Effect of Metformin on the Prognosis of Gastric Cancer Patients with Type 2 Diabetes Mellitus: A Meta-Analysis Based on Retrospective Cohort Studies

Background Metformin is one of the most common drugs for type 2 diabetes mellitus (T2DM) treatment. In addition, metformin intends to have a positive effect on the prognosis of several cancers. However, the therapeutic effect of metformin on gastric cancer (GC) remains controversial. This study explores and updates the therapeutic effect of metformin in GC patients with T2DM. Methods We searched through PubMed, Embase, Web of Science, and the Cochrane Library for relevant articles by July 2022. The relationship between metformin therapy and the prognosis of GC patients with T2DM was evaluated based on the hazard ratio (HR) at a 95% confidence interval (95% CI). Overall survival (OS), cancer-specific survival (CSS), and progression-free survival (PFS) were the primary outcomes analyzed. Results Seven retrospective cohort studies with a combined 2,858 patients met the inclusion criteria. OS and CSS were reported in six studies, and PFS was reported in four studies. Pooled results showed that, compared to the nonmetformin group, the prolonged OS (HR = 0.72, p = 0.001), CSS (HR = 0.81, p = 0.001), and PFS (HR = 0.70, p = 0.008) of the experimental group may be associated with the exposure to metformin. Conclusion Metformin may have a beneficial effect on the prognosis of GC patients with T2DM.


Introduction
Gastric cancer (GC) is one of the most common cancers and the fourth leading cause of cancer-correlated death in the world [1,2]. Every year, approximately 990,000 individuals are diagnosed with GC, resulting in 738,000 deaths. Central and South America, Eastern Europe, and East Asia have the highest incidences of GC [3]. In addition to genetic and dietary factors, helicobacter pylori (H. pylori), a Gramnegative bacteria promotes the development of GC [4]. Surgery combined with adjuvant chemotherapy is the only potentially curative treatment for GC [5,6]. However, the 5year-related survival rate of GC patients after surgery is between 10% and 30% [7]. Consequently, there is an urgent need to explore and develop other universal and efective therapies for GC treatment.
Type 2 diabetes mellitus (T2DM) is the 8 th most prevalent disease globally, and the prevalence rate increased from 2.88% in 1990 to 5.89% in 2019 worldwide [8]. A recent review estimated that in 2012, about 293,000 cancer cases globally were attributed to T2DM [9]. Accordingly, the treatment of T2DM might be benefcial for improving the prognosis and relieving the burden of cancer [10]. Among them, metformin, a frst-line pharmacologic treatment for T2DM, poses anticancer properties in an article which formally proposed that metformin could impair the metabolic plasticity and growth of tumor with the combination of hypoglycemia by modulating the PP2A-GSK3β-MCL-1 Axis at the molecular biochemical level [11][12][13][14]. Studies have shown that metformin has a favorable efect on the prognosis of several cancers including lung cancer [15][16][17], prostate cancer [18,19], colorectal cancer [20,21], breast cancer [19,[22][23][24], and endometrial cancer [25].
Nevertheless, whether metformin could improve the prognosis of GC patients with T2DM still remains controversial. A study claimed that metformin decreases the recurrence, all-cause mortality, and cancer-specifc mortality rates of GC patients with diabetes after gastrectomy, more signifcantly with the increase of dose [26]. Another study stated that the use of metformin improves the overall mortality of GC patients with T2DM, but no association was found in a cancer-specifc survival (CSS) [27]. A recent study revealed that metformin does not improve the prognosis of GC patients with T2DM [28]. Herein, the aim of our study is to render an update on the new data on the therapeutic value of metformin on the prognosis of GC patients with T2DM by exploring and updating correlative research systematically.

Search Strategy.
Tis meta-analysis was conducted in accordance with the principles of Preferred Reporting Items for Systematic Review and Meta-analyses (PRISMA) [29]. We systematically searched through PubMed, EMBASE, Web of Science, Medline, and the Cochrane Library to identify clinical articles from inception to July 2022 on the therapeutic efect of metformin on the prognosis of GC patients with T2DM. Te search terms were as follows: ("Neoplasm, Stomach" OR "Stomach Neoplasm" OR "Neoplasms, Stomach" OR "Stomach Neoplasms" OR "Gastric Neoplasms" [MeSH Terms] OR "Gastric Neoplasm" OR "Neoplasm, Gastric" OR "Neoplasm, Gastric" OR "cancer of stomach" OR "Stomach cancers" OR "Gastric cancer" OR "cancer, gastric" OR "cancers, gastric" OR "gastric cancers" OR "Stomach cancer" OR "Cancer, Stomach" OR "Cancers, Stomach" OR "Cancer of the Stomach" OR "Gastric Cancer, Familial Difuse") and ("Metformin") [MeSH Terms] OR "Dimethylbiguanidine" OR "Glucophage" OR "Metformin Hydrochloride" OR "Hydrochloride, Metformin" OR "Metformin HCL" OR "HCL, Metformin") and ("Prognosis" [MeSH Terms] OR "Prognoses" OR "Prognostic Factors" OR "Prognostic Factor" OR "Factor, Prognostic" OR "Factors, Prognostic" OR "Comes" OR "outcome" OR "outcomes" OR "Survival" OR "Mortality" OR "Overall Survival" OR "OS" OR "objective response rate" OR "ORR" OR "progression-free survival" OR "PFS" OR "time to progress" OR "TTP" OR "disease-free survival" OR "DFS" OR "Recurrence-free survival" OR "RFS"). Studies were searched primarily in English. Relevant articles in references of the selected studies were retrieved as well to avoid omissions. Te searches were conducted independently by two investigators. Divergences would be tackled at the discretion of a third investigator.
Te surplus selection criterion was defned as follows: (1) observational studies on the prognostic efect of metformin on GC patients with T2DM; (2) exposure to metformin for patients in the experimental group and nonmetformin for patients in the control group; (3) available tumor-related survival data. If the data from diferent publications is from the same participation, priority will be given to the most recent or systematic one.
Exclusion criterion was set as below: (1) studies without full text; (2) data results that cannot be counted; (3) protocols without specifc results; (4) studies that explored multiple factors but lacked interesting outcomes.
Te observational outcome indicators of prognosis included in this study were overall survival (OS), progressionfree survival (PFS), and CSS.

Quality Assessment and Data
Extraction. Te quality of the selected studies was evaluated using the Newcastle-Ottawa Scale (NOS), which consists of three components: selection, comparability, and outcome. Selection included four parts: representativeness of the exposed cohort; selection of the nonexposed cohort; ascertainment of exposure; and outcomes not present at the start of the study. Outcomes included three items: assessment of outcomes; length of follow-up; and adequacy of follow-up. Studies with a total evaluation score ≥7 were of high quality [30].
Data extracted from the selected articles included author, year of publication, country of research, sample size, sex, age, body mass index (BMI) of the study participants, tumor node metastasis stage (TNM stage), metformin therapeutic scheme and mode of administration, interventions in the control group, GC therapy, and type of gastrectomy. Any disagreement between the two investigators was resolved through arbitration by a third investigator.

Statistical Analysis.
Data analysis was performed using Stata software, version 12.0. Te association between metformin therapy and the prognosis of GC patients with T2DM was analyzed based on the hazard ratio (HR) at a 95% confdence interval (95% CI) by utilizing a random-efects model to increase reliability, giving consideration to potential heterogeneity between studies: the diference in the dosage of treatment received, diferences at baseline, inconsistent GC staging, and nonuniform treatment of the control groups. Heterogeneity was evaluated using Chisquare test in order to determine if there was a signifcant diference in the selected data. I 2 < 50% indicated low heterogeneity, while I 2 ≥ 50% indicated high heterogeneity [31]. If the number of articles included is more than ten, sensitivity analysis was performed to examine the stability of the pooled results, and Begg's regression test was conducted to examine the publication bias of the results. P value less than 0.05 indicated a statistically signifcant diference.

Search Results.
Te initial literature search from the online databases revealed 24,367 studies. Te additional searches of the reference materials in the retrieved studies generated 36 studies. After removing duplicates, 21,893 studies were retained. After reading the titles and abstracts and removing irrelevant articles, we read the full text of 586 studies for potential inclusion. Of these articles, 579 articles were further excluded: 164 were fundamental research on molecular, biochemical, or animal studies; 361 did not capture relevant outcomes; 46 were nonoriginal articles; fve were duplicates; and three had unavailable data. A total of seven studies [26,27,[32][33][34][35][36] were included for quantitative analysis. Te fowchart of the entire selection process is presented in Figure 1. Table 1 shows the basic characteristics of the seven included studies included in our metaanalysis. Two studies were conducted in China, two in Korea, and one each in America, Belgium, and Sweden. All the included articles were cohort studies published in the last six years involving a total of 2,858 patients. Te metformin exposure group consisted of 1,620 patients, while the control group contained 1,238 patients. A total of 1,141 patients had stage I-II tumors, while 1,134 patients had stage III-IV tumors. All the studies adjust for factors including but not limited to sex, age, BMI, TNM stages, and gastrectomy. Detailed data on the study population are shown in Supplementary Table 1. 3.3. Quality Assessment. Te quality assessment for the selected studies was evaluated using NOS. All the studies had NOS scores ranging from 7 to 9, indicating high quality and low risk of bias. A specifc assessment with perspective aspects is shown in Supplementary Table 2.

Prognostic Analysis.
Six studies reported the HRs of the relationship between metformin and the OS of the GC patients with T2DM. Pooled results reveal that compared with the control group, the OS rate of the metformin exposure group was signifcantly high (HR � 0.72, 95% CI: 0.59-0.87, p � 0.001). Te heterogeneity of the pooled results was signifcant (I 2 � 85.6%). Te detailed results are shown in Figure 2.
Six studies reported the relationship between metformin and CSS in GC patients with T2DM. Te pooled results revealed that the CSS rate was higher in the metformin exposure group than in the control group (HR � 0.81, 95% CI: 0.71-0.91, p � 0.001). Te heterogeneity among the studies was very low (I 2 � 29.8%). Te details are shown in Figure 3.
Four studies reported on the PFS. Pooled analysis indicated that metformin increased the PFS of GC patients with T2DM compared with the control group (HR � 0.70, 95% CI: 0.54-0.91, p � 0.008). However, the heterogeneity among the studies was very high (I 2 � 82.2%) (Figure 4).

Discussion
Metformin, a frst-line treatment for T2DM, has been considered to have an improved efect on the prognosis of cancer patients. However, before this study, the role of metformin in GC remained controversial. Tis study summarized the current data on the efect of metformin on the prognosis of GC patients with T2DM. Our study showed that metformin might be benefcial for the OS of GC patients with T2DM. Since all the included patients had T2DM, the prognostic efect of metformin may stem from the treatment of diabetes. CSS was further analyzed to rule out this potential factor, implying that metformin may have a positive efect on the CSS of GC patients with T2DM. Additionally, we found that metformin might contribute to a longer PFS in GC patients with T2DM.
To explore mechanisms for the infuence of metformin on GC, potential causes were listed as follows: First, T2DM promotes GC progression and related mortality [33]. Terefore, treating T2DM may improve GC prognosis. In T2DM patients, the hyperglycaemic environment provides sufcient energy for the growth and proliferation of GC cells and inhibits apoptosis [37]. A common complication of diabetes mellitus is nerve injury, and persistent hyperglycemia will lead to neuronal damage and promote the perineural metastasis of cancer cells [38]. So metformin reduces hyperglycaemic to limit the proliferation and growth of GC cells by inhibiting hepatic gluconeogenesis and stimulating peripheral glucose uptake. Otherwise, metformin lowers insulin levels in GC patients. Tis inhibits insulin binding to the insulin receptor (IR) [39] and refrains from activating the insulin-like growth factor (IGF) receptor signalling axis [40,41], therefore inhibiting tumor growth.
Second, from the cellular metabolism perspective, metformin inhibits the development of GC cells by inhibiting cellular autophagy, glutaminase activity, mitochondrial function, nicotinamide adenine dinucleotide (NADH) dehydrogenase, and inducing phosphorylation of epigenetic enzymes.
(1) Metformin inhibits cellular autophagy that supports tumor survival and growth by reducing GC intracellular metabolism in the terminal stage [42,43] and protects against hypoxia and nutrient defciency [44]. (2) Metformin impairs glutaminase activity [45] and reduces the proliferation of GC cells. A key metabolic characteristic of many cancer cells is their high dependence on glutamine. (3) Metformin impairs mitochondrial activity, inhibits cellular respiration, and subsequently induces GC cell death [46].
International Journal of Endocrinology (4) Under hypoglycemic conditions, metformin strongly induces cell death by inhibiting the NADH dehydrogenase. Tis breaks the electron transport chain, and the anticancer efect may be greatly enhanced [47].
More importantly, in terms of molecular pathways, adenosine 5′-monophosphate (AMP)-activated protein kinase (AMPK) is a crucial pathway that regulates the development of GC cells. Te AMPK pathway is activated in two ways. First, metformin directly activates AMPK, thereby inhibiting the downstream Akt/mTOR signalling pathway and resulting in the generation of cancer cells [54][55][56].
Terefore, inhibition of the PI3K/Akt/mTOR pathway disrupts cancer cell proliferation [57,58]. Metformin also activates AMPK by inhibiting and impairing the function of mitochondrial complex I [59] and inhibits cellular cell respiration and later triggers GC cell death [46].
Activation of AMPK triggers a series of reactions: (1) the expression of p53, an antioncogene in humans [60]; (2) inhibition of the expression of the fatty acid synthase (FAS) gene, modulating lipogenesis and inhibiting proliferation [57,61], stemness, and chemoresistance [62] of GC cell; and (3) regulation of cyclin D1 and the cyclin-dependent protein kinases p21 and p27, which contribute to their anticancer efects [63,64]. Te antitumor efects of metformin on GC and peritoneal metastases in vitro relies on pathways related to the AMPK pathway [65].
Tis study presented that metformin may improve the prognosis of GC patients through the above mechanisms. Although one study reported no association between the use of metformin and GC mortality in GC patients [66], metformin tended to exert a positive efect at diferent stages of GC. At the stage of gastric disease, metformin can slow down the progression of gastric disease due to H. pylori infection, suggesting that metformin may be benefcial for early gastric disease stage [67]. During the chemotherapy period for GC, metformin, in combination with oxaliplatin could inhibit the proliferation of GC cells and further induce their apoptosis [68]. In a GC cell line model, metformin in coordination with curcumin increased the cytotoxic efects of       International Journal of Endocrinology anticancer drugs on GC cells [69]. Metformin can also replace specifc small interfering RNA (siRNA) to inhibit the migration of human gastric adenocarcinoma (AGS) cell lines, a type of GC cells [70]. Generally, numerous studies showed positive results and demonstrated the benefcial efect of metformin on GC patients. Tis article is an updated meta-analysis on the prognostic impact of metformin on GC patients with T2DM. Te results of our study, including three prognostic indicators-OS, CSS, and PFS, are all statistically signifcant. However, this study has some limitations as well.
First, the pooled results of OS and PFS are of high heterogeneity. As for the reasons, we have considered the following aspects: (1) In terms of exposure time and dose of metformin, the specifc dose and length of exposure time in the experimental group may have infuenced the results. Unfortunately, some articles did not mention the dose of metformin given to the experimental group, and two articles that studied the dose relationship had diferent dose classifcations and exposure time was not mentioned, so we could not conduct an indepth study.
(2) In aspects related to blood glucose, the control of blood glucose by metformin may afect the prognosis of patients. Metformin may have a better efect on controlling the glycemic situation better than other hypoglycemic drugs. Unfortunately, due to the insufcient number of articles and the lack of relevant data, we cannot further study the relationship between the prognosis of gastric cancer patients and the blood glucose of patients, and we cannot exclude the possibility that it is because the metformin group had probably a shorter diabetes duration. If available, insulin resistance and the administration of other hypoglycemic drugs were also supposed to be analyzed.
(3) Eternal time bias may also explain the source of heterogeneity. For patients assigned to the experimental group, if gastric cancer progression or even death occurred before taking metformin, their progression and mortality from gastric cancer would be underestimated, and the prognostic efectiveness of metformin would be overestimated. Terefore, studies need to be designed to ensure that the three time points (follow-up initiation, compliance with inclusion criteria, and exposure allocation) are uniform. Unfortunately, the seven included cohort studies did not provide sufcient data for us to calculate the magnitude of the immortal time bias, so we could not know the specifc impact of the immortal time bias on the pooled results. Terefore, we used a random-efects model to combine efect sizes in the analysis of the data to improve the precision of the estimated confdence intervals and increase the power of the test.
Secondly, in perspective of study regions, three of these seven included articles are from China, two are from Korea, and two are from Europe. So the pooled results tend to show that metformin may improve OS, CSS, and PFS in Asian populations, and more studies are still needed to verify the PFS in European populations.

Conclusion
Te pooled results of this meta-analysis showed that metformin may have a positive efect on the prognosis of GC patients with T2DM. However, further multicenter studies of larger-scale clinical trials are still needed to validate the fndings of this meta-analysis.

Data Availability
Te data supporting the fndings reported in this study are available in the supplementary materials.
International Journal of Endocrinology 7