Age at Diagnosis of Type 2 Diabetes Mellitus and Associations With Cardiovascular and Mortality Risks

Introduction

Given the rising prevalence of obesity, especially in younger people over the past 3 to 4 decades in high-income countries,¹ type 2 diabetes mellitus (T2DM) is now more frequently diagnosed in young adults and adolescents.^2–4 This is a worrying trend because, as we⁵ and others⁶ have shown, these individuals have worse risk factor profiles (body mass index [BMI], lipids, and glycemia levels) at the point of diagnosis relative to those diagnosed at older ages, and worse clinical outcome trajectories^5,7 over time, as well. It is therefore likely that younger-onset T2DM may pose relatively greater excess cardiovascular mortality and morbidity risks in comparison with later-onset T2DM. In line with this notion, having T2DM at younger ages is associated with greater losses of expectancy⁸ and greater mortality risks, as well,⁹ relative to age-similar controls, and such losses may be related more to premature cardiovascular deaths than other causes. However, these latter studies did not compare mortality risks from the time of diagnosis and they did not fully adjust for the impact of diabetes mellitus duration, which is independently associated with greater cardiovascular risks. Furthermore, few previous studies have compared risks for cause-specific mortality (cardiovascular versus noncardiovascular) at the same time as examining a wide variety of nonfatal outcomes (eg, acute myocardial infarction, heart failure) by the age at onset of T2DM. Finally, few studies have been adequately powered to allow investigation across the entire age range of age at T2DM diagnosis, which is important, given that the treatment and screening options at either end of the age spectrum are increasingly debated.

The aim of this study was therefore to evaluate, in relation to the age at diagnosis: (1) detailed mortality and cardiovascular disease (CVD) mortality risks; (2) cardiovascular risks inclusive of acute myocardial infarction, stroke, and heart failure outcomes; (3) noncardiovascular mortality risks; and (4) life expectancy relative to nondiabetes counterparts. We repeated our analyses in a cohort without any previous CVD, given the clinical importance of this question, and looked at the entire cohort, and separately in both sexes.

Methods

Data Sources

Methods used in this investigation are available to any researcher worldwide. Patient data are not readily available because of Swedish and European privacy laws. We welcome any inquiries regarding data acquisition, which is allowed under Swedish and European law (visit https://ndr.nu for more information). The Swedish National Diabetes Register (NDR), started in 1996, is a nationwide register and includes patients with diabetes mellitus (both type 1 and type 2) aged ≥18 years. The data are collected by trained nurses and physicians and include information obtained in primary care and at hospital outpatient clinics. Patient data are either continuously reported via electronic patient clinical records, or registered directly online into the NDR.¹⁰ The study was approved by the ethics review board at the University of Gothenburg. Each patient provided informed consent (verbal or written) for inclusion in the register, and >90% of all patients with T2DM in Sweden are included.

T2DM is defined using an epidemiological definition: recorded T2DM in clinical records plus treatment with diet with or without the use of oral antihyperglycemic agents or treatment with oral antihyperglycemic agents with or without the use of insulin. The latter category only applied to patients who were ≥40 years of age at the time of diabetes mellitus diagnosis. The study includes all individuals with at least one record in the NDR between January 1, 1998, and December 31, 2012. For each individual with T2DM in the NDR, 5 matched controls were identified, matched for age, sex, and county, randomly selected from the general population, as reported before.^11,12 Primary analyses were performed in persons without preexisting CVD, including coronary heart disease (CHD), acute myocardial infarction, stroke, heart failure (HF), atrial fibrillation (AF), or dementia; analyses in the entire cohort with or without prevalent CVD were also performed. To reflect current conditions and management of T2DM, the analysis was restricted to persons with a diabetes mellitus duration of <10 years when first registered in the NDR.

Information regarding coexisting conditions, cardiovascular outcomes, and deaths was retrieved from the Swedish Inpatient and Cause of Death Registries. Data linkage is virtually complete because of the unique personal identification numbers, which are assigned to all Swedes from birth or immigration. Sensitivity and specificity for all outcomes have been validated in the Swedish Inpatient Register.¹³

The Longitudinal Database for Health Insurance and Labor Market studies provided information on individual incomes, country of birth, marital status, and highest educational level.

Outcomes

We assessed all-cause mortality, cardiovascular mortality, acute myocardial infarction, stroke, CHD, hospitalization for HF, and AF. The outcomes were identified from the Inpatient Registry using the International Classification of Diseases, Ninth and Tenth Revision, codes. The specific codes were as follows: acute myocardial infarction, 410 and I21; CHD, 410 to 414 and I20 to I25; HF, 428 and I50; AF, 427D, I48; and stroke, 431 to 434, 436, I61 to I64. The codes are listed in Table I in the online-only Data Supplement.

The last date for NDR registration (ie, inclusion) in the present analysis cohort was December 31, 2012. Patients were followed until December 31, 2013 for all outcomes, with the exception of mortality, for which follow-up ended December 31, 2014.

Statistical Methods

The associations between age at diagnosis of T2DM and mortality, and cardiovascular-related outcomes, as well, were analyzed using the Cox proportional hazards model with age as the underlying time scale. The models contain both persons with T2DM and diabetes mellitus–free matched controls coded such that the baseline hazard models the hazard function for the controls. The association between being diagnosed with diabetes mellitus at a certain age and outcomes is captured by a main effect term, and the effect of living with diabetes mellitus is captured by an effect of the yearly time-updated duration of diabetes mellitus. The matched controls are persons without T2DM diagnosis (ie, their diabetes mellitus duration equals zero) and therefore have no contribution in the model from the terms for duration of T2DM or age at T2DM diagnosis for the outcomes assessed. The persons with diabetes mellitus are modeled to get an instant increase in hazard at the date of diagnosis and have an effect of time-updated duration that captures the gradual increase in hazard beyond the effect of aging shared with the controls (please see online-only Data Supplement for more explanation and Figures I and II in the online-only Data Supplement). For people with diabetes mellitus, we centered diabetes mellitus duration around the population mean (2.52 years) by subtracting the mean from each individual’s diabetes mellitus duration. An individual with T2DM who had 2.52 years of duration would thereafter have 0 years of duration and can therefore be compared with controls. Consequently, we estimated the excess risks in T2DM after 2.52 years of duration. Additional model details are provided in the online-only Data Supplement. The main analyses included all patients with T2DM without previous CVD, with parallel analyses repeated in the entire cohort, including those with CVD, in which case all analyses were adjusted for CVD.

A robust statistical method was developed to analyze the associations between duration of diabetes mellitus and outcomes as detailed in the online-only Data Supplement. This is based on the principle that a patient with T2DM ages differently once T2DM develops in comparison with a person without T2DM.

The assumption of proportional hazard was evaluated by fitting a smoothing spline function for the duration of T2DM. The best fit was achieved for log-transformed duration but the choice of transformation for duration of T2DM had only a minor influence on the analyses results. Descriptive statistics were based on the means and SDs for age and income, and absolute and relative frequencies for discrete variables.

The predicted conditional survival functions were derived from Cox regression models with T2DM versus controls as the only independent variable. Age was used as the time scale, with left censoring at age of inclusion into the analysis cohort. The conditional median survival was estimated from the middle of each age interval, with the exception of 0 to 20, where age 15 is used, and 90, where 95 is used. These analyses were based on all persons without previous CVD without any restriction on duration of T2DM. The estimated cumulative hazard was subsequently converted to conditional survival.

Because of the explorative nature of the study, no adjustments for multiple comparisons were made and conclusions should be based on overall patterns rather than single-hypothesis tests of confidence intervals.

The statistical analysis was performed using R 3.4.0.

Role of the Funding Source

The Swedish Association of Local Authorities Regions provided financial support for NDR. We also recognize funding from the Swedish Heart and Lung Foundation, and the Swedish Research Council (2013–5187, SIMSAM). They had no role in the conduct of this study. Dr Sattar had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results

Baseline Characteristics

Data were initially analyzed for 214 278 patients with T2DM without previous CVD and 1 363 612 controls matched on age, sex, and county of residence, and repeated in the entire cohort when the respective numbers were 318 083 and 1 575 108. Their baseline characteristics are shown in Table 1 (cohort without prevalent CVD) and Table II in the online-Data Supplement (entire cohort), respectively, and their distribution of age at T2DM diagnosis is presented in Figure 1, demonstrating a mean age at diagnosis of 61.79 years and a spread of ages spanning from 9 to 101 years of age. By matching design, mean ages and sex distribution of cases and controls were similar for cases and controls.

• Download figure
• Download PowerPoint

Within the primary analysis subset of patients with T2DM without previous CVD, as presented in Table 1, patients with T2DM in comparison with controls had lower average income, were more commonly born beyond the European Union, had higher prevalence of amputation and of previous cancer, but lower prevalence of dementia at baseline. The same data in the entire cohort (Table II in the online-only Data Supplement) showed similar patterns, but more patients with T2DM had prevalent CVD, although prevalence of previous dementia remained lower.

Baseline risk factor profiles by the age at diagnosis of T2DM are presented in Table 2. As anticipated, there was an inverse relationship between the age at diagnosis of T2DM and BMI, with BMI in those diagnosed with diabetes mellitus at <40 years of age being around 8 units higher than if they developed diabetes mellitus in their 90s (Table 2). Likewise, hemoglobin A_1c at diagnosis declined with rising age at diabetes mellitus diagnosis, being ≈5 mmol/mol (0.47%) higher in those <40 years of age in comparison with those 71 to 80 years of age, although hemoglobin A_1c levels were somewhat higher in those ≤91. In terms of risk factors, triglyceride concentrations were higher and high-density lipoprotein cholesterol levels lower in patients with younger age at diagnosis of T2DM, whereas blood pressure levels rose with rising age at diagnosis (Table 2). Total and low-density lipoprotein cholesterol were slightly higher in middle ages but declined at older ages of diagnosis, whereas low-density lipoprotein cholesterol–lowering treatment was least at the extremes of age and highest in participants 61 to 80 years old (Table 2). Similar patterns were seen in the entire cohort (Table III in the online-only Data Supplement).

Mortality and Adverse Cardiovascular Outcomes

Median follow-up was 5.63 years. Overall, a total of 194 197 death events, 66 184 cardiovascular death events, 51 837 myocardial infarctions, 60 346 strokes, 61 501 events of incident HF, and 83 283 incident AF events were captured for analysis. Mortality and cardiovascular outcomes over time for T2DM versus controls stratified by the age at diagnosis of T2DM are presented in Figure 2 for the overall cohort and for the subset of patients who have T2DM without prevalent CVD at registration. There was a higher risk of all cardiovascular-related outcomes in the T2DM cohort regardless of the age at diagnosis of T2DM, with notable and consistent associations between age at diagnosis of T2DM and all outcomes analyzed, where those with T2DM diagnosed at ≤40 had the highest excess relative risk for most outcomes with adjusted hazard ratios (HRs; 95% CI) in those without previous CVD for total mortality, for cardiovascular-related mortality, and for noncardiovascular mortality well above one. HRs for those diagnosed with T2DM at ≤40 years for other CVD outcomes were even higher, greatest for those with HF and CHD, as noted below:

• Download figure
• Download PowerPoint

• 2.05 (1.81–2.33) for total mortality

• 2.72 (2.13–3.48) for cardiovascular-related mortality

• 1.95 (1.68–2.25) for noncardiovascular mortality

• 4.33 (3.82–4.91) for CHD

• 3.41 (2.88–4.04) for acute myocardial infarction

• 3.58 (2.97–4.32) for stroke

• 4.77 (3.86–5.89) for HF

• 1.95 (1.56–2.44) for AF

Thereafter, incremental risks generally declined with each higher decade age at diagnosis of T2DM. For participants with T2DM diagnosed at >80 years of age, adjusted relative risks reversed with most <1.0 for T2DM versus controls for all 3 of total mortality (0.83 [0.80–0.86]), cardiovascular mortality (0.75 [0.71–0.79]), and noncardiovascular mortality (0.87 [0.83–0.91]) outcomes. Adjusted risks for other outcomes that include nonfatal outcomes were generally >1.0, but all HR estimates substantially attenuated in comparison with relative incremental risks in those diagnosed with T2DM at younger ages. In those with T2DM diagnosis at age >90, the only outcome for which risk seemed to be appreciably higher for those with T2DM versus controls was stroke (HR, 1.56 [1.13–2.14]). Figure 3 presents causes of death in each age-of-onset group, along with matched controls. Noncardiovascular mortality among people with T2DM was primarily driven by external causes and cancers.

• Download figure
• Download PowerPoint

Results Stratified by Sex

When outcomes were examined separately by sex, doing so in those without all previous CVD (Figure III in the online-only Data Supplement), the results were broadly comparable in terms of patterns or risk by age at diagnosis, although HRs were usually higher in women for most CVD-related outcomes, and, in particular, for outcomes related to CHD, stroke, and HF (P<0.0001 for sex by age interactions for all outcomes). However, the higher excess risk in women was age dependent, being stronger among younger individuals and not present above the age of 70.

Results in the Entire Cohort

All analyses were repeated in the entire cohort so that all participants with or without previous CVD were included with multivariable analyses additionally adjusted for baseline CVD (Figure 2). In this case, the main patterns of results were broadly identical to those already presented in the cohort without all previous CVD. The only minor difference noted was a slightly earlier (in terms of decade of age) attenuation to the null for mortality outcomes, whereas hazard ratios for HF were a little higher for the comparison of T2DM versus control.

Median Survival in Those With Previous CVD by Age at Diagnosis of T2DM

Differences in survival analyses in individuals with versus without T2DM, stratified by age at diagnosis, were estimated (Figure 4, Table IV in the online-only Data Supplement), with a median loss of life being near 12 years when T2DM was diagnosed at ≈15 years of age, ≈6 years when diagnosed at 45 years of age, 2 years when diagnosed at 65 years of age, and no accelerated loss of life after 80 years or so. Life-years lost appeared somewhat greater in younger women than in men of the same age (data available on request), although wider confidence intervals prevent meaningful comparisons.

• Download figure
• Download PowerPoint

Discussion

In this evaluation of mortality and CVD outcomes associated with T2DM analyzed by age at diagnosis using data from a national registry, younger age at diagnosis is associated with higher subsequent risk for all outcomes analyzed. Diagnosis of T2DM at a younger age was also associated with the greatest loss of life-years. The risks for a range of nonfatal CVD outcomes are even more markedly elevated in those with T2DM diagnosed at a younger age, in particular, for CHD and HF, where incremental relative risks approach 4 to 5 times higher than matched controls. Incremental risks associated with T2DM are attenuated by the age at diagnosis of T2DM, so that by the time T2DM is diagnosed at the age of >80, adjusted mortality risks are <1.0 for both CVD mortality and non-CVD mortality, and excess risks for other outcomes are substantially attenuated; analysis of life-years lost with diagnosis of T2DM appears null at age >80. Finally, we show that these patterns are robust whether you consider only those without previous CVD, the entire cohort with or without previous CVD, or separately in men and women in those without CVD, although there was some evidence of greater excess risks in younger women developing T2DM.

At a clinical level, our findings may offer 2 important considerations: (1) a need to rethink risk factor treatment recommendations in those diagnosed with T2DM when <40 years of age (an age threshold commonly considered in guidelines), and (2) a need to reassess and discuss treatment goals and aggressiveness of interventions in people diagnosed after 80 years of age, in particular, in asymptomatic individuals. Whether it is cost-effective to screen for diabetes mellitus or prediabetes to identify people suitable for diabetes mellitus prevention programs in those >80 years of age is also questioned by our data.

In analyses evaluating associations between age at diabetes mellitus onset and CVD risks, analyzing data from just <8000 patients with newly diagnosed diabetes mellitus in the late 1990s,¹⁴ and before widespread preventative statin or antihypertensive use, they found a 14-fold higher myocardial infarction risk in those diagnosed with diabetes mellitus at <45 years old, whereas risks were 4-fold higher in those diagnosed at >45 years of age in comparison with age-matched controls.¹⁴ In a more recent study, prevalent diabetes mellitus at younger ages (<55) was associated with ≈3-fold greater mortality risks, whereas such risks were null in patients with diabetes mellitus over the age of 75.⁹ Data from the Emerging Risk Factor Collaboration likewise showed that life-years lost were significantly greater with diabetes mellitus present at younger ages, although, once again, this study did not examine risks by age at diagnosis.⁸ More recently, using only 2 age groups, a Chinese study suggested a near double the risk of nonfatal CVD in those with early (<40 years) versus later diagnosis of T2DM,⁶ whereas work from Australia showed higher mortality risk, in particular, CVD mortality, in younger patients diagnosed with diabetes mellitus in a study of just under three-quarters of a million Australians with diabetes mellitus.¹⁵ However, this latter study lacked access to individual controls and did not consider nonfatal outcomes. We were also able to show that a greater excess risk for CVD outcomes and mortality commonly attributed to women seemed to be far less evident in those with T2DM diagnosed at an older age (roughly >70 years of age), a novel finding extending considerable previous work on sex differences.

The present results meaningfully extend the relevant published data by: (1) evaluating the question in a large, national, contemporary cohort well characterized at baseline and with complete outcomes data capture; (2) evaluating an extensive range of all-cause mortality and cardiovascular outcomes; and (3) extending analyses of age at diagnosis of T2DM into the 10th decade, an important asset given the rising numbers of people living well beyond 80 years of age, especially in some countries like Japan. They also have the advantage over results from many previous studies by having age- and sex-matched controls and by adjusting for diabetes mellitus duration. Our findings are also potentially generalizable given that the trends in diabetes mellitus risks previously reported in Sweden seem to broadly match the findings in other high-income countries.

The potential mechanisms for incremental risk reductions and adverse effects on life-years lost at younger ages at T2DM diagnosis warrants some discussion. From Table 2, it is notable that diabetes mellitus diagnosis at younger ages is associated with considerably higher BMI. Beyond early T2DM incidence, these patients also have worse lipid profiles and higher glycemia levels than people developing T2DM when older.⁵ Although control data for BMI are lacking, population BMI in high-income countries is lower at younger ages and rises to a maximum at ≈50 to 70 years of age,¹⁶ suggesting that younger patients developing diabetes mellitus must be more obese than their nondiabetic counterparts. By contrast, BMI levels in older individuals developing diabetes mellitus must be closer to their counterparts without diabetes mellitus, so that other risk factors related to adiposity (eg, lipids, blood pressure) would also be less different. In other words, obesity and related risk factor perturbances more strongly accompany the development of diabetes mellitus at younger ages, leading, in turn, to greater relative increases in CVD risks. Younger patients developing diabetes mellitus also seem to smoke more and to have a lower socioeconomic status, both strong independent CVD risk factors.

Why might older (>80 years of age) patients diagnosed with diabetes mellitus have little or no difference in mortality rates relative to their nondiabetes counterparts? One potential is that such patients are better treated for CVD risk factors than those of similar age but without diabetes mellitus. However, because non-CVD mortality is also lower in diabetes mellitus diagnosed at >81 years of age in all analyses, other factors must be at play. One possibility is that to develop diabetes mellitus at older ages requires individuals to retain their weights better than their nondiabetes counterparts. A better weight retention would increase chances of developing diabetes mellitus, and keep blood pressures and other vascular risk factors (eg, cholesterol) slightly higher, as well, than controls not developing diabetes mellitus. However, some of those not developing diabetes mellitus may be at higher mortality risk because they are losing weight unintentionally because of comorbidities (eg, dementia, Table V in the online-only Data Supplement) that are linked to an earlier death. If this reasoning is correct, developing diabetes mellitus beyond the 9th or 10th decades of life may, in part, and in some patients, represent a part of the aging process. Of course, other unmeasured factors may also be at play, such as better health-seeking behavior in those developing diabetes mellitus at older ages.

Irrespective of the mechanisms, in stark contrast with younger patients, mortality risk in much older patients diagnosed with diabetes mellitus are not meaningfully elevated in comparison with their nondiabetes counterparts in Sweden, and even if nonfatal risks remain marginally elevated, there was no apparent loss in life-years (Figure 4). This finding needs to be replicated in other high-income countries, and, if confirmed, suggests that an upper age threshold of ≈80 years or so for diabetes mellitus could be helpful in diabetes mellitus–screening programs to enable better targeting of individuals with younger-onset diabetes mellitus (or prediabetes) who have much more to lose in terms of CVD risks and years of life.

These findings hold important clinical implications for CVD clinical management and prevention guidelines because they emphasize the need to be more aggressive both in population screening for T2DM and for consideration of more intensive cardiovascular risk modification among younger/newly diagnosed persons with T2DM. Currently, guidelines (eg, UK Joint British Societies 3, European Society of Cardiology cardiovascular prevention guidelines)^17,18 are less aggressive/prescriptive in management of risk factors in individuals developing diabetes mellitus at <40 years of age, and this group tends to have far slower uptake of preventative therapies. More recently, the UK National Institutes of Clinical Excellence guideline recommended a 10-year risk score to determine statin allocation, and, in this case, younger patients are likely to miss out on statins simply because of their low age. These and other guidelines need reconsideration because our data show that such patients are at highest relative risk of mortality and HF from younger ages (so highest lifetime risks) and stand to gain most from preventative therapies (and perhaps diabetes mellitus drugs with potential HF prevention benefits). It is well established than earlier interventions in those at excess risk yield greater lifetime benefits.¹⁹

Although this study has many strengths, including the ability to track people from the age at diagnosis of T2DM (or within 1 year of this date), and to match to controls of similar age, sex, and county, risk factor capture in controls was not systematic, precluding the ability to completely account for such risk factors. These data derived exclusively from the Swedish registry comprising almost exclusively white individuals, so further studies in different countries and in more heterogeneous populations are needed. We also acknowledge that classification of diabetes mellitus types is a complicated matter. Although a small percentage of the cohort may have late-onset type 1 diabetes mellitus, based on epidemiological estimates, the vast majority would have T2DM, and we believe that the overall conclusion about the importance of the age at onset would not meaningfully change. Of interest, we have recently shown that age at onset for type 1 diabetes mellitus also seems to have prognostic implications for subsequent adverse clinical outcomes.²⁰ We also acknowledge that some controls may have developed diabetes mellitus after baseline but would anticipate this to be a minority and to not meaningfully influence the results. Finally, correction for multiple testing was not performed, and thus caution is needed with respect to the interpretation of significance tests.

In conclusion, these nationwide data suggest that the mortality and cardiovascular harm associated with T2DM differs markedly by the age at diagnosis, with the highest mortality and especially CHD and HF risks in those with early diagnosed T2DM, whereas a slight survival benefit (both CVD and non-CVD related) appears present in patients with T2DM >81 years at diagnosis, reflected in no loss of life-years. These findings, in turn, reiterate the notion that the pathogenicity of T2DM differs markedly by age at diagnosis, highlighting, perhaps better than ever, the importance of age as an important risk stratifier in the management, screening, and preventative strategies for this chronic condition.

Acknowledgments

Drs Sattar, Araz Rawshani, and Gudbjörnsdottir contributed to concept of study. Drs Franzén and Araz Rawshani conducted the statistical analyses. All authors contributed to the writing and review of the manuscript. We thank Liz Coyle (University of Glasgow) for her excellent assistance in the preparation of this article.

Footnotes