Anastomotic techniques for oesophagectomy for malignancy: systematic review and network meta‐analysis

Background Current evidence on the benefits of different anastomotic techniques (hand‐sewn (HS), circular stapled (CS), triangulating stapled (TS) or linear stapled/semimechanical (LSSM) techniques) after oesophagectomy is conflicting. The aim of this study was to evaluate the evidence for the techniques for oesophagogastric anastomosis and their impact on perioperative outcomes. Methods This was a systematic review and network meta‐analysis. PubMed, EMBASE and Cochrane Library databases were searched systematically for randomized and non‐randomized studies reporting techniques for the oesophagogastric anastomosis. Network meta‐analysis of postoperative anastomotic leaks and strictures was performed. Results Of 4192 articles screened, 15 randomized and 22 non‐randomized studies comprising 8618 patients were included. LSSM (odds ratio (OR) 0·50, 95 per cent c.i. 0·33 to 0·74; P = 0·001) and CS (OR 0·68, 0·48 to 0·95; P = 0·027) anastomoses were associated with lower anastomotic leak rates than HS anastomoses. LSSM anastomoses were associated with lower stricture rates than HS anastomoses (OR 0·32, 0·19 to 0·54; P < 0·001). Conclusion LSSM anastomoses after oesophagectomy are superior with regard to anastomotic leak and stricture rates.


Introduction
Despite improvements in perioperative care over recent decades, which have led to improved patient selection, reduced operative morbidity and mortality, and prolonged postoperative survival 1,2 , anastomotic leak remains the most serious technical complication after oesophagectomy. Patients who experience anastomotic leakage suffer high morbidity, have a high postoperative mortality rate, ranging between 21 and 35 per cent, incur high hospital costs [3][4][5][6][7] . They also suffer long-term effects, such as an increased risk of anastomotic stricture and poorer long-term survival, compared with patients who recover uneventfully 8 . Many perioperative factors are thought to be responsible for anastomotic integrity after oesophagectomy, such as surgical approach, tumour location (cervical or thoracic) and technique of oesophagogastric anastomosis 9 .
Several meta-analyses [10][11][12][13][14] have compared stapled and hand-sewn anastomotic techniques. These studies have included both randomized and non-randomized trials, and have found no significant differences in anastomotic leak rates between the two anastomotic techniques. Most individual comparative studies, however, chose either to look at two types of stapled anastomosis or to group all stapled anastomoses together. There is a paucity of literature comparing all anastomotic techniques described in this study. Anastomotic techniques can include hand-sewn (HS), circular stapled (CS), linear stapled/semimechanical (LSSM) 15,16 and triangulating stapled (TS) 17,18 . There are encouraging reports of low anastomotic leak rates when linear stapled techniques are employed 19 . The strength of performing a network meta-analysis is that it allows the evaluation of treatments that have not been compared directly (for example, comparison of B versus C, using data from studies comparing A versus B and A versus C). Network meta-analysis ranks multiple treatments based on their efficacy, and pools together direct and indirect evidence within mixed comparisons, improving the precision of estimates 20,21 .
The aim of this systematic review was to evaluate current evidence and perform a network meta-analysis to identify techniques associated with superior perioperative outcomes in patients undergoing oesophagectomy for oesophageal cancer.

Methods
This was a systematic review and network meta-analysis. The study was registered with the PROSPERO database (Registration CRD42018106086) and reported according to the PRISMA guidelines 22 .

Search strategy
A systematic search of PubMed, EMBASE and Cochrane Library databases was conducted by two independent investigators on 22 April 2019, to include studies up to 31 March 2019. Search terms included 'oesophageal cancer' or 'esophageal cancer' or 'gastro-oesophageal cancer', and 'anastomosis' or 'hand-sewn' or 'linear stapler' or 'circular stapler' individually or in combination (Table S1, supporting information). The 'related articles' function was used to broaden the search, and all citations were considered for relevance. A manual search of reference lists in recent reviews was also undertaken. After excluding duplicates, two researchers independently reviewed the titles and abstracts of studies identified by the literature search. If a study was considered to be potentially relevant to the research question, the full publication was reviewed. Reference lists of all included studies were hand-searched to identify other potentially relevant studies. Any areas of disagreement between the two primary researchers were resolved through discussion with all authors.

Inclusion and exclusion criteria
Inclusion criteria included studies reporting the comparison of anastomotic technique (by any method) in patients with oesophageal cancer who underwent oesophagectomy, published in the English language. Exclusion criteria included conference abstracts, review articles, case reports (fewer than 5 patients), and publications with mixed populations, in which the outcomes of patients with either benign disease or cancer at another site could not be separated from those of patients with oesophageal cancer.

Study outcomes
The primary outcome measures were anastomotic complications, including anastomotic leak or stricture 23 . Secondary outcome measures were surgery-specific complications (pulmonary, cardiac) and death (30-day and in-hospital mortality).

Data extraction
One researcher extracted data on study characteristics (author, year of publication, country of origin, study design, patient number), patient demographics (age, sex), tumour stage (AJCC T category and AJCC stage), method and details of anastomotic technique, and reported clinical outcomes.

Definitions
Oesophageal cancer was defined as a malignancy of any portion of the oesophagus. Anastomotic technique was defined as any method of oesophagogastric anastomosis including HS, CS, LSSM 15,16 and TS 17,18 anastomoses. These anastomotic techniques may be employed in either the thoracic or the cervical phase of the operation. Subtle variations of the LSSM technique were described. The nomenclature includes a (modified) Collard technique 24 and a side-to-side semimechanical technique, which both refer to a combination of a linear stapled and hand-sewn technique.

Assessment of methodological quality
Methodological quality and standard of outcome reporting within included studies were assessed by two independent researchers. Methodological quality was assessed formally using the Newcastle-Ottawa Scale (NOS) 25,26 for cohort studies and the Cochrane risk-of-bias tool for RCTs.

Statistical analysis
Dichotomous outcomes were compared between anastomotic formation techniques using odds ratios (ORs), produced using random-effects DerSimonian-Laird meta-analytical models. Both randomized and non-randomized studies were pooled into a network meta-analysis comparing the above anastomotic formation techniques. Sensitivity analyses were performed for type of study (RCTs only, RCTs and prospective cohort studies (PCSs) and all RCT and cohort studies with a NOS score of 8 or above), study year (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018), and level of anastomosis (cervical versus thoracic). For each outcome, graphical representations of treatments (nodes) and comparisons (lines) were mapped. Network maps were then analysed for closed loops to be entered into network analyses. Networks were then examined for the presence of inconsistency, allowing for comparisons between direct and indirect treatment effects. Initially, this was assessed by checking for overall inconsistency throughout the entire network. A further check was then performed by fitting node side-splitting models, to identify loop inconsistency, within all three-way treatment comparison loops, as described by Dias and colleagues 27 . When P values were greater than 0⋅050, representing acceptance of the null hypothesis, consistency was assumed and networks were entered into consistency modelling. Consistency modelling utilized a restricted maximum likelihood model, generating network forest plots. Heterogeneity was examined by calculating the value of τ 2 . Hand-sewn anastomosis was used as the common reference treatment for all comparisons. These were supplemented with interval plots of pooled effect estimates.
Anastomotic techniques were then ranked using the P-score provided by the netmeta package (RStudio ® 3.2.1, Boston, Massachusetts, USA; https://CRAN.R-project .org/package=netmeta). The surface under the cumulative ranking areas for all outcomes assessed the probability of the superiority of each treatment [28][29][30] . The probability of ranking of a treatment (that a treatment ranks as the best treatment, second best treatment, third best treatment) for each outcome of interest was calculated. A probability of ranking below 90 per cent was not considered to be high enough to be confidently reported as the correct ranking position of a surgical technique for that outcome of interest 27,31 . Statistical significance was considered when P < 0⋅050. Statistical analyses were performed using R Foundation statistical software (RStudio ® 3.2.1).

Results
Of 4192 studies screened, 40 studies comparing different anastomotic techniques were eligible (Fig. 1). Details of these studies [15][16][17][18] are shown in Table 1. The majority of included studies reported on open oesophagectomy. Three studies 55,66,67 compared different types of hand-sewn layers and were therefore excluded from meta-analysis. The remaining 37 studies comprised 8571 patients. Fifteen RCTs, five non-randomized prospective and 17 retrospective studies were included. Within the non-randomized studies, the mean NOS score was 7 (range 5-9) ( Table 2). In RCTs, reporting of the blinding of participants and outcomes was unclear, but the risk of bias was mainly low for other domains. The anastomotic techniques analysed most commonly were HS (35 studies) and CS (26), with these two techniques being directly compared in 24 studies. LSSM was analysed in 16 studies, of which ten included comparisons with HS, and a further five studies reported comparisons with both HS and CS. In addition, five studies analysed TS anastomoses. One study 52 described a linear stapled technique without the use of a hand-sewn component. A summary of technical details of anastomosis is presented in Table S2 (supporting information).

Network meta-analysis
Network meta-analyses for the two primary outcomes, anastomotic leak and stricture, were conducted comparing all anastomotic techniques described in two or more studies including HS, TS, LSSM and CS. Initially, visual representations of the network of studies used for each outcome were generated (Figs 2a and 3a). For anastomotic leak, 36 studies were included, consisting of 2623 patients with a CS anastomosis, 1876 with a LSSM anastomosis, 3922 receiving a HS anastomosis, and 197 who had a TS anastomosis. For anastomotic stricture formation, 27 studies were included, comprising 1775 patients with a CS anastomosis, 1602 with a LSSM anastomosis, 3224 with a HS anastomosis, and 197 receiving a TS anastomosis.

Cohort studies
Perrachia et al. Level of bias was determined as: low, indicating a low risk of bias; unclear, indicating an uncertain risk of bias, and high, indicating a high risk of bias. NOS, Newcastle-Ottawa Scale; PCS, prospective cohort study; RCS, retrospective cohort study; n.a., not applicable.  and LSSM (OR 0⋅50; P = 0⋅001) anastomoses were associated with lower anastomotic leak rates than HS anastomosis. Anastomotic leak rates were no different for CS and LSSM anastomoses (OR 1⋅37; P = 0⋅198). There were no significant differences in leak rates between TS and HS. Overall, LSSM was ranked the best technique regarding anastomotic leakage with high probability, followed by CS and TS (Table S3, supporting information).

Sensitivity analysis of anastomotic leak
For cervical anastomosis, no technique was superior with regard to anastomotic leakage ( Table 3). For thoracic anastomosis, LSSM (OR 0⋅42) and CS (OR 0⋅56) anastomoses were superior to HS with regard to anastomotic leakage. There were no differences in anastomotic leaks between CS and LSSM (OR 1⋅33). In the analyses split by study type, only LSSM anastomoses (OR 0⋅56) had a lower anastomotic leak rate than HS in the 'RCT and cohort studies with a NOS score of 8 or above' subgroup only. In the subgroup analysis of RCTs there were no statistically significant differences. For studies published in 2005-2018, only LSSM (OR 0⋅50) was superior to HS anastomosis.

Intraoperative outcomes
Duration of surgery was reported in 16 studies. There were no differences in operating times between techniques ( Table S4, supporting information). LSSM was ranked first for the entire cohort and for cervical anastomosis only. Blood loss was reported in 11 studies. LSSM had significantly lower blood loss than HS (mean difference 24 ml; P = 0⋅024). LSSM was ranked first for the entire cohort and for cervical anastomosis only. There were insufficient studies in the thoracic anastomosis subgroup only for analysis.

Other postoperative complications
Cardiac complications rates were reported in ten studies (CS versus HS, 7 studies; LSSM versus HS, 2; CS versus TS, 1) ( Table S4, supporting information). There were no significant differences in cardiac complications between the different techniques. LSSM was ranked first for the overall and cervical anastomosis only subgroup. There were not enough studies in the thoracic anastomosis only subgroup for analysis.
Pulmonary complications were reported in 12 studies (CS versus HS, 8 studies; LSSM versus HS, 2; CS versus TS, 2). There were no significant differences in pulmonary complications between the different techniques. TS was ranked first in the overall group. LSSM was ranked first in the cervical anastomosis only subgroup. There were not enough studies in the thoracic anastomosis only subgroup for analysis.
Thirty-day mortality was reported in 11 studies (CS versus HS, 7 studies; LSSM versus HS, 3; CS versus TS, 1). LSSM was associated with lower rates of 30-day mortality than HS (OR 0⋅33; P = 0⋅016) and CS (OR 0⋅18; P = 0⋅002) anastomoses. CS was not associated with higher mortality rates than HS anastomosis. LSSM was ranked the best technique with high probability, followed by TS. For cervical anastomosis, LSSM was ranked first. There were not enough studies in the thoracic anastomosis only subgroup for analysis.
In-hospital mortality was reported in several pairwise comparisons: CS versus HS, 16 studies; LSSM versus HS, five; CS versus TS, three; LSSM versus HS, four studies. LSSM was associated with lower rates of in-hospital mortality than HS (OR 0⋅32; P < 0⋅001), CS (OR 0⋅15; P < 0⋅001) and TS (OR 0⋅15; P = 0⋅001) anastomoses. CS was associated with higher in-hospital mortality rates than HS anastomosis. LSSM was ranked the best technique with high probability, followed by HS. For cervical anastomosis, HS was ranked first. There were not enough studies in the thoracic anastomosis only subgroup for analysis.

Discussion
This study demonstrates that stapled anastomoses, specifically using an LSSM technique, are associated with lower anastomotic leak rates than HS anastomoses following oesophagectomy. The LSSM technique was associated with a lower rate of anastomotic stricture than CS, TS and HS anastomoses. This effect was consistent across the majority of subgroups in sensitivity analyses. LSSM anastomoses were associated with lower rates of 30-day mortality. Overall, the results indicate superiority of the LSSM technique for oesophagogastric anastomosis following oesophagectomy. Previous systematic reviews and meta-analyses 10-14 have examined the impact of stapled versus HS anastomoses following oesophagectomy (Table S5, supporting information). These, however, did not distinguish between CS and LSSM stapling techniques. Honda and colleagues 10 did not include LSSM anastomoses, looking only at the differences between HS and CS anastomoses. They reported no differences in anastomotic leak rates but an increased risk of anastomotic stricture with CS. Wang and co-workers 13 only compared HS with CS anastomoses. All anastomoses were performed in the neck, and no significant differences were demonstrated with regard to anastomotic leak, stricture or mortality. Markar et al. 14 compared HS with stapled oesophagogastric anastomoses, but did not separate the types of stapled anastomosis further. They did not observe significant differences in anastomotic leakage or 30-day mortality. Anastomotic stricture occurred more frequently with stapled than with HS anastomoses. Another systematic review 11 examined eight RCTs, all comparing HS with CS anastomoses. No meta-analysis was performed, and the authors concluded that there was insufficient evidence to recommend either technique. Liu and colleagues 12 grouped all stapled anastomoses together and compared them with HS anastomoses. Although a number of subgroup analyses were performed, the overall results demonstrated no significant differences in anastomotic leak rates or 30-day mortality between HS and stapled anastomoses.
Although the data from the present study are interesting, it is not known precisely why LSSM anastomoses may have a reduced anastomotic leak rate. Theories include: the wider anastomosis, hence also reducing the risk of anastomotic stricture; anastomosis performed near the greater curve arcade on the best perfused part of the stomach allows for improved healing; and the side-to-side orientation reduces traction-related tension 24,58 .
The present study has some limitations. The studies included in the review span a large time scale of over 28 years, and used slightly different definitions of anastomotic leakage. Future trials and studies in this area should adhere to the Esophagectomy Complications Consensus Group definitions of anastomotic leakage 68 , which classify leaks into three types in relation to severity and treatment needs. There is only limited published evidence on the use of the LSSM anastomoses, and this tends to be from more recently conducted studies. The included studies are heterogeneous in that they included different levels of anastomosis, suture material used, stapling device types and sizes, and approaches employed for oesophagectomy (open versus minimal access).
In the absence of large, high-quality, randomized trial data, this network meta-analysis provides the most up-to-date evidence base for comparing HS versus LSSM and CS techniques. Although triangular stapling showed promising results in the overall network, the encouraging results were not consistent in the sensitivity analysis. There is limited literature on the TS method, with only two papers describing outcomes for this technique; therefore it is difficult to give recommendations. The TS technique may be better viewed as a variation of the LSSM technique. The current multicentre international Oesophago-Gastric Anastomosis Audit 69 (https://www.ogaa.org.uk) is collecting data on outcomes after oesophagectomy, and includes intraoperative details regarding anastomotic techniques. Data from this large database will further inform surgical teams about the benefits of different anastomotic techniques.