Study characteristics

Characteristics of included studies are summarised in Table 1. Cases underwent operations over a period from 1984 to 2015 and articles were published between 2002 and 2016, reflecting the relatively low incidence of Zenker's diverticulum. Articles were exclusively from developed-world healthcare systems in Australia, North America and Northern Europe. Each article described cases operated on at different centres, but two articles had common authorship.^{Reference Visser, Hardillo, Monserez, Wieringa and Baatenburg de Jong14,Reference Verhaegen, Feuth, van den Hoogen, Marres and Takes17} Three of nine articles compared only endoscopic stapler-assisted diverticulotomy to endoscopic carbon dioxide laser diverticulotomy^{Reference Murer, Soyka, Broglie, Huber and Stoeckli16,Reference Verhaegen, Feuth, van den Hoogen, Marres and Takes17,Reference Adam, Paskhover and Sasaki19} ; three articles included a treatment group that had undergone non-specified transcervical surgery^{Reference Veivers11,Reference Visser, Hardillo, Monserez, Wieringa and Baatenburg de Jong14,Reference Leibowitz, Fundakowski, Abouyared, Rivera, Rudman and Lo15} ; and three compared endoscopic stapler-assisted diverticulotomy, endoscopic carbon dioxide laser diverticulotomy and open diverticulectomy with or without cricopharyngeal myotomy.^{Reference Yeo and Mackenzie12,Reference Shah, Slaughter, Fedore, Huang, Deal and Buckmire13,Reference Gutschow, Hamoir, Rombaux, Otte, Goncette and Collard18} We have not commented on transcervical surgery for Zenker's diverticulum as it does not relate to the hypothesis being tested.

Table 1. Study characteristics

ESD = endoscopic stapler-assisted diverticulotomy; ELD = endoscopic carbon dioxide laser diverticulotomy

Four articles did not describe length of follow up.^{Reference Veivers11,Reference Yeo and Mackenzie12,Reference Visser, Hardillo, Monserez, Wieringa and Baatenburg de Jong14,Reference Verhaegen, Feuth, van den Hoogen, Marres and Takes17} Among the remaining articles, minimum follow-up period varied significantly. Data collection was by case note review in all articles; in addition, two articles used telephone interviews^{Reference Shah, Slaughter, Fedore, Huang, Deal and Buckmire13,Reference Gutschow, Hamoir, Rombaux, Otte, Goncette and Collard18} and two used postal questionnaires^{Reference Murer, Soyka, Broglie, Huber and Stoeckli16,Reference Verhaegen, Feuth, van den Hoogen, Marres and Takes17} to gain information about post-operative symptoms.

Surgical techniques were broadly similar between endoscopic stapler-assisted diverticulotomy and endoscopic carbon dioxide laser diverticulotomy treatment groups. Differences between studies with regards to stapler device and diverticuloscope are described in Table 1.

Study bias

Two reviewers (DE, CT) independently used the Risk of Bias in Non-Randomised Studies of Interventions^{Reference Sterne, Hernan, Reeves, Savovic, Berkman and Viswanathan20} tool to subjectively assess for risk of bias in each included article. Where reviewers disagreed on the level of bias, the more critical assessment was accepted. The Risk of Bias in Non-Randomised Studies of Interventions tool allows assessors to classify risk of bias as ‘low’, ‘moderate’, ‘serious’ or ‘critical’. Risk of bias in selected articles ranged from ‘moderate’ to ‘serious’. A number of articles demonstrated baseline confounding because in some cases the rationale for selection of patients to each treatment group was not specified. In other cases, it was based on factors also considered to be prognostic variables, for example, size of Zenker's diverticulum,^{Reference Shah, Slaughter, Fedore, Huang, Deal and Buckmire13} failure of alternate surgical technique,^{Reference Murer, Soyka, Broglie, Huber and Stoeckli16} or frailty and medical co-morbidity.^{Reference Veivers11} Discrepancy in post-intervention management between treatment groups was another frequently encountered source of bias. Although some studies described a standard post-operative regimen for all patients,^{Reference Veivers11,Reference Visser, Hardillo, Monserez, Wieringa and Baatenburg de Jong14} one reported routine post-operative imaging of endoscopic carbon dioxide laser diverticulotomy but not endoscopic stapler-assisted diverticulotomy cases,^{Reference Verhaegen, Feuth, van den Hoogen, Marres and Takes17} and another reported differences in post-operative fasting period between groups.^{Reference Adam, Paskhover and Sasaki19} Although all articles had at least a ‘moderate’ risk of bias, we considered this to be expected in a group of retrospective non-randomised studies of intervention and will describe the likely direction of bias in our discussion.

Pharyngeal perforation

The incidence of post-operative pharyngeal leak or perforation was reported in each article, but definitions varied. Four studies did not specify how a leak was diagnosed.^{Reference Veivers11,Reference Yeo and Mackenzie12,Reference Visser, Hardillo, Monserez, Wieringa and Baatenburg de Jong14,Reference Murer, Soyka, Broglie, Huber and Stoeckli16} Verhaegen et al. defined leak by radiological findings.^{Reference Verhaegen, Feuth, van den Hoogen, Marres and Takes17} Three studies defined presence of leak based on clinical findings,^{Reference Shah, Slaughter, Fedore, Huang, Deal and Buckmire13,Reference Gutschow, Hamoir, Rombaux, Otte, Goncette and Collard18,Reference Adam, Paskhover and Sasaki19} and Leibowitz et al. described the clinical sequelae of leak, namely pneumomediastinum and retropharyngeal abscess.^{Reference Leibowitz, Fundakowski, Abouyared, Rivera, Rudman and Lo15} Management of leak was not consistently described. Pharyngeal leak was described in 12 of 417 endoscopic stapler-assisted diverticulotomy cases (2.9 per cent) versus 13 of 413 endoscopic carbon dioxide laser diverticulotomy cases (3.1 per cent). The risk of leak in the study group compared with the control group had an odds ratio of 0.83 (95 per cent CI = 0.34 to 2.01). We did not appreciate a difference between treatment groups based on these results. In each study, the confidence interval included an odds ratio equal to one. Six of the studies had zero events in at least one treatment arm and proportional continuity correction was applied. Inverse of variance for random effects was applied (I² = per cent of variation because of heterogeneity; I² = 0 per cent). The heterogeneity was less than 25 per cent. Funnel and forest plots for odds ratio of pharyngeal perforation in each study are shown in Figure 2. Egger test showed a lack of publication bias (p > 0.1).

Fig. 2. Comparison of endoscopic stapler-assisted diverticulotomy (ESD) and endoscopic carbon dioxide laser diverticulotomy (ELD) showing: (a) funnel and (b) forest plots showing difference in odds ratio of pharyngeal perforation following ESD compared with ELD; (c) funnel and (d) forest plots showing difference in odds ratio of re-operation following ESD compared with ELD; and (e) funnel and (f) forest plots showing difference in odds ratio of major complications following ESD compared with ELD. OR = odds ratio

Recurrence and need for re-operation

Two studies reported rate of recurrence of Zenker's diverticulum. Recurrence was not specifically defined by Visser et al.^{Reference Visser, Hardillo, Monserez, Wieringa and Baatenburg de Jong14} but was defined as persistent symptoms requiring a change in diet or re-operation by Leibowitz et al.^{Reference Leibowitz, Fundakowski, Abouyared, Rivera, Rudman and Lo15} Rate of re-operation was reported by 7 studies;^{Reference Veivers11–Reference Visser, Hardillo, Monserez, Wieringa and Baatenburg de Jong14,Reference Murer, Soyka, Broglie, Huber and Stoeckli16,Reference Verhaegen, Feuth, van den Hoogen, Marres and Takes17,Reference Adam, Paskhover and Sasaki19} 51 of 317 (16.1 per cent) and 46 of 293 (15.7 per cent) endoscopic stapler-assisted diverticulotomy and endoscopic carbon dioxide laser diverticulotomy cases, respectively, required re-operation. The re-operation risk in the study group compared with the control group had an odds ratio of 1.13 (95 per cent CI = 0.57 to 2.25). Based on these results, a difference between groups was not appreciated. In each study, the confidence interval included an odds ratio equal to one. Two of the studies had zero events in a treatment group and proportional continuity correction was applied. Inverse of variance for random effects was applied (I² = per cent of variation because of heterogeneity; I² = 19.3 per cent (95 per cent CI = 0 per cent to 61.0 per cent). The heterogeneity was less than 25 per cent. Funnel and forest plots for odds ratio of re-operation in each study are shown in Figure 2. Egger test showed a lack of publication bias (p > 0.1).

Other major complications

The rate of dental complication was not subject to meta-analysis because of the heterogenicity in diverticuloscope used between studies. The review identified no cases of post-operative mortality. Major complications reported included post-operative bleeding,^{Reference Veivers11,Reference Murer, Soyka, Broglie, Huber and Stoeckli16,Reference Verhaegen, Feuth, van den Hoogen, Marres and Takes17} quadriparesis,^{Reference Leibowitz, Fundakowski, Abouyared, Rivera, Rudman and Lo15} and one incidence of vocal fold paralysis in each of the endoscopic stapler-assisted diverticulotomy and endoscopic carbon dioxide laser diverticulotomy groups.^{Reference Veivers11,Reference Visser, Hardillo, Monserez, Wieringa and Baatenburg de Jong14} These were grouped with cases of perforation and their sequelae for analysis of odds ratio of major post-operative complication. The risk of major complication in the study group compared with the control group had an odds ratio of 0.81 (95 per cent CI = 0.43 to 1.54). A difference between groups was not appreciated. In each study, an odds ratio of one was within confidence intervals. Two of the studies had zero events in at least one treatment arm and proportional continuity correction was applied. Inverse of variance for random effects was applied (I² = per cent of variation because of heterogeneity; I² = 0 per cent). The heterogeneity was less than 25 per cent. Funnel and forest plots for odds ratio of major post-operative complication in each study are shown in Figure 2. Egger test showed a lack of publication bias (p > 0.1). The findings of the meta-analysis are summarised in Table 2.

Table 2. Summary of findings

* GRADE (Grades of Recommendation, Assessment, Development and Evaluation) certainty of evidence scoring was low given lack of prospective or randomised studies. CI = confidence interval; ESD = endoscopic stapler-assisted diverticulotomy; ELD = endoscopic carbon dioxide laser diverticulotomy; OR = odds ratio

Certainty of evidence

The Grades of Recommendation, Assessment, Development and Evaluation (‘GRADE’) working group system^{Reference Guyatt, Oxman, Vist, Kunz, Falck-Ytter and Alonso-Coello21} was used to grade the certainty of evidence. For all articles, initial level of certainty rating was ‘low’, as established for retrospective non-randomised studies of interventions. The level of certainty was not upgraded; although no significant evidence of inconsistency, indirectness, imprecision or publication bias was present, bias because of confounding has already been discussed and was thought to favour endoscopic stapler-assisted diverticulotomy treatment groups.