Introduction
Persons with diabetes have an approximately 20–25% higher cancer incidence compared with persons without diabetes, though this varies by cancer site [
1]. In particular, persons with diabetes have been shown to have an elevated incidence of liver, pancreatic, colorectal, endometrial and kidney cancer [
1]. Conversely, a decreased risk of prostate cancer has been reported in men with diabetes [
2‐
5].
The association between type 1 diabetes and cancer is not well described. The majority of previous studies assessing the link between diabetes and cancer have not made a distinction between the two major types of diabetes. As type 2 diabetes is far more prevalent than type 1 diabetes, study populations have been primarily composed of persons with type 2 diabetes, thus hindering the generalisation of findings to persons with type 1 diabetes. It is possible that the relationship between type 1 diabetes and cancer is different from that observed between type 2 diabetes and cancer as a result of differences in the underlying disease characteristics, drug therapies and patterns of risk factors, such as obesity.
The observed excess risk of cancer in persons with diabetes may, to some extent, be a consequence of anti-diabetic drug therapies, such as exogenous insulin [
6]. There is some evidence that insulin-treated patients have a higher cancer incidence compared with non-insulin-treated patients [
7]. If exogenous insulin use is associated with increased cancer risk, then an excess risk of cancer should be apparent in persons with type 1 diabetes, and this excess risk could potentially be larger than that observed in persons with type 2 diabetes. Alternatively, hyperglycaemia has also been suggested as a possible mechanism through which diabetes is associated with cancer [
8,
9]. If hyperglycaemia does contribute to the observed elevated risks of cancer among persons with type 2 diabetes then similar associations would be expected for both type 1 diabetes and type 2 diabetes patients.
The small number of existing studies investigating cancer occurrence among persons with type 1 diabetes [
10‐
13] have been underpowered to provide accurate risk estimates for site-specific cancers and have subsequently reported heterogeneous findings [
14]. Further limitations of previous studies include unrepresentative samples, difficulties in defining type 1 diabetes, short follow up and inadequate evaluation of the potential influence of ascertainment bias on diabetes duration.
In this large multinational study, we compared cancer incidence among persons with type 1 diabetes and the general population using population-based registries in five countries. The effect of diabetes duration on the association with cancer was also explored.
Discussion
In this large multi-country study, we compared the cancer incidence in persons with type 1 diabetes with that observed in the general population, and investigated the influence of diabetes duration on the risk estimates.
We found that women with type 1 diabetes had a marginally elevated incidence of overall cancer compared with the general population. Men with type 1 diabetes did not have an elevated incidence of overall cancer, although the substantial inverse association between type 1 diabetes and prostate cancer incidence in men contributed to the overall result and to these sex differences. When analyses were restricted to non-sex-specific cancers, the excess cancer incidence was similar in both men and women with type 1 diabetes: there were elevated HR estimates of about 15% in the first 20 years after diagnosis. We found that the cancer incidence was substantially higher in both men and women with type 1 diabetes during the first year of follow up compared with a longer duration of diabetes. The cancer incidence subsequently decreased to that of the general population after approximately 20 years of follow up for men and after 5 years of follow up for women.
We found that type 1 diabetes conferred an excess risk of cancer of the stomach, liver, pancreas, endometrium and kidney and a reduced risk of prostate cancer. We also report a lower incidence of breast cancer in women with type 1 diabetes compared with the general population.
There was evidence of heterogeneity in HRs of some cancers by country, though a lack of plausible biological mechanisms leads us to believe these differences are likely to be an artefact.
Previous studies investigating the association between type 1 diabetes status and cancer incidence have reported mixed findings [
14], with estimates for the overall cancer risk among persons with type 1 diabetes ranging between 5% lower to 25% higher compared with the general population, although reported 95% CIs include the estimates reported here [
10‐
13,
19]. The discrepancies in these findings may, in part, be due to substantial differences in the criteria used to define type 1 diabetes diagnoses. For example, a UK-based cohort study which identified 214 incident cancers among 23,834 persons with diabetes diagnosed between 1972 and 1986 and reported a standardised incidence ratio of 0.95 (95% CI 0.84, 1.08), used insulin treatment as a proxy for type 1 diabetes [
12]. In contrast, in the Swedish cohort study that identified 258 cancer diagnoses among 24,052 persons between 1964 and 2006 and reported an HR of 1.17 (95% CI 1.04, 1.33), ICD codes from hospital admissions plus age were used to ascertain type 1 diabetes status [
11]. The authors acknowledged the lack of specific type 1 diabetes diagnostic codes in earlier ICD versions as a possible source of misclassification bias because persons with type 2 diabetes might have been included in the type 1 diabetes study population.
Our finding of a significantly elevated cancer incidence among people with type 1 diabetes in the first year after diagnosis of diabetes may be attributable to ascertainment bias (i.e. earlier detection of pre-existing cancers) and to a lesser extent to reverse causation (i.e. the cancer itself causing diabetes). Previous studies investigating the relationship between type 1 diabetes and cancer have provided only sparse information on the influence of diabetes duration on this association. These studies typically applied wide intervals of diabetes duration in the analyses because of their limited sample sizes. As a consequence, they were unable to perform a detailed analysis of the differences in cancer occurrence according to time since diabetes diagnosis. In one Swedish cohort study including data from 29,187 persons who were hospitalised with type 1 diabetes between 1965 and 1999, HR estimates were stratified within two post-diagnostic periods of 1–14 or ≥15 years following diabetes diagnosis [
13]. These analyses yielded standardised incidence ratios of 1.1 (95% CI 1.0, 1.3) and 1.2 (95% CI 1.0, 1.3) for 1–14 and ≥15 years, respectively, which are largely consistent with our results but do not include rates in the first year after diabetes diagnosis.
HRs for site-specific cancers in persons with type 1 diabetes were similar to those observed among persons with all types of diabetes and with type 2 diabetes [
1,
20]. This finding suggests a potential common mechanism among persons with type 1 and type 2 diabetes, for example obesity, insulin treatment or hyperglycaemia.
HR estimates for site-specific cancers were highest for cancer types in which obesity is an established risk factor, namely stomach, colorectal, kidney, endometrial and pancreatic cancers [
21]. While obesity is less prevalent in persons with type 1 diabetes than in persons with type 2 diabetes, obesity is increasing among persons with type 1 diabetes and may contribute to an increased risk of certain cancers [
22,
23]. Further research is therefore required to compare obesity prevalence in people with and without type 1 diabetes.
The influence of glucose-lowering medications on the observed association between diabetes and cancer has been extensively investigated [
24,
25]. Our finding of a smaller excess incidence of cancer among persons with type 1 diabetes than was previously observed in persons with type 2 diabetes does not support the notion that insulin therapies contribute to the observed elevated incidence. If exogenous insulin did contribute to the observed association, the strength of the relationship between type 1 diabetes and cancer incidence would be expected to be stronger than that observed among persons with type 2 diabetes because all people with type 1 diabetes are treated with insulin and a minority of people with type 2 diabetes receive insulin treatment. Furthermore, the absence of an association between overall or site-specific cancer risk and increasing duration of diabetes in our study does not support a dose–response relationship between exogenous insulin use and cancer incidence. We did not have sufficiently detailed data to evaluate associations between type 1 diabetes and cancer incidence according to specific types of insulin.
Diabetes-specific metabolic deficiencies such as hyperglycaemia may provide an alternative explanation for the observed excess risk of some cancers among persons with diabetes, given that these deficiencies are common in both type 1 and type 2 diabetes. Hyperglycaemia may be a plausible explanation given the identification of a dose–response relationship between glycated haemoglobin levels and the risk of certain cancers [
8,
26,
27]. However, further work is required to elucidate whether obesity or hyperglycaemia in persons with type 1 diabetes is responsible for the observed association with some cancer types.
We report a 10% lower incidence of breast cancer in women with type 1 diabetes compared with the general population. Previous studies assessing the influence of type 2 or unspecified diabetes on breast cancer have displayed a similar or elevated risk compared with the general population, although these findings are largely limited to post-menopausal women [
28]. Our contradictory finding may therefore reflect our younger cohort (containing fewer post-menopausal women) or may be confounded by other breast cancer risk factors such as parity.
Our finding of an increased incidence of pancreatic and liver cancer among persons with type 1 diabetes is consistent with some earlier studies. Results from a meta-analysis based on nine studies (but with only a total of 39 cases of pancreatic cancer) indicated that persons with type 1 diabetes were at a twofold increased risk of pancreatic cancer compared with persons without diabetes [
29]. A Danish study (distinct from the Danish data in this study) observed an HR of 4.8 (95% CI 2.8, 7.7) of liver cancer among diabetes patients younger than 50 years at diagnosis. Although this definition of type 1 diabetes is not strictly comparable with ours, the result is consistent with our findings [
19]. However, the associations seen for pancreatic and liver cancer may be a consequence of reverse causation.
We found an increased risk of thyroid cancer in persons with type 1 diabetes compared with the general population, particularly among women. The worldwide incidence rates of thyroid cancer have increased steeply in recent decades [
30,
31], partly due to enhanced detection through the use of more sensitive diagnostic equipment. Consequently, our observation of an increased incidence of thyroid cancer among persons with type 1 diabetes may be a diagnostic artefact because persons with type 1 diabetes receive considerably higher levels of medical attention than those without diabetes.
Although a reduced risk of prostate cancer in persons with type 2 diabetes has been reported in numerous studies [
2‐
5], only one previous study has identified such an association among persons with type 1 diabetes (although some of the latter study data were included in the present study) [
10]. The similarity in prostate cancer incidence between type 1 and type 2 diabetic patients suggests a similar underlying mechanism in both types of diabetes. One proposed mechanism involves the lower testosterone levels present in men with diabetes, obesity or both [
32]. Higher testosterone levels were previously shown to lead to an increased risk of prostate cancer [
33], while hyperglycaemia has also been shown to inhibit testosterone production [
34].
Strengths and limitations
This study used five national cohorts of persons with type 1 diabetes, and is therefore the largest study to date to investigate the relationship between type 1 diabetes and cancer incidence. By combining cases across countries we assembled a sufficiently large sample size to investigate associations with site-specific cancers. The largest previous study included less than one-third of the number of cancers in people with type 1 diabetes compared with our study [
14]. The population-based registry approach also minimised selection bias.
A further strength of our study was the inclusion of information on duration of diabetes, which enabled us to assess the influence of ascertainment bias and the potential impact of reverse causation. For example, it is well established that pancreatic cancer can induce diabetes [
35].
Possible misclassification of type 2 as type 1 diabetes may have led to an overestimation of the relationship between type 1 diabetes and cancer because of the higher levels of well-established risk factors for cancer (including obesity and smoking) among persons with type 2 diabetes. However, the prevalence of type 2 diabetes below 40 years of age is low and we therefore believe that the influence of misclassification bias is negligible, as shown by our interaction analysis (see
http://bendixcarstensen.com/DMCa/T1D/T1D-Ca.pdf (accessed 8 January 2016), section 8.5).
Moreover, we compared cancer incidence rates among persons with type 1 diabetes with those of the general population, instead of only with those of persons with diabetes. This may have diluted the impact of type 1 diabetes on HR estimates, although this effect will be small because of the low prevalence of type 1 diabetes (1.5–4.8 per 1,000 people) in the study populations [
36].
In addition, the large number of comparisons performed introduced the potential for chance findings.
Finally, the study population was relatively young. Thus, we were not able to meaningfully evaluate potential changes in cancer incidence among older persons with type 1 diabetes.
Implications
Our findings do not support changing the policy for cancer screening in persons with type 1 diabetes. Similar recommendations for lifestyle approaches to reduce cancer risk such as weight management, increasing physical activity and avoiding smoking apply to persons with type 1 diabetes as for the general population.
Future work should be directed at ascertaining whether the increased incidence of some cancers among persons with type 1 diabetes leads to a raised risk of cancer mortality among persons with type 1 diabetes.