Conclusions
In this study of 378 253 adults with 21 172 incident cancer events over an average 7 years of follow-up, we found no consistent evidence that higher HbA1c was associated with incident cancer risk among nearly all of the 16 cancer types investigated apart from pancreatic cancer. A novel finding was an inverse association between HbA1c and premenopausal breast cancer, which persisted when people with diagnosed diabetes or on glucose-lowering medications were excluded. This study is the largest study to date to investigate the association between HbA1c and multiple specific cancers adjusting for a wide range of demographic, lifestyle, and clinical factors. There was no suggestion that any observed associations (or lack thereof) with cancer incidence were explained by reverse causality.
We examined the associations between cancer risk and having a diagnosis of type 2 diabetes at baseline and found positive associations with colorectal, pancreatic, uterine and bladder cancer, and a negative association with prostate cancer. This is consistent with previous research, alleviating concerns around UK Biobank participants being healthier than the general population.13 Interpretation of previous studies of individual cancer incidence across the entire glycemic spectrum has been limited by small numbers. Much of our understanding of this association therefore comes from dichotomized comparisons of people with and without diagnosed type 2 diabetes. We were able to take these analyses further, exploring associations across the full glycemic range. Exclusion of people with diagnosed diabetes or prescribed metformin for other indications addresses the uncertainties around effects of glucose-lowering medication.3 For the most part, this had little impact on associations. Thus, apart from known associations with pancreatic cancer,14 we were not able to demonstrate any consistent link between higher HbA1c and risk for cancer that would support a hyperglycemia associated mechanism for cancer risk.
We found a novel inverse association between HbA1c and premenopausal breast cancer, whereby increased HbA1c was associated with decreased risk of cancer, that were not attributed to differences in demographic characteristics, BMI, or lifestyle factors. There have been previous suggestions that glucose-lowering therapies, particularly metformin, are associated with lower risk of breast cancer though concerns over bias were raised.15 16 Subsequent studies have found no such evidence.17 In our current analysis, we demonstrated the relationship between HbA1c and premenopausal breast cancer could not be attributed to glucose-lowering therapy since the association was also found in a model excluding those with diagnosed diabetes or on metformin. Previous studies have found that younger maternal age at first birth increases risk of hyperglycemia and diabetes.18 19 In contrast, older maternal age at first birth is one of the key risk factors for premenopausal breast cancer, a factor that does not appear to contribute to postmenopausal cancer.20 Low normal glycemia could be acting as a proxy for older age at first birth, potentially accounting for the inverse association between hyperglycemia and premenopausal breast cancer risk. However, after adjusting estimates for maternal age at first live birth in post hoc analyses, the inverse association observed in primary analyses remained. Another possible mechanism could be that women with diabetes or higher BMI (ie, those with elevated HbA1c) are more engaged with healthcare services resulting in earlier detection of breast cancer; however, this information is not available in UK Biobank.
We had several incidental findings in sensitivity analyses. For example, after excluding individuals with diagnosed diabetes or on glucose-lowering medications, there was a suggestion that increased HbA1c was associated with elevated risk of incident stomach cancer. This finding is consistent with a previous study of 2603 individuals that accounted for confounding factors including BMI and alcohol intake21 but contradicts null or potentially underpowered findings from two larger studies by Travier et al22 and Miao Jonasson et al23 (n=46 575 and n=25 276, respectively). Caution should be used when interpreting this and findings from other sensitivity analyses since they were not observed in primary analyses. There remains the possibility that the observations occurred simply by chance due to the number of comparisons we performed in sensitivity analyses or perhaps due to collider bias introduced by excluding populations.
The most recent systematic review of associations between HbA1c and cancer risk included 19 studies,4 the largest of which analyzed 46 575 individuals with 634 cancer events.22 While the authors found no evidence of an association between HbA1c and any of the cancers also included in the present study, the number of site-specific cancer events were low. A more recent study by Goto et al24 followed 29 629 individuals with HbA1c measurements of whom only 6% had an existing diabetes diagnosis. The authors found an overall increased risk of cancer with increasing HbA1c, which was not observed after excluding events in the first 3 years of follow-up. A large study by Dankner et al7 included 440 000 individuals with diagnosed diabetes from a large insurance database. With over 26 000 cancer events, they found no association between HbA1c and overall cancer risk but did observe a positive association with pancreatic cancer and negative association with prostate cancer. We confirm an association with pancreatic cancer; however, we show that any protective effect of HbA1c on prostate cancer is lost after adjustment, namely for diagnosed diabetes. Danker et al were not able to adjust for variables included in the UK Biobank, including BMI, physical activity, and underlying cardiovascular disease, and only included individuals with diagnosed diabetes thus effects may be confounded by exposure to glucose-lowering therapies or some other unmeasured factor.
Our study has several key strengths, particularly a large sample size, the inclusion of individuals with and without diagnosed diabetes and as such not exposed to any glucose-lowering medications, a long follow-up window, and highly detailed covariate data with small proportions of missingness. We ascertained cancer outcomes from highly reliable linked UK cancer registry data covering all treated cancers in the National Health Service. We also acknowledge some important limitations. We approximated menopausal status, meaning that there may be some misclassification of premenopausal or postmenopausal breast cancers; however, previous studies have suggested that such proxies produce valid incidence rates.10 The overall healthier population contributing to the UK Biobank13 meant that we observed relatively low cancer incidence rates, and as such, the power of our analysis was likely affected. Although the underlying incidence rates and exposure distributions may not be representative of the UK population, the associations observed between HbA1c and cancer incidence are unlikely to be biased.25 This is supported by our sensitivity analysis replicating associations between type 2 diabetes and cancer incidence that are consistent with a recent umbrella review.2 Lastly, it is possible that although we adjusted for a wide range of confounders, some residual confounding may remain. We were also unable to look at the effect of medication use in those with diagnosed diabetes as we lacked longitudinal prescription data.
In conclusion, apart from pancreatic cancer, we did not demonstrate any independent positive association between HbA1c and risk for cancer in this large sample of UK adults. We identified an inverse association between HbA1c and premenopausal breast cancer unlikely to be attributed to antidiabetic therapies. These findings suggest that concerns around the potential for a cancer-inducing, direct effect of hyperglycemia may be misplaced. Future research should explore other potential mechanisms that have been hypothesized, including genetic risk factors and chronic inflammation.