Additional gastrectomy in early-stage gastric cancer after non-curative endoscopic resection: a meta-analysis

Abstract Background and objective The role of additional gastrectomy after non-curative endoscopic resection remains uncertain. The present meta-analysis aimed to explore the risk factors for early-stage gastric-cancer patients after non-curative endoscopic resection and evaluate the efficacy of additional gastrectomy. Methods Relevant studies that reported additional gastrectomy after non-curative endoscopic resection were comprehensively searched in MedLine, Web of Science and EMBASE. We first investigated the risk factors for residual tumor and lymph-node metastasis after non-curative endoscopic resection and then analysed the survival outcome, including 5-year overall survival (OS) and 5-year disease-free survival, of additional gastrectomy. Results Twenty-one studies comprising 4870 cases were included in the present study. We found that residual tumor was associated with larger tumor size (>3 cm) (odds ratio [OR] = 2.81, P < 0.001), undifferentiated tumor type (OR = 1.78, P = 0.011) and positive horizontal margin (OR = 9.78, P < 0.001). Lymph-node metastasis was associated with larger tumor size (>3 cm) (OR = 1.73, P < 0.001), elevated tumor type (OR = 1.60, P = 0.035), deeper tumor invasion (>SM1) (OR = 2.68, P < 0.001), lymphatic invasion (OR = 4.65, P < 0.001) and positive vertical margin (OR = 2.30, P < 0.001). Patients who underwent additional gastrectomy had longer 5-year OS (hazard ratio [HR] = 0.34, P < 0.001), 5-year disease-free survival (HR = 0.52, P = 0.001) and 5-year disease-specific survival (HR = 0.50, P < 0.001) than those who did not. Moreover, elderly patients also benefited from additional gastrectomy regarding 5-year OS (HR = 0.41, P = 0.001). Conclusions Additional gastrectomy with lymph-node dissection might improve the survival of early-stage gastric-cancer patients after non-curative endoscopic resection. However, risk stratification should be performed to avoid excessive treatment.


Introduction
Although the incidence of gastric cancer is decreasing, it remains the fourth most common malignancy worldwide, with an estimated 951,600 new cases in 2012 [1]. With a widespread increase in medical checkups and endoscopic screenings, the proportion of early-stage gastric cancer (EGC) is increasing, particularly in Japan [2] and Korea [3].
Therefore, this meta-analysis was performed to review the additional gastrectomy for non-curative endoscopic resection, to investigate the risk factors for residual tumor and LNM after non-curative endoscopic resection, and to explore whether EGC patients benefit from additional gastrectomy regarding survival outcome.

Search strategy
On 4 October 2017, the studies on gastric cancer and endoscopic resection with restrictions in English language were systematically searched in MedLine, Web of Science and EMBASE databases. The following terms in [Title/Abstract] were used: "endoscopic resection", "endoscopic submucosal dissection", "endoscopic mucosal resection", "early gastric cancer", "early gastric carcinoma", "early gastric neoplasm", "early stomach cancer", "early stomach neoplasm", "gastrectomy", "surgery" and "surgical procedure". We also screened the reference lists of the retrieved studies manually to supplement the related studies.

Inclusion and exclusion criteria
According to the Japanese Gastric Cancer Association (JGCA) [7], the endoscopic resection is considered curative if the resection conforms to the following criteria: en bloc resection, tumor size 2 cm, differentiated type, pT1a, negative horizontal and vertical margins, no lymphatic invasion and no venous invasion. Non-curative endoscopic resection was defined as resection that does not meet the above criteria.
All relevant studies that compared additional gastrectomy with non-gastrectomy after non-curative endoscopic resection were included. Studies that explored the risk factors of residual tumor and/or LNM after additional gastrectomy were also included. Animal studies, case reports, letters, comments, meeting abstracts, reviews and clinical studies with fewer than 15 patients were excluded. If multiple articles reported the same patient cohort, the articles containing more recent and complete data were selected. Multiple articles were included if they reported different endpoints of the same patient cohort.

Data extraction
Two authors (R.C.N. and Y.Q.Y.) extracted the following information from each study independently: author, publication year, country, study period, study design, patient number and median follow-up period. We also collected the clinicopathological characteristics of the two groups. Survival outcomes of these studies included 5-year OS, disease-free survival (DFS) and DSS. If the studies did not report the 5-year survival rate, we calculated the hazard ratio (HR) from the survival curves, as described by Parmar et al. [26]. The two senior authors (Y.F.L. and Y.B.C.) resolved any disagreements in the extracting data.

Quality assessment
All studies in this meta-analysis were retrospective; thus, the quality of the included studies was assessed with the Newcastle-Ottawa Scale (NOS) [27] using a scale from 0 to 9. Studies with a score higher than 5 were considered high-quality studies in this analysis.

Statistical analysis
All the data in this meta-analysis were analysed using STATA/ SE 12.0 (StataCorp LP, College Station, TX, USA). The weighted mean difference and odds ratio (OR) with reported 95% confidence intervals (CIs) were used to analyse continuous and dichotomous data, respectively. HRs were used to assess the treatment effect of additional gastrectomy compared with nongastrectomy. In the present study, we investigated the related risk factors with tumor residual and LNM for the patients in the additional gastrectomy group. All statistical analyses were performed with the v 2 test using the I 2 statistic to assess heterogeneity, with the level of significance set at 10%. The fixed-effects model was used when there was no significant heterogeneity. Otherwise, the random-effects model was used. Publication bias was tested using funnel plots.

Characteristics of the included studies
After the systematic search, a total of 21 studies [8-20, 22, 24, 25, 28-32] comprising 4,870 cases were included in the final analysis ( Figure 1). The characteristics of the included studies are shown in Table 1. All the included studies were retrospective and primarily considered Japanese and Korean patient populations. Eleven studies [9,13,14,16,18,22,24,25,29,31,32] performed survival data collection. These studies explored the survival differences between gastrectomy and observation but did not compare gastrectomy with other treatment strategies (such as adjuvant treatment and re-ESD). Five studies [8,11,19,28,30] reported the risk factors for residual tumor after additional gastrectomy and 11 studies [8-12, 14-17, 19, 20] reported the risk factors for LNM. All included studies were considered high-quality.

Risk factors for residual tumor
As shown in Table 2, patients with larger tumor size (>3 cm) or positive horizontal margin (OR ¼ 9.78, 95% CI ¼ 6.30-15.18, P < 0.001; I 2 ¼ 44.3%, P ¼ 0.145) exhibited higher rates of residual tumor than their controls. Gross type, ulcer formation, tumor depth (SM1 stage), lymphatic invasion and vertical margin were not associated with residual tumor.

Publication bias
The funnel plot indicated that there were no obvious publication biases in OS, DFS and DSS in the total population (Supplementary Figure 1). No obvious publication bias was observed in OS in the elderly patient population (Supplementary Figure 1).

Discussion
In the present meta-analysis, we reviewed the additional gastrectomy for non-curative endoscopic resection and found that additional gastrectomy with LN dissection can prolong the 5-year OS, DFS and DSS of patients with EGC after non-curative endoscopic resection. In addition, elderly patients can benefit from additional gastrectomy regarding 5-year OS. Several risk factors have been reported to be associated with residual tumor after non-curative endoscopic resection [8,11,19,28,30], including lymphovascular invasion and positive margins. Ito et al. [8] showed that tumor invasion and positive horizontal margins were associated with residual tumor. Sunagawa et al. [11] reported that positive horizontal and vertical margins were risk factors for residual tumor. Because of the small sample sizes of previous studies and the low incidence of residual tumor after non-curative endoscopic resection, the pooling results of our meta-analysis increased our ability to identify relevant risk factors. This meta-analysis showed that larger tumor size (>3 cm), undifferentiated tumor type and positive horizontal margins were predictors for residual tumor. Other factors, including gross type, ulcer, tumor invasion, lymphatic invasion and even vertical margin, were not predictors for residual tumor.
This study showed that the rate of LNM varies from 4.3% to 12.7% after additional gastrectomy with LN dissection. The pooled incidence of LNM was 8.0% (240/2989). These pooled results showed that patients with larger tumor size, elevated gross type tumors, deeper tumor invasion (>SM1 stage), lymphatic invasion and positive vertical margins had a higher risk of LNM. According to JGCA [7], the lack of ulcer formation is one of the best indications for EMR or ESD. Ulcer formation is  considered a criterion for non-curative endoscopic resection. However, the present meta-analysis demonstrated that ulcer formation predicted neither residual tumor nor LNM. Therefore, if the resected specimen conforms to other curative criteria without histological evidence of residual tumor in the ulcer, then it is possible to regard endoscopic resection as curative.
The therapeutic strategy of a positive horizontal margin is debated. Sunagawa et al. [11] reported that patients with positive horizontal margins have a higher risk of residual tumor but no increased risk of LNM. Several investigators also reported that the rate of LNM is quite low when positive horizontal margins are the only indicator of non-curative surgery [14,19]. This meta-analysis showed that positive horizontal margins are associated with a higher rate of residual tumor. Interestingly, a positive horizontal margin was negatively associated with LNM. Numata et al. [33] reported that all local recurrent tumors with positive horizontal margins were intramucosal lesions and could be resected by re-ESD. Given the low incidence of LNM in cases of positive horizontal margins, we consider that repeated ESD or strict endoscopic surveillance is feasible if the positive horizontal margin is the only indicator of non-curative surgery.
In this meta-analysis, we identified several risk factors, including elevated gross type, >SM1 stage and lymphatic invasion, which were associated with LNM, but not with residual tumor. Abe et al. [34] reported that a combination of ESD and laparoscopic LN dissection without gastrectomy for EGC patients with a risk of LNM can be effective, with minimally invasive treatments maintaining long-term outcomes. However, the sample in this previous study was very small (21 patients). Moreover, LN dissection without gastrectomy is a complex surgery. Therefore, it was still unclear whether LN dissection without gastrectomy could be performed in these patients.
In clinical practice, the decision to perform additional gastrectomy with LN dissection for non-curative endoscopic resection is influenced not only by the risk of residual tumor and LNM, but also by age, comorbidities, the wishes of the patients and their families, and the predicted quality of life after surgery. This meta-analysis showed that patients who underwent additional surgery were younger than those who did not (Supplementary Table 1). Moreover, patients who underwent additional surgery had a higher proportion of >SM1 stage tumors and lymphatic invasion, indicating that clinicians were also involved in the treatment strategy. Suzuki et al. [16] reported that patients who underwent additional surgery had a longer 5-year OS. However, 5-year DSS levels were similar between the two groups. Toya et al. [24] also found that patients who underwent additional surgery had a longer 5-year OS. However, of the patients who did not undergo additional surgery, none died of recurrent gastric cancer. Nevertheless, a multicenter retrospective study that included 1969 patients with non-curative ESD showed a larger discrepancy between OS and DSS in these two groups [18]. After pooling the survival outcome from 11 studies [9,13,14,16,18,22,24,25,29,31,32], our study revealed that patients with non-curative endoscopic resection could benefit from additional gastrectomy in terms of 5-year OS, DFS and DSS. Moreover, our study showed that elderly patients who underwent additional gastrectomy had a longer 5-year OS. However, additional gastrectomy should be performed selectively. Hatta et al. [20] established a simple but useful scoring system ('eCura system') to predict DSS in EGC patients after non-curative ESD. For patients at low risk of LNM categorized by the 'eCura system', observation may be an optimal choice because of their excellent survival outcome (5-year DSS rate, 99.6%).
There are several limitations of this meta-analysis. First, all included studies were retrospective and were performed in Japan and Korea, where the incidence of EGC is very high and endoscopic technology is advanced. Thus, whether the result of this meta-analysis applies to Western countries remains uncertain. Second, there are selection biases about treatment strategy choices before and after endoscopic resection. This bias will probably include the health of the patient and the surgeons' estimation of the risk of LNM. Third, some of the considered studies investigated non-curative ESD patients only, whereas others reported the results of both non-curative ESD and EMR. Fourth, we could not compare the outcome between gastrectomy and other treatment strategies, such as re-ESD and adjuvant chemotherapy. Fifth, the median follow-up for non-curative endoscopic resection of the included studies was shorter than 15 years, which is considered the most favorable surveillance time for gastric cancer after LN dissection [35]. Finally, several analyses showed significant heterogeneity.

Conclusions
In conclusion, after systematically reviewing and pooling the outcomes of additional gastrectomy, the present study showed that additional gastrectomy with LN dissection can prolong the 5-year OS, DFS and DSS of EGC after non-curative endoscopic resection. In addition, elderly patients can benefit from additional gastrectomy regarding 5-year OS. Therefore, additional gastrectomy with LN dissection might improve the survival of EGC after non-curative endoscopic resection. However, risk stratification should be performed to avoid excessive treatment. Re-ESD, even observation, might be the optimal choice for some populations.

Supplementary data
Supplementary data is available at Gastroenterology Report online.