Minimally invasive techniques for transthoracic oesophagectomy for oesophageal cancer: systematic review and network meta‐analysis

Background Oesophagectomy is a demanding operation that can be performed by different approaches including open surgery or a combination of minimal access techniques. This systematic review and network meta‐analysis aimed to evaluate the clinical outcomes of open, minimally invasive and robotic oesophagectomy techniques for oesophageal cancer. Methods A systematic literature search was conducted for studies reporting open oesophagectomy, laparoscopically assisted oesophagectomy (LAO), thoracoscopically assisted oesophagectomy (TAO), totally minimally invasive oesophagectomy (MIO) or robotic MIO (RAMIO) for oesophagectomy. A network meta‐analysis of intraoperative (operating time, blood loss), postoperative (overall complications, anastomotic leaks, chyle leak, duration of hospital stay) and oncological (R0 resection, lymphadenectomy) outcomes, and survival was performed. Results Ninety‐eight studies involving 32 315 patients were included in the network meta‐analysis (open 17 824, 55·2 per cent; LAO 1576, 4·9 per cent; TAO 2421 7·5 per cent; MIO 9558, 29·6 per cent; RAMIO 917, 2·8 per cent). Compared with open oesophagectomy, both MIO and RAMIO were associated with less blood loss, significantly lower rates of pulmonary complications, shorter duration of stay and higher lymph node yield. There were no significant differences between surgical techniques in surgical‐site infections, chyle leak, and 30‐ and 90‐day mortality. MIO and RAMIO had better 1‐ and 5‐year survival rates respectively compared with open surgery. Conclusion Minimally invasive and robotic techniques for oesophagectomy are associated with reduced perioperative morbidity and duration of hospital stay, with no compromise of oncological outcomes but no improvement in perioperative mortality.


Introduction
Oesophageal cancer remains a challenging disease worldwide, with over 570 000 new cases in 2018 1 . In managing this disease, oesophagectomy remains the mainstay of radical treatment with curative intent, with the transthoracic approach the most commonly employed. However, variation exists in surgical access techniques, with approximately 40 per cent of oesophagectomies in the UK now employing minimally invasive approaches 2 . The most common procedure is hybrid oesophagectomy where a laparoscopic gastric mobilization is performed with an open thoracotomy; a thoracoscopic-open abdominal hybrid procedure is uncommon. Less commonly both thoracoscopic and laparoscopic techniques are used in totally minimally invasive oesophagectomy (MIO).
The use of robotic surgery for oesophagectomy is also increasing.
Given the limited evidence and understanding of the potential benefits of different minimally invasive techniques for oesophagectomy, this systematic review and network meta-analysis aimed to compare oncological safety and perioperative outcomes between these different surgical approaches and transthoracic oesophagectomy for cancer, along with impact on long-term survival.

Search strategy
This study was conducted according to PRISMA guidelines 19 . A systematic and comprehensive search was undertaken of the MEDLINE, Embase and Cochrane Library databases, for studies published up to 25 February 2019. Search terms included the following, individually or in combination: 'oesophagectomy' or 'oesophagectomy' and 'minimally invasive surgical procedures' or 'laparoscopy' and 'anastomotic leak' or 'postoperative complications' or 'lymph nodes examined' or 'survival' and 'oesophageal cancer' or 'esophageal cancer'. The full search strategy with all included search terms is shown in Table S1 (supporting information). Manual scoping of reference lists in recent reviews was also undertaken. The protocol for this study was registered with the prospective PROSPERO database (CRD42019125848).

Inclusion and exclusion criteria
Inclusion criteria were: comparative studies comparing any approach to two-or three-stage transthoracic oesophagectomy in human subjects with cancer of the oesophagus or gastro-oesophageal junction, and studies published in the English language. Exclusion criteria were: review articles; conference abstracts; studies with non-comparative analyses of surgical approach including case reports; studies reporting transhiatal or left thoracoabdominal approaches; studies using a non-gastric replacement conduit; and studies reporting pharyngolaryngoesophagectomy. After performing the literature search and removing all duplicates, two researchers screened the remaining titles and abstracts independently. Where a study was considered for inclusion, the full text was obtained. Discrepancies between the judgement of the two primary researchers were resolved through consensus with the other authors. Additionally, during full-text review, authors of papers with mixed groups of both hybrid and totally minimally invasive techniques were contacted for separate data regarding each technique. Where multiple studies analysed the same data set or population, the most recent article was selected unless different outcomes were reported.

Study outcomes
Outcome measures were: oncological -lymph node yield, R0 resection margins; intraoperative -blood loss and duration of operation; postoperative -duration of hospital stay, 30-and 90-day mortality, overall, pulmonary, gastrointestinal and cardiac complications, anastomotic leak and chyle leak, and 1-, 3-and 5-year overall survival. The Esophageal Complications Consensus Group definitions of complications were used 20 . R0 status was defined using both College of American Pathologists 21 and Royal College of Pathology 22 definitions: absence of residual tumour at or within 1 mm of the resection margin respectively.

Data extraction
Two researchers extracted data on study characteristics (author, year of publication, country, study interval, number of participants), patient characteristics (age, sex, BMI, overall TNM stage, location of anastomosis (cervical, thoracic), anastomotic technique (stapled versus handsewn), details of surgical approach and reported clinical outcomes.

Definitions
Open oesophagectomy was defined as oesophagectomy carried out with laparotomy and open thoracotomy 23,24 . MIO was defined as total MIO where laparoscopy was used for the abdominal phase and thoracoscopy for the thoracic phase. Laparoscopically assisted hybrid oesophagectomy (LAO) was defined as a laparoscopic abdominal phase combined with open thoracotomy. Thoracoscopically assisted hybrid oesophagectomy (TAO) was defined by an open abdominal phase combined with a thoracoscopic chest phase. Robotic MIO (RAMIO) was defined as oesophagectomy where either the abdominal or thoracic phase was performed using a robotic platform, including hybrid approaches 25,26 . Regardless of access approach, two-and three-stage oesophagectomies, with intrathoracic and cervical anastomoses respectively, were included, and a subgroup analysis was planned based on location of the anastomosis.

Assessment of study quality
Methodological quality and standard of outcome reporting was assessed in each study by two independent researchers. Disagreements were settled through discussion between these researchers or consensus with all authors. For cohort studies, the Newcastle-Ottawa Scale 27,28 was used to formally assess quality, whereas the Cochrane risk-of-bias tool 29 was used for RCTs.

Statistical analysis
This systematic review and meta-analysis was conducted in accordance with the recommendations of the Cochrane Library and PRISMA guidelines, as reported previously 30 .
Dichotomous outcomes were compared using risk ratios (RRs), produced by meta-analysis using random-effects DerSimonian-Laird models. Heterogeneity between studies was assessed using the I 2 value, with values of less than 25, 25-75 and over 75 per cent considered to represent low, moderate and high degrees of heterogeneity respectively. Both randomized and non-randomized studies were pooled into a network meta-analysis comparing the above surgical approaches with transthoracic oesophagectomy. 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 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 *For RCTs, the risk of bias was determined as low, high or of some concern. U, upper; M, middle; L, lower; NOS, Newcastle-Ottawa Scale; LAO, laparoscopically assisted oesophagectomy; n.r., not reported; MIO, minimally invasive oesophagectomy; TAO, thoracoscopically assisted oesophagectomy; RAMIO, robotic minimally invasive oesophagectomy; PCS, prospective cohort study; RCS, retrospective cohort study. 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 31 . If P exceeded 0⋅050, representing acceptance of the null hypothesis, consistency was assumed and networks were entered into consistency modelling. Consistency models used a restricted maximum likelihood model, generating network forest plots. Heterogeneity was examined by calculation of τ 2 . These were supplemented with interval plots of pooled effect estimates. Surgical approaches were then ranked using P-scores, whereby a P-score greater than 0⋅900 was considered to indicate the best technique with high probability. Subgroup analyses were conducted according to location of anastomoses, either cervical or thoracic, and for a more recent time cohort (2010 onwards). Statistical analyses for network meta-analysis were undertaken using R version 3.2.1 (R Foundation for Statistical Computing, Vienna, Austria), with the netmeta packages, as described previously 32,33 .

Study characteristics
The review identified 98 studies 6,7,34-129 comparing surgical approaches for oesophagectomy, involving 32 315 patients (Fig. 1). Of these, 55⋅2 per cent (17 824 Tables S2-S5 (supporting  information). Table 2 shows the results of pairwise comparisons between intraoperative outcomes, and network maps are presented in Fig. S1 (supporting information). Duration of operation was reported in 77 studies. Open surgery resulted in significantly shorter operating times than MIO (mean Values in parentheses are 95 per cent confidence intervals. TAO, thoracoscopically assisted oesophagectomy; LAO, laparoscopically assisted oesophagectomy; MIO, minimally invasive oesophagectomy; RAMIO, robotic minimally invasive oesophagectomy. Table 3 Ranking of surgical techniques for intraoperative, oncological and postoperative outcomes according to P-scores  Table 2). Open surgery had the shortest operating time, with a high probability, followed by hybrid operations then MIO and RAMIO ( Table 3). Open oesophagectomy was ranked first for cervical anastomosis, whereas LAO was ranked first for thoracic anastomosis (Tables S2 and S3, supporting information). Blood loss was reported in 65 studies. Open oesophagectomy had significantly higher blood loss than TAO (MD 91 ml; P < 0⋅001), LAO (MD 84 ml; P = 0⋅016), RAMIO (MD 163 ml; P < 0⋅001) and MIO (MD 173 ml; P = 0⋅001). MIO was ranked first for lowest blood loss, with a high probability, followed by RAMIO ( Table 3). MIO was ranked first for both cervical and thoracic anastomosis, followed by RAMIO (Tables S2 and S3, supporting information).

Postoperative outcomes
The results of all pairwise comparisons of each surgical approach for postoperative complications are shown in Tables 4 and 5, and network maps in Fig. S2 (supporting information). There were no significant differences between surgical approaches for surgical-site infections, chyle leak and 30-or 90-day mortality.

Pulmonary complications
Pulmonary complications were reported in 79 studies. MIO (RR 0⋅52; P < 0⋅001) and TAO (RR 0⋅60; P = 0⋅004) were associated with significantly lower rates of pulmonary complications than open surgery. MIO was ranked the best technique in terms of pulmonary complications overall ( Table 3), and in subgroups of cervical and thoracic anastomoses (Tables S3 and S4, supporting information).

Anastomotic leaks
Anastomotic leak was reported in 74 studies. LAO was significantly associated with a higher rate of anastomotic leak than TAO (RR 1⋅69; P = 0⋅048) and MIO (RR 1⋅63; P = 0⋅030) ( Table 4). TAO was ranked first for anastomotic leak ( Table 3). In terms of anastomotic leakage, TAO was ranked first for thoracic anastomosis, whereas RAMIO was ranked first for cervical anastomosis (Tables S3 and S4, supporting information).

Overall survival
One-year overall survival was reported in 53 studies. The open approach was associated with significantly lower 1-year survival than TAO (RR 1⋅62; P = 0⋅043) and MIO (RR 1⋅35; P = 0⋅035) ( Table 5). Overall, TAO was ranked first for 1-year survival ( Table 3). However, MIO and LAO were ranked first for cervical and thoracic anastomosis respectively (Tables S3 and S4, supporting information). Three-year overall survival was reported in 46 studies. There were no significant differences in outcomes between any techniques. Five-year overall survival was reported in 34 studies. Open oesophagectomy was associated with significantly lower 5-year survival than RAMIO (RR 4⋅00; P = 0⋅042) ( Table 5). Overall, RAMIO was ranked the best technique, with high probability ( Table 3). A sensitivity analysis for 1-and 5-year survival including studies from 2010 onwards yielded similar results. Values in parentheses are 95 per cent confidence intervals. TAO, thoracoscopically assisted oesophagectomy; LAO, laparoscopically assisted oesophagectomy; MIO, minimally invasive oesophagectomy; RAMIO, robotic minimally invasive oesophagectomy.

Lymph nodes examined
The results of all pairwise comparisons of oncological outcomes for each surgical approach technique are shown in Table 6, and network maps in Fig. S3 (supporting information). Lymph node assessment was reported in 77 studies. LAO (mean difference 3⋅53; P = 0⋅031) and open surgery (mean difference 3⋅11; P = 0⋅024) were associated with significantly lower numbers of lymph nodes examined than RAMIO. RAMIO was ranked as the best technique, with high probability, followed by MIO ( Table 3).

Discussion
This network meta-analysis compared all combinations of open, minimally invasive and robotic approaches to transthoracic oesophagectomy. The analysis demonstrated that minimally invasive surgery for oesophagectomy was associated with increased operating time, but decreased operative blood loss, fewer pulmonary complications and shorter length of hospital stay, compared with open approaches. In addition, the review identified significantly decreased overall postoperative complications with minimally invasive surgery compared with the open approach. Importantly, no significant differences in perioperative mortality (either 30 or 90 day) were observed between any surgical approach. In addition, MIO and RAMIO were associated with significantly higher 1-and 5-year survival rates respectively than open oesophagectomy. These findings were not altered in a sensitivity analysis including studies from 2010 onwards. Based on the present evidence, no one approach demonstrates clear overall superiority over all others, but there is increasing evidence of the specific benefits related to minimally invasive techniques. Network meta-analysis allows assessment of different surgical techniques by combining direct evidence within studies and indirect evidence across studies. Hence, it enables indirect comparisons of surgical techniques that have not been studied directly in a head-to-head fashion 130 . By including evidence from both direct and indirect comparisons, a network meta-analysis may increase the precision in estimates of the relative effects of treatments and improve power compared with standard pairwise meta-analyses that include only direct evidence 131 . Network meta-analysis may yield more reliable and definitive results, and allows visualization and interpretation of a wider picture of the available evidence, and to calculate treatment rankings with probabilities, compared with a standard pairwise meta-analysis 130 .
This study has some limitations. The majority of the studies included in this network meta-analysis subject it to heterogeneity owing to patient selection criteria and demographics, such as age, sex, BMI and different disease stages. The amount of evidence a treatment carries and the number of comparisons available between treatments determines the diversity and strength of a network meta-analysis. Imbalance in terms of the amount of evidence available may affect the power and reliability of the network meta-analysis as inferences may be driven largely from the evidence from few treatments and comparisons 132 . Some of the studies assessed new techniques or technologies and may have incorporated a learning curve in the novel arm.
Previous standard pairwise meta-analyses [8][9][10][11][12][13][14][15][16][17][18] and RCTs 6,7,34,35 comparing open versus minimally invasive resection for oesophagectomy demonstrated that, although laparoscopic surgery increased operative time, it resulted in significantly reduced blood loss and wound infection, increased R0 resection rate and shorter hospital stay. In addition, the present review identified significantly decreased overall postoperative complications with minimal access compared with open surgery, and this may be related to the lower wound infection rate and pulmonary complications of the minimally invasive approach.
This network meta-analysis identified that minimally invasive surgery was associated with significantly more examined lymph nodes compared with open surgery, specifically with RAMIO and MIO techniques. Evidence from RCTs 6,7 is limited as none have demonstrated the superiority of either laparoscopic or open techniques. This network meta-analysis also showed that rates of R0 resection were better with MIO compared with open surgery. This is an important point as one of the barriers to adoption of the minimally invasive approach in routine clinical practice over conventional open oesophagectomy was concern over oncological clearance as R0 resections are recognized to be an important prognostic marker of long-term survival following surgery 133,134 . It is also important to note that differences in R0 resection rates may also be attributed to differences in the R0 classification systems used.
Both RAMIO and MIO techniques were associated with significantly lower rates of pulmonary complications and shorter length of hospital stay compared with conventional open oesophagectomy. However, there were no significant differences in outcomes between robotic and conventional MIO techniques. No significant differences between MIO and open techniques in rates of wound or diaphragm complications, gastrointestinal complications and chyle leak were identified. Operative blood loss is difficult to measure accurately, and the clinical relevance of the small differences in operative blood loss between the surgical techniques is debatable. However, previous studies [135][136][137] have suggested that volume of blood loss is an independent risk factor for postoperative adverse events, cancer recurrence and poorer overall survival. Furthermore, the potential advantages of the MIO approach, and especially the robotic approach, in decreasing operative trauma and blood loss, and improving postoperative recovery, may allow greater preservation of immune function, reduce the risk of tumour progression and allow earlier access to adjuvant treatment [138][139][140][141][142][143] .
A recent meta-analysis 18 reported that minimally invasive approaches for oesophagectomy significantly improved long-term survival of patients compared with conventional open surgery. However, that review did not address the impact of the different techniques on long-term outcomes given the heterogeneity of each approach as identified by the present review. In this network meta-analysis, TAO and MIO were only associated with a significant survival benefit compared with open surgery at 1 year, and not 3-or 5-year survival. This may reflect higher rates of negative resection margins and number of lymph nodes examined with MIO and RAMIO, as identified by this review.
Based on current evidence, no single approach demonstrates clear overall superiority over all others, but there is increasing evidence of the clinical benefits of minimally invasive over open surgery.