Meta-analysis: banding ligation and medical interventions for the prevention of rebleeding from oesophageal varices

Authors

  • M. Thiele,

    Corresponding author
    1. Department of Gastroenterology and Hepatology, Copenhagen University Hospital Koege, Koege, Denmark
    • Department of Internal Medicine, Copenhagen University Hospital Gentofte, Hellerup, Denmark
    Search for more papers by this author
  • A. Krag,

    1. Department of Internal Medicine, Copenhagen University Hospital Gentofte, Hellerup, Denmark
    2. Department of Gastroenterology and Hepatology, Copenhagen University Hospital Hvidovre, Hvidovre, Denmark
    Search for more papers by this author
  • U. Rohde,

    1. Department of Gastroenterology and Hepatology, Copenhagen University Hospital Hvidovre, Hvidovre, Denmark
    Search for more papers by this author
  • L. L. Gluud

    1. Department of Internal Medicine, Copenhagen University Hospital Gentofte, Hellerup, Denmark
    Search for more papers by this author

  • As part of AP&T's peer-review process, a technical check of this meta-analysis was performed by Dr. Y. Yuan.

Correspondence to:

Dr M. Thiele, Department of Gastroenterology and Hepatology, Copenhagen University Hospital Koege, Lykkebaekvej 1, DK-4600 Koege, Denmark.

E-mail: majath@dadlnet.dk

Summary

Background

In patients with oesophageal varices, the combination of endoscopic variceal ligation (EVL) and medical therapy is recommended as standard of care for prevention of rebleeding. The results of previous meta-analyses on this topic are equivocal.

Aim

To assess the effects of EVL plus medical therapy vs. monotherapy (EVL or medical therapy alone) for secondary prevention in oesophageal varices.

Methods

Electronic and manual searches were combined. The primary outcome measures were overall rebleeding (variceal and nonvariceal) and mortality. Random-effects meta-analyses were performed with subgroup, sensitivity, regression and sequential analyses to identify sources of intertrial heterogeneity and the robustness of the results.

Results

Nine randomised trials were included. In total, 442 patients were randomised to combination therapy and 513 to monotherapy. Combination therapy reduced rebleeding (RR = 0.68; 95% CI = 0.54–0.85; number needed to treat eight patients). The result was confirmed in sequential and regression analyses, but not when limiting the analysis to trials with adequate selection bias control. No effect on overall mortality was identified (RR = 0.89; 95% CI = 0.65–1.21). Combination therapy reduced bleeding-related mortality (RR = 0.52; 95% CI 0.27–0.99; number needed to treat 33 patients) and the risk of rebleeding from oesophageal varices. Combination therapy increased the risk of serious adverse events in fixed, but not in random-effects meta-analyses.

Conclusions

The combination of endoscopic variceal ligation and medical therapy reduce the risk of rebleeding, but not overall mortality. Additional research is needed to determine why reduced rebleeding rates do not lead to reduced mortality.

Introduction

Oesophageal varices are seen in approximately 50% of patients with cirrhosis.[1] Without treatment, about one-third of patients with varices will experience their first bleeding episode within 2 years after the diagnosis is made.[1, 2] Rebleeding occurs in about 60% of these patients during the first 2 years after the initial bleeding episode.[3] In a large proportion of cirrhotic patients, acute upper gastrointestinal bleeding is nonvariceal, and may be fatal.[4] The prognosis of patients with variceal bleeding is severe with a 6-week mortality rate of 20–30%.[5-7] Interventions for the prevention of rebleeding from oesophageal varices are therefore important.

Medical therapy, including nonselective beta-blockers (BB) alone or combined with isosorbide-mononitrate (IsMn) is recommended for the prevention of variceal rebleeding.[8] This treatment may be combined with endoscopic interventions. Endoscopic injection sclerotherapy is no longer recommended as it is associated with severe adverse events including oesophageal strictures.[9] Endoscopic ligation of the varices (EVL) is associated with fewer adverse events and lower rebleeding rates.[10, 11] Current guidelines for the prevention of rebleeding from oesophageal varices recommend medical therapy combined with EVL.[12-14] Previous meta-analyses found that the combination of medical therapy and endoscopic therapy reduces both overall and oesophageal variceal rebleeding,[15-17] but when analysing the subgroup of trials on EVL and medical therapy alone, the results were inconclusive. A meta-analysis on EVL and medical therapy[18] found no difference between the combination therapy and monotherapy in either rebleeding or mortality, but this may reflect the number of trials included. Since a number of subsequent trials have been published, we have conducted the present updated systematic review.

Methods

The present study is based on a protocol developed in accordance with the Cochrane Handbook for Intervention Reviews (http://www.Cochrane.org). The objective was to evaluate the beneficial and harmful effects of the combination of EVL with medical therapy (BB alone or with IsMn) vs. monotherapy (EVL or medical therapy) for patients with previous bleeding from oesophageal varices. We included trials investigating combination therapy with monotherapy in adult patients with endoscopically verified oesophageal varices and a history of oesophageal variceal bleeding. We excluded trials in which only patients with gastric varices were included.

Randomised trials were included irrespective of publication status or language.

Information sources

All authors participated in the selection of trials for inclusion. The electronic searches were performed in The Cochrane Hepato-Biliary Group Controlled Trials Register, the Cochrane Library, MEDLINE, EMBASE, and Science Citation Index Expanded. See Appendix S1 published online for detailed search strategy. The search update was performed in November 2011. Additional trials were identified through scanning of reference lists in relevant papers, conference proceedings, correspondence with experts and pharmaceutical companies, and online trial registers through the World Health Organization International Clinical Trials Registry Platform (http://apps.who.int/trialsearch/). Authors of included trials were approached for additional information when necessary.

Data

Two authors (MT, UR) extracted data in an independent manner. Discrepancies were resolved through discussion before the analyses. Data on all patients were sought to allow for intention-to-treat analyses. Extracted data included outcome measures, risk of bias and characteristics of trials, patients, and interventions (type, dose, frequency, and duration).

The primary outcome measures were upper gastrointestinal rebleeding and mortality. The outcome measure of upper gastrointestinal rebleeding was defined as the combination of variceal (oesophageal and gastric) and nonvariceal bleeding. Accordingly, trials that only reported variceal rebleeding were not included in our analysis of upper gastrointestinal bleeding.

Secondary outcome measures were rebleeding from oesophageal varices, bleeding-related mortality, quality of life and adverse events.

Methodological quality

Based on previous evidence,[19, 20] the assessment of bias control focused on randomisation methods assessed by the allocation sequence generation and allocation concealment. The allocation sequence was classed as adequate if based on computer generated random numbers, a table of random numbers or similar. The allocation concealment was classed as adequate if patients were randomised through a central independent unit, serially numbered opaque sealed envelopes or similar. Additional measures of bias control included blinding (detection and performance bias), reporting bias (whether the most clinically relevant outcome measures were reported), handling of missing outcome data (attrition bias classed based on whether all patients randomised were accounted for) and sample size calculations (whether a sample size calculation was performed and the required sample size was reached).

Statistical analysis

Data analyses were performed in RevMan version 5 (The Nordic Cochrane Centre, Copenhagen, Denmark) and STATA version 11 (Stata Corp, College Station, TX, USA). Due to expected clinical heterogeneity (differences between patient inclusion criteria and intervention regimens) our primary meta-analyses were performed using random-effects models. Results were expressed as risk ratios (RR) with 95% confidence intervals (CI) and I2-values as a marker of heterogeneity. I2-values below 25% were considered low risk of heterogeneity. For meta-analyses showing a clear intervention benefit (when the CI does not cross the value 1), the number needed to treat to prevent one unwanted event was calculated based on the inverse of the risk difference.

For the primary outcome measures, regression analysis (Egger's test of bias) was performed to assess evidence of bias. Sensitivity analyses including fixed effect meta-analyses were also performed. The robustness of the primary meta-analysis showing a clear intervention benefit (rebleeding) was assessed in sequential analysis adjusting for multiple testing.[21] The sequential analysis was performed using a random-effects model with alpha set to 5% and power to 80% and including the incidence rates and the intervention effects identified in the meta-analyses. Subgroup analyses were performed to evaluate intervention effects in trials comparing combination therapy vs. EVL or medical therapy, trials with adequate bias control (assessed through randomization methods) and publication status (excluding trials published in abstract form).

Results

The initial searches lead to the identification of 932 potentially eligible references (Figure 1). Among these 913 were excluded because they were duplicates or clearly irrelevant (did not refer to clinical trials or assessed primary prevention of variceal bleeding or did not assess the interventions specified in the protocol). Nineteen references referred to trials that were listed for further evaluation. Four of these trials were excluded, since they turned out not be randomised[22] and did not assess the interventions specified in the present protocol[23, 24] or were ongoing trials (no available data).[25] Nine trials (described in 15 references) were included in our qualitative analyses (Table 1).[26-40] All authors were approached for additional data. Authors from two trials responded,[31, 39] one of which supplied us with unpublished data.[39] For remaining trials, analyses are based on data retrieved from the published reports.

Figure 1.

Flow chart of trials.

Table 1. Table of included randomised trials
Name, yearType of publicationnGender, M/FMean age, yearsChild-pugh score A/B/CInclusion criteriacAetiology of cirrhosis, alcohol/viral/other
  1. T, treatment; C, control; EIS, endoscopic injection sclerotherapy.

  2. a

    n = 39 in BB group, n = 35 in BB+IsMn group, n = 39 in EVL group.

  3. b

    a) BB group, b) BB+IsMn group, c) EVL group.

  4. c

    Apart from age 18–75.

  5. d

    Correspondence with primary investigator.

Abdel-Rahim, 2000[27]Abstract

T: 25

C: 25

Not statedNot statedNot statedRecent oesophageal variceal bleedingNot stated
Ahmad, 2009[28]Article

T: 37

C: 113a

T: 30/7Cb:

a) 25/14

b) 21/14

c) 25/14

T: 50.5Cb:

a) 52.6

b) 51.9

c) 52.6

T: 4/27/6Cb:

a) 6/23/10

b) 2/19/14

c) 7/23/9

Endoscopy proven oesophageal variceal bleedingCirrhosis1/149/0
De la Pena, 2005[29]Article

T: 43

C: 37

T: 33/10

C: 27/10

T: 60

C: 60

T: 6/25/12

C: 6/20/11

Endoscopy proven oesophageal variceal bleedingCirrhosis53/20/7
Garcia-Pagan, 2009[31]Article

T: 80

C: 78

T: 65/15

C: 53/25

T: 57

C: 56

T:16/46/18C:18/42/18Endoscopy proven oesophageal variceal bleedingCirrhosis

81/43/15

(19 patients with both viral hepatitis and alcohol)

Kumar, 2009[35]Article

T: 84

C: 83

T: 75/13

C: 78/11

T: 42

C: 41

T: 35/31/10

C: 26/34/15

Endoscopy proven oesophageal variceal bleeding

63/33/55

26 without cirrhosis

Lo, 2000[37]Article

T: 60

C: 62

T: 45/15

C: 49/13

T: 53

C: 51

T: 11/30/19

C: 12/28/22

Endoscopy proven oesophageal variceal bleedingCirrhosis37/82/3
Lo, 2001[38]Article

T: 37

C:40

T: 31/6

C: 35/5

T: 57

C: 55

T: 3/18/16

C: 4/21/15

History of oesophageal variceal bleedingPortal hypertension due to cirrhosis23/47/7
Lo, 2009[36]Article

T: 60

C: 60

T: 42/18

C: 45/15

T: 54

C: 52

T: 20/29/11

C: 21/31/8

Endoscopy proven oesophageal variceal bleedingCirrhosis36/71/13
Sollano, 2001[39]Abstract

T: 16

C: 15

T: 13/3d

C: 11/4

T: 47d

C: 53

Not statedEndoscopy proven oesophageald variceal bleedingCirrhosis6/25/0d

The included trials were conducted in Spain, Taiwan, India, Pakistan, Egypt, and the Philippines. Seven trials were published as full paper articles[28, 29, 31, 35-38] and two in abstract form.[27, 39] Eight trials were designed with two and one trial with four intervention arms (Table 2). Seven trials compared combination therapy vs. EVL and three trials combination vs. medical therapy. The proportion of patients with cirrhosis was 100% in seven trials and 85% in one trial.[35] One trial did not report the proportion of patients with cirrhosis.[27] Exclusion criteria were comparable between trials, focusing on previous drug or endoscopic treatment, contraindications to treatment, malignancy or other debiliating disease and former surgery for portal hypertension. Two trials excluded patients with refractory ascites,[29, 38] four trials excluded patients with hepatic encephalopathy[29, 37-39] and one trial patients excluded patients who did not achieve variceal eradication.[39]

Table 2. Table of interventions in included randomised trials
Name, yearEndoscopic intervention at index bleedFollow-up, monthsTreatmentControlType and mean dose of beta blockerMean EVL sessions to obliterateRate of obliterationTime to obliterate, daysVariceal recurrence rateNumber of patients lost to follow-up
  1. T, treatment; C, control; EIS, endoscopic injection sclerotherapy.

  2. a

    Correspondence with primary investigator.

Abdel-Rahim, 2000[27]Not stated

T: 3

C: 3

EVL+BBEVLPropranololT: 115 mg/dNot statedNot statedNot statedNot statedNot stated
Ahmad, 2009[28]EVL or EIS

T: 9.5

C: 8.4

EVL+BB+IsMn

a) EVL

b) BB

c) BB+IsMn

Propranolol

T: 53 mg/d

C: 52 mg/db) 50 mg/dc)

T: 3

Ca): 3.5

T: 100%

Ca): 100%

T: 54

Ca): 62

T: 32%

Ca): 26%

2 lost to follow-up, 8 excluded due to adverse events
De la Pena, 2005[29]EVL or EIS

T: 17.5

C: 15

EVL+BBEVL

Nadolol

T: 58 mg/d

T: 3

C: 3

T: 100%

C: 100%

Not stated

T: 68%

C: 97%

3 excluded due to adverse events, 1 excluded due to malignancy
Garcia-Pagan, 2009[31]EVL or EIS

T: 15.3

C: 14.4

EVL+BB+IsMnBB+IsMn

Nadolol

T: 90 mg/d

C: 102 mg/d

T: 2T: 75%T: 21T: 33%4 lost to follow-up, 2 excluded after randomization
Kumar, 2009[35]EVL

T: 15

C: 15

EVL+BB+IsMnEVL

Propranolol

T: 120 mg/d

T: 4.6

C: 4.6

T: 52%

C: 40%

Not stated

T: 36%

C: 36%

10 lost to follow-up, excluded
Lo, 2000[37]EVL

T: 22

C: 21

EVL+BB+SucralfateEVL

Nadolol

T: 60 mg/d

T: 3.3

C: 3.6

T: 75%

C: 70%

T: 51

C: 55

T: 26%

C: 50%

1 lost to follow-up, 3 did not comply with protocol
Lo, 2001[38]EVL

T: 17

C: 17

EVL+BBEVL

Propranolol

T: 96 mg/d

T: 3.3

C: 3.6

T: 84%

C: 80%

T: 42

C: 46

T: 38%

C: 43%

5 did not comply with protocol
Lo, 2009[36]EVL

T: 23.2

C: 22.7

EVL+BB+IsMnBB+IsMn

Nadolol

T: 40 mg/d

C: 40 mg/d

T: 3.8T: 75%Not statedNot stated2 lost to follow-up, 16 did not comply with protocol
Sollano, 2001[39]EVL

T: 9.6

C: 9.2

EVL+BBEVLPropranolol Dose not statedNot stated

T: 100%

C: 100%

Not statedNot stated2 deaths, excludeda

The duration of follow-up ranged from 3 to 23 months. The mean age of included patients ranged from 41 to 60 years.

Bias assessment

The allocation sequence generation was classed as adequate in seven trials and the allocation concealment as adequate in five trials (Table 3). Three trials[36-38] used blinded outcome assessment (blinding of endoscopist when evaluating the source of rebleeding), but none used blinding of participants or investigators. Three trials were classed as having incomplete reporting of outcome data. Selective reporting of outcome measures was registered for two trials. Six trials performed sample size calculations,[29, 35, 36, 38, 39, 41] three of which met the required sample size.[35, 38, 41]

Table 3. Risk of bias summary
 Random sequence generation (selection bias)Allocation concealment (selection bias)Blinding (performance bias and detection bias)Incomplete outcome data (attrition bias)Selective reporting (reporting bias)Other bias
  1. *Other bias refers to whether a sample size calculation was done and whether the sample size was met.

Abdel-Rahim 2000????+?
Ahmad 2009++++?
De la Pena 2005+?+++
Garcia-Pagan 2009+++++
Kumar 2009++++
Lo 2000+?+++
Lo 2001?+?
Lo 2009+?++?
Sollano 2001++?

Overall and oesophageal variceal rebleeding

Six trials with a total of 724 patients reported upper gastrointestinal rebleeding (Figure 2).[29, 31, 35-37] Rebleeding was diagnosed for 88 of 364 patients randomised to combination therapy and 131 of 360 patients in the control group. In total, 60 rebleeding episodes (27%) did not stem from oesophageal varices, but from portal hypertensive gastropathy, gastric varices, postligation ulcers and peptic ulcers. Random-effects meta-analysis showed that combination therapy reduced the risk of upper gastrointestinal rebleeding (RR = 0.68; 95% CI = 0.54–0.85; I2 = 1%). The corresponding number needed to treat with combined therapy to prevent one case of rebleeding was eight (95% CI = 5–25). The result was confirmed when the meta-analysis was repeated using a fixed effect model. There was no evidence of bias in regression analyses (Egger's test P = 0.460). A sequential analysis was performed with a relative risk reduction of 25.87, a control group incidence rate of 38% and model based heterogeneity correction. The analysis suggested that the overall result was robust to adjustment for multiple testing since the monitoring and alpha spending boundary crossed in 2009 after the initial five trials were published. In subgroup analyses, combination therapy reduced upper gastrointestinal rebleeding compared with EVL (RR = 0.59; 95% CI = 0.41–0.85, 4 trials), but not when compared with medical therapy (RR = 0.76; 95% CI = 0.56–1.04, 2 trials). Likewise, the difference between combination therapy and monotherapy was not seen in an analysis that only included trials with adequate randomisation (RR = 0.67; 95% CI = 0.42–1.09). Repeating the meta-analysis with inclusion of full paper articles confirmed the overall result.

Figure 2.

Random effect meta-analysis of overall rebleeding in randomised trials on combination therapy [medical interventions (MI) and banding ligation (EVL)] vs. monotherapy (MI or EVL) on secondary prevention in oesophageal varices.

Based on peer review comments we performed a post hoc analysis of upper gastrointestinal rebleeding including data from trials that only reported variceal rebleeding. This did not change the overall results (RR = 0.67; 95% CI = 0.54–0.84; I2 = 0%, 9 trials).

Eight trials with a total of 905 patients reported rebleeding from oesophageal varices (Figure 3).[28, 29, 31, 35-39] The combination of EVL and medical therapy reduced the risk of rebleeding compared with monotherapy (RR = 0.61; 95% CI = 0.47–0.80; I2 = 0%). The number needed to treat was 10 patients (95% CI = 7–20). The result remained stable in analysis using a fixed effects model. The result was confirmed in subgroup analyses comparing trials on combination therapy vs. monotherapy with EVL (RR = 0.65; 95% CI = 0.45–0.93) or medical interventions (RR = 0.61; 95% CI = 0.44–0.86). Subgroup analyses with trials published as full paper articles and trials with adequate selection bias control did not change the results.

Figure 3.

Random effect meta-analysis of oesophageal variceal rebleeding in randomised trials on combination therapy [medical interventions (MI) and banding ligation (EVL)] vs. monotherapy (MI or EVL) on secondary prevention in oesophageal varices. *Overall risk estimate not shown, referred in text.

Overall and rebleeding-related mortality

Eight trials with 878 patients reported overall mortality (Figure 4).[27-29, 31, 35-37, 39] A total of 59 of 405 patients in the combination therapy and 81 of 473 in the control group died (RR = 0.90; 95% CI = 0.66–1.23). There was no evidence of intertrial heterogeneity (I2 = 0%) and no evidence of bias in the regression analyses (Egger's test P = 0.588). The result was confirmed when the meta-analysis was repeated using a fixed effect model. In subgroup analyses, mortality was not different in the combination and control group when analysing trials comparing combination therapy vs. EVL (RR = 0.71; 95% CI = 0.45–1.11) or medical interventions (RR = 1.08; 95% CI = 0.73–1.60). The overall result remained stable in subgroup analyses including trials with adequate randomisation and trials published as full paper articles.

Figure 4.

Random effect meta-analysis of mortality in randomised trials on combination therapy [medical interventions (MI) and banding ligation (EVL)] vs. monotherapy (MI or EVL) on secondary prevention in oesophageal varices. *Overall risk estimate not shown, referred in text.

Six trials described the causes of death.[28, 29, 31, 36, 37, 39] The most common cause was liver failure (32% of deaths), followed by bleeding-related mortality and infections.

In a total of seven trials,[28, 29, 31, 35-37, 39] mortality was classed as bleeding-related for 11 of 380 patients in the combination therapy and 29 of 448 patients in the control group. Combination therapy reduced the bleeding-related mortality (RR = 0.52; 95% CI = 0.27–0.99) with number needed to treat 33 patients (95% CI = 17–135).

None of the included trials provided data on the quality of life.

Adverse Events

Six trials reported adverse events.[28, 29, 31, 35-37] The combination therapy and monotherapy did not differ regarding the total number of adverse events (RR = 1.42; 95% CI = 0.94–2.13; I2 = 69%) or serious adverse events (RR = 1.93; 95% CI = 0.94–3.99; I2 = 22%). However, there was clear evidence of intertrial heterogeneity. Repeating the meta-analyses using a fixed effect model, the combination therapy caused more adverse events (RR = 1.38; 95% CI = 1.13–1.68) and serious adverse events (RR = 2.02; 95% CI = 1.14–3.56).[28, 29, 31, 35-37]

The most common serious adverse events related to EVL were clinically important bleeding from postbanding ulcers (13 cases).[28, 29, 31, 36, 37] One case of oesophageal stricture necessitating dilation was registered.[29] The most common causes for withdrawal of BB were bradycardia, hypotension, dizziness, asthenia, lethargy or fatigue.[28, 29, 31] Three cases of hypoglycaemia were reported including one that required hospitalisation.[29, 31] Two cases of clinically important cardiac arrhythmias were reported.[28, 29] Infections including sepsis, bacteriaemia, pneumonia, spontaneous bacterial peritonitis, and postligation fever were registered in six trials (27 of 364 vs. 21 of 433; RR = 1.35; 95% CI = 0.80–2.29).[28, 29, 31, 35-37]

Discussion

The present review suggests that the combination of EVL and medical therapy reduces the risk of upper gastrointestinal rebleeding and rebleeding from oesophageal varices compared with EVL or BB used alone. The reduced risk of rebleeding was associated with a reduction in bleeding-related but not overall mortality. These results suggest that the combination of EVL and medical therapy may be the best treatment choice for most patients with oesophageal varices and previous bleeding. However, the number of losses to follow-up and adverse events may reflect that a number of patients do not tolerate the combination regimen. Furthermore, the characteristics of included patients suggest that the patients analysed and the general population of patients with oesophageal varices and previous bleeding may differ. Accordingly, the best treatment strategy for this patient group still involves a considerable degree of clinical judgment.

When the overall analysis of rebleeding was assessed, the benefit of combining EVL with medical therapy was only seen when compared with EVL alone and not when compared with medical therapy. The reason may be related to the number of patients and the number of events in the two subgroup analyses. The fact that there was no clear difference between combination therapy and medical therapy could reflect inadequate statistical power. On the other hand, the difference could be related to potential nonhemodynamic effects of BB such as a reduction in the drive of the sympathic nervous system and a reduced risk of bacterial infection through improved immune response and reduced bacterial translocation.[42, 43] Indeed, a previous meta-analysis suggests that medical therapy reduces mortality compared with EVL.[44] Reduced bacterial translocation is thought to further improve portal pressure reduction[45] and protect against spontaneous bacterial peritonitis,[46] overall improving both rebleeding episodes[47] and survival.[44, 48] This protection from bacterial translocation with BB is, however, only seen in patients with ascites.[49, 50]

There is some evidence suggesting that the patients who are treated with the combination of EVL and BB have an increased risk of serious adverse events. Unfortunately, we were unable to perform in-depth assessments of this potential question as the number and definition of adverse events varied across trials. Additional evidence is needed to identify the subgroup of patients that may benefit from combination therapy, and potential subgroups that may be treated with medical interventions alone.

In contrast to previous meta-analyses with fewer trials, we did not find evidence of intertrial heterogeneity or funnel plot asymmetry and our overall results were confirmed in sequential analysis. On the other hand, only one of the included trials was classed as having overall adequate bias control (not counting blinding, which could not be expected due to the type of intervention). Four trials were classed as having adequate selection bias control. Although, there is evidence suggesting that objective outcome measures are more stable to the potential influence of bias,[20] the risk of bias should be considered. Our analyses were limited to the data that were reported in the included articles. For patients lost to follow-up, we were unable to perform intention-to-treat analysis if the data were not presented in the published report. Four trials reported that patients were randomised and lost to follow-up without specifying their allocation group. We were unable to include these patients in our analyses. We cannot exclude the possibility that this has lead to bias.

Few data on infections were available in included trials. Among 129 patients who died, the reported cause of death was infections in 24 cases. These data are difficult to interpret as competing events are likely to have occurred. For example, in a number of patients who died, infections may have led to liver failure or rebleeding. We found that mortality was lower in the combination therapy compared with the EVL group. However, the result was not statistically significant (RR = 0.69; 95% CI = 0.43–1.09). Furthermore, even planned subgroup analyses in meta-analyses should be viewed with caution.

One observational study suggests that BB have deleterious effects in refractory ascites.[51] Two of the trials in our meta-analysis excluded patients with refractory ascites,[29, 38] and one trial excluded patients with Child-Pugh score above 13.[31] In the remaining trials 27% to 87% of patients had ascites. The proportion of patients with refractory ascites was not reported, but the life expectancy in refractory ascites suggests that these patients were not included in trials with up to 2 years of follow-up. Additional analyses on this issue would have been interesting, but may require individual patient data. Unfortunately, these data were not available to us.

Overall, our findings support current recommendations regarding oesophageal variceal rebleeding prophylaxis.[12, 14] On the other hand, it may be argued that we still need evidence from high quality trials and that the lack of effect on overall mortality raises important questions. The frequency of adverse events, and the effect of combination therapy and monotherapy in specific subgroups still need to be assessed. Important questions include the potential effect of combination therapy in patients with severe or end-stage liver disease, patients with more than one oesophageal variceal rebleeding episode, patients already receiving secondary prophylaxis when experiencing an oesophageal variceal rebleeding episode and methods that may increase the effects of the interventions assessed (such as hepatic venous pressure gradient-guided therapy).

Acknowledgements

Thank you to Sarah Klingenberg who helped to prepare the search strategy. Thank you to Hongyun Chen who helped translate the article by Cheng et al. Thank you to authors Sollano, Garcia-Pagan and Bosch, who kindly responded to our request for additional information. Declaration of personal and funding interests: None.

Ancillary