Age at Type 2 Diabetes Diagnosis and Cause-Specific Mortality: Observational Study of Primary Care Patients in England

In 2021, the number of people worldwide with diabetes was estimated to be 537 million, which is projected to increase to 783 million by 2045 (1). Mortality rates associated with diabetes, both with and without complications (e.g., cardiovascular disease, chronic kidney disease) are high, resulting in approximately 6.7 million deaths globally in 2021 (1).

The most common form of diabetes is type 2 diabetes (90% of all those with diabetes), which was traditionally considered a disease of mid to late adulthood (1,2). However, over the past few decades the prevalence of type 2 diabetes among younger adults (i.e., diagnosed when younger than 40 years, here “early-onset adult type 2 diabetes”) has greatly increased, now representing 15–20% of all adults with type 2 diabetes globally (2,3). This is a significant public health issue because the diagnosis at a younger age has been associated with a more adverse cardiovascular risk profile and an increased relative risk of diabetes-associated complications (2,4,5). In addition, because individuals with early-onset adult type 2 diabetes represent a working-age population, the socioeconomic impact of diabetes-related morbidity is likely to be greater among these individuals compared with those with later-onset type 2 diabetes (6).

Previous large cohort studies have explored the association between age at type 2 diabetes diagnosis and mortality, with several reporting higher relative risks of both all-cause and cardiovascular mortality in individuals with type 2 diabetes diagnosed at younger versus older ages (7–10). However, from a public health perspective, it is essential to complement the relative risk with the absolute risk because large relative risks may potentially translate into small absolute risks, which should be used to guide the implementation of preventive strategies for common chronic diseases (11). Although the absolute risk of all-cause and cardiovascular and noncardiovascular mortality by age at type 2 diabetes diagnosis has been reported in a few previous studies (8,10), other causes of death have not been investigated in this way. Because early-onset adult type 2 diabetes seems to be characterized by a more aggressive phenotype, it is possible that the complication profile may differ from that observed in later-onset type 2 diabetes, which could translate into differences in the causes of death (6,12).

In this study, we examined the relationships between age at diagnosis and the relative and absolute risks of all-cause mortality and cause-specific mortality (cancer-, cardiorenal-, and noncancer or cardiorenal-related mortality), using routinely collected primary and secondary care data in England.

This retrospective cohort study used the Clinical Practice Research Datalink (CPRD) GOLD and Aurum data. CPRD contains electronic primary care records relating to patients’ characteristics, diagnoses, and prescriptions, for a nationally representative sample of more than 60 million patients, of whom 16 are currently registered (13–15). CPRD is linked to the Hospital Episodes Statistics (HES) database for patients in England and to the Office of National Statistics (ONS) death registration data. HES comprises information on diagnoses, procedures, and demographic data for hospitalized patients, and ONS contains data on date and cause of death. This study was approved by the CPRD Independent Scientific Advisory Committee (protocol 20_000288). All codes used to identify the study cohort, exposure, outcomes, and confounders, as well as the study protocol and the statistical code, are publicly available on GitHub (marymbarker22).

All individuals aged 16–50 years with their first-ever diagnostic code for type 2 diabetes recorded in CPRD between 1 January 2000 and 31 October 2020 (study end date) and with available linkage to HES and ONS constituted the exposed subcohort (i.e., individuals with type 2 diabetes). These patients were matched—by year of birth, sex, and general practice—to up to 10 patients with available HES and ONS linkage who did not have a diagnostic code for type 2 diabetes prior to the study end date, which constituted the nonexposed subcohort (i.e., individuals without type 2 diabetes). Any patients diagnosed with type 1 diabetes prior to the study end date were excluded (16). The index date for individuals in the exposed cohort was the date of type 2 diabetes diagnosis; the corresponding index date for the nonexposed cohort was the same as that of their matched, exposed individual. All individuals (exposed or nonexposed) with a diagnosis of cancer or cardiorenal disease (chronic kidney disease, end-stage renal disease, dialysis, kidney transplant, myocardial infarction, cerebrovascular disease) in HES on or before their index date were excluded. Supplementary Figs. 1 and 2 display the cohort flow diagrams for GOLD and Aurum, respectively.

Outcomes were all-cause and cause-specific mortality. Date and cause of death were extracted from ONS, with ICD–10 codes used to define the underlying cause of death as cancer, cardiorenal, or noncancer/cardiorenal. We combined cardiovascular and renal causes in view of the continuum between cardiovascular and renal disease (17). All participants were followed up from their index date to the earliest of death or the study end date.

Demographic confounders included sex, ethnicity, and deprivation status. Sex was extracted from CPRD; ethnicity data were extracted from HES (n = 11 categories). Linked Index of Multiple Deprivation 2010 data were used to assess deprivation status (range: quintile 1 [least deprived] to quintile 5 [most deprived]) (18). Further characteristics from CPRD included smoking status, systolic blood pressure, estimated glomerular filtration rate (eGFR), BMI, LDL cholesterol, and comorbidities (hypertension, anxiety, and depression identified using medical/diagnostic codes). All clinical and comorbidity variables were identified in CPRD using values recorded at or up to 5 years before the individual’s index date.

Baseline characteristics were summarized by exposure status. All-cause and cause-specific number of deaths and mortality rates, in the whole cohort and stratified by age group (16–27, 28–31, 32–35, 36–39, 40–43, 44–47, and 48–50 years) were estimated. Royston–Parmar parametric models were used to quantify both relative (i.e., hazard ratio [HR]) and absolute (i.e., risk difference) risk of all-cause and cause-specific mortality, comparing individuals with versus without type 2 diabetes (19). Complete-case regression models were adjusted for age at index date (i.e., age at type 2 diabetes diagnosis), sex, ethnicity, deprivation status, smoking status, hypertension, and included an interaction between exposure status (type 2 diabetes/no type 2 diabetes) and age at diagnosis. Ten-year adjusted cumulative incidences of competing causes of death and their differences by exposure status were estimated for varying ages at diagnosis (16–50 years) using standsurv in Stata (20).

In a sensitivity analysis, we investigated the relative and absolute risks in a model that included BMI as a further confounder or use of antihypertensive medications instead of diagnosis of hypertension. We also quantified all-cause and cause-specific mortality rates in a subcohort of individuals with nonmissing BMI, LDL, eGFR, and systolic blood pressure data (cohort 2). Analyses were conducted in Stata (18.0 MP) and results are reported with 95% CIs.

The patient-level data used in this study cannot be shared but can be accessed via CPRD. All codes used to identify the study cohort, exposure, outcomes, and confounders, as well as the study protocol and the statistical code, are publicly available on GitHub (marymbarker22).

A total of 938,007 individuals (53.1% male) were included: 108,061 with type 2 diabetes and 829,946 without (Table 1). The median age at diagnosis of individuals with type 2 diabetes and matched index date for those without was 44 (interquartile range [IQR] 39–47) years. Only 3.2% of individuals with type 2 diabetes were aged between 16 and 27 years at diagnosis. Over the study period, the total number of individuals with a diagnosis of type 2 diabetes increased progressively (from 3,312 in 2000 to 5,648 in 2019), whereas the proportions across age groups remained relatively stable (Supplementary Fig. 3).

White race was the most common (79.6%) yet less in those with versus without type 2 diabetes (68.2% vs. 81.0%). The proportion of individuals from the most deprived quintile was slightly higher in those with versus without type 2 diabetes (33.4% vs. 29.1%), whereas a larger proportion of individuals with type 2 diabetes reported active smoking (23.8% vs. 12.5%). Values of BMI, systolic blood pressure, LDL cholesterol, and eGFR were all higher among individuals with type 2 diabetes, and comorbidities were more common in this group. In individuals with type 2 diabetes, 35.3% were prescribed any glucose-lowering medication in the 12-month period before the diagnostic code of type 2 diabetes appeared: the most common was metformin (31.7%), and prescriptions of glucagon-like peptide 1 (GLP-1) agonists and/or sodium–glucose cotransporter-2 (SGLT2) inhibitors were very low (0.1%). The use of lipid-lowering (18.5% vs. 1.9%) and antihypertensive (29.6% vs. 5.2%) medications, antidepressants (22.3% vs. 7.7%), and benzodiazepines (5.5% vs. 2.3%) was considerably more common in individuals with type 2 diabetes.

During a median follow-up of 9.5 (IQR 4.8–14.5) years (Supplementary Table 1), 34,235 deaths (3.6%) occurred; the follow-up was shorter in younger participants (8.2 [IQR 3.9–13.6] years in the age group 16–27 years) and progressively increased in older participants (9.8 [5.0–14.6] years in the age group 48–50 years). The number of all-cause and cause-specific deaths in the main cohort and two subcohorts, with corresponding crude mortality rates, are shown overall in Supplementary Table 2 and stratified by age at diagnosis in Supplementary Tables 3–6.

The relative risk of both all-cause and cause-specific mortality comparing individuals with versus without type 2 diabetes declined with increasing age at diagnosis (Fig. 1). The relative risk (95% CI) of all-cause mortality was highest in the 16- to27-year age group (HR 4.32; 3.35–5.58) and decreased to 1.53 (1.46–1.60) in individuals aged 48–50 years. Similarly, the relative risk of cancer-specific mortality was highest among those 16–27 years old (3.74; 1.86–7.55), decreased to 1.28 (0.90–1.82) in those 32–35 years old, and was stable in older age groups (48–50 years old; 1.28; 1.18–1.38). The relative risk (95% CI) of cardiorenal mortality was 10.67 (3.97–28.65) in the 16- to 27-year age group and 2.01 (1.75–2.32) in individuals aged 48–50 years; corresponding estimates (95% CI) were 4.08 (3.06–5.45) and 1.66 (1.55–1.76) for noncancer or cardiorenal mortality. In individuals with type 2 diabetes, the HR (95% CI) per 5-year increase in age at diagnosis was 1.30 (1.27–1.33) for all-cause, 1.67 (1.58–1.77) for cancer, 1.35 (1.26–1.45) for cardiorenal, and 1.18 (1.15–1.21) for noncancer or cardiorenal mortality.

For all ages at diagnosis, the 10-year survival was lower in individuals with versus without type 2 diabetes; in individuals with and those without type 2 diabetes, 10-year survival decreased with age at diagnosis (and matched age in those without type 2 diabetes; Fig. 2), whereas the difference in survival between individuals with and without type 2 diabetes increased with age (Supplementary Fig. 4). At 16 years of age, the 10-year survival rate (95% CI) was 98.8% (98.6–99.0) in individuals with and 99.7% (99.7–99.7) in those without type 2 diabetes. Corresponding estimates (95% CI) were 97.6% (97.4–97.8) and 99.1% (99.1–99.1) at 30 years of age, and 93.6% (93.3–93.8) and 95.7% (95.6–95.8) at 50 years of age. The corresponding 10-year survival differences (95% CI), comparing individuals with versus without type 2 diabetes, were −0.9 (−1.1 to −0.7) percentage points at 16 years, −1.5 (−1.7 to −1.3) at 30 years, and −2.1 (−2.4 to −1.9) at 50 years (Supplementary Fig. 4).

For all investigated causes of death, the mortality risk was higher in individuals with versus without type 2 diabetes (Fig. 3). The 10-year risk (95% CI) of cancer mortality for individuals diagnosed at 16 years of age was 0.1% (0.0–0.1) and 0.0% (0.0–0.0) in individuals with and without type 2 diabetes, respectively, resulting in a difference of 0.0 (0.0–0.1) percentage points (Supplementary Fig. 5). At 30 years of age, the 10-year cancer mortality risks (95% CI) were 0.3% (0.2–0.3) and 0.2% (0.2–0.2) in individuals with and without type 2 diabetes, respectively, corresponding to a difference of 0.1 (0.0–0.1) percentage points. Estimates (95% CI) at 50 years of age were 2.2% (2.0–2.3) in individuals with type 2 diabetes and 1.8% (1.7–1.8) in those without, resulting in a mortality difference of 0.4 (0.3–0.6) percentage points.

The 10-year risk of cardiorenal mortality at 16 years of age was 0.1% (95% CI 0.1–0.1) and 0.0% (95% CI 0.0–0.0) in individuals with and without type 2 diabetes, respectively (Fig. 3). Corresponding estimates (95% CI) were 0.2% (0.2–0.3) and 0.0% (0.0–0.1) at 30 years of age; and 0.7% (0.6–0.8) and 0.4% (0.3–0.4) at 50 years of age. These risks resulted in mortality differences (95% CI), comparing individuals with versus without type 2 diabetes, of 0.1 (0.0–0.1) percentage points at 16 years, 0.2 (0.1–0.2) at 30 years, and 0.3 (0.2–0.4) at 50 years (Supplementary Fig. 5).

For noncancer or cardiorenal mortality, the 10-year risk (95% CI) at 16 years of age was 1.2% (1.0–1.5) and 0.3% (0.3–0.3) in individuals with and without type 2 diabetes, respectively (Fig. 3). Corresponding estimates (95% CI) at 30 years of age were 2.0% (1.8–2.1) and 0.7% (0.7–0.7); at 50 years, they were 3.7% (3.5–3.9) and 2.3% (2.2–2.3), respectively. The mortality risk differences (95% CI) comparing individuals with versus without type 2 diabetes were 0.9 (0.7–1.1) percentage points at 16 years, 1.3 (1.0–1.4) at 30 years, and 1.4 (1.3–1.6) at 50 years (Supplementary Fig. 5).

The sensitivity analysis including BMI as a further confounder beyond sex, ethnicity, deprivation status, smoking status, and hypertension (main analysis) resulted in virtually identical relative (Supplementary Fig. 6) and absolute (Supplementary Fig. 7) risk differences for both all-cause and cause-specific mortality. Likewise, replacing medical history of hypertension with use of antihypertensive medications resulted in very similar estimates (Supplementary Figs. 6 and 7), yet differences at age 50 years were slightly attenuated: the difference in 10-year survival comparing individuals with versus without type 2 diabetes was −1.3 percentage points in the sensitivity analysis (vs. −2.1 in the main analysis) as a result of a smaller noncancer or cardiorenal 10-year mortality risk difference (1.4 vs. 0.8 in the main and sensitivity analysis, respectively; Supplementary Fig. 7).

In this study, we investigated the association between a wide range of ages at type 2 diabetes diagnosis and both the relative and absolute risk of multiple causes of death. Aligning with findings from previous research (Supplementary Fig. 7) (7–10), we found that the relative risk of all-cause mortality in individuals with versus without type 2 diabetes was highest at younger ages of diagnosis. This trend was also observed for cardiorenal and noncancer/cardiorenal–related mortality. Conversely, the relationship between age at diagnosis and cancer-specific mortality was less clear; although a younger age at diagnosis was associated with a higher relative risk of cancer-specific mortality between 16 and 32 years, the magnitude of the association was roughly stable thereafter, with an HR of approximately 1.3. Differing associations between age at type 2 diabetes diagnosis and mortality according to the cause of death have previously been reported, whereby a younger age at diagnosis was associated with a higher risk of all-cause and cardiovascular mortality but lower cancer mortality (21). The 10-year absolute mortality risk was higher in individuals with versus without type 2 diabetes, irrespective of age at diagnosis and cause of death; however, the risk differences over the 10 year period were small.

All previous cohort studies exploring the relative risk of mortality by age at type 2 diabetes diagnosis found increased relative risks of mortality (all-cause and cardiovascular-related) at younger ages at diagnosis (7–10,21), except one (Supplementary Fig. 7) (22). The current study provides further support for these findings and extends the available knowledge by investigating the relative risk of multiple causes of death. While some previous studies have explored both all-cause and cardiovascular mortality by age at diagnosis (8,10), to our knowledge none has also estimated the absolute risk of multiple, competing causes of death, including cancer.

Previous evidence would suggest the implementation of strategies aimed at identifying type 2 diabetes at younger ages and tailoring screening and therapeutic approaches for the prevention of diabetes–related complications and death in these individuals (6,12). However, we observed a striking divergence between the relative and absolute mortality risk associated with type 2 diabetes across ages. The magnitude and age-specific trends in the relative risks, which were higher at younger ages (up to an HR of 11 at 16–27 years of age at diagnosis) and progressively smaller at older ages, and in the absolute risks—conversely smaller at younger ages and progressively higher at older ages (up to a 10-year mortality risk difference of 21 more deaths per 1,000 individuals at 50 years)— highlight the greater effect of age, rather than the relative risk associated with type 2 diabetes, on the risk of death within this range of ages. These results underline the necessity to report both relative and absolute risks, particularly when investigating the risk of complications and death in individuals at a low absolute risk of events, such as in individuals with type 2 diabetes diagnosed at young age.

The magnitude of the 10-year mortality risk observed in this study would therefore suggest that preventive strategies aiming to reduce the risk of death may translate into small absolute mortality benefits during this time horizon (23). However, such interventions could have a larger impact on the morbidity experienced by these individuals. Indeed, previous research has reported a higher risk of non–fatal cardiorenal complications and other comorbidities, including psychiatric disorders, in adults with early-onset type 2 diabetes (4,24,25). The high level of morbidity observed in these adults, who represent a working age population at their “peak” time of productivity, is also likely to have large socioeconomic effects, in addition to effects on individuals’ quality of life, all of which could potentially be reduced by effective interventions aiming to reduce morbidity. However, as this study investigated mortality only and the trajectories of subclinical chronic diseases may last several decades before they become clinically evident, further research is required to comprehensively examine the absolute risk of morbidity by diagnostic age and inform the design of age-specific interventions.

We investigated the effect of age at diagnosis on the risk of death over 10 years, as this is the suggested time frame for informing decisions regarding patients’ treatment needs for the prevention of cardiovascular disease (23). In our study, the median follow-up time was 9.5 (IQR 4.8–14.5) years, making our estimation at 10 years robust and less related to model-based extrapolation. It may be argued that the excess mortality in this phenotype of type 2 diabetes can be better delineated only over a longer follow-up time; however, given the starting age of observation and current life expectancies, studies following individuals for several decades would be required to estimate the mortality risk for time horizons greater than 10 years. Our findings also underline that further investigations are required to disentangle the relative contribution of age at diagnosis, attained age, and diabetes duration on the absolute risk of diabetes complications and mortality in young individuals with type 2 diabetes, as the literature is limited in this respect (26,27).

This study has several strengths. It is one of the first to investigate the association between a wide range of ages at diagnosis and the relative and absolute risk of multiple causes of death. The use of a large, multiethnic, nationally representative cohort (13,14), from medically accurate data sources (28), enhances both the generalizability and validity of the findings. Additionally, the analysis of multiple causes of death, accounting for competing risk, allowed us to identify age-related differences in the causes of death, which may vary given the heterogeneous phenotypes of diabetes-related complications across different ages at diagnosis (12). The mortality excess observed in individuals with type 2 diabetes was mainly related to noncardiorenal or cancer deaths, with diseases of the digestive system being the most common underlying cause in individuals with type 2 diabetes. In contrast, cardiorenal diseases and cancer represent the primary causes of death in adults with later-onset type 2 diabetes (29). As such, our findings extend the current knowledge on the early-onset type 2 diabetes phenotype by suggesting that differences exist not only in the clinical characteristics at diagnosis but also in the primary causes of death. Of note, previous studies have reported a change in the causes of mortality in subjects with type 2 diabetes in the last two decades in the U.K. (30): to what extent the increasing prevalence of young type 2 diabetes has contributed to this transition remains unclear.

Some limitations of the study should also be acknowledged. First, although our findings can be applied to England, owing to the representative nature of the data, generalization to other global regions must be exercised with caution, particularly in countries with different health care systems. Although both CPRD and HES data have high medical accuracy (28), errors in data collection are possible, potentially reducing the validity of the study’s conclusions. The electronic health records used in this analysis have been collected for administrative purposes and, following the introduction of the U.K. Quality and Outcomes Framework pay-for-performance scheme in 2004, the frequency and quality of coding may have changed (31). However, the observation of virtually no prescriptions of GLP-1 agonists or SGLT2 inhibitors in the year before the index date for type 2 diabetes suggests no meaningful lag-time bias in the ascertainment of type 2 diabetes. Moreover, we used clinical codes to identify individuals with type 2 diabetes; while algorithms based on clinical-demographic characteristics, medications, and laboratory tests have been suggested (32), the accuracy of clinical codes to identify type 2 diabetes against algorithms based on multidimensional data are very high (33). To assess the robustness of our estimations in relation to data quality, we quantified the mortality rates in two nested cohorts: cohort 1, with nonmissing data for well-established confounders and prognostic factors of death; cohort 2, a subcohort with further nonmissing data for other factors (i.e., blood-based biomarkers and BMI). As expected, the subcohort resulted in a smaller number of individuals; however, the mortality rates were largely overlapping across the two cohorts. In a trade-off between data availability and clinical plausibility, we extracted clinical characteristics in CPRD within 5 years before the index date; however, we recognized that shorter time windows could also be considered for some time-varying characteristics (i.e., systolic blood pressure). We accounted for differences between individuals with and without type 2 diabetes by design (matching) and analytically (regression adjustment): an alternative approach would have been to match for a larger set of confounders, yet any residual imbalance potentially resulting from the matching process was accounted for in the regression models (34). To define the cause of death, we relied on the ICD-10 coding system used in the ONS database; although this is the standard approach used by ONS and the World Health Organization in their official nationwide mortality reports, previous evidence suggests geographical and temporal variations in the completeness of ICD-10 codes in death certificates as well as incorrect reporting of causal sequences (35,36). Finally, like all observational studies, it is not possible to infer causality from our findings.

In this large, population-based cohort study, we assessed the association between age at diagnosis and mortality in individuals with type 2 diabetes diagnosed between 16 and 50 years of age. We found that the relative risk of mortality in younger individuals with versus without type 2 diabetes ranged from 4 to 11, depending on the cause of death. These results highlight the substantial burden associated with type 2 diabetes in this age group. However, the 10-year mortality risk differences between the two groups were small. To further detail the long-term risk of morbidity and mortality in individuals with early-onset type 2 diabetes, studies with a follow-up of several decades are required.

Funding. This study was supported by the National Institute for Health and Care Research (NIHR) under its Grants for Applied Research Program (NIHR201165), as well as by the NIHR Leicester Biomedical Research Centre and the NIHR Applied Research Collaboration East Midlands. M.J.D. is cofunded by the NIHR Leicester Biomedical Research Centre and University of Leicester. K.K. and F.Z. are supported by the NIHR Applied Research Collaboration East Midlands and the NIHR Leicester Biomedical Research Centre.

The views expressed are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health and Social Care. 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. J.C.N.C. reported receiving grants and/or honoraria for consultancy or giving lectures from AstraZeneca, Bayer, Boehringer Ingelheim, Celltrion, Eli Lilly, Hua Medicine, Lee Pharm, Merck Serono, Merck Sharp & Dohme, Pfizer, Sanofi, Servier, and Viatris Pharmaceutical. J.C.N.C. is the chief executive officer (pro bono) of the Asia Diabetes Foundation that designed and implemented the JADE platform; is the co-founder of GemVCare, a biotech start-up company, with partial support from the Hong Kong Government, which uses biogenetic markers and information technology in pursuit of prevention and precision care in diabetes through partnerships. K.K. has acted as a consultant or speaker or received grants for investigator-initiated studies for AstraZeneca, Novartis, Novo Nordisk, Sanofi, Lilly, Merck Sharp & Dohme, Boehringer Ingelheim, and Bayer. M.J.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.J.D. has 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. M.M.B. contributed to the study protocol, data preparation, statistical analysis, and drafting the manuscript. M.J.D. and J.A.S. contributed to the study protocol, funding, and manuscript revision. J.C.N.C., J.A.S., E.W.G., and K.K. contributed to data interpretation and revision of the manuscript; S.S. contributed to data preparation and revision of the manuscript; and F.Z. contributed to the study protocol, drafting the manuscript, and study overview. M.M.B. 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.