Cancer incidence and mortality in 23 000 patients with type 1 diabetes in the UK: Long‐term follow‐up

Abstract Type 2 diabetes is associated with raised risk of several cancers, but for type 1 diabetes risk data are fewer and inconsistent We assembled a cohort of 23 473 UK patients with insulin‐treated diabetes diagnosed at ages <30, almost all of whom will have had type 1 diabetes, and for comparison 5058 diagnosed at ages 30 to 49, of whom we estimate two‐thirds will have had type 2, and followed them for an average of 30 years for cancer incidence and mortality compared with general population rates. Patients aged <30 at diabetes diagnosis had significantly raised risks only for ovarian (standardised incidence ratio = 1.58; 95% confidence interval 1.16‐2.11; P < .01) and vulval (3.55; 1.94‐5.96; P < .001) cancers, with greatest risk when diabetes was diagnosed at ages 10‐14. Risks of cancer overall (0.89; 0.84‐0.95; P < .001) and sites including lung and larynx were significantly diminished. Patients diagnosed with diabetes at ages 30 to 49 had significantly raised risks of liver (1.76;1.08‐2.72) and kidney (1.46;1.03‐2.00) cancers, and reduced risk of cancer overall (0.89; 0.84‐0.95). The raised ovarian and vulval cancer risks in patients with type 1 diabetes, especially with diabetes diagnosed around pubertal ages, suggest possible susceptibility of these organs at puberty to metabolic disruption at diabetes onset. Reduced risk of cancer overall, particularly smoking and alcohol‐related sites, might reflect adoption of a healthy lifestyle.

especially if diabetes was diagnosed at ages 10 to 14, that is, likely around puberty, but were significantly decreased for cancer overall.

| INTRODUCTION
Diabetes mellitus is one of the major chronic diseases of Western and increasingly of other populations. It is, however, the consequence of two very different disease processes, type 1 (insulindependent) and type 2 (non-insulin-dependent), (and several much rarer types), with different metabolic and hormonal characteristics and treatments that might affect cancer risk. Most studies of cancer in patients with diabetes have concerned type 2 diabetes, far the more common type, or have been of patients with diabetes overall and therefore effectively of type 2. [1][2][3][4] These have shown raised risks of several tumours including, liver, pancreatic, colorectal, renal and endometrial cancers, 1-5 but could reflect confounding by alcohol consumption and obesity, which are causes of this type of diabetes. There has been far less research on cancer incidence [5][6][7][8][9][10][11][12] and mortality 5,13 in patients with type 1 diabetes, and these have been difficult to interpret because of uncertainty about the proportion of subjects who were truly type 1. Proxy measures have been used to categorise diabetes type, for instance by including as type 1 all patients with diabetes incident under age 40 or 50 irrespective of treatment type. 8,12 However, most patients with diabetes incident at ages 30 and above are actually type 2, [14][15][16] and hence cancer risks in several published cohorts may substantially reflect contamination by type 2 subjects.
Few data have been published on cancer risks by duration since diagnosis of type 1 diabetes, 9,10,12 and these have been problematic both because of the diagnostic misclassification described above, and because hospitalisation or unspecified events have been used to categorise when diabetes was diagnosed, rather than using actual date of diagnosis. Duration is important because most known cancer causes act to raise risk over decades, not in the early years after first exposure, and this is the more so for exposures starting in childhood, when type 1 diabetes incidence is greatest. Also, in the immediate period after diabetes diagnosis, there is potential for reverse causation or detection bias, which could be uncovered by duration analyses.
We have therefore analysed cancer incidence and mortality risks in the Diabetes UK (formerly British Diabetic Association [BDA]) cohort, which includes over 20 000 patients with type 1 diabetes for whom date of type 1 diagnosis was recorded, and hence duration of diabetes known. The original cohort also included over 5000 insulin-treated patients diagnosed at ages 30 to 49, and we analysed follow-up of these too, to provide a comparison, albeit not as large, of mainly type 2 patients from the same sources. Cancer follow-up in the cohort was last published 17 years ago. 17 The follow-up now includes over three times as many cancers and over twice as many cancer deaths as in that publication.

| MATERIALS AND METHODS
The study cohort was formed by combining registrations from several registers of UK-resident patients with insulin-treated diabetes identified during 1972 to 1993. The largest was a register of 12 891 children compiled by the BDA during 1972 to 1986. The remainder were population-based geographical registers from several parts of the UK. 18  For initial assessment of cancer risks we analysed separately subjects with diabetes onset at ages under 30, and aged 30 to 49, because at the period of diabetes incidence in this cohort, this approximately divided those who would have been almost entirely type 1 diabetes from those who would have been largely type 2. 20 We also assessed cancer risks by duration since diagnosis and by 10-year age group of diabetes diagnosis, and for females we analysed risk separately for ages 10-14 and 15-19 at diagnosis, because in the UK menarche occurs at ages 10-14 in about 90% of girls. 21 Additionally, following the observation by Leete et al 22 that patients with diabetes incident at ages under 30 can be subdivided into three immunologically separate groups based on age at diagnosis, 0-6, 7-12 and 13-29, we analysed these three age-groups separately. To examine whether the cancer diagnosis might have been made as a consequence of the clinical work-up after diabetes was diagnosed (or vice versa), we conducted analyses of cancer incidence risks in the first year after diabetes diagnosis.
We used the above standard statistical methods for cohort analysis, without smoothing or "correction" of SIRs, and used standard quinquennial and decennial cut-points (except where others have reported specific immunologically relevant cut-points, 22 or we were examining for diagnostic artefact at the time of diagnosis), to allow comparison with the literature and allow potential meta-analysis, and to avoid data-driven sub-analyses. We calculated 95% confidence intervals and tested for trend assuming a Poisson distribution. 23 All P values presented are 2-sided. All statistical analyses were performed using Stata IC Version 16.0. 24

| RESULTS
There were 29 321 patients on the BDA cohort register aged under 50 years at diagnosis, of whom we excluded 29 because their diabetes was secondary to another disease such as cystic fibrosis, Numbers in mortality analyses. The numbers in the cancer incidence analyses are similarly distributed but slightly smaller and with no subjects from Northern Ireland, total 27 682 (see Section 3).
Overall cancer incidence risk in the cohort (   Patients diagnosed with diabetes at ages 30 to 49 also had significantly reduced risks of cancers of the lung and prostate as well as non-melanoma skin cancer, but significantly raised risks only of liver and kidney cancers. There were, however, also somewhat raised risks, albeit not significantly, in this older age group but not the younger, for several cancers that have been shown increased in patients with diabetes generally (ie, preponderantly type 2) 1-5 -liver, pancreas, endometrium, colorectum and non-Hodgkin lymphoma.
However, these studies have varied greatly in the parameters used to define diabetes type and in the length of follow up after diabetes diagnosis, and these variations may explain some or all of the inconsistency between results and the differences from our findings. Most of the studies have been relatively small, but studies from Sweden 9,10,13 and Australia 5 have been large, as has an analysis combining heterogeneous data from these countries plus Finland, Denmark and Scotland, 12 the latter likely with a modest overlap with the Scottish patients within our whole UK study.
In our cohort, unlike several 8-10,12,13 but not all [5][6][7]12,25 12,13 or 45 5 or older, 6,8 the proportion who were type 2 will have been considerable. This is the more so if the subjects were from more recent diagnosis periods: the great majority of patients with diabetes incident at ages ≥30 are type 2 [14][15][16] and the proportion who are type 2 has been increasing over time. 15,28,29 Furthermore, since the subjects who enter follow-up older, and hence are more likely to be type 2, will have had more of their follow-up at ages at which cancer incidence rates are greatest, they will disproportionately contribute to cancer incidence in a cohort.
For instance, although only 8.5% of our cohort were aged 40 to 49 at diabetes diagnosis, this age group contributed 24.1% of the cancers incident (Table 4).
Differences between our results and those previously may therefore reflect differences in the extent to which nominally type 1 cohorts inadvertently included type 2 patients (assuming that cancer risks truly differ between these two diabetes types, which seems likely if only because the latter is strongly associated with a major known cancer risk factor [obesity] whereas the former is not).
The two cancers at significantly raised risk in our young-onset diabetes cohort, were both of the female reproductive tract-ovarian and vulval cancers. Most previous cohort studies have found raised ovarian cancer risks in nominally type 1 patients, 5,10-12 and the exceptions were small with wide confidence intervals. 8,9 While the risk has not previously been examined by age at diabetes diagnosis, we found greatest risk for women whose diabetes was diagnosed at ages 10 to 14, the age of menarche for 90% of women in the UK. 21 However, these are subgroup analyses which need to be interpreted cautiously. Although we do not have data directly on age at menarche in our cohort, the results imply that the raised ovarian cancer risk might be due to greater susceptibility of the ovary around puberty to the metabolic disruption accompanying the initial disturbance leading to diabetes diagnosis, in the same way as breast cancer risk is more raised after radiation exposure at puberty than at other times. 30 Vulval cancer too showed greatest (6-fold) risk for those whose diabetes had been diagnosed at ages 10 to 14, although also with greatly raised risk for those diagnosed at younger ages than this (0-9 years); the reasons are inapparent, but for both tumours an analysis of risk in relation to actual age at menarche, if there were a cohort in which this was ascertained, would be desirable.
We can find no previous cohort analyses of vulval cancer risk in patients with type 1 diabetes. Two cohorts have examined risk of a wider category, "other female genital cancers", with non-significant results, but based on only four and three cases, respectively. 8,10 One cohort examined risk of non-cervical anogenital tumours overall, with non-significantly raised risk for cancers and significantly raised risk for intraepithelial neoplasms. 31 No cohorts have published data on ovarian or vulval cancer risks by age at diabetes onset. Vulval cancer at younger ages is often HPV related, 31,32 as are anal, vaginal and cervical cancers, but there was no raised risk of these other tumours in our cohort. An alternative potential mechanism is that glycosuria may cause chronic itching, inflammation, and lichen sclerosis, which are posited as potential risk factors for vulval cancer. 32,33 Of the other significant findings in the patients with diabetes diagnosed under age 30, the diminished risks for larynx and lung cancers presumably reflect diminished smoking, which has been found in some but not all studies of young people with type 1 diabetes. 34,35  The diminished risk of prostate cancer that we found has been seen also in other large nominally type 1 diabetes cohorts 5,12 although not in a cohort in Taiwan. 11 It might be a consequence of reduced free testosterone levels. 36,37 There was also a significantly diminished risk of Hodgkin lymphoma, primarily among patients whose diabetes onset had been before age 10. Previous studies of nominally type 1 cohorts have found non-significantly reduced risks of Hodgkin lymphoma 5,10 or risk decreased in females but not males, 12  The cancer risks in our cohort for patients with diabetes diagnosed at ages 30 to 49, likely to be mainly type 2 cases, were much more similar to those in previous cohorts. The significantly raised risk of liver and kidney cancers, and non-significantly raised risks of colorectal, pancreatic and endometrial cancers and non-Hodgkin lymphoma, accord with raised risks of these tumours generally found in type 2 cohorts 1-5 and, with the exception of non-Hodgkin lymphoma, generally also in nominally type 1 cohorts. 5,[8][9][10][11][12] Likewise, the significantly reduced risk of prostate cancer in patients with diabetes diagnosed at older ages in our cohort accords with reduced risk in type 2 diabetes cohorts 1-5 and might, as noted above, relate to testosterone levels. The significantly reduced risk of lung cancer in our older onset diabetes patients, as in the younger onset group, seems likely to be due to reduced smoking. There were not reduced risks of lung cancer in other large type 1 cohorts, but perhaps smoking patterns in those cohorts may have been different.
Our results for cancer mortality showed similar patterns to those for cancer incidence, although less marked based on smaller numbers, and with a modest tendency to greater SMRs than SIRs, perhaps indicating poorer survival from cancer in the patients with diabetes than in the general population. The only previous cohort study of sitespecific cancer mortality in patients with type 1 diabetes 5 likewise found somewhat greater SMRs than SIRs. Our study of type 1 diabetes patients had several strengths-large size; information to allow analyses by age at diagnosis of diabetes (not published for any other cohort, except for one publication on a twoway split of ages for three cancer sites 10 ); high completeness of follow-up; longer follow-up than any study previously; information on cancer mortality as well as incidence; and restriction to insulin-treated patients. However, it is a weakness that, like most other cohorts, we did not have direct data on diabetes type: for the diabetes onset under age 30 group, the effect should have been minor, since we estimate about 94% will have been type 1, but for the comparison group of patients diagnosed at ages 30 to 49, the effect will have been of greater impact since only about 68% would have been type 2. We used standard quinquennial and decennial cut-points for age and duration (plus immunologically relevant cut-points reported by others, 22 and a 1-year cut-point examining for potential diagnostic artefact at the time of diagnosis) to avoid data-driven analyses. However, since these are subgroup analyses, they need to be interpreted cautiously, especially where numbers are sparse.
In summary, our large cohort of patients with type 1 diabetes has shown significantly raised risks of ovarian and vulval cancers, greatest in patients diagnosed with diabetes at ages 10 to 14, the usual ages at puberty in the UK, and an overall diminished risk of cancer, reflecting diminished risks of smoking and alcohol-related cancers, likely because of healthier lifestyles in these respects of the diabetes patients, and of prostate cancer, which might be connected to the reduced testosterone levels in men with diabetes.