Robotic versus Laparoscopic versus Open Gastrectomy: A Meta-Analysis.

PURPOSE: To define the role of robotic gastrectomy for the treatment of gastric cancer, the present systematic review with meta-analysis was performed.
MATERIALS AND METHODS: A comprehensive search up to July 2012 was conducted on PubMed, EMBASE, and the Cochrane Library. All eligible studies comparing robotic gastrectomy versus laparoscopic gastrectomy or open gastrectomy were included.
RESULTS: Included in our meta-analysis were seven studies of 1,967 patients that compared robotic (n=404) with open (n=718) or laparoscopic (n=845) gastrectomy. In the complete analysis, a shorter hospital stay was noted with robotic gastrectomy than with open gastrectomy (weighted mean difference: -2.92, 95% confidence interval: -4.94 to -0.89, P=0.005). Additionally, there was a significant reduction in intraoperative blood loss with robotic gastrectomy compared with laparoscopic gastrectomy (weighted mean difference: -35.53, 95% confidence interval: -66.98 to -4.09, P=0.03). These advantages were at the price of a significantly prolonged operative time for both robotic gastrectomy versus laparoscopic gastrectomy (weighted mean difference: 63.70, 95% confidence interval: 44.22 to 83.17, P<0.00001) and robotic gastrectomy versus open gastrectomy (weighted mean difference: 95.83, 95% confidence interval: 54.48 to 137.18, P<0.00001). Analysis of the number of lymph nodes retrieved and overall complication rates revealed that these outcomes did not differ significantly between the groups.
CONCLUSIONS: Robotic gastrectomy for gastric cancer reduces intraoperative blood loss and the postoperative hospital length of stay compared with laparoscopic gastrectomy and open gastrectomy at a cost of a longer operating time. Robotic gastrectomy also provides an oncologically adequate lymphadenectomy. Additional high-quality prospective studies are recommended to better evaluate both short and long-term outcomes.

Introduction

Minimally invasive surgery (MIS) has enjoyed wide acceptance since the first laparoscopic gastrectomy (LG) for cancer was reported in 1994.1,2 Numerous approaches to this procedure have been developed, and many patients have benefited from their effectiveness. This is particularly salient as recently demonstrated by phase III clinical trials for early gastric cancer.3-5 However, laparoscopic surgery has certain limitations, such as two-dimensional imaging, restricted range of motion of the instruments, and poor ergonomic positioning of the surgeon. Moreover, indications to treat advanced gastric cancers by MIS still remain to be defined.3,6,7

The robotic surgery system was introduced as a solution to minimize the shortcomings of laparoscopy. This emerging method provides undoubted technical advantages over conventional laparoscopy,8 but its role for gastric cancer is still unclear.9-14 Since robotic gastrectomy (RG) has been reported,8,15 the application of this new approach is increasing in experienced centers. Yet, to date only a limited number of studies have focused on this new method. Thus, we have performed a meta-analysis of studies that have compared RG with open gastrectomy (OG) and with LG for gastric cancer. In the era of rapid penetration of robotic technology, we specifically aimed to systematically and objectively assess the value of RG for the treatment of gastric cancer.

Materials and Methods

1. Search strategy

Publications were identified by searching the major medical databases, such as PubMed, EMBASE, and the Cochrane Library for all articles published until July 19, 2012. We adopted the MeSH form strategy for PubMed and the Cochrane Library as follows: (stomach cancer or gastric cancer or stomach neoplasm or gastric neoplasm or stomach carcinoma or gastric carcinoma or stomach tumor or gastric tumor or stomach tumour or gastric tumour or stomach neoplasms or gastric neoplasms) and (robot or robotic or robotics). We used the Emtree form strategy for EMBASE as reported: ('stomach cancer' or 'gastric cancer' or 'stomach neoplasm' or 'gastric neoplasm' or 'stomach carcinoma' or 'gastric carcinoma' or 'stomach tumor' or 'gastric tumor' or 'stomach tumour' or 'gastric tumour') and (robot or robotic or robotics).

2. Inclusion and exclusion criteria with quality of literature

Inclusion criteria were: (1) observational studies that compared RG versus OG or LG or both with lymph node dissection for the treatment of gastric cancer; (2) patients with primary gastric adenocarcinoma; and (3) data which included most of the following: operation time, blood loss, number of harvested lymph nodes, hospital stay, and morbidity. Exclusion criteria were: (1) gastrointestinal stromal tumors or benign gastric diseases; and (2) duplicate publication or the publication that did not provide sufficient data for the analyses.

Two of the authors (M.A., Y.Y.C.) evaluated the eligibility of all studies independently collected from the databases based on the selection criteria. The risk of bias in each study was assessed according to the Newcastle-Ottawa Scale for observational studies.16

3. Data extraction

Two researchers (M.A., Y.Y.C) extracted data from each study using a structured sheet and entered the data into a database. Another author (J.K.) reviewed this process. The following factors were examined: author, year of publication, source journal, sample size, characteristics of the patients, research design, operative time, blood loss, number of harvested lymph nodes, morbidity, and length of hospital stay.

4. Statistical analysis

We conducted a meta-analysis with the software Review Manager (RevMan) ver. 5.1 by the Cochrane Collaboration (Nordic Cochrane Center, Copenhagen, Denmark). We calculated the effect sizes by odds ratio (OR) for dichotomous variables and weighted mean differences (WMD) for continuous outcome measures with 95% confidence intervals (CIs). The Higgins I2 and chi-square tests were used to evaluate the heterogeneity of the results across studies. We considered a P-value of chi-square less than 0.10 with an I2-value of greater than 50% as indicative of substantial heterogeneity. The inverse variance method with fixed-effects model was applied if no heterogeneity was considered, whereas the random effects model was used in opposite cases. The correction of the zero cell counts was automatically performed by the RevMan program with exclusion of the outcome from the meta-analysis if there was no event in both arms in one study. We obtained the standard deviation from the article by contacting the authors by e-mail if it was not reported or by extracting the estimated standard deviation from the P-value ⌜Cochrane Handbook for systematic reviews of interventions version 5.0.2⌟ if it was not directly available from the authors. The symmetry of Begg's funnel plot was used to evaluate the presence of publication bias in the studies.

Results

From a total of 190 articles seven17-23 were selected according to our inclusion and exclusion criteria (Fig. 1). All are non-randomized, non-blinded, retrospective cohort studies. The characteristics of the seven selected articles are shown in Table 1.

Fig. 1

Flow diagram for identification of selected articles. Seven cohort studies were finally included in the meta-analysis. RG = robotic gastrectomy; OG = open gastrectomy; LG = laparoscopic gastrectomy.

Table 1

Selected studies

TG = total gastrectomy; SG = subtotal gastrectomy; LN = lymph node; RG = robotic gastrectomy; LG = laparoscopic gastrectomy; OG = open gastrectomy; NA = not available; EGC = early gastric cancer; AGC = advanced gastric cancer.

Next to the intraoperative outcomes, this analysis was directed at the short-term results of RG because long-term outcomes were not sufficiently addressed by the included studies. A total of 1,967 patients affected by gastric cancer were analyzed, including 404 who underwent RG, 718 who underwent OG, and 845 who underwent LG. The risk of bias of included studies is shown in Table 2, whereby two received seven stars, one received five stars, and the remaining received six stars. The mean number of stars for the risk of bias was 6.1. Each meta-analysis was performed twice; with all available studies and those including only complete data which was analyzed excluding the study showing results with high risk of bias.23 The forest plots of each outcome, subdivided in two different groups, are shown in Fig. 2 and 3.

Table 2

The assessment of the risk of bias in each study using the Newcastle-Ottawa scale

REC = representativeness of the exposed cohort; SNEC = selection of the non-exposed cohort; AE = ascertainment of exposure; DO = demonstration that outcome of interest was not present at start of study; SC = study controls for age, sex, marital status; AF = study controls for any additional factors; AO = assessment of outcome; FU = follow-up long enough for outcomes to occur; AFU = adequacy of follow-up of cohorts.

*Asterisk means that the study is satisfied the item, and no asterisk means the opposite situation.

Fig. 2

Forest plots of studies comparing robotic gastrectomy (RG) vs. open gastrectomy (OG) for gastric cancer. (A) Operative time. (B) Blood loss. (C) Number of harvested lymph-nodes. (D) Overall complication rate. (E) Hospital stay. SD = standard deviation; CI = confidence interval; df = degree of freedom.

Fig. 3

Forest plots of studies comparing robotic gastrectomy (RG) vs. laparoscopic gastrectomy (LG) for gastric cancer. (A) Operative time. (B) Blood loss. (C) Number of harvested lymph nodes. (D) Overall complication rate. (E) Hospital stay. SD = standard deviation; CI = confidence interval; df = degree of freedom.

1. Comparing RG with OG

We found that three out of seven selected articles compared RG to OG (Table 1). The operative time, amount of blood loss, number of lymph nodes retrieved, overall complications, and hospital stay were analyzed. Detailed information about these outcomes was available in all three selected studies.

The operative time in the robotic group was significantly longer than in the open group (WMD: 95.83, 95% CI: 54.48 to 137.18, P<0.00001) in a random effect model (P=0.002, I2: 84%). After exclusion of insufficient data the operative time was still significantly longer in the robotic group (WMD: 100.91, 95% CI: 37.72 to 164.11, P=0.002) in a random effect model (P=0.0008, I2: 91%). The blood loss was less in RG than OG, but without statistical significance (WMD: -225.58, 95% CI: -478.88 to 27.72, P=0.08) in a random effect model (P<0.0001, I2: 98%). By considering only complete data, the analysis revealed the same results (WMD: -114.88, 95% CI: -251.90 to 22.13, P=0.10) in a random effect model (P=0.002, I2: 90%). The number of lymph nodes retrieved was higher in the open group compared with the robotic group, though without a statistically significant difference (WMD: -2.68, 95% CI: -5.74 to 0.38, P=0.09) in a fixed effect model (P=0.88, I2: 0%). The same non-significant outcomes in favor of OG (WMD: -3.28, 95% CI: -7.48 to 0.91, P=0.13) were noted when including only complete data in a fixed effect model (P=0.75, I2: 0%). The overall complication rate between the two comparative surgical approaches did not show any significant difference (OR: 0.93, 95% CI: 0.51 to 1.68, P=0.81) in a fixed effect model (P=0.44, I2: 0%). Additionally, when considering only valid data, the robotic and open groups did not differ significantly in regards to morbidity (OR: 0.84, 95% CI: 0.38 to 1.86, P=0.66) in a fixed effect model (P=0.22, I2: 33%). However, the hospital length of stay for the robotic group was significantly shorter than for the open group (WMD: -2.92, 95% CI: -4.94 to -0.89, P=0.005) in a random effect model (P=0.04, I2: 69%). Analysis of data extracted only from the articles including complete data 17,22 also revealed an outcome with a significant difference in favor of the robotic group (WMD: -2.54, 95% CI: -4.67 to -0.41, P=0.02) in a random effect model (P=0.03, I2: 78%). Fig. 4 shows the symmetry of the funnel plots in each outcome. Estimation of publication bias by funnel plot was difficult as there were only three publications which compared RG to OG. However, publication bias was suspected among the overall complications.

Fig. 4

Begg's funnel plots for identifying publication bias for five perioperative outcomes regarding robotic vs. open and laparoscopic gastrectomy. (A) Operative time. (B) Blood loss. (C) Number of harvested lymph nodes. (D) Overall complication rate. (E) Hospital stay. RG = robotic gastrectomy; OG = open gastrectomy; LG = laparoscopic gastrectomy; SE = standard error; MD = mean difference; OR = odds ratio.

2. Comparing RG to LG

We found that six out of the seven selected articles compared RG to LG (Table 1). The operative time, amount of blood loss, number of lymph nodes retrieved, overall complications, and hospital length of stay were examined. All six articles reported their respective data regarding these particular parameters; however, one report19 compared the amount of blood loss by the change in the hemoglobin level, and this parameter was excluded from this specific analysis. We found that the operative time for RG was significantly longer than for LG (WMD: 63.70, 95% CI: 44.22 to 83.17, P<0.00001) in a random effect model (P=0.002, I2: 74%). The results were the same when considering only complete articles, with similar statistical significance (WMD: 65.82, 95% CI: 43.27 to 88.37, P<0.00001) in a random effect model (P=0.0007, I2: 79%). The amount of blood loss was also significantly less for RG than for LG (WMD: -35.53, 95% CI: -66.98 to -4.09, P=0.03) in a random effect model (P=0.003, I2: 75%). By including only valid data, analysis of the pooled data revealed that the estimated blood loss was not significantly less for RG than for LG (WMD: -27.82, 95% CI: -63.60 to 7.95, P=0.13) in a random effect model (P=0.002, I2: 79%). Likewise, our analysis of the number of lymph nodes retrieved demonstrated no significant difference between RG and LG (WMD: 0.50, 95% CI: -3.30 to 4.30, P=0.80) in a random effect model (P=0.0001, I2: 80%). Analysis of this variable by considering only valid articles also revealed that there was no significant difference in the number of lymph nodes retrieved (WMD: 0.54, 95% CI: -4.51 to 3.42, P=0.79) in a random effect model (P=0.0006, I2: 80%). The overall complication rate of each surgical approach was not different between the groups (OR: 0.87, 95% CI: 0.59 to 1.27, P=0.46) in a fixed effect model (P=0.69, I2: 0%). Inclusion of only the articles that matched our quality assessment did not change the final results, as even in this analysis there was no significant difference in the overall complication rate (OR: 0.85, 95% CI: 0.57 to 1.28, P=0.44) in a fixed effect model (P=0.56, I2: 0%). Likewise, the duration of hospital stay was not significantly different between the RG and LG groups (WMD: -0.60, 95% CI: -1.66 to 0.45, P=0.26) in a random effect model (P<0.00001, I2: 87%). Similar results were observed when considering properly sufficient data (WMD: -0.43, 95% CI: -1.46 to 0.61, P=0.42) in a random effect model (P<0.00001, I2: 89%). Fig. 4 shows the symmetry of the funnel plot for each outcome. Asymmetry was suspected for the amount of blood loss, the overall complications, and the hospital length of stay.

Discussion

The aim of this meta-analysis was to compare RG to OG and LG by means of a thorough evaluation of the available data. Most studies demonstrated similar advantages to RG, including a reduced intraoperative blood loss and shorter hospital stay compared with OG and less intraoperative blood loss compared with LG. These results suggest that robotic surgery can be especially useful for selected patients affected by gastric cancer with severe anemia, as some authors have already published.20 This is important as transfusion-related problems and risk of gastric cancer recurrence could be reduced.24-26

Lymph node dissection is a fundamental part of a radical gastrectomy. In this meta-analysis the overall number of lymph nodes retrieved was high, which is explained by many studies with D2 lymphadenectomies.27 Evaluation of the pooled data revealed that the number of lymph nodes dissected with RG was less than with OG but greater than with LG, though these differences were not significant. A possible explanation for the high number of lymph nodes retrieved with OG might be that comparative groups of two22,23 of the three pooled studies were heavily unbalanced by centers highly skilled with OG. Moreover, most patients enrolled in these two open groups were affected by advanced gastric cancer. Therefore, since a more extended lymphadenectomy is recommended for advanced stages,28,29 one might expect the number of lymph nodes retrieved to be higher than in early stages, which are more common in the robotic group. Nonetheless, even in comparison with the open approach, RG guarantees the removal of at least 15 lymph nodes which still allows for proper staging and prognostication. This is important because it favors greater rates of survival.30,31 All told, these results demonstrate that RG can be safely performed with adequate oncological principles compared to either OG or LG. RG could also be used for D2 lymphadenectomies, although it was not possible to analyze whether D2 lymphadenectomies were validated by any approach.

These positive results derive from some technical and evidence-based aspects to robotic surgery, such as superior 3-dimensional visualization, more degree of freedom of the wristed-tip instrument, and tremor-filtered control of the four robotic arms.8,13 These emerging tools may improve a surgeon's dexterity when fine manipulation of tissues in a close, fixed operating field is required, or when hand-sewn sutures and knot-tying are a necessity. The robotic technique provides an easier lymph node dissection, a reduced inflammatory response, less manipulation of the tissues, and a quicker bowel recovery which are reflected by a reduced intraoperative blood loss, a shorter hospital stay, and an easier lymphadenectomy compared with other approaches.17

Some limitations were present in our meta-analysis. The quality assessment of all seven selected articles is shown in Table 2. One work had a score of <6 points,23 whereas the remaining studies all had a valuation of ≥6 points.17-22 Our purpose was to obtain a good quality meta-analysis, but the pooled studies were limited in number. Thus, we decided to compare all selected variables both considering all seven reports as well as excluding only the work that did not match our quality assessment.23 This decision was aimed at reducing potential bias and to assess for any differences in the results between the two specific groups. According to this strategy, the only difference considering the complete data was regarding the estimated blood loss, which was less in RG vs. LG, though without statistical significance (Fig. 3B).

Another limitation is that significant heterogeneity was recognized in some characteristics of the selected articles. Most of the pooled studies included patients who underwent both subtotal gastrectomy and total gastrectomy without distinction. Furthermore, surgeons performed both robotic gastrectomies and robotic-assisted gastrectomies. The stage of gastric cancer was also not the same for all of the enrolled patients. For instance, inclusion of patients was specifically limited to early gastric cancer in one study,21 three reports considered patients with both early and advanced gastric cancer,18,22,23 and three articles17,19,20 included patients with a preoperative clinical stage Ia or Ib disease. With the exception of one report,19 the intended extent of lymph node dissection was, however, defined for the RG, LG, and OG groups in all studies (namely D1+β or D2) (Table 1). Finally, the majority of the studies were from Asian populations and accounted for 1,748 of the patients,17,19-21,23 whereas the remaining two reports were European, consisting of 219 patients.18,22 This is relevant as gastric cancer is less common in Western countries where it is diagnosed at later stages since screening in Europe is not routine.32 Additionally, Asian surgeons have been the first to embrace the new robotic system, and to date they continue to perform many robotic gastrectomies in experienced centers. Therefore, it is difficult to generalize the findings of this meta-analysis to worldwide populations of patients with gastric cancer.

To date, only one Chinese meta-analysis comparing RG vs. LG has been published.33 The authors selected three studies from an initial collection of 52 articles up to December 2011 according to specific criteria. This resulted in a comparison of 268 patients undergoing RG vs. 650 undergoing LG. This Chinese meta-analysis did have some limitations which are acknowledged by the authors, including a limited number of published studies, small sample size, and short duration of follow-up. Indeed, each of these weaknesses is also present in our work, though ours differs in some aspects. First, the aim of our meta-analysis was to compare RG with both OG and LG, for this aspect it can be considered the first one. Second, we collected all 190 articles present in the literature up to July 2012 using an accurate research strategy. We contacted each author if some data of interest were lacking, whereas if data were unavailable, we noted this as incomplete in the database. Finally, we decided to perform a strategic double comparison in order to minimize possible bias since we realized that not all of the seven selected articles had enough quality level data and since the number of patients overall was relatively small.

Nevertheless, our meta-analysis still has some important shortcomings. The pooled studies are all retrospective and obviously limited. The number of patients entered is relatively small and sometimes comparative groups are unbalanced, which may gloss over true differences among some variables. Moreover, some characteristics are not mentioned in all of the pooled studies, such as type of surgical reconstruction, stage of the tumor, short and long-term outcomes, and follow-up period. For instance, in two of the articles,22,23 the groups do not have the same period of recruitment. Additionally, in another of the two studies, the patients were drawn from the same database;19,21 although, in one report, only patients who underwent subtotal gastrectomy were enrolled, while in the other report, only those who had received a total gastrectomy were considered. A final important deficiency in our work is the complete lack of information regarding costs, which could represent a real and incontrovertible problem. So far, only one article is present in the literature 21 that has reported some information regarding this aspect. Even if in the near future some other authors do focus on this topic it will remain very difficult to evaluate the cost of the robotic approach and compare it to that of more traditional methods. This is relevant as more than one insurance system is often present in each country and because insurance systems differ between countries. Furthermore, the price of the instruments and the devices is also not equivalent worldwide. Since each type of uniformity cannot be guaranteed, it is likely that the tools generally recognized for the assessment of costs will not be provided.

Many high quality prospective studies must be performed in order to assess the safety and efficacy of RG for the treatment of gastric cancer in the near future. In addition to the outcomes reported in this article, future studies should be directed at more detailed oncologic findings, survival, and costs. Currently, a multicenter prospective analysis regarding the advantage of RG vs. LG (NCT01309256) is in process in Korea, representing a good effort to assess the validity of RG.

In conclusion, our meta-analysis is the most accurate and updated work present in the literature and it can be considered as the starting point for subsequent investigations. Our study demonstrates that RG for gastric cancer is a promising new surgical technique that reduces blood loss and postoperative hospital stays compared with LG and OG, though at the price of longer operating times. RG also allows for an oncologically correct lymphadenectomy for gastric cancer treatment.