Metformin Use and All-Cause and Prostate Cancer–Specific Mortality Among Men With Diabetes

Prostate cancer (PC) is the most common male malignancy in the western world and the second leading cause of death.¹ Diabetes is also common and termed “the millennium epidemic.”² Metformin (1,1-dimethylbiguanide hydrochloride) is an insulin sensitizer that belongs to the biguanide oral hypoglycemic family. There is emerging evidence linking metformin use to decreased cancer risk and improved cancer-related outcomes.^3,4

Metformin may influence cancer cells indirectly by decreasing insulin levels or directly by influencing cancer cell proliferation and apoptosis.⁴ Metformin is a potent adenosine monophosphate–activated protein kinase (AMPK) activator.^4,5 Once activated, AMPK inactivates enzymes involved in adenosine triphosphate consumption such as fatty acid and protein synthesis.⁵ Furthermore, AMPK activation inhibits the mammalian target of rapamycin complex 1 pathway and S6K1 phosphorylation implicated in carcinogenesis.⁶ Metformin may also be associated with autophagic cell death.⁷

Studies examining the impact of metformin exposure on PC risk had inconsistent results.^8–10 Because metformin is not believed to influence transformation of benign cells to malignant cells but rather to modulate cellular energy, metformin may have a greater impact on cancer survival than incidence. One study reported a beneficial association between metformin and overall survival¹¹; however, this was a single-institution study and did not measure PC-specific mortality.

Because PC is a slow-growing disease, medication exposures postdiagnosis may have an impact on disease progression and survival and thus may be ideal for secondary prevention strategies. We tested the hypothesis that increasing duration of exposure to metformin after PC diagnosis is associated with lower PC-specific and all-cause mortality among diabetic men.

We conducted a population-based, retrospective cohort study approved by the institutional review boards of Sunnybrook and Princess Margaret Hospitals, Toronto, Ontario, Canada.

During the study period, 105,245 men older than age 66 years were diagnosed with incident diabetes in Ontario. Linkage with the Ontario Cancer Registry yielded 4,736 patients (4.5%) who were later diagnosed with PC. Of these, 3,837 (81%) had pathology reports. The median age at PC diagnosis was 75 years (interquartile range, 72 to 79 years; Table 1). During a median follow-up of 4.64 years (interquartile range, 2.7 to 7.1 years), 1,343 (35%) died, and 291 patients (7.6%) died of PC. At presentation, 976 patients (25.4%) had high-grade tumors (Gleason score ≥ 8), and 2,167 (57%) had high-volume tumors (> 30%).

Overall, 1,251 (32.6%) and 1,619 (42.2%) were exposed to metformin before and after PC diagnosis, respectively (Table 2). Patients were exposed to metformin for a median of 19 months (range, 6.3 to 40 months) before PC diagnosis and 8.9 months (range, 3.6 to 12.3 months) after PC diagnosis. Among metformin users, 53% (n = 858) continued to take metformin once the medication was initiated, whereas 47% (n = 761) of metformin users had periods of nonuse.

Diabetes and PC are common. In the United States, the most commonly diagnosed cancers in men are prostate, lung/bronchus, and colon/rectum.¹ Of the world adult population, an estimated 285 million (6.6%) have diabetes.² In 2007, diabetes prevalence in the United States was 10.7%, with an estimated 1.6 million new cases per year.² Therefore, an increasingly large number of men will be diagnosed with both diabetes and PC. Our population-based study demonstrated that increased cumulative use of metformin after PC diagnosis is associated with a significant improvement in all-cause and PC-specific survival among older men with diabetes and PC. We have shown that for every additional 6 months of metformin treatment, there is a 24% decrease of PC-specific mortality and a significant decrease in all-cause mortality that declines over time.

Most studies of the association of metformin with PC focused on cancer incidence. Wright et al⁸ demonstrated that among whites with diabetes, metformin resulted in a 44% reduced risk of PC. Other larger studies did not report similar findings.^9,10 One meta-analysis²⁶ concluded that current data do not support an association between decreased risk of PC incidence and use of metformin.

Considering the hypothesized biologic mechanisms of the effects of metformin in cancer, it remains plausible that cancer progression may be the most relevant outcome. Furthermore, in the current era, PC incidence is mainly dependent on prostate-specific antigen screening.¹ He et al¹¹ report a single institutional retrospective cohort of 233 patients with PC in whom both thiazolidinedione and metformin exposure were predictors of improved overall survival. However, this study did not report cancer-specific death and did not consider drug exposure as a time-dependent covariate, which may overestimate its effects.²⁷

We used cumulative use of antidiabetic medications as our main exposure. We believe that because metformin may work by preventing progression, simply analyzing whether a patient was exposed to metformin or not may not capture its effect. This method also allows evaluating a dose-response effect, strengthening evidence for a causality.²⁸ Furthermore, our approach incorporated cumulative metformin as a time-varying covariate, thus circumventing potential survival bias. In addition, we analyzed all drug exposures in the same manner. If our results were attributed to longer duration of survival among patients using medications for a longer time, we should have noticed a significant effect for all medications. However, the only antidiabetic drug associated with improved outcome was metformin.

Healthy-user/adherer biases are a major concern in observational studies.^29,30 Our database offers a unique opportunity to minimize these biases. Most diabetes cohorts are captured by using prescriptions.^8–10 Per guidelines for diabetics, metformin is considered first-line therapy.³¹ Thus, beneficial effects of metformin in these databases may be partly due to use of metformin among healthier patients. Because our diabetes database uses diagnostic codes rather than medication to capture diabetes, our cohort has a large proportion of patients (n = 1,702) who are not medically treated but are receiving diet control. Although some of these patients may be untreated because of nonadherence or end-stage cancer, many have early-stage diabetes and may be healthier than even metformin users. In the sensitivity analyses, we demonstrated that metformin use was associated with decreased mortality, even when compared with diet control patients.

We used several other methods to minimize healthy-user effects.²⁵ First, we included patients with incident diabetes who subsequently had PC. Although this restriction limited follow-up time, an incident population compared with a prevalent one is more homogenous, because it is less likely to include patients who tolerate metformin and are adherent to their metformin dosage regimen.²⁵ This argument is even stronger since diabetes itself is correlated with PC.^8–11 Diabetes is associated with a lower risk of PC but higher risk of high-grade disease and mortality. Including a prevalent cohort would make it hard to disentangle the effect of metformin use from that of diabetes. Furthermore, because our drug exposures are modeled as duration of cumulative exposure after PC diagnosis, the comparison is not exclusively between users and nonusers but rather it is between users who have had different durations of cumulative exposure. This should help mitigate the healthy-user effects and other indication biases.

We observed that cumulative use of statins is also associated with decreased mortality (data not shown). Prior studies of the association of statins and PC demonstrate that statin use is associated with decreased advanced and fatal PC.^32,33 Our study was not designed to test the association of statin use and mortality, but we believe that our data add to the evidence that statin use is associated with a reduced risk of PC-related mortality. Unfortunately, the complexity of analysis with cumulative use time-varying covariates precluded us from testing the effect of an interaction between statin and metformin.

Several limitations merit mention. First, because of its observational nature, treatment was not randomly assigned, and differences between individuals prescribed different drugs for differing durations may be related to the outcomes independent of any metformin-modifying effects. Still, our methodology helped minimize many of the potential biases. Second, we used administrative data and were not able to retrieve information on reason for drug discontinuation, severity of diabetes, laboratory data, body mass index, exercise, smoking status, and PC stage. However, our large sample size enabled us to adjust for many other potential confounders. We also acknowledge that excluding patients without pathology reports may have introduced selection bias, and patients without pathology reports may have, on average, a more advanced cancer that was diagnosed clinically. However, the percentage of patients who had pathology reports is similar to that in other studies that used the Ontario Cancer Registry.^34,35 In addition, pathology abstraction allowed us to adjust for important cancer parameters: Gleason score and tumor volume. Furthermore, we performed several sensitivity analyses to address differences in severity of diabetes and cancer stage, which all revealed similar point estimates. Residual confounding still exists because individual-level data on reason for drug choice, personal habits (smoking, diet, exercise), body size, prostate-specific antigen, and PC stage are lacking. Third, because our population was older, all had diabetes and many were treated initially with ADT; thus, generalizability to a contemporary PC cohort is tempered. However, all eight sensitivity analyses demonstrated similar point estimates. Finally, because our cohort was limited to patients with diabetes, we cannot conclude whether similar effects of metformin would be seen in a nondiabetic population. Thus, our study results do not demonstrate a survival benefit for diabetic men who use metformin compared with men who do not have diabetes.

There are several clinical implications of our findings. First, consistent with current guidelines,³¹ metformin should be considered first-line therapy among patients with PC and diabetes, not only for diabetes control but possibly to improve cancer prognosis. Second, we found that metformin was associated with benefit regardless of cancer treatments. These results suggest that metformin may further improve survival as an adjunct therapy, even among those already receiving optimal cancer treatments. Finally, metformin may be ideal for secondary prevention because it is inexpensive, safe, and well tolerated.³¹ We believe that these data can serve as proof-of-concept to design interventional studies of metformin to prevent or delay progression in PC. Indeed, large-scale phase III breast cancer studies are underway.³⁶

There is some evidence to suggest benefits of metformin among patients without diabetes. Metformin has a demonstrable safety profile among nondiabetics and is used in polycystic ovary syndrome³⁷ and nonalcoholic fatty liver.³⁸ In patients who did not have diabetes but who did have breast cancer, metformin decreased tumor proliferation markers.⁵ Metformin may also have other benefits for nondiabetic patients who require ADT through its insulin-sensitizing effects. ADT is commonly used as therapy for men with advanced PC. It can be associated with insulin resistance and metabolic syndrome.³⁹ Recently, a small randomized study demonstrated improvement in metabolic parameters among nondiabetic patients with PC who were randomly assigned to metformin and lifestyle changes.⁴⁰

Our study is among the first to report that cumulative metformin use after PC diagnosis is associated with improved all-cause and PC-specific survival among elderly diabetic men. We believe an interventional study of the use of metformin to delay progression in PC is warranted.

The author(s) indicated no potential conflicts of interest.

Supported by Prevention Grant No. 2011-701003 from the Canadian Cancer Society Research Institute, by the Ontario Ministry of Health and Long-Term Care, by a Canadian Institutes of Health Research (CIHR) and Canadian Patient Safety Institute Chair in Patient Safety and Continuity of Care (C.M.B.), by a Career Investigator Award from the Heart and Stroke Foundation of Ontario (P.C.A.), and by a Canadian Diabetes Association/CIHR-Institute of Nutrition, Metabolism and Diabetes Clinician Scientist Award (L.L.L.).