Abstract
Previous studies demonstrate mixed evidence regarding the association between metformin and skin cancer risk. To synthesize prior evidence and evaluate the association between metformin and skin cancer risk in patients with diabetes/prediabetes, we conducted a meta-analysis. A systematic literature search was performed up to March 23, 2020 to identify randomized controlled trials (RCT) and observational studies of metformin that reported any event of squamous cell carcinoma (SCC), basal cell carcinoma (BCC), and melanoma. In a meta-analysis of 6 trials involving 8,541 patients (Peto method), compared with controls, metformin was not significantly associated with decreased risk of melanoma [OR, 0.82; 95% confidence interval (CI), 0.27–2.43], BCC (OR, 0.75; 95% CI, 0.36–1.57), SCC (OR, 0.98; 95% CI, 0.06–15.60), total nonmelanoma skin cancer (NMSC; OR, 0.69; 95% CI, 0.38–1.24), or total skin cancer (OR, 0.71; 95% CI, 0.42–1.20). This nonsignificant association pattern was consistent with the random-effects meta-analysis of 4 cohort studies with 354,746 patients (melanoma: RR, 0.91; 95% CI, 0.62–1.33; NMSC: RR, 0.65; 95% CI, 0.35–1.18; total skin cancer: RR, 0.83; 95% CI, 0.59–1.16). In conclusion, meta-analyses of both RCT and cohort studies showed no statistically significant association between metformin and skin cancer risks, although suggestive evidence of modestly reduced risks of skin cancer among metformin users was observed. Further studies are needed.
Prevention Relevance: Meta-analyses of RCT and cohort studies showed no significant association between metformin and skin cancer, although suggestive evidence of modestly reduced skin cancer risks among metformin users was observed. These findings suggest metformin use should not influence current medical decision making for diabetes patients at risk of developing skin cancer.
Introduction
Metformin is the first-line treatment for type 2 diabetes mellitus (T2DM), significantly reducing hyperinsulinemia and hyperglycemia (1). Because T2DM has been associated with increased risk for numerous cancers, including colorectal, breast, endometrial, and pancreatic cancers, there is interest in the potential use of antidiabetic treatments for cancer prevention (2). Data from cohort studies and meta-analyses suggest that metformin may reduce cancer risk (3), although findings have been inconsistent (4). Proposed mechanisms for anti-carcinogenic effects of metformin include activation of the LKB1/AMP-activated kinase (AMPK) pathway, decreased insulin and insulin growth factor-1 signaling, and reduction of endogenous reactive oxygen species formation and associated DNA damage (4, 5).
Prior work has identified obesity and metabolic syndrome as potential risk factors for melanoma and non–melanoma skin cancer (NMSC) (6, 7). However, few studies have investigated the relationship between metformin and skin cancer risk, and their association remains unclear. Because skin cancer incidence is rising, there is a need for preventive therapeutics. Metformin may have benefit due to its proposed anticarcinogenic effects (6, 7).
To date, several observational cohort studies in California, Taiwan, and the United Kingdom have examined the impact of metformin on both melanoma and NMSC risk. These studies have demonstrated mixed findings of both increased and decreased skin cancer risk (8–10). In addition, they were limited by sample size and lack of participant diversity. Given the fewer number of melanoma cases, it is also difficult to ascertain melanoma risk from the results of individual randomized controlled trials (RCT). Thus, we conducted a meta-analysis of multiple RCTs and observational cohort studies to comprehensively synthesize evidence with the goal of quantitively summarizing any association between metformin and risk of skin cancer, including squamous cell carcinoma (SCC), basal cell carcinoma (BCC), and melanoma, in patients with T2DM and prediabetes.
Methods
This meta-analysis was conducted according to the standards for systematic reviews and meta-analyses outlined by the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statements (11).
Search strategy and study selection
PubMed, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL) were searched from inception to March 23, 2020, using relevant terms without language restrictions. These phrases included metformin, skin cancer, melanoma, basal cell carcinoma, and squamous cell carcinoma. ClinicalTrials.gov was searched to identify eligible trials. Two reviewers (K. Y and J. X) independently selected studies according to the following inclusion criteria: (a) RCTs that included adult patients with T2DM or prediabetes; (b) metformin compared to placebo or other oral hypoglycemic agents regardless of background treatments; (c) explicitly reported event data on melanoma, BCC and/or SCC as adverse outcomes; (d) observational studies of metformin and skin cancer risk.
Data extraction
Two reviewers (K.Y. and J.X.) independently extracted the following information: first author (publication year), study characteristics (study design and population, sample size, data source of outcome, ClinicalTrials.gov registry number, region, treatment, maximum daily dose of metformin, duration of follow-up), patient characteristics (age, sex, race), and outcomes of interest (melanoma, NMSC, SCC events). Numbers of skin cancer events were extracted from publications or the “Serious Adverse Events” section on ClinicalTrials.gov when there was no relevant publication. When more than two arms were included in a trial, each group was considered as a separate comparison.
Risk of bias and quality assessment
Cochrane Collaboration's tool was used for assessing risk of bias and the quality of individual RCTs (12). Two authors (K. Yang and J. Xue) independently reviewed and judged each RCT as having low, high, or unclear risk of bias in terms of random-sequence generation, allocation concealment, blinding of participants and investigators, incomplete outcome data, selective reporting, and other bias, if any.
The Newcastle-Ottawa Scale (NOS) was used to assess the quality of observational studies (13). The scoring system categorized 8 aspects of each study, including representativeness of the cases, selection of controls, ascertainment of exposure, outcome of interest not present at the start of the study, comparability of cases and controls on the basis of the design or analysis, assessment of outcome, follow-up long enough for outcomes to occur, and adequacy of follow up of cohorts. Two authors (K.Yang and J. Xue) independently reviewed and judged each study on the criteria for selection, comparability, and outcome per manual instructions.
Statistical analysis
A meta-analysis of included RCTs was performed to calculate the pooled ORs with 95% confidence intervals (CI) between metformin and controls for risk of melanoma, BCC, SCC, NMSC, and total skin cancer, respectively. Given low event rates of these outcomes, Peto method was used (14). A random-effects model was used to obtain pooled estimates of cohort studies and RRs with 95% CIs were reported. We assessed heterogeneity across studies using I² statistic (<25%, 25%–75%, and >75% indicating low, moderate, and high levels, respectively; ref. 15), as well as Cochran Q statistic (16). Two prespecified sensitivity analyses were performed to examine the robustness of the estimations, including removing one trial (NCT01804049) that had patients with prediabetes (17) and one observational study that included recurrent NMSC cases (18). Prespecified subgroup analyses of RCTs were further performed according to: type of control (placebo or other drugs), maximum daily dose of metformin (< 2,000 mg, 
2,000 mg), duration of trial (<2 years, 
2 years), baseline age on average (<55 years, 
55 years), sex (<50%, 
50% female), race (<80%, 
80% white). Considering the small number of included observational studies (n = 4) and the unavailability of some study characteristics, subgroup analyses were only conducted in the meta-analysis of RCTs. The potential for small-study effects (publication bias) for total skin cancer was tested using Begg test and funnel plots (19, 20). Statistical analyses were performed with STATA (Version 15; Stata Corp).
Results
Study selection and study characteristics
We retrieved 1,085 records through electronic databases and ClinicalTrials.gov. Of 23 eligible articles that underwent review, 6 RCTs and 4 cohort studies were included in this meta-analysis (Fig. 1; refs. 8–10, 19–29). Regional distribution encompassed the United States, United Kingdom, Taiwan, and international multicenter trials. Basic characteristics of included studies are presented in Table 1. In RCTs, 8,541 patients with diabetes/prediabetes (1.3%–55.2% female) received either metformin or control treatments (placebo or other active antidiabetic therapeutics). The median trial duration was 108 weeks (range: 24–312 weeks). Patients were predominantly middle-aged (median = 57.0 years) and white (5/5 trials, where reported). Patients were prescribed a median maximum daily dose of 2,000 mg of metformin (range: 500–2,550 mg). Measured outcomes included melanoma (5 RCTs with 6 comparisons), BCC (2 RCTs with 3 comparisons), SCC (2 RCTs with 2 comparisons), and total NMSC (4 RCTs with 5 comparisons). In the meta-analysis of observational data, 354,746 patients with diabetes were included from 4 cohorts, among which 3 reported melanoma outcomes and 2 reported NMSC outcomes.
Flow diagram of study identification, selection and inclusion.
Characteristics of included studiesa.
Risk of bias and quality of studies
Results of bias assessment of included studies are presented in Supplementary Table S1. Of the RCTs, 5 of 6 (83%) featured adequate random-sequence generation and 3 of 6 (50%) had adequate allocation concealment. Moreover, 4 of 6 (67%) confirmed that the trial was double-blinded and the outcome assessment was also blinded. All other trials were otherwise classified as unclear risk in these categories. All trials were free of bias from incomplete outcome data and selective reporting. For cohort studies, in terms of selection, 4 of 4 (100%) reported adequate representativeness of the exposed cohort and selection of the nonexposed cohort. All studies met the full criteria for ascertainment of exposure and ensuring the outcome of interest was not present at the start of the study. Three of 4 (75%) of cohort studies were strongly comparable on the basis of design or analysis. Finally, for the quality of the outcome, all studies had sufficient assessment of outcomes, duration of follow-up and adequacy of cohorts.
Primary analyses
Data on the association of metformin with melanoma and NMSC risk was available from 6 RCTs (Fig. 2; refs. 20–28). The crude event rate of melanoma was 0.12% (6 events/4,811 patients) in the metformin group and 0.15% (7 events/4,750 patients) in the control group. The crude event rate of BCC was 0.39% (12 events/3,090 patients) and 0.52% (16 events/3,073 patients) in the metformin and control groups, respectively. The crude event rate of SCC was 0.32% (1 event/312 patients) and 0.33% (1 event/304 patients) in the metformin and control groups, respectively. Total NMSC crude event rate was 0.41% (18 events/4,342 patients) and 0.60% (26 events/4,304 patients) in the metformin and control groups, respectively. Finally, the total skin cancer crude event rate was 0.47% (24 events/5,123 patients) and 0.65% (33 events/5,054 patients) in the metformin and control groups, respectively.
Meta-analysis (Peto method) of metformin and skin cancer risk based on RCTs.
Individually, metformin was not significantly associated with decreased risk of melanoma (OR, 0.82; 95% CI, 0.27–2.43), BCC (OR, 0.75; 95% CI, 0.36–1.57), or SCC (OR, 0.98; 95% CI, 0.06–15.6), compared with the control groups, with no between-trial heterogeneity for melanoma and BCC (P = 0.48, I2 = 0.0% for melanoma; P = 0.46, I2 = 0.0% for BCC) and moderate between-trial heterogeneity for SCC (P = 0.16, I2 = 50.5%). In addition, these meta-analyses are consistent when examining for summative skin cancer risk, with a null association for total NMSC (OR, 0.69; 95% CI, 0.38–1.24) and total skin cancer (OR, 0.71; 95% CI, 0.42–1.20) relative to controls, with low between-trial heterogeneity for both (P = 0.27, I2 = 22.5% for total NMSC; P = 0.25, I2 = 22.0% for total skin cancer; Fig. 2). Moreover, data on the association of metformin with risk for melanoma and NMSC was available from 4 cohort studies including 354,746 patients (Fig. 3; refs. 8–10, 21). The RRs (95% CIs) of melanoma, NMSC and total skin cancer were 0.91 (0.62–1.33), 0.65 (0.35–1.18), and 0.83 (0.59–1.16), respectively, with moderate between-study heterogeneity (P = 0.16, I2 = 45.5% for melanoma; P = 0.12, I2 = 58.5% for NMSC; P = 0.06, I2 = 60.6% for total skin cancer).
Meta-analysis (random-effects) of metformin and skin cancer risk based on observational cohort studies.
Subgroup and sensitivity analyses
Subgroups analyses were performed according to selected characteristics of the individual RCTs and participants. Overall, the reduced risks of total skin cancer, melanoma and NMSC associated with metformin were not statistically significant in any subgroups. The association pattern was in general similar when analyzing by treatment comparison group (placebo, sitagliptin, rosiglitazone/thiazolidinedione, and gibenclamide), maximum metformin dose/day (<2,000 mg, ≥ 2,000 mg), RCT duration (<2 years, ≥2 years), age (<55 years, ≥ 55 years), sex (<50%, ≥50% female), or race (<80%, ≥80% white) of participants. There was low-to-moderate heterogeneity between study among the subgroup meta-analyses (Table 2).
Subgroup analysis according to selected characteristics of RCTs and participants.
Overall, risk estimates determined by the sensitivity analysis of RCTs were largely similar to those of the primary analysis, even when factoring the removal of one trial (NCT01804049) that had patients with prediabetes and reported melanoma outcomes (17). The estimates for melanoma risk in the sensitivity analysis of the RCTs were similar to those in the primary analysis (OR, 0.71; 95% CI, 0.23–2.21), as were the estimates for total skin cancer (OR, 0.69; 95% CI, 0.41–1.17). Risk estimates from the sensitivity analysis of the cohort studies were also nearly identical to those of the primary analysis, even when accounting for the removal of one cohort study that assessed risk of recurrent NMSC among patients with diabetes who had a prior SCC or BCC (18). Risk estimates for total skin cancer remained largely similar (RR, 0.81; 95% CI, 0.53–1.24), relative to primary analyses.
Publication bias
No evidence of publication bias was observed based on funnel plots for meta-analyses of metformin and total skin cancer based on RCTs (P = 0.90) and cohort studies (P = 0.73; Supplementary Fig. S1).
Discussion
We completed a meta-analysis of 6 RCTs involving 8,541 patients and 4 cohort studies involving 354,746 patients to comprehensively assess the risks of skin cancer associated with metformin use. Our meta-analysis did not find a significant association between metformin and skin cancer risk in patients with T2DM or prediabetes, although suggestive evidence of modestly reduced risks of skin cancer among metformin users was observed in meta-analyses of both RCTs and cohort studies. Further analysis delineating skin cancer type demonstrated no significant association with metformin use.
There are proposed biological mechanisms underlying the potential chemopreventive role of metformin in skin cancer. For example, a preliminary study found that mice fed with metformin experienced inhibited development of SCCs in a dose-dependent manner (30). This effect on skin tumor development was associated with a reduction in tetradecanoylphorbol-13-acetate (TPA)-induced epidermal hyperproliferation, as well as increased AMPK activation and reduced signaling through the mTOR complex (mTORC)-1 and ps70S6K (30). However, in humans, observational studies and RCTs demonstrate mixed evidence on how metformin may impact skin cancer risk at the population level, and our null findings reflect these results (2, 8–10).
The null results of this study are clinically relevant. Both metformin and skin cancers are highly common events in the general population. Metformin is the first-line therapeutic for T2DM and polycystic ovarian syndrome (31), and now increasingly used in patients with prediabetes (32). Skin cancer is also the most common cancer in the United States (33, 34). Thus, metformin users and patients with skin cancer encompass such a large population that teasing out any association, even with a relatively small effect size, would have a tangible impact on public health outcomes.
Our study has several strengths. First, we conducted a comprehensive search strategy to identify all available RCTs and cohort studies, thereby maximizing statistical power. Our study incorporated an additional 4 RCTs and 3 cohort studies that were not previously accounted for in similar work (4), thereby providing the best evidence on this topic. In addition, we performed multiple prespecified subgroup analyses to ensure no major sources of between-study heterogeneity. Furthermore, we performed multiple sensitivity analyses to test and confirm the consistency and robustness of our findings. To our knowledge, this is the first meta-analysis that specifically addresses the risks of both melanoma and NMSC associated with metformin.
We acknowledge several limitations. First, our estimates may be subject to reporting bias, despite thorough efforts to ensure their integrity, because skin cancer was not a prespecified endpoint in RCTs assessing efficacy of metformin among patients with diabetes. This may have limited the number of identified skin cancer events, which may explain the wide 95% CIs of ORs observed in some trials. Second, unmeasured confounding and confounding by indication may exist in cohort studies such that additional risk factors for skin cancer may differ between metformin users and nonusers, although a list of potential confounders have been adjusted in most of those studies and multivariable-adjusted estimates were used in our meta-analysis. Third, detection bias may exist between exposed/unexposed groups in observational studies. However, this potential risk is minimized in our study as the comparison group consists of patients with diabetes on other antidiabetic medications who likely receive similar monitoring by physicians compared with the metformin group. Finally, although point estimates in our meta-analysis suggested there may be an inverse relationship of metformin and skin cancer risk, the modest number of included studies and their limited duration of follow-up limited our statistical power to detect any significant association. However, our results provided the most comprehensive evidence to date for any possible association of metformin and skin cancer risk.
In conclusion, our meta-analyses based on current evidence of both RCTs and cohort studies show no significant association between metformin and risk for skin cancer. Future investigation through longitudinal post-marketing surveillance of RCTs as well as long-term follow-up of large cohort studies would complement the extensive and most current findings in this meta-analysis.
Authors' Disclosures
R.I. Hartman reports grants from American Skin Association outside the submitted work. No disclosures were reported by the other authors.
Authors' Contributions
M. Chang: Conceptualization, investigation, writing-original draft, writing-review and editing. R.I. Hartman: Conceptualization, investigation, writing-original draft, writing-review and editing. J. Xue: Investigation, writing-review and editing. E.L. Giovannucci: Investigation, writing-review and editing. H. Nan: Investigation, writing-review and editing. K. Yang: Conceptualization, formal analysis, investigation, methodology, writing-review and editing.
Acknowledgments
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Footnotes
Note: Supplementary data for this article are available at Cancer Prevention Research Online (http://cancerprevres.aacrjournals.org/).
Cancer Prev Res 2021;14:77–84
- Received July 14, 2020.
- Revision received August 26, 2020.
- Accepted September 15, 2020.
- Published first September 21, 2020.
- ©2020 American Association for Cancer Research.
References
Volume 14, Issue 1
