Metformin use and the Risk of Gastrointestinal Malignancies in Diabetic Populations: A Meta-Analysis

Diabetes Mellitus (DM) is a chronic disease which affects approximately 10.5% of the United States population to date, and has been recognized as a risk factor for the development of a variety of malignancies [1,2]. A recent review of observational studies on cancer incidence and Type 2 Diabetes (T2DM) found a signifi cant association between T2DM and cancers of the breast, bile duct, colon, and endometrium, but revealed the strongest associations between T2DM and hepatocellular and pancreatic cancer risk [2]. Various mechanisms have been hypothesized to explain the association between diabetes and Abstract


Introduction
Diabetes Mellitus (DM) is a chronic disease which affects approximately 10.5% of the United States population to date, and has been recognized as a risk factor for the development of a variety of malignancies [1,2]. A recent review of observational studies on cancer incidence and Type 2 Diabetes (T2DM) found a signifi cant association between T2DM and cancers of the breast, bile duct, colon, and endometrium, but revealed the good safety profi le, weight neutrality or weight loss, lack of associated hypoglycemia, reduced cardiovascular mortality and low cost [5]. Metformin normalizes serum glucose levels by reducing gluconeogenesis and improving insulin mediated glucose uptake by liver and muscle cells thus increasing insulin sensitivity and decreasing serum insulin levels [6]. It accomplishes this through the Adenosine Monophosphate-Activated Kinase (AMPK) signaling pathway [6]. Generally, metformin inhibits mammalian Target of Rapamycin (mTOR) activity by activating ATM (ataxia telangiectasia mutated), Liver Kinase B1 (LKB1) , and then AMPK, which ultimately prevents protein synthesis and cell growth [6]. Aside from metformin's role in glucose control, studies have shown that it might also reduce the risk and mortality of certain malignancies, and improve diabetics response to certain cancer treatments [4].
On the contrary, exogenous insulin and insulin secretagogues such as sulfonylureas and meglitinides have been thought to increase the risk of malignancy in diabetic patients [7]. This association, however, has not been wellestablished with evidence from meta-analysis conducted on randomized controlled trials showing no evidence of association of sulfonylurea and the risk of malignancy at any site [7].
Although their exact mechanism is not fully understood, the effects of high serum insulin levels are predicted to contribute to increased malignancy in diabetes. This hypothesis may also explain why metformin, an insulin sensitizer, thereby reducing serum insulin levels, has been found to be associated with a decreased cancer risk [2].
Based on current literature, metformin has been shown to be associated with a decreased risk of pancreatic, breast cancer and malignancies overall [2]. Additionally, a retrospective study conducted by our group, from 2002-2014, by Zhu, et al. (2017) revealed that while the diagnosis of diabetes had no impact on CRC survival or disease, diabetic patients on metformin had signifi cantly increased survival when compared to diabetic without metformin use and non-diabetic populations [8].
However, the effect of metformin on colorectal cancer risk has been inconsistent in the literature to date [9]. In this metaanalysis, we aim to assess the effect of metformin on the risk of gastrointestinal cancers, specifi cally pancreatic and colorectal cancers among individuals with T2DM.

Literature search and data sources
An electronic search was performed by two investigators using the PUBMED and Cochrane library databases at the library of the University Hospital of Brooklyn. We used a time (from 2000 to 2018) and language restricted to English. Utilizing this strategy, we were able to identify 2,258 articles ( Figure   1). Three sets of search terms were used to ensure an adequate and comprehensive literature review. Search terms utilized included "Metformin and Cancer risk," "Metformin and pancreatic cancer," "Metformin and colorectal cancer." After title and abstract review, the bibliographies of studies that met our inclusion criteria were examined for suitable additional literature. We excluded studies in which suffi cient information to allow adequate estimation of hazard ratio, relative risk or odds ratios, and 95% confi dence interval was not available. We excluded studies that were not published to maintain a highquality meta-analysis.

Study selection and quality assessment
The inclusion was limited to studies that were (1) casecontrol in design, (2) performed after 2000, (3) compared cancer odds in diabetic patients on metformin versus those not treated with metformin, (4) cancer sites limited to the colon and pancreas. No study was excluded based on the size of the study population, but duplicate articles were excluded. For each study, we applied the Newcastle-Ottawa Scale to assess its quality. The quality scale included three components: selection of the study groups with points ranging from 0 to 4; compatibility of the study groups, with points from 0 to 2; and ascertainment of study groups exposure, points from 0 to 3. The maximum for each study is 9, studies with less than 5 points carry a high risk of bias and were excluded (N=2).

Data extraction
A total of 2,250 studies were excluded for not meeting the inclusion criteria. Eight (8) studies met the inclusion criteria and were reviewed in full; fi ve (5) case control studies were removed due to unclear fi ndings, and an additional eight cohort studies were removed because the results for their control groups were unclear. The Newcastle-Ottawa Scale for casecontrol studies was used to assess the quality of the selected articles.

Statistical analysis
The meta-analysis was performed using Review Manager, version 5.3 (Cochrane Collaboration). Mean differences were evaluated along with 95% confi dence intervals (95% CIs). Each included cancer was summarized using the random-effects model. Heterogeneity, Cochran's Q, tau-squared test, and I2 index were assessed for each study.

Results
A total of 8 studies with 487,591 subjects were included in our meta-analysis; 484,368 subjects were included in the analysis of colorectal carcinoma (CRC) incidence, and 3,223 in the analysis of pancreatic carcinoma incidence. All subjects included were previously diagnosed with T2DM and were age and sex matched. All studies included were case-control in study design. Four studies were from the United States, two from the United Kingdom, one from Korea and Denmark respectively. Some of the following subject qualities were excluded from the studies: prior diagnosis of PCOS, Infl ammatory Bowel Disease (IBD), type 1 diabetes, prior cancer, and pancreatic disease. The key points of the individual studies are represented in (Table 1).
Our meta-analysis showed that metformin use did not have any effect on the risk of pancreatic cancer (OR .98; 95% CI 0.82-1.17, p=. 83) in the diabetic population (Table 2A).
Metformin was associated with a decreased risk of colorectal cancer; (OR .84; 95% CI 0.81-0.87, p < 0.01) in patients with diabetes (Table 2B). Table 2A shows association of metformin exposure with pancreatic cancer risk. Table 2B shows association of metformin exposure with colorectal cancer risk. Solid square represents the odds ratio for the risk of cancer on diabetic patients who were exposed to metformin. Solid diamond represents point estimate and confi dence interval after analyzing and averaging all combined individual studies.

Metformin and pancreatic cancer
Pancreatic cancer is the fourth most common cause of death from malignancy in the United States [10]. T2DM has been implicated as a predisposing factor for Pancreatic Cancer (PC) along with cigarette smoking and obesity [11]. In fact, those with T2DM are at 1.5-2.0 times increased risk of developing PC [11]. Although an association between T2DM and PC has been established, the causality of the relationship is complex because while T2DM has been identifi ed as a risk factor for PC, it can also be seen as an early manifestation of PC. Individuals who are diagnosed with DM within two years of PC diagnosis are often considered to have 'secondary diabetes' or pancreatic cancer-associated diabetes in which the diabetes developed as a result of PC [11]. It has been hypothesized that precancerous pancreatic cells are incapable of producing insulin, which might contribute to this phenomenon [12]. The prevalence of diabetes and impaired glucose tolerance in individuals with PC has been reported to be as high as 50-80% [11]. However, given that there are no clinical or laboratory methods to accurately determine exact timing of the onset of diabetes or distinguish T2DM from diabetes caused by PC, many of these cases are likely to be misclassifi ed.
Despite the fact that DM can result from PC itself, there is still evidence indicating that long term DM is an independent risk factor for PC [11]. This is likely due to the insulin resistance and resulting hyperinsulinemia and elevated levels of insulinlike growth factors (IGFs) characteristic of individuals with T2DM. Insulin, a growth promoting hormone, works to promote cellular proliferation and increase glucose consumption.
Both of these actions are needed for cancer cell growth and proliferation. Additionally, islet cell turnover, characteristic of individuals with T2DM and insulin resistance promotes pancreatic carcinogenesis [11]. IFG-1, has a tumor promoting effect and is overexpressed in pancreatic cancer cells [13].
As previously mentioned, the use of antidiabetic medications may play a role in the association between T2DM and PC. Treatments such as exogenous insulin and sulfonylureas, which induce high levels of systemic insulin, have been thought to increase risk of PC, but this has not been confi rmed, with evidence providing inconclusive results [10] .
On the other hand, Metformin, has been shown to have both a protective effect on PC risk and survival among diabetics with pancreatic tumors [7,14]. Various mechanisms have been proposed to explain metformin's protective effect on PC, the majority of which center around its ability to increase insulin sensitivity thereby reducing plasma insulin and IFG-1 levels and normalizing the rate of islet cell turnover [10]. Metformin also stimulates AMP-activated protein kinase (AMPK), which inhibits mTOR, a signaling molecule that regulates cell growth and cycling, including pancreatic cancer cells [15].   Table 2A shows association of metformin exposure with pancreatic cancer risk. Table 2B shows association of metformin exposure with colorectal cancer risk. Solid square represents the odds ratio for the risk of cancer on diabetic patients who were exposed to metformin. Solid diamond represents point estimate and confi dence interval after analyzing and averaging all combined individual studies.  [10,13]. Individuals with a new PC diagnosis within that time period were excluded in order to control for reverse causality due to DM cases, which resulted from PC [10,13]. In contrast, the study conducted by Lu and colleagues in 2015, did not impose a lag-period, and revealed a positive association of anti-diabetic medication use with a risk of pancreatic cancer [19]. This association cannot be differentiated from the positive association between diabetes and pancreatic cancer which might have introduced both time related bias and misclassifi cation bias into their study [19].
Duration of DM diagnosis is important to consider because longer duration of DM often infers greater disease severity, which has been hypothesized to increase one's risk for PC [11].
In fact, the case-control study conducted by Lu and colleagues revealed that individuals with elevated HbA1c levels were at three times greater risk of developing PC [19]. However, despite this signifi cance, the other studies in our analysis did not account for DM severity or HbA1c levels. Therefore, in order to maintain homogeneity among studies included, we did not account for this factor, which is one of the major limitations of our analysis.
The dose and duration of metformin exposure could be an independent factor for PC risk, but it could also correlate with disease severity. The case-control study conducted by Walker, et al. adjusted for diabetes duration, and stratifi ed by metformin duration in months [13]. However, stratifi cation of participants by never-ever metformin use or duration of metformin exposure did not reveal a statistically signifi cant relationship between metformin and PC risk [13]. On the other hand, the study conducted by Li and colleagues revealed a statistically signifi cant reduction risk of PC in metformin users, which was greater with increased duration, specifi cally those with >5 years of use [10]. While some of the studies included in our meta-analysis did in fact calculate ORs stratifi ed by metformin exposure duration or dosages, we decided to limit our study to ratios calculated on ever or never use in order to maintain homogeneity among studies. This might have caused a source of error in our analysis as well.
Our analysis has strengths and limitations, which are important to note. One limitation is the possibility that the studies included in our analysis are too heterogenous to provide reliable pooled data. This heterogeneity in part can be explained by differences in each of the studies analytic approach and study design, which will be discussed later on.
Additionally, it is important to note that when compared to the analysis of the effect of metformin use on CRC risk, that of PC risk included a notably smaller sample size.
Another limitation of our meta-analysis is that we used unadjusted ORs for our analysis, and thus did not adjust for potential confounding variables, specifi cally those known or suspected to predispose to PC, despite the fact that the majority of the studies included did adjust for a number of these potential confounding variables in their analysis in order to maintain heterogeneity among studies.

Metformin and colorectal carcinoma
Colorectal Cancer (CRC) is the third major cause of cancerrelated deaths worldwide, with a large geographical variation in incidence and mortality across the world. According to recent literature, individuals with type II diabetes are at an increased risk of developing colorectal cancer when compared to non-diabetics [20]. Hyperglycemia, insulin resistance, and the resulting hyperinsulinemia and advanced glycation end-products are thought to lead to CRC through the direct stimulation of colorectal tumor cell growth, DNA synthesis and enhanced resistance to apoptosis [20,21]. Additionally, decreased peristalsis and constipation characteristic of patients with diabetes mellitus increases exposure to bowel toxins including potential carcinogens which might also contribute to carcinogenesis [12].
As discussed previously, metformin improves glucose preclinical models [23]. In addition, metformin plasma and tissue concentrations were closely correlated, suggesting that the exchange between blood and tissue occurs rapidly [23].  [28][29][30]. In contrast, Demb, et al. addressed specifi c anatomic sites and found that risk reduction associated with metformin exposure was mainly limited to rectal cancer [31]. Of note, there is little to no published literature that is currently capable of separating out site-specifi c effects and no mechanistic studies, which could explain a site-specifi c effect of metformin on risk of developing rectal cancer.
Similar to the point made in the discussion of pancreatic cancer, another important limitation to consider is the duration and dosage of metformin use. For instance, a major limitation of the study conducted by Demb, et al. is that they examined metformin as a binary single time exposure such that duration and accumulation were not considered; all of this may underestimate the true protective effect of metformin on CRC [31]. The study conducted by Shin, et al. did not account for dosage of metformin, but did include duration of metformin use and distributed the results as 90-180, 180-360,360-730, and 730> days. Sehdev, et al. accounted for both duration and dosage of metformin use but no signifi cant dose response relationship was found between metformin dose, duration or total exposure (dose × duration) and odds of developing CRC [28]. Finally, the study conducted by Cardel et al. showed a statistical signifi cant dose response association between the use of metformin and a reduced risk of CRC; but the protective effect of metformin ceased after nine years of treatment [29].

Conclusion
Our analysis revealed that metformin use in patients with T2DM signifi cantly decreases the odds of developing CRC, but did not demonstrates an association between metformin use and PC odds. The proposed role of metformin in its protection against CRC is multifactorial and includes both insulindependent and insulin-independent mechanisms. This supports the use of metformin in diabetic patients, particularly those with risk factors for CRC such as those with hereditary CRC syndromes, personal or family history of sporadic CRC or adenomatous polyps, infl ammatory bowel disease, and obesity. Further investigations, especially large-scale clinical studies and experiments at the cellular level are needed to confi rm the relationship between colorectal cancers and metformin treatment. Given that the pancreas is the center of insulin production, the effect of metformin on pancreatic cancer odds/risk in people with type II diabetes is more complex, as pancreatic cancer has been shown to be a risk factor for developing diabetes and vice versa. Although the mechanisms by which metformin infl uences tumor growth have recently been reviewed, it is uncertain whether the use of metformin can reduce the odds and/or risk of developing pancreatic cancer. Future research should focus on exploring the effect of metformin on non-diabetics and those with pre-diabetes (impaired fasting glucose or impaired glucose tolerance) as well. Moreover, it would be helpful to conduct future research on particular tumor biomarkers in order to elucidate which patients and types of cancers would be most responsive to metformin therapy for the prevention of gastrointestinal cancers.