Systematic review: secondary prevention with band ligation, pharmacotherapy or combination therapy after bleeding from oesophageal varices

Authors


Dr J. Cheung, Zeidler Ledcor Centre, 130 University Campus, Edmonton, Alberta, Canada, T6G 2X8.
E-mail: justin.cheung@ualberta.ca

Summary

Background  Variable methods are available for secondary prevention after oesophageal variceal bleeding (EVB).

Aim  To compare band ligation (BL), pharmacotherapy (PT) and BL+PT for EVB secondary prevention.

Methods  A systematic search of databases, references and meeting abstracts was conducted for randomized trials of BL, PT or BL+PT. The outcomes were mortality, rebleeding and adverse events. A random-effects model was used for meta-analyses.

Results  Twelve trials were included (6 BL vs. PT, 4 BL+PT vs. BL, 2 BL+PT vs. PT). All trials used beta-blockers ± isosorbide mononitrate (ISMN) as PT. Mortality was not significantly different among trials. Rebleeding was not significantly different for BL vs. PT (RR 1.00, 95% CI 0.73–1.37). BL reduced rebleeding compared with PT for trials with mean beta-blocker dose <80 mg/day (RR 0.67, 95% CI 0.49–0.91). There were nonsignificant differences in rebleeding for BL+PT vs. BL (RR 0.57, 95% CI 0.31–1.08) and BL+PT vs. PT (RR 0.76, 95% CI 0.56–1.03). There was no difference in adverse events between BL vs. PT, but was higher with BL+PT vs. BL.

Conclusion  Band ligation and PT alone are comparable for secondary prevention of rebleeding after EVB. Further trials with adequate PT dosing are required to determine the efficacy of combination BL+PT therapy.

Introduction

Oesophageal variceal bleeding (EVB) is a cause of significant morbidity in patients with liver cirrhosis. Approximately one-third of cirrhotics have oesophageal varices and one-third will experience EVB with a mortality rate of up to 30%.1–3 After EVB, up to 60% will experience a recurrent episode if secondary prevention measures are not initiated.1, 2

Therapies for secondary prevention include pharmacotherapy (PT) and endoscopic band ligation (BL). Pharmacotherapy with nonselective beta-blockers, such as nadolol and propranolol, reduce portal pressure by decreasing cardiac output and by causing splanchnic vasoconstriction.3–5 The combination of beta-blocker and isosorbide mononitrate (ISMN) is associated with an even greater reduction in portal pressure.6 Secondary prophylaxis with endoscopic band ligation involves repeated ligation of oesophageal varices until obliteration followed by regular surveillance for recurrence. Recent surveys have shown variability in how gastroenterologists practise secondary prevention for EVB.7, 8

Randomized trials comparing PT with BL have been inconclusive.9–13 Although a recent portal hypertensive consensus workshop suggested either BL or PT as suitable options,14 recent guidelines suggested combination BL plus PT as first line therapy.15 There is no published meta-analysis comparing BL with PT. A recent meta-analysis evaluated the role of combination endoscopic plus drug therapy in secondary prophylaxis, but a majority of trials that were included utilized sclerotherapy,16 a technique that is inferior to band ligation.15 The purpose of this systematic review was to determine the effectiveness of BL, PT and combination BL + PT in the secondary prevention of oesophageal variceal rebleeding in patients with liver cirrhosis.

Methods

Inclusion criteria

Studies were included with the following criteria: (1) randomized trials comparing endoscopic band ligation vs. nonselective beta blockers (±nitrates), or combination therapy with beta blockers (±nitrates) plus band ligation vs. either band ligation or beta blockers (±nitrates); (2) age ≥18 years; (3) previous oesophageal variceal bleeding; (4) liver cirrhosis. Both manuscripts published in full and abstracts were searched for.

Search strategy

An electronic search strategy used the following terms: beta-blocker, nadolol, propranolol, ligation and variceal bleeding. No language restrictions were used. The databases included MEDLINE (1950-Feb 2008), EMBASE (1988-Feb 2008), PubMed (last 180 days to obtain ‘in process’ citations), Web of science (1900-Feb 2008), Scopus (1945-Feb 2008), Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Database of Abstracts of Reviews of Effects, ClinicalTrials.gov, CenterWatch Clinical Trials Listing Service, OCLC PaperFirst. Handsearching of conference abstracts (Digestive Disease Week, American College of Gastroenterology, American Association for the Study of the Liver) for the past 2 years was carried out. Additional trials were identified through review of reference lists in relevant reviews and trials. Also, the top 30 citation results from Google Scholar were searched. Personal communication with authors of included trials for other relevant trials was also attempted.

Trial selection

Two independent reviewers performed study selection and quality assessment (JC and MZ). Disagreement was resolved through consensus decision, or by consulting a third party (PuT) if no consensus could be reached. The study risk of bias was assessed using the Jadad scale, concealment of allocation and intention-to-treat analysis.17–19

Data extraction

The following data were extracted: (a) study population (age, gender, cause of liver disease, Child Pugh score); (b) intervention; (c) outcomes (all-cause mortality, upper gastrointestinal rebleeding, variceal rebleeding, and adverse events); (d) study risk of bias (e) follow-up and (f) loss to follow-up, withdrawal/intolerance due to adverse events, and non-compliance.

Statistics

RevMan software (Review Manager v4.2.10. Copenhagen, Denmark) was used for analysis. Data were analysed using a Random-effects model with the Mantel-Haenszel method. For dichotomous variables, relative risks (RR) and 95% confidence interval (CI) were calculated. Number-needed-to-treat (NNT) was calculated for significant outcomes as the inverse of the pooled absolute risk reduction. The I2 statistic indicates the percentage of total variation across studies and was used for assessment of intertrial heterogeneity. An I2 of >60% was considered significant heterogeneity.

A priori subgroup analysis was performed for trials comparing BL with PT by contacting authors for Child-Pugh (CP) scores of individual patients. Patients were arbitrarily divided into two groups to compare CP A with CP B-C. Post hoc subgroup analyses were performed to assess whether length of follow-up (< or >20 months), majority alcoholic vs. majority viral hepatitis cirrhosis and mean beta-blocker dose (< or >80 mg/day) did influence the results. Follow-up length was selected based on a primary prophylaxis meta-analysis showing possible improved outcomes with beta-blockers in trials with follow-up >20 months.20 Sensitivity analyses were performed to assess the effect of study quality (high vs. low risk of bias) and fixed vs. random effects modelling. For evaluation of study bias risk, the validated Jadad score (blinding 2 points, method of random allocation 2 points, loss of follow-up 1 point), concealment of allocation (adequate, unclear, inadequate, not used) and intention-to-treat analysis (adequate, unclear, inadequate) were used. The concealment of allocation scale18 was defined as follows: adequate (centralized randomization; on site computer only accessed after entering the characteristics of an enrolled; sequentially numbered sealed opaque envelopes); unclear (sealed envelopes, but not sequentially numbered or opaque; open list of random numbers read by someone entering patient into trial; description suggests adequate concealment, but features are suspicious) and inadequate (any allocation procedure transparent before assignment). Blinding was defined as investigator assessing the intervention outcome being unaware of the intervention the participant received. A study with either Jadad score <3 (as blinding is not possible comparing BL with PT) or allocation of concealment inadequate/not used or no intention-to-treat analysis (i.e. per protocol analysis) was considered higher risk of bias.

Results

Search

The search in April 2008 identified 1191 potential references. After reviewing the titles and abstracts, 1166 were excluded for duplications and nonrelevance. Further 13 were excluded for the following reasons: editorial,21, 22 case series,23 incorrect intervention24, 25 and preliminary results when updated study results were available.26–33 Six studies were included for analysis of band ligation (BL) vs. pharmacotherapy (PT).9–13, 34 Of these, five were full articles and one was as an abstract for which the thesis manuscript was obtained.33 Four studies were included for combination BL+PT vs. BL,35–38 two of which were published as abstracts.37, 38 Two study abstracts (currently submitted for full publication) were included for combination BL+PT vs. PT.39, 40

Patients

BL vs. PT.  Overall, 698 patients were randomized to BL or PT (Table 1). The majority were Child Pugh B or C (71%) and 46% had alcohol-related liver disease. One study enrolled 26% with noncirrhotic portal hypertension and these patients were excluded from analyses.13 Of the 698 patients, only 25 were lost to follow-up.

Table 1.   Patient characteristics in the included studies
Study Group I/IIChild-Pugh class A/B/C (%)F/U (mo)
NAge (years)Males (%)Alcoholic (%)Group IGroup II
  1. F/U, follow-up.

Band Ligation (I) vs. pharmacotherapy (II)
 Villanueva 200172/7258/6065/6060/6015/60/2526/54/1921
 Lo 200260/6152/5177/7727/3622/58/2021/57/2125
 Patch 200251/5151/5269/6963/7116/37/4713/36/5110
 Sarin 200551/5036/3672/6825/2331/49/2028/50/2212
 Shiha 200561/6050/5080/8327/2223/54/2320/53/277
 Romero 200657/5253/5167/6565/5832/58/1040/44/1612
Combination band ligation + pharmacotherapy (I) vs. ligation alone (II)
 Lo 200062/6051/5379/7532/2819/45/3518/50/3121
 Sollano 200116/15N/AN/AN/AN/AN/A10
 De la Pena 200543/3760/6077/7370/6314/58/2816/54/3016
 Jha 200779/92N/AN/AN/AN/AN/A16
Combination band ligation + pharmacotherapy (I) vs. pharmacotherapy alone (II)
 Lo 200560/60N/AN/AN/AN/AN/A23
 Garcia-pagan 200680/79N/AN/AN/AN/AN/A11

BL+PT vs. BL or PT alone.  Overall, 404 patients were randomized to BL+PT or BL (Table 1). Complete demographic data (age, gender, CP score, cause of liver disease) were available for two studies (n = 202).35, 36 One study, published in abstract form, included 18% of patients (n = 31) with noncirrhotic portal hypertension and the information available did not allow for the ability to exclude this subset of patients.37 The numbers lost to follow-up (n = 1) and withdrawal (n = 3) were small. Overall, 279 patients were randomized to BL+PT or PT (Table 1), of which complete demographic data were not available in the abstracts.39, 40

Band ligation

Endoscopic band ligation was performed by commercial multibanding devices only10, 11, 13, 34–36 or both single and multibanding devices.9, 12 The devices were not described in trials published as abstracts.37–40 Repeat endoscopy for patients randomized to band ligation varied between 7 days and 4 weeks until varices were obliterated. Surveillance endoscopy after varices were obliterated varied from no reported surveillance10, 12 to every 3–6 months. 9, 11, 13, 34–36

Pharmacotherapy

Treatments are summarized in Table 2. In all trials, the dose of beta-blockers was titrated to a reduction in heart rate of 25% or to <55–60 beats per minute. No trials documented what proportion of patients reached target heart rates. Only two trials reported heart rates achieved on follow-up.9, 13 Only one trial reported hepatic portal venous gradient (HVPG) values achieved on follow-up.13 Isosorbide mononitrate (ISMN) treatments doses ranged between 20 and 80 mg daily. In one trial, ISMN was added as a co-intervention in 41% of patients if beta-blockers failed to reduce the HPVG by 20% or to ≤12 mmHg.11

Table 2.   Intervention features of included studies
StudyBeta-blocker (BB)BB Dose (mean)Drug Co-intervention (mean)Other Co-interventionComparison Group
  1. ISMN, isosorbide mononitrate.

Monotherapy trials
 Villanueva 2001Nadolol96 mgISMN 60 mgLigation
 Lo 2002Nadolol48 mgISMN 30 mgLigation
 Patch 2002Propranolol80 mgISMN 20–40 mgLigation
 Sarin 2005Propranolol109 mgISMN 34 mgLigation
 Shiha 2005Propranolol58 mgISMN 34 mgLigation
 Romero 2006Nadolol88 mgISMN 58 mgLigation
Combination therapy trials
 Lo 2000Nadolol60 mgSucralfate 1 g QIDLigationLigation
 Sollano 2001PropranololN/ALigationLigation
 De la Pena 2005Nadolol58 mgLigationLigation
 Lo 2005NadololISMNLigationNadolol/ISMN
 Garcia-Pagan 2006NadololISMNLigationNadolol/ISMN
 Jha 2007Propranolol114 mgISMN 27 mgLigationLigation

Study design and risk of bias

Study risk of bias is summarized in Table 3. As expected, there was no double blinding of patients randomized to the endoscopy procedure. One trial was terminated early after an a priori interim analysis at 50% enrollment found a significant benefit in variceal rebleeding by BL+PT vs. BL.36 One combination therapy trial was published as a preliminary report.39

Table 3.   Quality analysis of included trials
StudyRCTPublication typeAllocation concealmentIntention-to-treatJadad Score*
  1. * Jadad scale17: one point for randomization, adequate method of randomization, and description of withdrawals/dropouts.

  2. † Manuscript obtained.

Pharmcotherapy vs. Band Ligation
 Lo 2002YesArticleAdequateAdequate3
 Patch 2002YesArticleAdequateAdequate3
 Romero 2006YesArticleAdequateAdequate3
 Sarin 2005YesArticleUnclearAdequate2
 Shiha 2005YesAbstract†AdequateAdequate3
 Villanueva 2001YesArticleAdequateAdequate3
Combination Band Ligation + Pharmacotherapy vs. Band Ligation
 Lo 2000YesArticleAdequateAdequate3
 Sollano 2001YesAbstractUnclearUnclear2
 De la Pena 2005YesArticleAdequateAdequate3
 Jha 2007YesAbstractUnclearUnclear2
Combination Band Ligation + Pharmacotherapy vs. Pharmacotherapy
 Lo 2005YesAbstractUnclearUnclear2
 Garcia-Pagan 2006YesAbstractUnclearAdequate2

Outcome measures

Mortality.  The mortality rates among all studies were similar and there were no significant differences pooling trials for BL vs. PT (25% vs. 21%, I2 0%), BL+PT vs. BL (13% vs. 15%, I2 45%) and BL+PT vs. PT (20% vs. 22%, I2 31%) (Figures 1–3).

Figure 1.

 Pooled meta-analysis for mortality and rebleeding in trials comparing band ligation (BL) with pharmacotherapy (PT) for secondary prophylaxis after oesophageal variceal bleeding.

Figure 2.

 Pooled meta-analysis for mortality and rebleeding in trials comparing combination band ligation and pharmacotherapy (BL+PT) with band ligation (BL) for secondary prophylaxis after oesophageal variceal bleeding.

Figure 3.

 Pooled meta-analysis for mortality and rebleeding in trials comparing combination band ligation and pharmacotherapy (BL+PT) with pharmacotherapy (PT) for secondary prophylaxis after oesophageal variceal bleeding.

Rebleeding.  In trials comparing BL with PT, there was no difference in all-cause rebleeding (39% vs. 40%), but heterogeneity was significant (I2 62%) (Figure 1). The pooled rates of oesophageal variceal rebleeding in the BL (26%) and PT (32%) groups were not significantly different, but again heterogeneity was marked (I2 79%). The other causes of rebleeding in cirrhotics (BL vs. PT) were described in four studies:9–12 oesophageal band ulcers (6% vs. 1%), portal hypertensive gastropathy (PHG) (5% vs. 3%), gastric varices (2% vs. 1%), undetermined (1% vs. 3%) and other (3% vs. 2%). After the initial EVB event, oesophageal band ulcer bleeding occurred more often with the use of BL vs. PT secondary prevention (RR 3.4, 95% CI 1.1–10.3, I2 0%).

Pooling all four studies comparing BL+PT with BL resulted in a nonsignificant trend for benefit in rebleeding with combination therapy (20% vs. 30%; RR 0.57, 95% CI 0.31–1.08), but heterogeneity was high (I2 60%) (Figure 2). Two full articles described the causes of rebleeding.35, 36 There was a significant reduction in rebleeding rates from oesophageal varices with BL+PT compared with BL (10% vs. 25%, RR 0.38 95% CI 0.19–0.76, I2 0%). The other causes were oesophageal band ulcers (3% vs. 8%), gastric varices (2% vs. 4%), PHG (2% vs. 2%) and peptic ulcers (4% vs. 4%) (all P > 0.05). Pooling the two studies of BL+PT vs. PT resulted in a nonsignificant trend for BL+PT in rebleeding (32% vs. 43%, RR 0.76, 95% CI 0.56–1.03) with minimal heterogeneity (I2 0%) (Figure 3). However, BL+PT significantly reduced rebleeding from varices compared with PT (22% vs. 38%, RR 0.58, 95% CI 0.40–0.85, I2 0%). Further data for nonvariceal causes of rebleeding were not available from the abstracts.

Adverse events.  In trials comparing BL with PT, adverse events were available for analysis in cirrhotic patients in five trials.9–12, 33 The rates were 27% and 31% for BL and PT respectively (RR 0.90, 95% CI 0.70–1.15, I2 0%). There was no intervention-related mortality. Common adverse events with BL were transient dysphagia (7%), chest pain (4%) and oesophageal ulcer bleeding (6%). With PT, common adverse events included headache (9%), hypotension (6%), bradycardia (4%), weakness (4%) and sexual dysfunction (1%). Overall withdrawal rates from the BL and PT groups were 1% and 11% respectively. Noncompliance rates were 2% (BL) and 4% (PT).

In trials comparing BL+PT with PT, the withdrawal rate was 5% for combination therapy. For full-text publications, the adverse event rate was higher in the BL + PT group (21%) compared with the BL group (6%), (RR 3.4, 95% CI 1.4–8.2, I2 74%), P = 0.006. The most common adverse events in the BL+PT group were dizziness (4%), sexual dysfunction (3%), fatigue (2%), arrhythmia (2%), and chest discomfort (2%). Data on adverse events were not provided from the two abstracts comparing BL+PT with BL alone.

Subgroup analyses

Subgroup analyses were performed for trials comparing BL with PT, but not for the combination therapy trials as only two of these trials were available in full-text. Subgroup analyses are summarized in Tables 4 and 5.

Table 4.   Within-study subgroup analysis by Child-Pugh class of patients from four trials comparing band ligation vs. pharmacotherapy (n = 475)
VariableBL (%)PT (%)Relative risk95% CIP value
  1. BL, band ligation; PT, pharmacotherapy; CP, Child-Pugh.

Mortality
 CP A5/57 (9)5/69 (7)1.460.47–4.520.51
 CP B-C56/178 (31)40/171 (23)1.280.91–1.790.12
Rebleeding
 CP A17/57 (30)21/69 (30)1.010.41–2.490.99
 CP B-C74/178 (42)77/171 (45)0.920.72–1.170.49
Variceal rebleeding
 CP A13/57 (23)17/69 (25)0.890.28–2.800.84
 CP B-C52/178 (29)62/171 (36)0.750.42–1.320.44
Table 5.   Between-study subgroup analyses for trials comparing band ligation with pharmacotherapy by follow-up length, majority of liver disease and mean beta blocker dose
 BL (n/N)PT (n/N)RR (95% CI)P valueI2 Statistic (%)
  1. BL, band ligation; PT, pharmacotherapy.

Mortality
 Follow-up < 20 mo.35/16437/1680.97 (0.65–1.45)0.890
 Follow-up > 20 mo.45/13231/1331.43 (0.98–2.08)0.070
 Majority Alcoholic29/17221/1791.15 (0.84–1.56)0.230
 Majority Viral Hepatitis57/17551/1801.38 (0.81–2.33)0.390
 Beta Blocker > 80 mg/day63/22655/2301.16 (0.86–1.57)0.320
 Beta Blocker < 80 mg/day23/12117/1211.33 (0.92–1.57)0.4741
Rebleeding
 Follow-up < 20 mo.77/21583/2180.95 (0.67–1.33)0.7646
 Follow-up > 20 mo.58/13259/1330.98 (0.46–2.12)0.9787
 Majority Alcoholic86/17570/1801.26 (0.97–1.63)0.0916
 Majority Viral Hepatitis49/17272/1710.68 (0.51–0.90)0.0070
 Beta Blocker > 80 mg/day95/22682/2301.19 (0.90–1.56)0.2226
 Beta Blocker < 80 mg/day40/12160/1210.67 (0.49–0.91)0.010
Oesophageal variceal rebleeding
 Follow-up < 20 mo.47/21565/2180.69 (0.34–1.40)0.3175
 Follow-up > 20 mo.44/13246/1330.88 (0.26–2.93)0.8391
 Majority Alcoholic67/17552/1801.31 (0.88–1.96)0.1841
 Majority Viral Hepatitis24/17259/1710.4 (0.27–0.63)<0.00010
 Beta Blocker > 80 mg/day71/22664/2301.06 (0.61–1.84)0.8569
 Beta Blocker < 80 mg/day20/12147/1210.43 (0.27–0.68)<0.0010

Child-pugh class.  Four of six authors provided individual patient Child Pugh (CP) data on request.9, 10, 12, 13 A total of 475 patients were divided into two groups (CP A and CP B-C). There were no differences in mortality, rebleeding and variceal rebleeding outcomes between BL and PT by CP subgrouping (Table 4).

Length of follow-up.  For trials with >20 months of follow-up, nonsignificant differences in mortality with PT (23%) compared to BL (34%) (RR 0.70, 95% CI 0.48–1.02, I2 0%).9, 10 For trials with <20 months follow-up, no differences in mortality were found (RR 0.97, 95% CI 0.65,1,45 I2 0%).11–13, 34 No differences in effect size and heterogeneity based on follow-up were found for rebleeding and oesophageal variceal rebleeding.

Aetiology of cirrhosis.  No differences in mortality between BL and PT were found in trials with a majority of alcoholic9, 11, 12 or viral hepatitis10, 13, 34 patients. For trials with majority alcoholics, there were also no differences in rebleeding or oesophageal variceal rebleeding between therapies.9, 11, 12 However, in trials with majority viral hepatitis patients, there was significantly reduced rebleeding with BL (28%) compared to PT (42%), improved heterogeneity (RR 0.68, 95% CI 0.51–0.90, I2 0%) and an NNT 7 in favour of BL.10, 13, 34 Similarly, oesophageal variceal rebleeding was significantly lower with BL (14%) vs. PT (35%) (RR 0.41, 95% CI 0.27–0.63) with minimal heterogeneity (I2 0%) and an NNT 5 in favour of BL.

Beta blocker dose.  No differences in mortality were found between BL and PT in trials with mean beta blocker doses > or <80 mg/day. In trials with a mean beta-blocker dose >80 mg/day, nonsignificant differences between BL and PT were found for rebleeding (42% vs. 36%, RR 1.19, 95% CI 0.90–1.56) or oesophageal variceal rebleeding (31% vs. 28%, RR 1.06, 95% CI 0.61–1.84).9, 11–13 However, in trials with a mean beta blocker dose of <80 mg/day, there was a significant reduction in rebleeding by BL (33%) vs. PT (50%), (RR 0.67, 95% CI 0.49–0.91) with minimal heterogeneity (I2 0%) and an NNT of 6 in favour of BL.10, 34 Likewise, there was a significant reduction in oesophageal variceal bleeding by BL (17%) vs. PT (39%) in trials with beta blocker doses <80 mg/day (RR 0.43, 95% CI 0.27–0.68) with minimal heterogeneity (I2 0%) and NNT of 5 in favour of BL.10, 34

Sensitivity analyses

Five studies comparing BL with PT were considered to have low risk of bias.9–12, 34 Only 1 study was considered to have higher risk of bias because of unclear method of randomization.13 Excluding this one study did not alter the overall treatment effect or heterogeneity for mortality, rebleeding and oesophageal variceal rebleeding.

For trials comparing BL+PT with BL, excluding two trials in abstract form37, 38 with higher risk of bias resulted in significantly lower rebleeding (19% vs. 44%, RR 0.45, 95% CI 0.29–0.70, NNT 7, I2 0%) and oesophageal variceal rebleeding with BL+PT (RR 0.38, 95% CI 0.19–0.76, NNT 7, I2 0%) compared with BL. Mortality, however, was not significantly different (15% vs. 24%, RR 0.61, 95% CI 0.33–1.13, I2 4%).

A sensitivity analysis with fixed effects modelling resulted in a statistically significant difference in rebleeding among four trials comparing BL+PT with PT (RR 0.62, 95% CI 0.44–0.87), but heterogeneity remained significantly high (I2 60%). There was no significant change in heterogeneity or effect size when fixed effect modelling was used to analyse mortality, rebleeding or oesophageal rebleeding.

Analyses conducted by excluding two trials10, 12 without reported surveillance endoscopy post-obliteration did not alter the overall effect or heterogeneity for outcomes of mortality, rebleeding, or oesophageal variceal rebleeding.

Discussion

The current review analysed randomized trials of band ligation (BL), pharmacotherapy (PT) and combination band ligation plus pharmacotherapy (BL+PT). Mortality rates in trials were similar (13–25%) over a follow-up period that ranged from 7 to 25 months. No differences in pooled mortality were observed. No differences in rebleeding were found between BL and PT, but subgroup analysis suggested that BL may be superior to PT for rebleeding in trials with lower doses of beta-blockers and higher proportion of patients with viral hepatitis liver disease. There was no statistically significant benefit for combination BL+PT therapy when compared with either BL or PT.

There was clinical heterogeneity among trials regarding causes of liver disease, pharmacotherapy regimens and beta-blocker doses. Trials from North America/Europe included a majority of alcoholic patients, while trials from Asia included a majority of viral hepatitis patients. For trials of combination therapy (BL+PT vs. either BL or PT alone), there was heterogeneity in the drug regimens. All combination BL+PT trials used beta-blockers, but the addition of ISMN was variable.37, 39, 40 In contrast, all trials of BL vs. PT alone used both ISMN and beta-blockers as dual PT. Adding ISMN reduces portal pressures more than beta-blockers alone and therefore several studies evaluating combination BL+PT may not be obtaining optimal portal pressure reduction with pharmacotherapy.6 Additionally, although all studies titrated beta-blockers to heart rate, there was variation in the mean doses achieved and few trials reported measurements to confirm portal pressure reduction. Only one study reported HVPG (mean 16 mmHg) values achieved with PT on follow-up.9 Additionally, only two trials reported follow-up heart rates and none described how many patients achieved target heart rates.

Among trials comparing BL with PT, there was no difference in mortality. This finding was robust and not affected by subgroup analyses by Child-Pugh class, major cause of liver disease, or beta-blocker dosage. All-cause rebleeding with PT was also not different from that with BL. The most common causes of rebleeding were oesophageal varices and postbanding oesophageal ulcers. The frequency of oesophageal ulcer bleeding was significantly higher with band ligation secondary prophylaxis (6% vs. 1%), but there was no evidence that this was completely offset by significantly lower oesophageal variceal bleeding by BL.

There was significant heterogeneity, when pooling BL vs. PT trials, which was not accounted for by study risk of bias, follow-up length or liver disease severity. However, post hoc subgroup analyses of trials by the proportion of patients with alcoholic liver disease and by the dose of beta-blockers did reduce heterogeneity. In trials with majority viral hepatitis cirrhosis, band ligation reduced both overall rebleeding and oesophageal variceal rebleeding. However, there was no benefit for ligation among trials with majority alcoholic cirrhosis. It is possible that the increase in portal pressures from ongoing alcohol use may reduce the mechanical effect of ligation.41, 42 Further studies are required to verify this hypothesis. Secondly, among trials with mean beta-blocker dose <80 mg/day, BL was associated with reduced rebleeding compared to PT. Therefore, beta-blocker doses <80 mg/day may be hypothesized to be insufficient for secondary prevention. However, as patients are likely to have some variation in response to the same dose of beta-blockade, information regarding heart rate reduction or ideally HVPG reduction needs to be reported to better understand the reason for these differences.

Among combination therapy trials (BL+PT vs. BL and BL+PT vs. PT), there were no differences in mortality. In addition, although trends were present favouring combination therapy for rebleeding, there were also no significant overall differences for combination therapy when compared separately with either BL or PT alone. An analysis of BL+PT vs. BL trials showed a benefit for combination therapy when two trials as abstracts were excluded.37, 38 However, one trial as abstract is the largest study available and the factors that may account for these heterogenous results are unknown.37 Additionally, combination therapy trial outcomes may have been affected by heterogeneity in the PT regimens. Only 3 of 6 combination BL+PT trials used dual beta-blockers and ISMN as the PT regimen despite dual drug therapy having been shown to improve portal pressure reduction.6 However, this did not limit the comparison of BL + PT with PT because both trials used dual drug therapy. For BL+PT vs. BL, only one of four trials used dual drug therapy for PT.37 Interestingly, this trial was the largest trial and the one trial that favoured BL alone compared with BL+PT.37 The results from this trial did not support the expected finding that combination BL+PT would be more effective than BL alone. This analysis makes clear that further studies investigating the efficacy of combination therapy are warranted.

In terms of adverse events, trials comparing BL (27%) with PT (31%) had similar event rates. Although most of the BL adverse events appear to be transient events after the procedure, PT events were likely persistent effects associated with continued pharmacotherapy use. Despite the high rates of adverse events, patients tolerated therapies well since study withdrawal (1–11%) and noncompliance was low (2–4%). For trials comparing BL+PT with BL, adverse events were significantly higher with combination therapy (21% vs. 6%). However, the pooled rates were based on a limited number of studies with available information and were highly influenced by one study where the adverse event rates were 33% (BL+PT) vs. 2.7% (BL).36

A recent meta-analysis on combination therapy demonstrated improved efficacy in rebleeding with combination therapy compared with either BL or PT alone.16 However, the present meta-analysis is different for several reasons. First, the present meta-analysis included a pooled analysis of BL vs. PT alone which has not been investigated previously. Second, the interpretation of the overall results of the former meta-analysis is influenced by the inclusion of a majority of trials utilizing endoscopic sclerotherapy. Sclerotherapy is less effective and has a higher complication rate than BL and is not currently recommended for secondary prophylaxis.15 Third, we used a priori the more conservative statistical method of random-effects modelling, which assumes underlying trial heterogeneity. This was appropriate as there was clinical heterogeneity among trials. In contrast to the former meta-analysis which used fixed-effects modelling, the current review used a different statistical approach and found no statistically significant differences for trials of BL+PT vs. BL, although a trend was present. In addition, there was significant outcome heterogeneity for BL+PT vs. BL trials which is unexplained and suggests that pooling the trial results should be interpreted with caution. Fourth, the current meta-analysis separately pooled the only two trials of BL+PT vs. PT and also found no statistically significant differences in rebleeding. Fifth, the current review reports a pooled analysis of adverse event rates.

One of the limitations of the current review was the limited number of full-text publications available for combination therapy trials.34, 36 This limited further evaluation of the precise nature of the causes of rebleeding, such as from varices, band ulcers, or PHG, etc. Additionally, although multiple subgroup analyses were attempted to explain trial heterogeneity, there is an increased risk of statistical type I errors with multiple comparisons. Another limitation is that none of the studies was blinded and there is therefore a higher risk for bias for all trials. Blinding is understandably difficult for trials comparing an endoscopic treatment with a drug treatment. The interpretation of adverse event rates is problematic and should be used with caution given that there were no standard definitions of adverse events. In unblinded combination therapy trials, it is possible that adverse events are more likely to be reported for patients receiving two therapies than for those undergoing single therapy.

In summary, no mortality differences have been demonstrated among therapies of BL, PT and BL+PT. Both BL and PT are equally effective in preventing rebleeding in patients with a previous oesophageal variceal bleed. Combination therapy trials are limited in number, have included variable drug regimens and have heterogenous outcomes. Further published trials with appropriate bias control are required to investigate the efficacy of combination BL+PT vs. either BL or PT alone. Future studies should document the proportion of alcoholic patients and aim for optimal dosing and reporting of portal pressure and heart rate targets achieved with beta-blockers.

Acknowledgement

Declaration of personal interest: The authors would like to thank Dagmara Chojecki for her assistance in the literature search. They would also like to thank the following for providing individual patient data on request: GH Lo, C Villanueva, G Romero, SK Sarin, A Kumar, and G Shiha. Declaration of funding interest: None.

Ancillary