gms | German Medical Science

Background: Many observational studies linked metformin use in type 2 diabetes to a lower risk of overall cancer, and in meta-analyses, cancer risk was assessed to be about 30% lower in metformin users than in non-users [1], [2], [3]. Moreover, some meta-analyses were done to assess associations between metformin use and incidence of selected cancer sites, and reduced cancer risks were reported for breast, colorectal, pancreatic, lung, hepatocellular, liver, and stomach cancer [3], [4]. Recently, however, shortcomings of these observational studies have been addressed, and some major biases in studies on metformin and cancer risk have been discussed including time-related biases (i.e., immortal time bias, time-lagging bias, time-window bias), prevalent user bias, and confounding by indication. For example, Suissa and Azoulay claimed that in 26 observational studies published until 2011, 23 were afflicted with time-related bias [5].

We aimed to compare the incidence rate of any cancer and some selected cancer sites in metformin, sulfonylurea and insulin users, and to reduce some major biases common in observational studies.

Methods: This study was based on data from the Disease Analyzer database (IMS HEALTH, Frankfurt / Main, Germany). Disease Analyzer allows anonymous access to a nationwide panel of general and internal medicine physicians’ practices and their patients. The data are generated directly from the computers in the physicians’ sites via standardized interfaces and provide daily routine information on patients’ diseases and therapies.

Altogether, the database contains data from 1,143 general practitioner sites in Germany and 253 in the UK. Patients were included into analyses if (1) they had a diagnosis of diabetes, (2) they were 30 to 89 years old at the time of diabetes diagnosis – so diabetes is most likely type 2 diabetes, (3) they had any prescription of diabetes drugs, (4) diabetes drugs were not prescribed prior to the first diagnosis of diabetes, and, (5) they had a diagnosis of invasive cancer neither prior to the first prescription of a diabetes drug nor in the first year after first drug prescription. These inclusion criteria were fulfilled by 22,394 German and by 65,948 British patients who were observed for a mean follow-up of 4.7 years after first antidiabetes medication.

In Cox regression models adjusted for age, sex, country, metabolic factors, diabetes duration, medication and comorbidity, patients who started using sulfonylurea and insulin, respectively, were compared to those who started using metformin (intention-to-treat type analysis), and, additionally, patients with sulfonylurea (and insulin, respectively) monotherapy were compared to those with metformin monotherapy. The initial twelve months of follow-up after first antidiabetes prescription were excluded.

Results: 5,094 (5.8%) incident cases of cancer were identified. Throughout all analyses, hazard ratios were close to the null for comparisons of sulfonylurea and insulin use with metformin use.

In intention-to-treat analyses comparing sulfonylurea to metformin use, hazard ratios were 0.97 (95%-CI: 0.91-1.04) for any cancer, 0.96 (0.78-1.19) for colorectal, 0.99 (0.79-1.25) for lung, 1.00 (0.79-1.26) for breast, and 1.15 (0.94-1.40) for prostate cancer. In intention-to-treat analyses comparing insulin to metformin use, hazard ratios were 0.89 (95%-CI: 0.78-1.02) for any cancer, 1.08 (0.72-1.64) for colorectal, 1.03 (0.65-1.64) for lung, 1.06 (0.69-1.65) for breast, and 0.88 (0.56-1.38) for prostate cancer.

In comparisons of sulfonylurea monotherapy with metformin monotherapy, hazard ratios were 1.00 (95%-CI: 0.89-1.14) for any cancer, 0.94 (0.64-1.39) for colorectal, 1.13 (0.75-1.72) for lung, 1.00 (0.62-1.62) for breast, and 0.87 (0.58-1.30) for prostate cancer. In comparisons of insulin monotherapy with metformin monotherapy, hazard ratios were 0.88 (95%-CI: 0.73-1.06) for any cancer, 0.81 (0.44-1.50) for colorectal, 0.99 (0.52-1.88) for lung, 1.19 (0.67-2.14) for breast, and 0.71 (0.38-1.35) for prostate cancer.

Discussion: For overall cancer and four selected cancer sites, we found no differences in cancer incidence between users of metformin, sulfonylurea, and insulin in the first years after initiation of diabetes drug therapy. A lack of association between metformin and cancer has no consequences for metformin use in patients with diabetes as metformin is well established as an initial glucose lowering drug. However, reduction of cancer risk of about 30% as previously suggested by many observational studies would imply that cancer risk in diabetes could even be brought below the risk level of persons without diabetes, and might suggest that metformin might be cancer protective also in people without diabetes. In the light of the results from RCTs and database studies including the present study, evidence for this assumption is getting weaker.