Identifying Optimal Surgical Intervention-Based Chemotherapy for Gastric Cancer Patients With Liver Metastases

Background This study aimed at evaluating the effects of surgical treatments-based chemotherapy in the treatment of gastric cancer with liver metastases (GCLM). It has not been established whether Liver-directed treatment (LDT) options such as hepatectomy and gastrectomy plus chemotherapy (HGCT), radiofrequency ablation and gastrectomy plus chemotherapy (RFAG), transarterial chemoembolization and gastrectomy plus chemotherapy (TACEG), gastrectomy plus chemotherapy (GCT) enhance the survival of GCLM patients. Methods We performed systematic literature searches in PubMed, EMBASE, and Cochrane library from inception to September 2021. We created a network plot to comprehensively analyze the direct and indirect evidence, based on a frequentist method. A contribution plot was used to determine inconsistencies, a forest plot was used to evaluate therapeutic effects, the publication bias was controlled by funnel plot, while the value of surface under the cumulative ranking curves (SUCRA) was calculated to estimate rank probability. Results A total of 23 retrospective studies were identified, involving 5472 GCLM patients. For OS and 1-, 2-, 3-year survival rate of all trials, meta-analysis of the direct comparisons showed significant better for HGCT treatments compared with GCT or PCT. In the comparison of the 5 treatments for 1-, 2-, 3-year survival rate, HGCT and RFAG were found to be more effective than GCT and PCT, respectively. By OS and 2-, 3-year survival rate analysis, RFAG was identified as the best option, followed by HGCT, TACEG, GCT and PCT. By 1-year survival rate analysis, HGCT and RFAG were identified as the most effective options. Conclusion HGCT and RFAG has remarkable survival benefits for GCLM patients when compared to TACEG, GCT and PCT. HGCT was found to exhibit superior therapeutic effects for GCLM patients for 1-year survival rate while RFAG was found to be a prospective therapeutic alternative for OS and 2-, 3-year survival rate. Systematic Review Registration identifier [10.37766/inplasy2020.12.0009].


INTRODUCTION
Globally, gastric cancer is the fourth most common malignant tumor and the second highest cause of cancer-related mortalities (1)(2)(3). Therapeutic options for advanced gastric cancers have been enormously improved. In the last two decades, the 5-year survival rate is up to 40%. However, gastric cancer with liver metastases is considered a late-stage disease. Systemic chemotherapy was recommended as standard cure, with a 5-year survival rate of less than 10% (4,5). The current standard management of GCLM is systemic chemotherapy with supportive care. Liver metastasis is a common phenomenon for many types of cancer (6-8). Liverdirected treatment (LDT) options such as hepatectomy and gastrectomy plus chemotherapy (HGCT), radiofrequency ablation and gastrectomy plus chemotherapy (RFAG), transarterial chemoembolization and gastrectomy plus chemotherapy (TACEG), gastrectomy plus chemotherapy (GCT) for GCLM is controversial (5,9,10). Compared to systemic chemotherapy, surgical treatment such as HGCT and RFAG of hepatic metastases presents favorable prognosis (11)(12)(13). According to the guidelines of The Committee of the Japan Gastric Cancer Association (JGCA) and National Comprehensive Cancer Network (NCCN), palliative management is recommended for stage IV gastric cancer, e.g. GCLM. In contrast, colorectal liver metastases are considered as suitable targets for radical surgery because they often present as liveronly metastatic disease, and R0 resection shows good prognostic outcomes, with a 5-year survival rate > 50% (14,15). Retrospective studies have presented that the combination of hepatectomy and gastrectomy has visible survival outcome superiority (16)(17)(18)(19)(20)(21). In the last two decades, along with the results of reported studies which demonstrated that radical surgery of primary gastric cancer and metastatic liver lesions had survival benefits, the Guidelines Committee of JGCA reconsidered the effect of surgical treatment in GCLM patients (22). Therefore, the role of LDT for GCLM is gradually being considered.
Previous therapeutic options for GCLM were HGCT, RFAG, TACEG, GCT and palliative chemotherapy (PCT). There are no randomized controlled clinical trials for GCLM therapies. In the present literature, majority of the studies are retrospective studies, which were performed at a single center, with a limited number of patients. Although some studies have confirmed the superior therapeutic outcomes of LDT, the clinical pathological characteristics of the involved patients reveal some selection bias, therefore, their results are difficult to accept. We performed a network meta-analysis to evaluate the survival benefits of LDT and systemic chemotherapy in the treatment of GCLM.

Study Protocol
This work was performed in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) of the Cochrane Handbook for Systematic Reviews of Intervention (23). The full protocol was registered and available on INPLASY (INPLASY2020120009).

Search Strategy
We retrieved literature published in between 1966 and September 1st, 2021 by searching PubMed, EMBASE, and Cochrane Library with the keywords (1) "stomach neoplasm" OR "gastric neoplasms" OR "cancer of stomach" OR "stomach cancers" OR "gastric cancer" AND (2) "liver metastases OR liver metastasis OR hepatic metastasis" AND (3) "operative surgical procedure" OR ablation OR liver resection OR hepatectomy OR gastrectomy OR chemotherapy OR "interventional therapy" and using the search strategies as illustrated in Supplementary Table 1. We selected and evaluated all relevant studies and review articles about GCLM and inquired the authors for unpublished raw data. Searches were limited to English-language publications. In addition, the reference lists of the retrieved articles were examined for potential eligible studies.

Study Selection
The inclusion criteria for the studies were: i. Systemic chemotherapy and surgical treatment; ii. Series of case control or cohort studies; iii. The number of patients were to be > 20; iv. Consists of available endpoints, such as overall survival, 1-, 2-, 3-, and 5-year survival rates, median survival time, and postoperative complications. The exclusion criteria for the studies were: i. studies with insufficient data or no related endpoints; ii. Missing control group.

Data Extraction
Two researchers (MS and ZZ) independently extracted results from the enrolled articles in a standardized form. In addition, a third researcher (TL) was consulted in case there were disagreements. The information extracted from each study included the first author, country, year of publication, number of cases, treatment, sex, median or mean age of patients, study design, follow-up, median survival time. If a study did not report the Hazard Ratio (HR) of overall survival, we estimated HR and their corresponding 95% confidence intervals (CIs) using the method described by Parmar et al. (24) and Tierney et al. (25). We recovered the data of Kaplan-Meier curves as recently described by us (26,27).

Quality Assessment
We used the Newcastle-Ottawa Scale (NOS) to assess the quality of each included study. Scores ≥ 7 were considered high quality. We used a "star system" for case-controlstudies (SupplementaryTable 2).

Publication Bias
The funnel plots were used to establish publication bias. The funnel plot that was symmetrical near zero represented no publication bias.

Statistical Analysis
The primary endpoint of this network meta-analysis (NMA) was overall survival (OS), defined as the time from random assignment to date of death from any cause or date of last follow-up. Secondary endpoints were 1-, 2-, 3-year survival rates. A pair-wise meta-analysis was performed by STATA 13.0 (Stata Corp, College Station, TX). R-3.6.3 and R packages gemtc were applied to conduct the Bayesian NMA, and 95% confidence intervals (CIs) were computed for HR in overall survival analysis. 1-, 2-, 3-year survival rates were analyzed while Odds Ratios (OR) with 95% confidence intervals (CIs) was calculated by fixed-effects or random-effects model (28,29). Z test was performed to evaluate the significance of overall effect size.
A network plot was then used to directly demonstrate the whole information of included studies (30). Depending on direct comparison and indirect comparison outcomes, we estimated the contribution of each direct treatment comparison in the whole network structure, which was presented in a contribution plot. The inconsistency factor (IF) was calculated to determine the possible inconsistency in network comparison. The 95% CIs of IF values close to zero or the p value of Z test higher than 0.05 demonstrated there being no statistically significant inconsistency (31). Summary effects and corresponding predictive intervals were used to conclude relative mean effects and impact of heterogeneity in the network forest plot.
Finally, we calculated the surface under the cumulative ranking curve (SUCRA) of each treatment, which transformed the relative effects to the probability (2). SUCRA values range from 0 to 100%. The treatment was more valuable if the SUCRA value was higher. According to the estimated probability value, the treatments were ranked, which showed the percentage of effectiveness a treatment achieves with reference to an imaginary ideal treatment. Small-study effects was adjusted by a model of network meta-regression, the variance of the log-odds ratios as covariation (32).

Study Characteristics
A total 6362 relevant articles were downloaded. The flow diagram documenting the search and inclusion of relevant studies is displayed in Figure 1. After considering the inclusion and exclusion criteria, a total of 23 retrospective studies involving 5472 GCLM patients were identified (12,. At least one of the following treatments were assessed by the study: HGCT, RFAG, GCT, PCT, and TACEG. Eight studies were three-arm trials while fifteen studies were two-arm trials.  Further characteristics and Newcastle-Ottawa scale results regarding the included studies are presented in Table 1 (Supplementary Table 2).

Direct Comparisons and Subgroup Analysis
For OS and 1-, 2-, 3-year survival rate of all trials, meta-analysis of the direct comparisons showed significant better for HGCT treatments compared with GCT or PCT, with the exception of RFAG ( Table 2). As to OS and 1-, 2-, 3-year survival rate of all trials, PCT predicted a significantly worse OS than GCT ( Table 2). For 1-, 2-, 3-year survival rate, the results showed that RFAG indicated a better survival rate than GCT ( Table 2). Analysis of Asian subgroups showed that HGCT were better than GCT in OS, and 1-, 2-, 3-year survival rate, RFAG were better than GCT in 1-, 2-, 3-year survival rate ( Table 2). Overall, statistical heterogeneity was moderate, although for most comparisons 95% CIs were wide and included values indicating very high or no heterogeneity, which portrayed the small number of studies available for every pair-wise comparison. In the meta-analyses of direct comparisons for OS and 1-, 2-, 3-year survival rate, I² values higher than 40% were recorded for the comparisons HGCT versus GCT and HGCT versus PCT ( Table 2).

Network Meta-Analysis
The network evidence plot is shown in Figure 2. Five treatments were included for analysis; HGCT, GCT, PCT, RFAG and TACEG, respectively. Comparing the studies with regards to their OS, 1-, 2-, 3-year survival rates, HGCT had the highest number of related studies and number of patients, while RFAG had the least number of patients and TACEG had the least number of related studies. The contribution plot is presented in Figure 3. Ten comparisons were made in the network analysis. All of them are mixed comparisons. In the overall contribution of network analysis, the remarkable influence evidence in the comparisons of 1-, 2-, 3-year survival rate is PCT vs. TACEG (19.9%), HGCT vs. RFAG (25.8%), GCT vs. PCT (22.2%), respectively.
There was no inconsistency between direct and indirect point estimates. In our network, there were 5 closed loops (Supplementary Figure 1). All confidence intervals for inconsistency factors (IFs) were compatible with zero inconsistency (IF=0) for all study outcomes (Supplementary Figure 1).

Network Comparison
The summary effects with 95% CI are shown in Figure 4. In the comparison of the 5 treatments for 1-year survival rate, HGCT and RFAG were found to be more effective than GCT and PCT, respectively. GCT and TACEG was found to be more effective than PCT while there was no difference between HGCT and RFAG ( Figure 4A). In the comparison of 2-year survival rates, HGCT and RFAG were found to be more effective than GCT and PCT, respectively. Other comparisons did not exhibit any significant differences ( Figure 4B). In the comparison of 3-year survival rate, HGCT and RFAG were found to be more effective than GCT and PCT, respectively. GCT and TACEG was found to be more effective than PCT while there was no difference between HGCT and RFAG ( Figure 4C).

Subgroup Analysis of Network Comparison in Asian Population
The summary effects in Asian population with 95% CI are shown in Figure 6. In the comparison of the 5 treatments for 1-, 3-year survival rate, HGCT and RFAG were found to be more effective than GCT and PCT, respectively. TACEG was found to be more effective than PCT while there was no difference between GCT and PCT ( Figures 6A, C). In the comparison of 2-year survival rates, HGCT and RFAG were found to be more effective than GCT and PCT, respectively. Other comparisons did not exhibit any significant differences ( Figure 6B). The SUCRA value rankings of 1-year survival rate were HGCT, RFAG, TACEG, GCT, and PCT ( Figure 6D). The SUCRA value rankings of 2-, 3year survival rate were RFAG, HGCT, TACEG, GCT, and PCT ( Figures 6E, F).

Publication Bias
The funnel plot for network meta-analysis is presented in Figure 7. In general, all the selected studies were symmetrically distributed between the vertical line (x = 0). Therefore, there was no noteworthy publication bias in our network meta-analysis.

DISCUSSION
In this network meta-analysis, we revealed that HGCT and RFAG and has remarkable survival benefits for GCLM patients when compared to TACEG, GCT and PCT. By OS and 2-, 3-year survival rate analysis, RFAG was identified as the best option, followed by HGCT, TACEG, GCT and PCT. By 1-year survival   (59) reported that gastrectomy or gastrectomy plus hepatectomy in GCLM patients has survival benefits when compared to chemotherapy. Tsujimoto et al. (60) showed that the 5-year survival rate of GCLM patients after hepatic resection was 31.5%, median survival time was 34 months. They also found that gastric tumor less than 6 cm and D2 lymphadenectomy were important factors for prognosis. Song's study (61) suggested that surgical hepatic resection is beneficial for long-term survival in selected patients, with a 3-year survival rate of 47.6%. Groundbreaking by survival benefits of combined conversion therapy with surgery in patients with colorectal cancer liver metastases, numerous general surgeons navigated HGCT or RFAG in GCLM, which was thought over as a crucial strategy to alleviation disease and to prolong patient life (62)(63)(64)(65)(66)(67)(68)(69)(70)(71)(72)(73)(74)(75)(76)(77)(78)(79). Liverdirected treatment (LDT) options for GCLM patients and surgical treatments were gradually attempted (12). If complete resection of liver metastases is possible, considering adequate hepatic reserve and surgical security, radical operations for primary gastric cancer and liver metastases lesions should be attempted (22,79).
Considering the retrospective nature of the included studies and different selection biases for choosing patients on whom to perform radical surgery, their outcomes can hardly be regarded as a rationale in the treatment of GCLM, but it broadens the horizon of radical surgery in the selected GCLM patients. Furthermore, its prognostic value is considerable. Hepatic resection for liver metastases from colorectal cancer has been recommended as a standard treatment, 5-year survival was almost 40% (80). When the number of liver metastasis tumor ≤ 3, the diameter of single metastasis lesion ≤ 3 cm, the resection of primary gastric cancer and liver metastasis can also offer survival benefits in the GCLM patient (40). The security of surgical treatments for GCLM patients has also been confirmed. It does not enhance postoperative mortality (37). Studies also reported that GCLM patients with hepatectomy and   In our network analysis, we adopted several methods to control potential bias. First, the quality of all included studies was assessed by the Newcastle-Ottawa scale. The contribution plot was then performed to seek for significant bias in the network analysis. HGCT and RFAG exhibited the most impact on the 1-, 2-and 3-year survival rates, with 19.6%, 25.8% and 22.2% respectively, which was attributed to the small number of included patients. We also applied the small-study effects to adjust the value of SUCRA to control for potential bias. There was a low risk of publication bias.
Our study had some limitations. The retrospective nature of the included studies enhances the possibility of selection bias between different centers. Patient characteristics such as the number and size of hepatic metastasis, the location of metastasis lesions, the postoperative supportive treatment and adjuvant chemotherapy, which are vital prognostic factors to influence the survival benefits in GCLM patients could hardly ensure balance. However, it is difficult to perform prospective cohort studies for this group of patients die to the small number of GCLM patients in single centers and dismal prognosis with systemic chemotherapy. Our results recommend the HGCT or RFAG treatment option for GCLM patients when resection of gastric cancer and liver metastases lesions is feasible. This recommendation is in tandem with those of the EORTC and JCOG studies. Liver resection or RFA is a favorable option for GCLM patients without extrahepatic metastases, peritoneal dissemination and multiple hepatic metastases (22). Meanwhile, the maximum liver metastatic tumor size for which RFA is safe and effective remains highly controversial (55,88).
To sum up, HGCT was found to exhibit superior therapeutic effects for GCLM patients while RFAG was found to be a prospective therapeutic alternative. Although we obtained data  from retrospective studies, we confirmed the role of RFAG and HGCT as a therapeutic option for GCLM. Large-scale prospective studies in multiple centers are needed to further evaluate the survival benefits of potential radical surgery or RFAG in selected patients.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.