Results of the search

We identified 3069 references through electronic searches of Cochrane Central Register of Controlled Trials (CENTRAL; (Wiley); N = 28), MEDLINE (OvidSP; N = 762), EMBASE (OvidSP; N = 1659), Science Citation Index expanded (N = 604), ClinicalTrials.gov (N = 9) and WHO Trials register (N = 7). After removing duplicate references, there were 1965 remaining. We excluded 1930 clearly irrelevant references through reading abstracts. We retrieved the full publication of 35 references for further detailed assessment. We excluded 24 studies (25 references) for the reasons listed in the Characteristics of excluded studies. Six non‐randomised studies (10 references) fulfilled the inclusion criteria (Characteristics of included studies). The reference flow is shown in Figure 1.

Figure 1

---

---

Study flow diagram.

Included studies

We included a total of six non‐randomised studies (Badessi 2003; Cash 2014; Ecker 2015; Maas 2012; Saha 2009; Valenti 2008). Five studies were retrospective studies (Badessi 2003; Cash 2014; Ecker 2015; Maas 2012; Saha 2009), while one of the studies was a prospective study (Valenti 2008). All studies were single institutional studies. Two studies compared laparoscopic oesophagectomy with historical controls who underwent open oesophagectomy (Cash 2014; Maas 2012). Three studies compared laparoscopic oesophagectomy with contemporary controls who underwent open oesophagectomy (Ecker 2015; Saha 2009; Valenti 2008). It was not clear whether one of the studies was a case‐control study or a cohort study (Badessi 2003).

One study included only patients with adenocarcinoma (Saha 2009). Three studies included a mixture of adenocarcinoma and squamous cell carcinoma but did not report the outcome data separately (Ecker 2015; Maas 2012; Valenti 2008). Two studies did not report the histological types of cancer (Badessi 2003; Cash 2014). One study included Stage I cancer only (Saha 2009). Three studies included Stages I to III cancer but did not report the outcome data separately for different stages (Cash 2014; Ecker 2015; Maas 2012). Two studies did not report the stages of cancer (Badessi 2003; Valenti 2008). Three studies indicated that the location of cancer was in the lower third (Maas 2012; Saha 2009; Valenti 2008). Information on the location of the tumours was not available in the remaining three studies. One study included patients with ASA I to III (Maas 2012). One study included patients with ASA I to IV; one patient in each group belonged to ASA IV category, while the remaining patients belonged to ASA I to III (Valenti 2008). Neither study reported outcome data separately for the different ASA stages. Information on ASA was not available in the remaining four studies. There was no restriction based on BMI in any of the studies. None of the studies reported the outcome data separately for healthy weight versus overweight or obese patients.

Five ports were used to perform laparoscopic oesophagectomy in all three studies that provided information on the number of ports (Maas 2012; Saha 2009; Valenti 2008). The remaining studies did not provide this information. A mini‐laparotomy of 7 cm was used to perform the anastomosis and retrieve the specimen in two studies that reported this information (Maas 2012; Valenti 2008). The remaining studies did not provide this information. None of the studies reported the size of incision in the open oesophagectomies. Drain use was not stated in either group, in any of the studies. The proportion of patients that were converted from laparoscopic to open oesophagectomy was 2/33 (6.1%), 4/36 (11.1%), 9/50 (18%), and 0/16 (0%) respectively, in the four studies that reported this information (Cash 2014; Ecker 2015; Maas 2012; Valenti 2008).

A total of 334 patients underwent laparoscopic (154 patients) or open (180 patients) transhiatal oesophagectomy. One study, which included eight patients, did not report any outcomes of interest for this review. Excluding this study, a total of 326 patients, undergoing laparoscopic (151 patients) or open (175 patients) transhiatal oesophagectomy, contributed to one or more outcomes in this review. The mean or median age in the studies ranged from 64 years to 74 years in the five studies that reported this information (Cash 2014; Ecker 2015; Maas 2012; Saha 2009; Valenti 2008). All the studies reported the proportion of females, which ranged from 20% to 37.5%.

The follow‐up period was not stated in three studies (Badessi 2003; Maas 2012; Valenti 2008). The median follow‐up period in the remaining studies were as follows.

• Cash 2014: 26 months for the laparoscopic oesophagectomy group and 64 months for the open oesophagectomy group (survival at 24 months was used to calculate proportion survived)

• Ecker 2015: 10 months

• Saha 2009: 44 months

The outcomes reported in the studies are summarised in Characteristics of included studies.

Excluded studies

Two studies were rejected because they included more than 10% of patients without cancer, but separate outcome data were not available for patients with cancer (Bernabe 2005; Perry 2009). Eight studies were excluded because the patients did not undergo laparoscopic transhiatal oesophagectomy (Blazeby 2011; Burdall 2015; Csendes 2013; Dolan 2013; Parameswaran 2013; Safranek 2010; Schoppmann 2010; Yamasaki 2011). One study was excluded because it was not clear whether patients underwent transhiatal oesophagectomy (Harrison 2013). One study was excluded because there was no control group of open oesophagectomy (Scheepers 2008). Six studies were excluded because separate data were not available for patients who underwent transhiatal oesophagectomy (Bresadola 2006; Fabian 2008; Kang 2013; Mamidanna 2012; Mao 2012; Messenger 2015). The remaining six studies were excluded because they were not primary research (e.g. review, editorial, letter to editor, comment, or cost‐effectiveness study with no primary research data; Cuesta 2012; Ferreira 2004; Lee 2013; Mariette 2012; Rice 2012; Ujiki 2013).

Bias due to confounding

There was no information for the risk of bias due to confounding for five studies (Badessi 2003; Ecker 2015; Maas 2012; Saha 2009; Valenti 2008). Although four studies reported that there were no baseline differences between the groups, the studies were not powered to measure the baseline differences, and did not assess the baseline difference for one or more confounding factors (Ecker 2015; Maas 2012; Saha 2009; Valenti 2008). In one study, the tumour size was smaller in patients undergoing laparoscopic surgery, although the proportion of patients who underwent neo‐adjuvant therapy was more in the laparoscopic surgery group (Cash 2014). This is likely to have introduced critical bias to the estimates of effect.

Bias due to the selection of participants

Two studies used historical controls (Cash 2014; Maas 2012). In these two studies, after a certain date, the authors only performed laparoscopic oesophagectomy, and compared the results of laparoscopic oesophagectomy with those of open oesophagectomy before this date. We considered these studies to have moderate risk of bias. In two studies, the decision to perform laparoscopic or open oesophagectomy was based on the surgeon's preference (Saha 2009; Valenti 2008). We classified these studies as providing 'no information'. The criteria used to perform oesophagectomy or open oesophagectomy was not stated in the remaining studies (Badessi 2003; Ecker 2015). We also classified these studies as providing 'no information'.

Bias due to departures from intended intervention

None of the studies reported whether the patient care other than laparoscopic or open procedure was identical in the two groups. We classified all studies as providing 'no information'.

Bias in the measurement of outcomes

Three studies clearly reported that the outcome assessors were not blinded (Maas 2012; Saha 2009; Valenti 2008). This might have introduced bias in the measurement of outcomes other than mortality. We considered these studies to have critical risk of bias. The information on outcome assessor blinding was not reported in the remaining studies (Badessi 2003; Cash 2014; Ecker 2015). We considered these studies as providing 'no information'.

Bias due to missing data

Four studies included all the patients who met the inclusion criteria; we considered them to be at low risk of bias due to missing data (Ecker 2015; Maas 2012; Saha 2009; Valenti 2008). It was not clear whether any patients were excluded from analysis in the remaining two studies (Badessi 2003; Cash 2014). We considered these studies as providing 'no information'.

Bias in selection of the reported findings

Only two studies reported mortality and morbidity adequately and could be considered at low risk of bias due to selective outcome reporting (Cash 2014; Maas 2012). We considered one study to be at critical risk of bias since neither mortality nor morbidity was reported (Badessi 2003). We considered the remaining studies to be at serious risk of bias as morbidity was not reported, since one would expect studies comparing laparoscopic with open oesophagectomy to report the data on mortality and morbidity in a detailed manner (Ecker 2015; Saha 2009; Valenti 2008).

Mortality

Five studies reported short‐term mortality (Cash 2014; Ecker 2015; Maas 2012; Saha 2009; Valenti 2008). There was no statistically significant difference in the short‐term mortality between the two groups (laparoscopic group: 0/151 (adjusted proportion based on meta‐analysis estimate: 0.5%) versus open group: 2/175 (1.1%); Risk Ratio (RR) 0.44; 95% CI 0.05 to 4.09; participants = 326; studies = 5; I² = 0%; Analysis 1.1). There was no change in results when we used a random‐effects model.

Two studies reported long‐term mortality between the two groups (Cash 2014; Maas 2012). There was no statistically significant difference in the long‐term mortality between the two groups (Hazard Ratio (HR) 0.97; 95% CI 0.81 to 1.16; participants = 193; studies = 2; I² = 0%; Analysis 1.2). The two‐year mortality was 30% and 35% in the laparoscopic and open groups respectively in Cash 2014, while the three‐year mortality was 64% and 62% in the laparoscopic and open groups respectively in Maas 2012. There was no change in results when we used a random‐effects model. In addition to the two studies included in the meta‐analysis, two other studies reported the mortality at the maximum follow‐up (Ecker 2015; Saha 2009). However, we did not include these studies in the meta‐analysis since there were no deaths after a median follow‐up of 10 months in the laparoscopic group (0/36) versus 6% dead in the open group (absolute numbers not available) in Ecker 2015, and no deaths after a median follow‐up of 44 months in the open group (0/4) versus 1/16 (6.3%) dead in laparoscopic group in Saha 2009.

Serious adverse events

Three studies reported the proportion of patients with serious adverse events (Cash 2014; Maas 2012; Saha 2009). The proportion of people with adverse events was statistically significantly lower in the laparoscopic group (10/99; adjusted proportion: 10.3%) compared to the open group (24/114 (21.1%); RR 0.49; 95% CI 0.24 to 0.99; participants = 213; studies = 3; I² = 0%; Analysis 1.3). There was no change in results when we used a random‐effects model.

Details of the serious adverse events were not available in Cash 2014. The serious adverse events in the other studies included complications which required re‐operations, such as re‐inspection of anastomosis, revision of anastomosis, and tracheal repair in Maas 2012, and anastomotic leaks in the Saha 2009.

Ecker 2015 reported anastomotic stenosis. There was no statistically significant difference in the proportion of people with anastomotic stenosis between the two groups (laparoscopic group: 4/36 (11.1%) versus open group: 3/37 (8.1%); RR 1.37; 95% CI 0.33 to 5.70; participants = 73; studies = 1; Analysis 1.4). Since there was only one study for this outcome, the issue of fixed‐effect versus random‐effects model did not arise.

None of the studies reported post‐operative dysphagia.

Health‐related quality of life

None of the studies reported health‐related quality of life at any time frame.

Recurrence

Saha 2009 reported short‐term recurrence within six months. There was no statistically significant difference in the proportion of people with short‐term recurrence between the two groups (laparoscopic group: 1/16 (6.3%) versus open group: 0/4 (0%); RR 0.88; 95% CI 0.04 to 18.47; participants = 20; studies = 1; Analysis 1.5). Since there was only one study for this outcome, the issue of fixed‐effect versus random‐effects model did not arise.

Two studies reported long‐term recurrence (Ecker 2015; Maas 2012). There was no statistically significant difference in the long‐term recurrence between the two groups (HR 1.00; 95% CI 0.84 to 1.18; participants = 173; studies = 2; Analysis 1.6). The 10‐month recurrence was 20% and 24% in the laparoscopic and open groups respectively in Ecker 2015, while the three‐year recurrence was 69% and 70% in the laparoscopic and open groups respectively, and the five‐year recurrence was 77% and 79% in the laparoscopic and open groups respectively in Maas 2012. There was no change in results when we used a random‐effects model. We excluded two other studies from the meta‐analysis (Cash 2014; Saha 2009). In Cash 2014, recurrence at the maximum follow‐up was 8/33 (24.2%) in the laparoscopic group and 15/60 (25%) in the open group. However, the patients in the laparoscopic group were followed for a median period of 26 months, while those in the open group were followed up for a median period of 64 months. So, it was inappropriate to compare the two proportions. In Saha 2009, there were no recurrences after a median follow‐up of 44 months in the open group (0/4) versus 1/16 (6.3%) recurrence in the laparoscopic group.

Adverse events

Three studies reported the proportion of patients with adverse events (Cash 2014; Maas 2012; Saha 2009). The proportion of people with adverse events was statistically significantly lower in the laparoscopic group (37/99 (39.9%) compared to the open group (71/114 (62.3%); RR 0.64; 95% CI 0.48 to 0.86; participants = 213; studies = 3; I² = 0%; Analysis 1.7). There was no change in results when we used a random‐effects model. One other study reported that the number of complications were fewer in the laparoscopic group compared to the open group, without providing information on the complications or statistical significance (Badessi 2003).

Perioperative blood transfusion requirements

One study reported the proportion of people who required perioperative transfusion (Ecker 2015). There was no statistically significant difference in the proportion of people who required perioperative transfusion between the two groups (laparoscopic group: 0/36 (0%) versus open group: 6/37 (16.2%); RR 0.08; 95% CI 0.00 to 1.35; participants = 73; studies = 1; Analysis 1.8). Since there was only one study for this outcome, the issue of fixed‐effect versus random‐effects model did not arise.

One study reported the quantity of blood transfused (Cash 2014). The median blood transfused was 0 in the laparoscopic group compared to 2.5 units in the open group (Analysis 1.9). The statistical significance was not clear since the P value presented in this study was for the comparison of three groups (only two of which were eligible for this review).

Measures of earlier postoperative recovery

Three studies reported the length of hospital stay (Cash 2014; Ecker 2015; Maas 2012). All three studies reported the median length of hospital stay so we did not perform a meta‐analysis. The median length of hospital stay was statistically significantly lower by three days in the laparoscopic group over the open group in all three studies (Analysis 1.10). One other study reported that the post‐operative hospital stay was shorter in the laparoscopic group compared to the open group without providing information on the length of hospital stay or statistical significance (Badessi 2003).

None of the studies reported time‐to‐return to normal activity or time‐to‐return to work.

Positive resection margins

Three studies reported the proportion of patients with positive resection margins (Ecker 2015; Maas 2012; Valenti 2008). There was no statistically significant difference in the proportion of people with positive resection margins between the two groups (laparoscopic group: 15/102 (adjusted proportion: 15.8%) versus open group: 27/111 (24.3%); RR 0.65; 95% CI 0.37 to 1.12; participants = 213; studies = 3; I² = 0%; Analysis 1.11). There was no change in results when we used the random‐effects model.

Number of lymph nodes harvested during surgery

Five studies reported the number of lymph nodes harvested during surgery (Cash 2014; Ecker 2015; Maas 2012; Saha 2009; Valenti 2008). Since four studies reported the median number of lymph nodes harvested during surgery, we did not perform a meta‐analysis (Cash 2014; Ecker 2015; Maas 2012; Saha 2009). Three studies reported that there was no statistically significant difference in the mean or median number of lymph nodes harvested during surgery (Ecker 2015; Maas 2012; Valenti 2008). One study did not report whether the median difference in lymph nodes harvested during surgery was statistically significant (Saha 2009). In the last study, the statistical significance was not clear, since the P value presented in this study was for the comparison of three groups (only two of which were included in this review; Cash 2014; Analysis 1.12).

Assessment of heterogeneity

There was no evidence of heterogeneity demonstrated by the I² statistic, the Chi² test for heterogeneity, or by visual inspection of forest plots to identify overlapping confidence intervals, for any of the outcomes for which we performed a meta‐analysis.

Assessment of reporting biases

We did not assess reporting biases using a funnel plot because we found fewer than 10 studies. There was some evidence of selective outcome reporting, as shown in the Characteristics of included studies.

Subgroup analysis

Sensitivity analysis

We did not perform any of the planned sensitivity analyses since none of the studies were at low risk of bias, standard deviation was not imputed for any of the outcomes, there were no cluster RCTs, and we either included studies with only two arms, or included the data from only two arms.