Results of the search

We identified 18,369 references through electronic searches of the Cochrane Central Register of Controled trials (CENTRAL), MEDLINE, EMBASE, Science Citation Index, ClinicalTrials.gov and WHO ICTRP (World Health Organization International Clinical Trials Registry Platform). After completion of manually removing duplicate references there were 11,953 references. We excluded 11,914 clearly irrelevant references through reading the abstracts. We sought 42 references in full text for further assessment. We did not identify any additional references to trials by searching the trial registry. We excluded 13 references (11 studies or comments) because of the reasons mentioned in the Characteristics of excluded studies tables and Excluded studies. We excluded five references (three trials) which were protocols of ongoing trials with no interim results available. Thirteen trials (24 references) met the inclusion criteria and were included in this review (Aoyama 2014; Cai 2011; Chen Hu 2012; Deng 2009; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013). The reference flow diagram is shown in Figure 3.

Figure 3

---

---

Study flow diagram.

Included studies

The thirteen trials compared laparoscopic with open gastrectomy (Aoyama 2014; Cai 2011; Chen Hu 2012; Deng 2009; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013). Twelve of the trials were two‐armed RCTs (Aoyama 2014; Cai 2011; Deng 2009; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013). The thirteenth was a four‐armed trial (Chen Hu 2012). Two of the arms involved laparoscopic surgery (fast‐track laparoscopic gastrectomy versus standard procedure laparoscopic gastrectomy) and two arms involved open gastrectomy (fast‐track laparoscopic gastrectomy versus standard procedure laparoscopic gastrectomy). The exact tumour stages included for each trial are reported in the Characteristics of included studies tables. Broadly, five trials included patients with early stage gastric cancer (Hayashi 2005; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013), three trials included patients with only advanced gastric cancer (Cai 2011; Kim 2013; Hu 2015), four trials included patients with early or advanced gastric cancer, a wide range of cancer staging (Aoyama 2014; Chen Hu 2012; Huscher 2005; Kim 2015), and one trial did not specify the cancer staging of included patients (Deng 2009).

Four of the trials included patients with an ASA risk score of III (Cai 2011; Hu 2015; Huscher 2005; Kim 2015), one trial did not include patients with ASA risk score III (Sakuramoto 2013), and the remaining eight trials did not specify their inclusion or exclusion (Aoyama 2014; Chen Hu 2012; Deng 2009; Hayashi 2005; Kim 2013; Kitano 2002; Lee 2005; Takiguchi 2013). None of the 13 trials specifically stated the inclusion or exclusion of patients with a BMI greater than 30.

Ten of the studies used laparoscopy‐assisted gastrectomy (Aoyama 2014; Cai 2011; Chen Hu 2012; Hayashi 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013), with three potentially using the totally laparoscopic method (Deng 2009; Hu 2015; Huscher 2005). There was no mention of an incision to remove the specimen in these trials, and the procedure was termed laparoscopic gastrectomy in these trials. However, it should be noted that some trials which were really laparoscopic‐assisted gastrectomy (based on the description of the procedure) reported the procedure as laparoscopic gastrectomy. So, it is not clear whether any of the trials used totally laparoscopic gastrectomy.

Twelve of the trials involved patients in whom subtotal (distal) gastrectomy was performed (Aoyama 2014; Chen Hu 2012; Deng 2009; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013); in Cai 2011, proximal, distal or total gastrectomy was performed. D1 nodal dissection was performed in both groups in one trial (Hayashi 2005). D2 nodal dissection was performed in both groups in four trials (Cai 2011; Deng 2009; Hu 2015; Kim 2013). D1 or more nodal dissections were performed in both groups in three trials (Aoyama 2014; Huscher 2005; Kim 2015). In two trials, selected groups of lymph nodes were dissected in both groups (Sakuramoto 2013; Takiguchi 2013). In one trial, selected nodes were dissected in the laparoscopic group, while D2 nodal dissection was performed in the open group (Lee 2005). Information on nodal dissection was not available in two trials (Chen Hu 2012; Kitano 2002). Seven of the trials used the Billroth I method of anastomosis alone (Aoyama 2014; Deng 2009; Hayashi 2005; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013), four of the trials used either the Billroth I or II anastomosis, Roux en Y anastomosis, oesophagogastrostomy or oesophageal jejunostomy (Cai 2011; Chen Hu 2012; Huscher 2005; Kim 2015), and two did not state the method of anastomosis used (Hu 2015; Kim 2013). Five trials used staples as the anastomotic method (Aoyama 2014; Hayashi 2005; Lee 2005; Sakuramoto 2013; Takiguchi 2013). In the remaining trials, either a combination of stapler and hand‐sewn anastomosis were used (Huscher 2005), or the information on stapler versus hand‐sewn anastomosis was not available (Cai 2011; Chen Hu 2012; Deng 2009; Hu 2015; Kim 2013; Kim 2015; Kitano 2002). Drains were routinely used in both groups in one trial (Sakuramoto 2013), and no routine drains were used in either group in one trial (Aoyama 2014). In one 2 x 2 factorial trial in which the safety and effectiveness of laparoscopic versus open gastrectomy and fast‐track surgery versus conventional surgery, drains were used in participants who underwent fast‐track surgery in both the laparoscopic and open groups (Chen Hu 2012). Two trials reported drains being used in the laparoscopic gastrectomy group, but did not report whether drains were used routinely in the open gastrectomy group (Lee 2005; Takiguchi 2013). The information on drain use was not available in the remaining trials (Cai 2011; Deng 2009; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002).

The follow‐up period was not available for one trial (Hu 2015). The follow‐up period in the remaining trials were as follows.

• Until discharge (Deng 2009; Sakuramoto 2013)

• 30 days (Aoyama 2014; Chen Hu 2012; Kim 2013; Kim 2015).

• 14 months (Lee 2005); 22 months (Cai 2011); 26 months (Kitano 2002).

• 42 months (Hayashi 2005); 52 months (Huscher 2005); 60 months (Takiguchi 2013).

In total, 2794 participants were randomised in the 13 trials included in this review. One trial which included 53 participants did not contribute any data for this review, because none of the outcomes included in the review were reported (Deng 2009). Two hundred and thirteen participants were excluded in the remaining 12 trials that contributed data, leaving a total of 2528 participants for whom data were available. Of these 2528 participants, 1288 were randomised to laparoscopic gastrectomy and 1240 to open gastrectomy (Aoyama 2014; Cai 2011; Chen Hu 2012; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013).

Excluded studies

We excluded a total of 13 references (11 studies or comments). We excluded three references because they were reports of a 'quasi‐randomised' control trial in which block allocation equivalent to alternate assignment was used (Kim 2008). We excluded two references because they were comments on Kim 2008 (Kim 2009; Liakakos 2009), and we excluded one reference because it was an editorial (Kanellos 2009). We excluded seven references because they were not RCTs (Han 2014; Kawamura 2008; Lee 2008; Lee 2009; Li 2014; Lin 2014; Sakuramoto 2009).

We excluded five other references because they were protocols for three trials which have not yet reported the results (Haverkamp 2015; Straatman 2015; Yoshikawa 2012). A summary of these trials is reported in Characteristics of ongoing studies.

Allocation

Four trials were free from selection bias (Aoyama 2014; Kim 2013; Kim 2015; Sakuramoto 2013). These trials had a low risk of bias in random sequence generation and allocation concealment. The remaining trials had unclear risk of bias in at least one of the aspects of random sequence generation or allocation concealment.

Blinding

Seven of the trials were at unclear risk of performance bias because of a lack of reported information (Aoyama 2014; Cai 2011; Deng 2009; Huscher 2005; Kim 2015; Kitano 2002; Lee 2005), with the remaining six being at high risk of performance bias, with patients and healthcare providers not being blinded (Chen Hu 2012; Hayashi 2005; Hu 2015; Kim 2013; Sakuramoto 2013; Takiguchi 2013).

Ten of the trials were at unclear risk of detection bias because of a lack of reported information (Aoyama 2014; Cai 2011; Chen Hu 2012; Deng 2009; Hayashi 2005; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005), with the remaining three being at high risk of detection bias, with outcome assessors not being blinded (Hu 2015; Sakuramoto 2013; Takiguchi 2013).

Incomplete outcome data

We classified three trials at low risk of attrition bias as they described no post‐randomisation drop‐outs (Hayashi 2005; Hu 2015; Takiguchi 2013). Four trials were at unclear risk of attrition bias as the reports did not describe the participant flow clearly (Aoyama 2014; Deng 2009; Kitano 2002; Lee 2005). Six trials were at high risk of attrition bias as they had post‐randomisation drop‐outs which were likely to affect the effect estimates (Cai 2011; Chen Hu 2012; Huscher 2005; Kim 2013; Kim 2015; Sakuramoto 2013).

Selective reporting

We classified six of the trials at low risk of reporting bias, with both postoperative mortality and morbidity reported (Aoyama 2014; Cai 2011; Chen Hu 2012; Hayashi 2005; Huscher 2005; Sakuramoto 2013). We classified seven of the trials at high risk of reporting bias, as one or both of these were not reported (Deng 2009; Hu 2015; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Takiguchi 2013).

Other potential sources of bias

In one trial, a more extensive procedure (subtotal gastrectomy) was performed in open gastrectomy group compared to laparoscopic group (distal gastrectomy) (Lee 2005). This could potentially favour the laparoscopic group in terms of decreased complications, but favour the open group in terms of decreased long‐term recurrence and mortality. We did not detect any other sources of bias in the remaining trials.

Short‐term mortality

Eleven trials reported short‐term mortality which is defined as mortality in hospital or within thirty days of treatment (Aoyama 2014; Cai 2011; Chen Hu 2012; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013). We pooled the trials using a fixed‐effect model. There was no significant difference in the proportion of participants who died within 30 days of treatment between laparoscopic gastrectomy (7/1188: adjusted proportion (based on meta‐analysis) = 0.6%) and open gastrectomy (4/1447: 0.3%) (RR 1.60, 95% CI 0.50 to 5.10; participants = 2335; studies = 11; I² = 0%; low quality evidence) (Analysis 1.1; summary of findings Table for the main comparison). There was no change to the conclusions when we used a random‐effects model or when we calculated the risk difference (RD 0.00, 95% CI ‐0.01 to 0.01; participants = 2335; studies = 11; I² = 0%).

Long‐term mortality

Three trials reported long‐term mortality (Cai 2011; Huscher 2005; Takiguchi 2013). In one of these trials, all the participants were alive after a follow‐up period of 60 months (Takiguchi 2013). We pooled the trials using a fixed‐effect model. There was no significant difference in long‐term mortality between the groups (HR 0.94, 95% CI 0.70 to 1.25; participants = 195; studies = 3; I² = 0%; very low quality evidence) (Analysis 1.2; summary of findings Table for the main comparison). There was no change to the conclusions when we used a random‐effects model. Approximately 55% to 60% of participants were alive after about 52 months (Huscher 2005).

Serious adverse events within three months

Eight trials reported serious adverse events within three months of treatment (Aoyama 2014; Cai 2011; Chen Hu 2012; Hayashi 2005; Huscher 2005; Kitano 2002; Lee 2005; Sakuramoto 2013). The serious adverse events in the trials included anastomotic leakage, anastomotic stenosis requiring balloon dilatation, pancreatic fistula, pancreatic injury, small bowel volvulus requiring adhesiolysis, bleeding requiring reoperation, abdominal abscess, myocardial infarction, acute respiratory distress syndrome, pleural effusion requiring puncture, and pneumonia. The type of serious adverse events were similar in nature between the groups. We pooled the trials using a fixed‐effect model. There was no significant difference in the proportion of participants who suffered a serious adverse event between laparoscopic gastrectomy (7/216: adjusted proportion = 3.6%) and open gastrectomy (13/216: 6%) within three months of treatment (RR 0.60, 95% CI 0.27 to 1.34; participants = 432; studies = 8; I² = 0%; very low quality evidence) (Analysis 1.3; summary of findings Table for the main comparison). There was no change to the conclusions when we used a random‐effects model or when we calculated risk difference.

Health‐related quality of life

Short‐ and medium‐term health‐related quality of life were not reported in any of the trials.

Short‐term recurrence

Three trials reported short‐term recurrence, which is defined as local recurrence, surgical wound recurrence or distal metastases within six months (Hayashi 2005; Kitano 2002; Lee 2005). No events were reported for either laparoscopic (52 participants) or open gastrectomy (51 participants). Therefore, we could not calculate an effect estimate (participants = 103; studies = 3) (Analysis 1.4).

Long‐term recurrence

Four trials reported long‐term recurrence (> six months) (Hayashi 2005; Huscher 2005; Kitano 2002; Lee 2005). Three of these three trials did not report any recurrence in either group (Hayashi 2005; Kitano 2002; Lee 2005). There was no significant difference in the hazard ratio for recurrence more than six months after treatment between the two groups (HR 0.95, 95% CI 0.70 to 1.30; participants = 160; studies = 4; very low quality evidence) (Analysis 1.5; summary of findings Table 2). Since only one trial contributed to the analysis, the issue of fixed‐effect versus random‐effects meta‐analysis and assessment of heterogeneity did not arise.

Adverse events within three months

Eleven trials reported adverse events within three months of treatment (Aoyama 2014; Cai 2011; Chen Hu 2012; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013). We pooled the trials using a random‐effects model. There were significantly fewer adverse events following laparoscopic gastrectomy (204/268: adjusted proportion = 16.1%) versus open gastrectomy (253/1222: 20.7%) (RR 0.78, 95% CI 0.60 to 1.01; participants = 2490; studies = 11; I² = 38%; very low quality evidence) (Analysis 1.6: summary of findings Table 2). However, on using the fixed‐effect model, there was no statistically significant difference between the groups (RR 0.78, 95% CI 0.66 to 0.92; participants = 2490; studies = 11; I² = 38%). Two large trials had narrow confidence intervals and had results in opposite directions, and this may account for the differences between the fixed‐effect and random‐effects models. In addition, we performed a sensitivity analysis excluding Lee 2005, in which the laparoscopic group underwent a less invasive procedure than the open group. There was no change in the results by excluding this trial.

Blood transfusion during or within a week of surgery

Two trials reported the proportion of patients requiring blood transfusion during or within a week of surgery (Aoyama 2014; Takiguchi 2013). None of the participants in either group in either of the trials required a blood transfusion. Therefore, we could not estimate an effect estimate (participants = 66; studies = 2) (Analysis 1.7). Since both trials reported the mean and standard deviation, we did not perform any sensitivity analysis excluding studies in which standard deviation was imputed.

Quantity of perioperative blood transfused

Two trials reported quantity of perioperative blood transfused (Cai 2011; Lee 2005): Lee 2005 reported the number of units of blood transfused; and Cai 2011 reported the amount of blood transfused in SI units. We pooled the trials using a fixed‐effect model. There was no significant difference in the amount of blood transfused between the laparoscopic and open gastrectomy groups (SMD 0.05, 95% CI ‐0.27 to 0.38; participants = 143; studies = 2; I² = 0%; very low quality evidence) (Analysis 1.8; summary of findings Table 2). There was no change to the conclusions when we used a random‐effects model.

Length of hospital stay

Eight trials reported length of hospital stay (Cai 2011; Chen Hu 2012; Hayashi 2005; Huscher 2005; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013). The length of hospital stay was significantly shorter in the laparoscopic group than the open group (MD ‐1.38 days, 95% CI ‐2.57 to ‐0.19; participants = 444; studies = 8; I² = 76%; very low quality evidence) using the random‐effects model (Analysis 1.9; summary of findings Table 2). There was substantial heterogeneity as noted by I² of 76% and Chi² test for heterogeneity P value of 0.0001. Using the fixed‐effect model did not alter the results (MD ‐0.86 days, 95% CI ‐1.27 to ‐0.44; participants = 444; studies = 8; I² = 76%). In two trials, median hospital stay rather than mean hospital stay was reported (Chen Hu 2012; Takiguchi 2013). In Takiguchi 2013, the standard deviation was calculated from the P value, while in Chen Hu 2012, the standard deviation was imputed as the highest standard deviation from the remaining trials. Excluding these two trials, there was no statistically significant difference between the two groups (MD ‐1.82 days, 95% CI ‐3.72 to 0.07; participants = 319; studies = 6; I² = 83%) using the random‐effects model, although there was still a statistically significant difference between the groups using the fixed‐effect model (MD ‐0.68 days, 95% CI ‐1.31 to ‐0.06; participants = 319; studies = 6; I² = 83%). The exclusion of Lee 2005, in which the laparoscopic group underwent a less invasive procedure did not alter the results.

Time to return to normal activity

This outcome was not reported in any of the trials.

Time to return to work

This outcome was not reported in any of the trials.

Positive resection margins at histopathological examination

One trial reported the number of patients with positive resection margins at histopathological examination (Kitano 2002). No events were reported for either laparoscopic or open gastrectomy. Therefore, we could not calculate an effect estimate (participants = 28; study = 1) (Analysis 1.10).

Number of lymph nodes harvested

Nine trials reported the number of lymph nodes harvested (Aoyama 2014; Cai 2011; Chen Hu 2012; Hayashi 2005; Huscher 2005; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013). We analysed the trials using a fixed‐effect model. There was no significant difference in the number of lymph nodes harvested between the two groups (MD ‐0.63, 95% CI ‐1.51 to 0.25; participants = 472; studies = 9; I² = 40%; very low quality evidence) (Analysis 1.11; summary of findings Table 2). There was no change to the conclusions when we used a random‐effects model. Since higher numbers of lymph nodes harvested is a better surgical marker, we switched the direction of the X‐axis in Analysis 1.11, with trials to the left of the equivalence line favouring open gastrectomy. The mean or standard deviation or both were calculated from other information such as median, standard error, or P value or imputed from the maximum standard deviation in the included studies in four trials (Aoyama 2014; Chen Hu 2012; Huscher 2005; Takiguchi 2013). Excluding these trials did not alter the conclusions (MD ‐0.62, 95% CI ‐1.55 to 0.31; participants = 262; studies = 5; I² = 70%).

Subgroup analysis

Of the planned subgroup analyses for the primary outcomes, only two were possible.

1. Different cancer stages (early gastric cancer and advanced gastric cancer). There was no short‐term mortality or long‐term mortality in laparoscopic or open gastrectomy groups in the early gastric cancer subgroup. So, we could not perform tests for subgroup differences (Analysis 2.1; Analysis 2.2). The test for subgroup differences was not statistically significant for subgroup analysis of serious adverse events, and there was a good overlap of confidence interval between the effect estimates of the different subgroups (Analysis 2.3).

2. Different types of gastrectomy (subtotal versus total gastrectomy). Of the 13 trials, 12 trials reported the use of subtotal gastrectomy (Aoyama 2014; Chen Hu 2012; Deng 2009; Hayashi 2005; Hu 2015; Huscher 2005; Kim 2013; Kim 2015; Kitano 2002; Lee 2005; Sakuramoto 2013; Takiguchi 2013). In Cai 2011, a mixture of different types of gastrectomies was performed. No separate data were available for the different types of gastrectomies. Excluding this trial did not alter the results (Analysis 2.4; Analysis 2.5; Analysis 2.6).

The remaining subgroup analyses were not possible for the following reasons.

1. Totally laparoscopic and laparoscopy‐assisted gastrectomy: it was not clear whether any of the trials used totally laparoscopic gastrectomy.

2. Different cancer stages (node‐positive and node‐negative gastric cancer): none of the trials reported the results separately for different nodal status.

3. Different methods of anastomoses (stapler versus hand‐sewn anastomoses): five trials used staples as the anastomotic method (Aoyama 2014; Hayashi 2005; Lee 2005; Sakuramoto 2013; Takiguchi 2013). In the remaining trials, either a combination of stapler and hand‐sewn anastomosis were used (Huscher 2005), or the information on stapler versus hand‐sewn anastomosis was not available (Cai 2011; Chen Hu 2012; Deng 2009; Hu 2015; Kim 2013; Kim 2015; Kitano 2002).

4. People with different anaesthetic risk (ASA I (a healthy person) or ASA II (a person with mild systemic disease) versus ASA III or more (a person with severe systemic disease or worse): none of the trials that included ASA III participants reported the results separately for ASA III participants.

5. Different BMI (healthy weight (BMI 18.5 to 25) versus overweight or obese (BMI 25 or greater)): none of the trials reported results separately for people with different BMI ranges.

Sensitivity analysis

The sensitivity analysis excluding trials in which either mean or standard deviation or both were imputed have been presented in the individual outcomes. Another sensitivity analysis excluding the trial in which the laparoscopic group underwent less invasive distal gastrectomy, and the open group underwent more invasive subtotal gastrectomy, did not alter the results. We performed the remaining sensitivity analyses for the following reasons.

1. All trials were at unclear or high risk of bias.

2. No cluster‐RCTs were included.

3. The only multi‐armed trial included was investigating fast‐track surgery. Since this variable was not of interest to this review, we considered the trial a two‐arm trial.

Reporting bias

Only two outcomes had 10 or more trials, namely, short‐term mortality and adverse events. We did not assess reporting bias using a funnel plot in the remaining comparisons. In the outcome, short‐term mortality, only three trials had one or more events in at least one of the groups (Hu 2015; Huscher 2005; Kim 2015). So, we did not assess reporting bias using a funnel plot for short‐term mortality either. Visual inspection revealed that studies with large variance were more prevalent in the laparoscopic group than the open group, suggesting potential reporting bias. However, the Egger's test was not statistically significant (P = 0.3144).