranolol. Although no individual patient data are reported in the article,23 it may be assumed that the acute hemodynamic response tends to be maintained. Superiority of carvedilol over propranolol in reducing HVPG is accredited to be a consequence of the alfa-1 blocking effect reducing intra-hepatic resistance. In this respect it remains unclear the better (although not statistically significant) effect of carvedilol in the trial by Lin,24 comparing carvedilol with propranolol plus isosorbide mononitrate. The authors adduce viral aetiology or the known hyper-responsiveness to alfa-1 blockers in cirrhosis as potential explanations, although equivalence of drug doses with respect to the expected hemodynamic effect should also be accounted for. On the other hand, the only RCT comparing the clinical efficacy of carvedilol with nadolol plus isosorbide mononitrate showed that the two treatments did not differ in the prevention of variceal rebleeding after a median follow-up of 30 months.19 Adverse events were satisfactorily reported in only one study.21 The most serious reported adverse event was orthostatic hypotension (14/65 carvedilol- and 9/60 propranolol-treated patients). Although not statistically significant, the higher incidence of this serious adverse Aliment Pharmacol Ther 2014; 39: 557-568 ª 2014 John Wiley & Sons Ltd
Systematic review with meta-analysis: carvedilol vs. propranolol for portal hypertension RR (95% CI)
%Weight
Banares (1999)
0.22 (0.06, 0.85)
8.00
De (2002)
0.82 (0.45, 1.48)
25.10
Subtotal (I 2 = 71.3%, P = 0.062)
0.48 (0.13, 1.85)
33.10
Banares (2002)
0.42 (0.18, 0.99)
16.09
De (2002)
0.97 (0.58, 1.62)
28.55
Hobolth (2012)
0.67 (0.34, 1.29)
22.27
Subtotal (I 2 = 35.1%, P = 0.214)
0.71 (0.44, 1.13)
66.90
Overall (I 2 = 39.9%, P = 0.155)
0.66 (0.44, 1.00)
100.00
Acute assessment
Long-term assessment
.2 .5 1 5 10 Better carvedilol Better propranolol Relative risk: random effects
Figure 3 | Summary relative risk (RR) of treatment nonresponse (reduction in HVPG of ≥20% of baseline or to
