Duration of Type 2 Diabetes and Incidence of Cancer: An Observational Study in England

Improvements in the risk stratification and personalized strategies for early identification and treatment of cardiovascular risk factors have contributed to the reducing mortality trends and increasing life expectancies in people with type 2 diabetes during the past two decades (1–6). These epidemographic changes have likely contributed to the phenotypical shift in the trajectories of morbidities and causes of death in people with type 2 diabetes: growing evidence has, indeed, shown that noncardiovascular diseases represent an increasingly more frequent cluster of complications in individuals with type 2 diabetes, particularly in older individuals (7,8). Among these complications, cancer is emerging as an important morbidity (9), possibly overtaking cardiovascular disease as the leading cause of excess deaths in individuals with diabetes (8). Although a robust and extensive observational literature has reported increased risks of several cancers in people with type 2 diabetes (10,11), the strength of associations is higher for some cancers (e.g., pancreatic, liver) than others (e.g., lung) (10). In line with preclinical studies (9,12), a higher BMI—one of the strongest risk factors for type 2 diabetes (13)—results in an insulin resistance and hyperinsulinemia milieu that could enhance mitogenesis and risk of oncogenesis (12).

A longer life expectancy in individuals with type 2 diabetes may also contribute to longer disease duration and prolonged exposure to insulin resistance and hyperinsulinemia, potentially explaining the higher risks of cancer incidence and mortality observed in older individuals with diabetes (7,14). However, because longer disease duration necessarily results in an older age, which is a risk factor for cancer, a higher risk in individuals with a longer disease duration may simply reflect ageing. Therefore, to investigate whether diabetes duration is a risk factor for cancer, both the duration itself and the (isochronous) increase in chronological age should be accounted for.

There is some evidence on the association between diabetes duration and risk of cardiorenal diseases or all-cause and cause-specific mortality, with longer durations generally linked to higher risks (15–19). However, the available evidence about diabetes duration and cancer incidence remains sparse (20–27); this information is important for developing and optimizing the design of early detection strategies for cancer. In this study, we aimed to disentangle the association of age and diabetes duration with the risk of incident cancer in people with type 2 diabetes. To corroborate our results, we also detailed associations with cancers that are currently considered causally linked to higher body fat, given the postulated role of insulin resistance and hyperinsulinemia in the risk of cancer in people with type 2 diabetes (9).

This study followed a preregistered protocol, approved by the Clinical Practice Research Datalink (CPRD) Independent Scientific Advisory Committee (No. 19_120Mn) before data were made available, and has been reported in line with the Reporting of Studies Conducted Using Observational Routinely Collected Data (RECORD) guidelines (Supplementary Material).

We identified a cohort of individuals with type 2 diabetes in the U.K. within the CPRD GOLD database, which routinely collects deidentified data on primary care patients; CPRD is representative of the national population in terms of age, sex, and ethnicity (28). Individual-level information is linked to the Hospital Episode Statistics (HES) and the Office for National Statistics (ONS) death registrations data to extract details on hospitalizations and date and cause of death, respectively, based on the International Classification of Diseases (ICD) codes (ICD-9 and ICD-10). Linkages are available only for patients in England.

The flowchart of the cohort definition is shown in Supplementary Fig. 1. In CPRD, we first extracted data on all individuals with a first-ever diagnosis code of type 2 diabetes between 1 January 1998 and 30 November 2018 (i.e., the date of type 2 diabetes diagnosis: index date), aged ≥35 years, and registered with an up-to-standard practice for a minimum of 1 year at the index date, and with linkages to HES and ONS. We then excluded individuals with a CPRD diagnosis code of type 2 diabetes after death, on the same day of death, or on the last day of the CPRD cohort definition (30 November 2018), as well as those who had a code of type 2 diabetes in HES and were younger than 35 years at the time of coding. To rule out potential diabetes misclassification by clinical coding, we further excluded individuals with a code of type 1 diabetes any time in either CPRD or HES. Last, because the main outcome was cancer incidence, we excluded individuals with a cancer diagnosis in HES before the first-ever diagnosis code of type 2 diabetes in CPRD or HES, leaving 130,764 individuals for the analyses.

Information on the two main exposures—age at type 2 diabetes diagnosis and BMI—was extracted from the CPRD. Age at diagnosis was estimated as the difference between the calendar date at type 2 diabetes diagnosis and the date of birth. Data on BMI were extracted using the values of the closest date before the index date; BMI was categorized as underweight (<18.5 kg/m²), healthy weight (≥18.5 to <25 kg/m²), overweight (≥25 to <30 kg/m²), and obesity class 1 (≥30 to <35 kg/m²), class 2 (≥35 to <40 kg/m²), or class 3 (≥40 kg/m²).

To identify incident cancers after the index date, we used the first-ever ICD codes (C00-C97, except nonmelanoma skin cancer [C44]) in any position in hospital (HES) or death (ONS) records. We investigated cancer incidence rates for all cancers, the four commonest cancers in the U.K. (lung [C34], colorectal [C18–C20], breast [C50], and prostate [C61]), and for cancers considered causally linked to higher body fat (namely, breast, colorectal, endometrial, gallbladder, gastric cardia, kidney, liver, multiple myeloma, esophagus, ovary, pancreas, and thyroid; Supplementary Table 1) (29). Individuals were followed from the index date until the occurrence of cancer, death, or the end of study (linkage date of HES data: 30 November 2018), whichever occurred first.

We reported the baseline characteristics at index date, stratified by sex and age at type 2 diabetes diagnosis (<40, ≥40 to <50, . . ., ≥70 to <80, and ≥80 years) as median and interquartile range for continuous variables and numbers and proportions for categorical ones; similarly, we estimated cancer incidence rates by sex and age at diagnosis.

During the follow-up from diabetes diagnosis until cancer incidence or right censoring, both duration of diabetes and age increase by the same amount: for example, in an individual diagnosed at 40.5 years, the attained age (i.e., the current or chronological age while the individual is followed up in the study) after 5.5 years of follow-up (or, equivalently, after 5.5 years of diabetes duration) is 46 years; similarly, the attained age is also 46 years in another patient diagnosed at 39 years with 7 years of diabetes duration. Because the follow-up occurs across two timescales, we modeled both scales to disentangle the association of age and diabetes duration with cancer incidence (30). We used a Royston–Parmar survival model with a continuous, nonlinear main effect for both timescales using a restricted cubic spline with five knots and an interaction between the timescales. This approach allowed us to investigate the (time-varying) association of diabetes duration (from diagnosis until age 20 years) on the cancer incidence rates by attained age (50–100 years) and age at diagnosis (50–80 years), thus separating the role of each of these three factors in terms of the risk of cancer. In the analyses of obesity-related cancers, we estimated incidence rates in all BMI groups except <18.5 kg/m² (given the very small number of events occurring in individuals in this group) and included an interaction between BMI and each of the two timescales, allowing for the association of BMI to vary across diabetes duration and attained age.

In sensitivity analyses, we re-estimated the incidence rates for 1) all cancers and the four commonest cancers, after excluding individuals with a BMI <18.5 kg/m² (to reduce the risk of diabetes misclassification) and a follow-up shorter than 2 years (to reduce the risk of reverse causation and detection bias) (27); and 2) obesity-related cancers, after excluding individuals with follow-up shorter than 2 years.

All analyses, conducted in Stata (18.0 MP, SPECTRE High Performance Computing Facility, University of Leicester), were stratified by sex; results are reported with 95% CIs.

Data access is through permission from the CPRD only (protocol 19_120Mn). Enquiries should be addressed to [email protected]. The electronic health record codes are either reported in the article or at https://github.com/frazac82/, alongside the statistical analysis underpinning this study.

The characteristics at type 2 diabetes diagnosis of the 130,764 individuals included in our study are summarized in Table 1: Women accounted for 44.3% of the cohort and the median age was 61.8 (interquartile range: 52.8–70.6) years in men and 65.6 (55.2–75.0) years in women. At diagnosis, the BMI was slightly higher in women (31.2 kg/m²) than men (30.2 kg/m²), with values of BMI ≥30 kg/m² in 51.9% and 57.9% of men and women, respectively. Regardless of sex, there was an inverse relationship between age at diagnosis and BMI, ranging from a median of 32.5 kg/m² in men younger than 40 years to 27.3 kg/m² in those aged ≥80 years; corresponding figures in women were 36.4 years and 27.4 kg/m².

During a median follow-up of 8.0 (interquartile range: 4.6–11.9) years and a total of 1,089,923 person-years of observation, 18,977 incident cancers (14.5%) were recorded: 11,146 (15.3%) in men and 7,831 (13.5%) in women. The crude incidence rate was 17.4 (95% CI 17.2–17.7) per 1,000 person-years, slightly higher in men (18.3) than women (16.3). Supplementary Table 2 reports the numbers of all cancers and cancer-specific incident events and rates stratified by sex and age at diabetes diagnosis. In both men and women, incidence rates of all cancers were progressively higher in individuals diagnosed at older ages, increasing from 1.9 (95% CI 1.4–2.6) per 1,000 person-years in men younger than 40 years at diagnosis to 49.4 (95% CI 46.7–52.3) in those aged 80 years or more; corresponding estimates in women were 3.5 (95% CI 2.6–4.5) and 28.7 (95% CI 27.1–30.3). This pattern was also observed for obesity-related and the four commonest cancers, albeit the absolute rates varied (generally higher for the two sex-specific cancers; Supplementary Table 2).

For all cancers, incidence rates decreased within the first 1 or 2 years from type 2 diabetes and progressively increased thereafter, particularly in individuals diagnosed at younger age, in both men (Fig. 1A) and women (Fig. 2A). Rates were also 1) higher for older attained age but largely overlapping across varying diabetes duration (men, Fig. 1B; women, Fig. 2B); 2) higher for older attained age and for age at diagnosis (Figs. 1D and 2D); and 3) for the same attained age, virtually identical across varying diabetes duration (Figs. 1C and 2C).

In men, incidence rates (95% CI) of all cancers at an attained age of 50 years were 2.9 (2.5–3.3) and 1.7 (0.4–7.5) per 1,000 person-years for diabetes durations of 5 and 20 years, respectively; at 60 years, 8.1 (7.6–8.7) and 9.5 (6.7–13.7); at 70 years, 20.0 (19.0–21.0) and 18.4 (15.0–22.5); at 80 years, 37.3 (35.3–39.4) and 33.6 (28.0–40.2); at 90 years, 50.5 (45.7–55.7) and 48.3 (36.9–63.2); and at 100 years, 59.6 (42.8–83.0) and 45.9 (18.9–111.6) (Supplementary Fig. 2). These estimates translated into incidence rate ratios (95% CI), comparing 20 to 5 years of diabetes duration, of 0.60 (0.14–2.63) at an attained age of 50 years; 1.18 (0.82–1.69) at 60 years; 0.92 (0.75–1.13) at 70 years; 0.90 (0.75–1.08) at 80 years; 0.96 (0.73–1.26) at 90 years; and 0.77 (0.31–1.92) at 100 years (Fig. 3).

Although the absolute rates differed, the pattern was mirrored in women, with rates largely related to the attained age (increasing in older individuals) but similar for the same attained age and varying durations. Incidence rates (95% CI) of all cancers at an attained age of 50 years were 4.3 (3.8–5.0) and 10.4 (3.1–34.8) per 1,000 person-years for diabetes durations of 5 and 20 years, respectively; at 60 years, 9.9 (9.1–10.7) and 10.6 (7.1–16.0); at 70 years, 16.2 (15.2–17.1) and 14.6 (11.3–18.8); at 80 years, 21.3 (20.1–22.6) and 17.8 (14.2–22.2); at 90 years, 27.3 (25.0–29.8) and 20.3 (15.2–27.1); and at 100 years, 35.3 (26.3–47.4) and 26.3 (10.3–66.7) (Supplementary Fig. 2). These estimates translated into incidence rate ratios (95% CI), comparing 20 to 5 years of diabetes duration, of 2.40 (0.70–8.25) at an attained age of 50 years; 1.07 (0.71–1.63) at 60 years; 0.90 (0.69–1.18) at 70 years; 0.84 (0.66–1.05) at 80 years; 0.74 (0.55–1.00) at 90 years; and 0.74 (0.28–1.95) at 100 years (Fig. 3).

As for all cancers, incidence rates of the four commonest cancers in the UK were greater in older individuals but overlapped across diabetes durations for the same attained age (Supplementary Fig. 2). In men, there was no evidence of higher incidence rates at 20 compared with 5 years of diabetes duration for colorectal, lung, and prostate cancer (i.e., all CIs of the rate ratios crossed 1; Fig. 3), regardless of the attained age. Likewise, incidence rates of the four commonest cancers in women were similar across diabetes duration (Supplementary Fig. 2), with no differences comparing 20 to 5 years of diabetes duration, except 1) for colorectal cancer: at an attained age of 80 years the incidence rate [95% CI] was higher at 5 (3.3 [2.8–3.9] per 1,000 person-years than at 20 years 0.8 [0.3–1.7]; Supplementary Fig. 2) of duration, equating to a rate ratio, comparing 20 with 5 years of 0.24 (95% CI 0.11–0.54; Fig. 3); and 2) for lung cancer: at an attained age of 90 years, corresponding figures were: 3.1 (2.4–4.1), 1.1 (0.4–3.0), and 0.34 (0.12–0.95) (Supplementary Fig. 2 and Fig. 3).

The pattern found for all and the four commonest cancers was also observed for obesity-related cancers: although the incidence rates were generally higher in individuals with a higher BMI and greater by attained age (men, Supplementary Fig. 3; women, Supplementary Fig. 4), for the same BMI and attained age, rates overlapped across varying diabetes duration (Supplementary Fig. 5). Nominal statistically significant associations, with wide CIs, were observed (Supplementary Fig. 6): 1) in the BMI group ≥18.5 to <25 kg/m², incidence rates were higher at 5 than at 20 years of diabetes duration, in men (rate ratio [95% CI] comparing 20 with 5 years: 0.34 [0.13–0.90] and 0.33 [0.13–0.84] at an attained age of 70 and 80 years, respectively) and in women (0.42 [0.18–0.97] and 0.40 [0.17–0.91] at 70 and 90 years, respectively); and 2) in the BMI group ≥35 to <40 kg/m², incidence rates were conversely lower at 5 than at 20 years in women at an attained age of 50 years (rate ratio comparing 20 with 5 years: 4.89; 95% CI 1.01–23.80).

After the exclusion of individuals with a BMI <18.5 kg/m² and a follow-up shorter than 2 years (n = 12,705), the results were virtually identical to those of the main analysis for all cancers (Supplementary Fig. 7) and the four commonest cancers (Supplementary Fig. 8). Similarly, the exclusion of individuals with a follow-up shorter than 2 years (n = 11,029) resulted in no difference in the incidence rates of obesity-related cancers (Supplementary Fig. 9).

In this study of approximately 130,000 individuals with newly diagnosed type 2 diabetes, we did not find evidence of an association between diabetes duration and risk of all cancers, the four commonest cancers (colorectal, lung, prostate, and breast), and obesity-related cancers; conversely, we observed a strong, positive association between age and cancer incidence irrespective of the diabetes duration. Together, these findings suggest that the previously reported higher risk of cancer in individuals with longer diabetes duration is likely explained by the older age associated with the longer duration. Although in contrast with previous studies showing associations between longer diabetes durations and higher risks of vascular diseases and death, our results have important implications for cancer risk stratification and possibly screening or new early detection strategies in relation to age and diabetes duration, and contribute to elucidate the mechanisms underpinning the temporal associations between type 2 diabetes and certain cancers.

The declining trends in CVD morbidity and mortality that occurred during the past 20 years have contributed to the increased life expectancy in individuals with type 2 diabetes (1–6), who, therefore, spend longer periods of their life with the condition. This longer duration results in a prolonged exposure to the metabolic abnormalities of type 2 diabetes, including insulin resistance and hyperinsulinemia, which could enhance the risk of carcinogenesis (9,12). Yet, a longer disease duration is necessarily associated with an older age; as such, other co-occurring, ageing-related pathophysiological changes can contribute to the higher risk of cancer in individuals with longer disease duration. Separating the association of diabetes duration from that of age is thus essential to disentangle their relative impact. Therefore, we used an incident diabetes study design, which has been recommended when the aim is to investigate the role of diabetes duration on the incidence of cancer: this design results in a cohort of individuals followed from their diagnosis, allowing us to examine associations for a wide range of ages at diagnosis and diabetes durations (31). Moreover, we modeled a time-varying association of age and duration using two timescales. This approach has been advocated as the methodology of choice, if not necessary (30–32), when the aim is specifically to explore the (time-varying) association of continuous age and disease duration with the risk of outcomes, particularly when the evidence suggests that the risk is expected to be related to both age and duration of the disease, a well-established notion in clinical diabetes (31–33).

Likely due to the heterogenous designs and analyses, published studies have described associations between diabetes duration and cancer incidence of variable directions and magnitudes (21–24). This limits the possibility of a comparison with our study. Conversely, our findings can be compared with previous investigations that used the same study design, with follow-up starting from diagnosis. In an incident cohort of people with type 2 diabetes in Canada (1994–2006) and in two cohorts with both type 1 and type 2 diabetes in Denmark (1995–2009) and Australia (1997–2008) (25–27), the investigators reported increased risks of several cancers associated with the diagnosis of diabetes, followed by a reduction and more stable risks thereafter. Notably, these patterns were identical to those observed in our study, with higher rates soon after the diagnosis and a subsequent reduction followed by a plateau until 20 years of diabetes duration, possibly related to a detection bias, whereby the suspicion of a cancer would trigger clinical investigations resulting in the diagnosis of diabetes or the diagnosis of diabetes could increase the clinical investigations resulting in a cancer diagnosis (31). By showing consonant results in a different geographical region (spatial validation) and in a more contemporary cohort (temporal validation, particularly relevant in view of the shift in the long-term complications in individuals with type 2 diabetes during more recent years (7,8)), our findings further clarify and corroborate the current evidence about the role of age and diabetes duration on the risk of cancer.

Our findings, which did not show an association between the duration of type 2 diabetes and the risk of incident cancers, are not contradictory to the numerous previous investigations that have shown an increased risk of several cancers in individuals with type 2 diabetes compared with those without (10,11). Instead, our results suggested that the elevation in cancer risk may occur earlier in the trajectory of metabolic abnormalities that ultimately lead to (the clinical diagnosis of) type 2 diabetes. In this respect, the results of the analyses examining the impact of obesity on the relationship of age and diabetes duration with the risk of cancer are informative. If excess body fat (and, in turn, greater insulin resistance) is among the potential mechanisms linking type 2 diabetes to cancer (12), higher incidence rates should be expected in individuals with higher BMIs. Although this was generally observed in our study, we did not find evidence of a greater association with a longer diabetes duration in individuals with higher BMIs; rates were rather constant up to 20 years of duration within each clinically relevant BMI group. This observation points toward a different comparative role of insulin resistance and hyperglycemia on the temporal relationship between diabetes duration and its complications and aligns with the current knowledge on the pathophysiological changes occurring before and after the clinical diagnosis of type 2 diabetes (34). The lack of an overall association with cancer incidence, and of a differential impact of duration across BMI levels, would, indeed, indicate that the insulin resistance–related mechanisms enhancing cancer risk would develop earlier in the continuum of the metabolic abnormalities in type 2 diabetes and exert most of their effects well before the clinical, glycemia-based diagnosis of type 2 diabetes, with a modest or null effect over the course of diabetes after its diagnosis. This would also suggest that the potential actionable window to reduce the legacy effects of an excess in body fat and higher insulin resistance occurs earlier than the interventional time frame to reduce the long-term effects of uncontrolled hyperglycemia (35), with important implications in the research and policy area of cancer prevention and early detection.

Our study has some limitations and strengths. Because CPRD data are not primarily collected for research, the quality of diabetes recording has changed over time (36). To establish the diagnosis date of type 2 diabetes, we relied on the date of the first-ever code associated with the condition; although free testing to evaluate the risk of type 2 diabetes is available for individuals aged 40–74 years in the U.K., there may still be instances where gaps exist between the actual diagnosis and the recording of the diagnostic code. We also identified incident cancers in hospital and death records; information from cancer registries, which was not available in our study, may further improve ascertainment for some cancers (37). Associations are also representative of primary care patients in England who were aged ≥35 years at type 2 diabetes diagnosis; caution should be exercised when generalizing these results to other countries or younger individuals. Although unlikely to fully explain the associations between diabetes and cancer, and notwithstanding the consistent results in the sensitivity analyses excluding the first 2 years of follow-up, an underlying subclinical cancer may cause hyperglycemia (reverse causality), probably driving some of the early increase in the cancer incidence rates associated with the diagnosis of diabetes (31). The heterogeneity of cancer biology would also suggest that directions and magnitudes of the observed associations could vary across cancers of different sites. We focused on cancer incidence; the pattern for cancer mortality may be different because several factors occurring after cancer diagnosis (e.g., treatment) could have a relatively greater impact than diabetes duration on the risk of all-cause and cancer-specific mortality. Last, we did not explore the role of other risk factors (e.g., smoking, socioeconomic status, HbA_1c, glucose-lowering medications) that could have confounded the association between duration and cancer. Strengths of our study include modeling relative and absolute differences, the latter being easier to interpret for the individual patient and more relevant from a public health perspective (38); the incident cohort design, with follow-up since diabetes diagnosis (31); the simultaneous use of two timescales (31); and the linkage to HES and ONS data to identify incident cancers (39).

In conclusion, using a large cohort of individuals with type 2 diabetes, we did not find evidence of an association between diabetes duration and incidence of cancer. The higher incidence rates observed in individuals with a longer disease duration are likely an epiphenomenon of ageing.

Acknowledgments. The Leicester Real World Evidence Unit is supported by the National Institute for Health and Care Research (NIHR) Applied Research Collaboration East Midlands (ARC EM) and the NIHR Leicester Biomedical Research Centre.

Funding. This study was partly funded by Hope Against Cancer (grant RM60G0690). M.D. is cofunded by the NIHR Leicester Biomedical Research Centre and University of Leicester. K.K. and F.Z. are supported by the NIHR ARC EM and the NIHR Leicester Biomedical Research Centre. S.L. is funded by the Cancer Research UK program Inequalities in Cancer Outcomes (grant EPNCZS34).

The funding bodies had no role in the study design, data collection, data analysis, interpretation of results, or writing of the report.

Duality of Interest. K.K. has acted as a consultant and speaker or received grants for investigator-initiated studies for AstraZeneca, Novartis, Novo Nordisk, Sanofi, Lilly, Merck Sharp & Dohme, Boehringer Ingelheim, and Bayer. M.D. has acted as consultant, advisory board member, and speaker for Boehringer Ingelheim, Lilly, Novo Nordisk, and Sanofi; an advisory board member and speaker for AstraZeneca; an advisory board member for Janssen, Lexicon, Pfizer, Medtronic, and ShouTi Pharma Inc.; and as a speaker for Napp Pharmaceuticals, Novartis, and Takeda Pharmaceuticals International Inc. M.D. has also received grants in support of investigator and investigator-initiated trials from Novo Nordisk, Sanofi, Lilly, Boehringer Ingelheim, AstraZeneca, and Janssen. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. F.Z. designed the study, defined the clinical codes, analyzed the data and drafted the article, and is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. K.B. and K.K. acquired research funding. S.L. extracted and cleaned the data. F.Z., S.L., K.B., M.D., and K.K. contributed to the interpretation of the data, critically revised the article, and approved the final version. S.L. and F.Z. had full access to all the data.