Presence and Determinants of Cardiovascular Disease and Mortality in Individuals With Type 1 Diabetes of Long Duration: The FinnDiane 50 Years of Diabetes Study

The determinants of morbidity and mortality have been thoroughly explored in individuals with type 1 diabetes (T1D) duration of <30 years, with hyperglycemia identified as a major driver of vascular complications (1,2). However, data from the Swedish national registry and Medalist cohorts suggested that 30–40% of individuals remain free of chronic complications despite living with T1D for >50 years and having suboptimal glucose control (3–6), implying that these individuals share some protective factors against the complications (7). This may also reflect survivorship bias, because individuals who survived with diabetes for >50 years were also less likely to develop chronic complications.

However, even the world’s best registries may not capture all of the data on diabetes complications in the population with extreme duration of T1D. This is reflected by, for example, a very low reported prevalence of severe retinopathy (16.9%) after living with T1D for >50 years (5), possibly because these complications may have arisen years before registries started to collect data (e.g., National Patient Register since 1987 and outpatient doctor visits since 2001 in Sweden). In contrast, the Joslin Medalists represent a selected sample of patients who may have lower rates of complications than the entire population of patients with diabetes and may be vulnerable to volunteer bias; those individuals with the most advanced complications may not have wanted to receive a medal for having diabetes and were therefore not included (4).

Cardiovascular disease (CVD) is recognized as the main cause of mortality in T1D (8). Understanding risk factors or possibly even protective factors for CVD in the population with T1D of very long duration becomes increasingly important as we face an improved life expectancy in T1D (9). Consequently, an optimal treatment strategy is needed. Therefore, we aimed to, firstly, assess the incidence rates of CVD and mortality in individuals with T1D of very long duration in one of the largest and most thoroughly characterized T1D cohorts, the Finnish Diabetic Nephropathy Study (FinnDiane) cohort. To assure the representativeness of our cohort, we compared the incidence rates of CVD with the incidence rates of all Finnish individuals with the same T1D duration. Secondly, we compared the CVD incidence rates with sex-, age-, and geographical region–matched control individuals without diabetes. Thirdly, we explored risk factors for different CVD events, including coronary artery disease (CAD), stroke, and peripheral artery disease (PAD), that occur in this specific population.

FinnDiane is an ongoing multicenter study of individuals with T1D for the purpose of uncovering the risk factors and mechanisms of diabetic complications. The study design and methodology have previously been published (10). For this study, of 5,396 participants, we examined those who had been diagnosed with T1D in the year 1967 or earlier and therefore had the possibility to reach 50 years of diabetes duration until the end of year 2017 (n = 1,004). This study includes those individuals who reached 50 years of diabetes duration, comprising the FinnDiane 50-year cohort (n = 729). The remaining 275 individuals died before reaching 50 years of life with T1D (the FinnDiane comparison cohort). All FinnDiane participants provided written informed consent. The study protocol was approved by the ethics committee of the Helsinki and Uusimaa Hospital District, and the study was performed in accordance with the Declaration of Helsinki.

T1D was defined as a diabetes diagnosis before 40 years of age, with insulin initiated within 1 year of the diagnosis. At the FinnDiane baseline visit, serum samples were analyzed for HbA_1c by standardized assays at each study center. In addition, serial HbA_1c values were obtained from medical files before the baseline visit, and intraindividual mean HbA_1c and HbA_1c variability were assessed by the coefficient of variation of HbA_1c. At the baseline FinnDiane visit, lipids and creatinine were assessed by routine enzymatic methods and high-sensitivity C-reactive protein was assessed by radioimmunoassay (Human C-Peptide RIA Kit; Linco Research, St Charles, MO) at the central laboratory of the Helsinki University Hospital. The estimated glomerular filtration rate (eGFR) was calculated using the Chronic Kidney Disease Epidemiology Collaboration equation (11). The estimated glucose disposal rate (eGDR) to measure insulin sensitivity was calculated by the equation eGDR = 24.4 − 12.97 × WHR − 3.39 × HT − 0.60 × HbA_1c, where WHR is waist-to-hip ratio and HT is hypertension (yes = 1, no = 0) defined as having antihypertensive treatment or blood pressure ≥140/90 mmHg. Also, each participant collected timed urine samples for the measurement of urinary albumin excretion rate (AER) by radioimmunoassay (Pharmacia, Uppsala, Sweden). Albuminuria status was determined at the baseline FinnDiane visit and regularly during follow-up from at least two timed overnight or 24-h urine collections. Normal albumin excretion (normoalbuminuria) was defined as AER <20 μg/min or <30 mg/24 h, microalbuminuria as AER ≥20 and <200 μg/min or ≥30 and <300 mg/24 h, macroalbuminuria as AER ≥200 μg/min or ≥300 mg/24 h, and end-stage renal disease (ESRD) as receipt of dialysis treatment or a kidney transplant. Diabetic kidney disease (DKD) was defined as the presence of microalbuminuria, macroalbuminuria, or ESRD or eGFR <60 mL/min/1.73 m². History of laser treatment was used as a marker of severe diabetic retinopathy (SDR). Although anti–vascular endothelial growth factor treatment can also be used for the treatment of SDR, it was less common in Finland before the year 2010, and therefore, adding anti–vascular endothelial growth factor therapy as a marker of SDR did not identify new SDR cases before the study baseline visit. Self-reported data were available for smoking, alcohol consumption, and socioeconomic status, which was based on education and professional status. Blue-collar workers included both unskilled (basic education) and skilled (upper secondary school; matriculation examination and/or vocational institutes) workers, and they had lower socioeconomic status than white-collar workers, including both lower (university of applied science) and upper white-collar (university) workers.

To ensure the representativeness of our study population, we compared the CVD incidence in the FinnDiane 50-year cohort with that in the Finnish population of all individuals diagnosed with T1D during the years 1965–1967 at <30 years of age (n = 1,764). Of these, 957 (54.3%) reached 50 years’ duration (national 50-year cohort), whereas 807 (45.7%) died before this (national comparison cohort). These individuals were identified from the National Institute for Health and Welfare database. Individuals diagnosed at <18 years of age (60%) were included in the Diabetes Epidemiology Research International Group study (12). In addition, individuals <30 years of age with permanent reimbursement entitlement for free-of-charge insulin identified from the Finnish Drug Reimbursement Register (established in 1964) were added after being linked to the hospital discharge register and after exclusion of cases with a code indicating type 2 or secondary diabetes. In Finland, all individuals with T1D receive permanent reimbursement entitlement for free-of-charge insulin. The study obtained permission from the Finnish National Institute for Health and Welfare (THL/786/6.02.00/2016) and Statistics Finland (TK53-26-16).

Finally, to study the potential excess risk of CVD in the FinnDiane 50-year cohort, the data were compared with data from individuals without diabetes. To maximize the statistical power, we used all available control individuals without diabetes who had previously been identified by the FinnDiane study (i.e., 2–3 control individuals per each FinnDiane individual; total n = 12,710). These controls were age, sex, and place of residence matched and identified from the Population Register Centre. To be comparable, the data were split regarding follow-up time for each individual into 1-year calendar year and 1-year age intervals by sex as described in Statistical Methods.

Follow-up data on CVD outcomes and mortality for the FinnDiane 50-year cohort and all other cohorts were collected at the end of 2017 from the Finnish health registers. Nonfatal CVD events were retrieved from the Finnish Care Register for Health Care, which collects data from 1969 to the present with personal identity codes and therefore enables linking of data (13), and fatal events from the Cause of Death Register. CVD was defined as having either CAD, any stroke, or PAD. CAD was diagnosed in individuals having had either myocardial infarction (MI) or coronary revascularization (percutaneous coronary intervention or coronary artery bypass graft). Stroke was defined as either ischemic or hemorrhagic. PAD was defined as a nontraumatic lower-extremity amputation at any level or revascularization of lower-extremity arteries (Supplementary Table 1). Data on parental survival status were obtained from the Population Register Centre in 2016.

The cumulative incidence of CVD events from the diagnosis of diabetes in the FinnDiane 50-year cohort was estimated retrospectively using the Kaplan-Meier method. However, in the comparison of different subgroups, the cumulative incidence of CVD was analyzed by the Fine and Gray method, considering death as a competing risk to make the comparison equivalent to the FinnDiane 50-year cohort without deaths before the baseline visit. In this analysis, follow-up in the FinnDiane 50-year cohort and in the national 50-year cohort started from the diagnosis of diabetes and ended at the date of occurrence of the studied event, death, or the end of 2017. Follow-up in the individuals without diabetes started on the same day as diabetes diagnosis in the FinnDiane participant.

In the FinnDiane 50-year cohort, at first, the 50- and 60-year cumulative incidences of CAD, stroke, PAD, and all combined as CVD were retrospectively estimated and illustrated from the diagnosis of diabetes. Then, the cause-specific proportional hazards for each of the CVD events were estimated from the baseline visit; in these analyses, the prevalent events were excluded (14). Predictors included those that have previously been shown to be associated with CVD in T1D: age, duration of diabetes, sex, smoking, BMI, HbA_1c, dyslipidemia, arterial hypertension, DKD, and SDR presence. For HbA_1c, we used the mean of serial HbA_1c measurements as well as the coefficient of variation of serial HbA_1c. Because of well-known collinearity, all lipid variables could not be included in the same model. Therefore, we chose HDL cholesterol and triglycerides (ratio of triglycerides to HDL cholesterol), because we have previously shown that HDL cholesterol and triglycerides are among the best lipid predictors of CAD, whereas LDL cholesterol is not as good a predictor (15). Participants were considered to have hypertension if the mean blood pressure at the baseline visit was ≥140/90 mmHg or they were taking antihypertensive medication. The assumption of proportional hazards was tested by plotting Schoenfeld residuals against time and testing a nonzero slope by including time-covariate interactions.

Because albuminuria poses a strong effect on CVD risk, analyses were also performed separately for the individuals with normoalbuminuria. In these analyses, a predefined set of variables was not included because of a smaller number of CVD events; however, significant predictors were assessed by a stepwise method, separately for each CVD type.

The comparison between CVD morbidity in the FinnDiane 50-year cohort and that in the individuals without diabetes was conducted by calculating standardized incidence ratios (SIRs) as ratios of observed and expected numbers of cases. The expected numbers were derived by multiplying the number of person-years at risk by sex-, age-, and period-specific incidence rates observed in the individuals without diabetes. Follow-up started from the baseline visit. The SAS lexis macro was used to split the follow-up time for each individual into 1-year calendar year and 1-year age intervals (16). The 95% CIs were calculated assuming Poisson distribution.

Variables that were associated with all-cause mortality in the FinnDiane 50-year cohort were also studied. The causes of death were grouped based on the primary cause of death into acute diabetes complications ICD-10 codes (E100–E101), MI (I21–I22), ischemic heart disease (IHD) (I25), stroke (I60–I65), diabetic nephropathy (E102, N18), multiple diabetic complications (E107), malignancy (C00–C97), neurological disease (G00–G99), and other causes of death. CVD death was primarily defined as death resulting from MI, IHD, or stroke. Secondarily, the CVD death included death resulting from any disease of the circulatory system (I00–I99), including acute rheumatic fever, chronic rheumatic heart disease, hypertensive disease, IHD, pulmonary heart disease, and cerebrovascular and other forms of heart disease.

The excess mortality in the FinnDiane 50-year cohort was assessed by calculating the standardized mortality ratio as the ratio of the numbers of observed and expected deaths. The data were split into time at risk for each sex, 1-year age, and calendar year groups, and analysis was performed after duration of diabetes of 50 years. Expected deaths were calculated by multiplying the number of person-years at risk by sex-, age-, and calendar year–specific mortality rates (Statistics Finland).

For all tests, P < 0.05 was considered statistically significant. We used SAS 9.4 (SAS Institute, Cary, NC) and R open-source (https://www.r-project.org) statistical software for the analyses.

The median duration of diabetes in the FinnDiane 50-year cohort (n = 729) was 39.3 (interquartile range 35.3–45.0) years at the baseline visit and 54.5 (51.8–58.5) years at the end of follow-up (Table 1). At baseline, 25.5% had an eGFR <60 mL/min/1.73 m², 18.5% had microalbuminuria, 20.5% had macroalbuminuria, 9.1% had ESRD, and 63.2% had SDR. Of those with eGFR <60 mL/min/1.73 m², only 24 individuals (19 women, five men) had normoalbuminuria. Also, 19.1% had experienced a CVD event: 11.4% CAD, 5.3% PAD, and 5.6% stroke (Table 1); 15.0% had no advanced diabetic complications at all. Less than one-third had received lipid-lowering therapy, and less than two-thirds had received blood pressure–lowering therapy. A total of 9.5% still had a detectable serum C-peptide concentration. Supplementary Table 2 shows the differences between the clinical variables in the FinnDiane 50-year cohort and the FinnDiane comparison cohort. Most of the clinical and lifestyle risk factors were unfavorable for those who did not reach 50 years’ diabetes duration.

During a median follow-up of 16.6 (95% CI 13.5–18.6) years from the baseline visit, there were 206 incident CAD, 122 PAD, and 93 incident stroke events in 314 (43.1%) participants, and 181 individuals died in the FinnDiane 50-year cohort. Figure 1 presents the cumulative incidence of CVD events from the diagnosis of diabetes. The 50-year cumulative risks of CAD, PAD, and stroke were 24.7% (22.3–27.0%), 14.1% (11.9–16.3%), and 12.6% (10.5–14.7%), respectively. Cumulative incidence rates at 60 years’ diabetes duration reached 48.6% (46.0–51.1%), 29.7% (26.3–33.6%), and 23.6% (20.4–26.7%) for CAD, PAD, and stroke, respectively. These cumulative incidence rates were similar in the national 50-year comparison cohort (P = 0.90) (Fig. 1A). The 50-year cumulative incidence of CVD in the national 50-year cohort was 39.7% (37.8–41.5%), and in the 50-year FinnDiane cohort, it was 38.7% (36.5–40.8%). Figure 1A also shows the cumulative incidence curves for CVD in the FinnDiane comparison cohort and the national comparison cohort (P = 0.43 for difference), showing the considerably earlier onset of CVD compared with the 50-year cohorts.

Incidence rates of the CVD events in the FinnDiane 50-year cohort were increased when compared with individuals without diabetes (Fig. 1 and Supplementary Table 3). The SIRs for any CVD, CAD, PAD, and stroke events were 7.4 (95% CI 6.5–8.3), 7.2 (6.2–8.3), 20.4 (16.9–24.4), and 3.3 (2.7–4.0), respectively.

Of note, because half of the individuals from the FinnDiane 50-year cohort had normoalbuminuria, we analyzed the cumulative incidence of CVD events separately for this cohort (Fig. 2). Individuals with normoalbuminuria were more likely to be older and female and less likely to be current smokers, with later diabetes onset, better socioeconomic status, and lower indices of insulin resistance. They had lower baseline HbA_1c but not lower mean serial HbA_1c during follow-up, and they had better lipid and blood pressure profiles as well as fewer microvascular diabetes complications (Supplementary Table 4). However, compared with those in the population without diabetes, the cumulative incidences were substantially increased, even for the individuals with normoalbuminuria (Fig. 2 and Supplementary Table 3).

The risk factor profiles for different types of CVD events combined in the FinnDiane 50-year cohort are shown in Table 2. Importantly, mean HbA_1c and HbA_1c variability were both significant and independent risk factors for incident CVD. A 1% point increase in HbA_1c was associated with a 26% increase in the hazard for a CVD event. HbA_1c variability, but not mean HbA_1c, was a risk factor for mortality as well. Also, dyslipidemia (higher ratio of triglycerides to HDL cholesterol and the use of lipid-lowering medication) was associated with an increased CVD risk. Age and duration of diabetes were risk factors for both an incident CVD event and mortality. Having DKD increased the CVD risk by 1.81-fold (95% CI 1.33–2.47). Of note, higher BMI decreased the CVD risk in this population. Interestingly, the presence of SDR was significantly associated with an increased mortality hazard (Table 2).

In studying the effects of risk factors on different types of CVD events in this population, in addition to DKD, we identified dyslipidemia as the only modifiable risk factor increasing the hazard ratio for all types of CVD events. Interestingly, mean HbA_1c and HbA_1c variability were independently associated with an increased risk of PAD. Moreover, a 1% point increase in the mean HbA_1c concentration increased the hazard for PAD by 50%. In addition to HbA_1c and dyslipidemia, arterial hypertension was also a significant risk factor for PAD. Presence of DKD was invariably associated with an increased risk of any kind of CVD event (Supplementary Table 5). However, even in those with normoalbuminuria, mean HbA_1c and HbA_1c variability were both independently and adversely associated with CVD event risk (Supplementary Table 6).

During follow-up, 181 individuals in the FinnDiane 50-year cohort died. The median age at the time of death was 65.9 (61.4–72.1) years. The standardized mortality ratio was 3.2 (95% CI 2.8–3.7). The most common cause of death was CVD: MI in 22.7%, IHD in 25.4%, and stroke in 5.0% of deaths. DKD was the primary cause of death in 7.2%. However, 28.7% of deaths were due to combined micro- and macrovascular diabetic complications, including CVD and DKD. When taking any disease of the circulatory system as a cause of death into account (I00–I99), 60.2% of deaths were CVD related.

Factors independently associated with mortality in the FinnDiane 50-year cohort were diabetes duration, ESRD treated with dialysis, decrease in eGFR, PAD, HbA_1c variability, increased insulin resistance, and high-sensitivity C-reactive protein (Supplementary Table 7).

Cardiovascular events are common and increase exponentially in individuals with long-term T1D. The cumulative incidence of CVD events after a median of 54.5 years’ T1D duration was 54%. Compared with the population without diabetes, the risk of CVD was more than triple, even in the absence of albuminuria. Common modifiable risk factors, such as hyperglycemia, arterial hypertension, and dyslipidemia, are also CVD risk factors in this population. Of note, an especially strong association was found between glycemic control and PAD, which is a relatively neglected complication of T1D.

Compared with cohorts with similar T1D duration, ours had a substantially higher prevalence of chronic complications. More than two-thirds had SDR, compared with less than half of the Joslin Medalists (17) and those in the U.K. Golden Years cohort (6) and less than one-fifth of those in the Swedish registry (5). In addition, one-third of our participants had an eGFR <60 mL/min/1.73 m², whereas only 12% of those from the Joslin cohort had an eGFR <45 mL/min/1.73 m² (17). Medalist data, either from the U.S., Canada, or the U.K., are subject to selection bias, not only because of survivorship bias but also because of volunteer bias. Swedish registry data, on the other hand, may not capture all the albuminuria data, as acknowledged by researchers from the National Diabetes Register (5), which may be one of the reasons for the discrepancies among different data sets. However, the CVD data were more similar, with a prevalence rate of ∼40% (5,17). Of note, different diagnoses were used to define a CVD event in the different cohorts, and some cohorts used mainly self-reported data (18). In our study, we used only hard CVD end points, as described in Supplementary Table 1. Moreover, we believe that our data reliably represent an average person with T1D, as also confirmed by the incident CVD event analysis in the Finnish national 50-year T1D cohort. Finnish registries had already started collecting data in the year 1969 and can therefore reliably capture outcome events over the entire period of follow-up.

To our knowledge, this is the first study to report SIRs for CVD in a population with longstanding diabetes. We here demonstrate that these individuals have a 7.4-fold greater risk of CVD compared with the population without diabetes. Moreover, a 4.9-fold greater risk of CVD exists even for the individuals with normoalbuminuria, and this risk is mainly associated with glycemic control.

Historically, it was suggested that individuals surviving decades with T1D might be protected from the adverse effects of hyperglycemia (4,6) and may therefore be treated less intensively. Moreover, although data from the Diabetes Control and Complications Trial/ Epidemiology of Diabetes Interventions and Complications (DCCT/EDIC) study show that after age, glucose control is the most important risk factor for CVD (19,20), its influence diminishes by time (19). This could mean that the legacy effect attenuates by time or, intriguingly, that survival after a certain time point confers CVD protection (19). Our results, however, suggest a strong association of HbA_1c with incident CVD (especially PAD), even after very long diabetes duration, implying that glycemic control is crucial not only for the prevention or initiation but also for the perpetuation of the atherosclerotic process. Similarly, recent data from the Joslin Medalists and the Swedish registry confirm this association (17). Therefore, our findings, strengthened by a comparison with a matched population without diabetes and by serial HbA_1c measurements, argue against glycemic protection in those surviving with longstanding diabetes. Moreover, our data again emphasize the need for lifelong meticulous glycemic control and suggest decreasing HbA_1c variability might be beneficial. With modern technology, newer therapies, and improvements in CVD treatment, a decrease in CVD events and mortality is anticipated in cohorts with a more recent diabetes onset, as already suggested (21).

Of note, fewer than one-third of the participants in the FinnDiane 50-year cohort received lipid-lowering therapy. This might reflect the fact that early statin trials were performed in the mid-1990s, and the conclusions from those studies were not yet fully implemented in routine clinical practice at the time of the baseline FinnDiane visits during that same time period. Nevertheless, a majority of the individuals from our T1D cohort had triglyceride and HDL cholesterol concentrations within the target ranges, as recommended by international guidelines (22); however, the event rate of CVD was high and independently associated with increasing lipid levels. Even more, in addition to DKD, dyslipidemia was the only modifiable risk factor that was independently associated with all CVD event subtypes in our study. This may suggest that lipid-lowering therapy should be much more aggressive in T1D and that lower lipid targets as well as therapies focused on additional lipid subclasses are needed (23). It may also be that a worse ratio of triglycerides to HDL cholesterol is a marker of increased insulin resistance, which is also a known CVD risk factor in T1D (24), and reflects an individual’s susceptibility to vascular damage, which cannot simply be reversed by the initiation of available lipid-lowering therapy. Intriguingly, lipid-lowering therapy was associated with a higher risk of CVD but showed a trend toward a lower risk of mortality. The association with CVD events is most likely due to reversed causality.

Our study is also the first to analyze how different modifiable risk factors are associated with specific CVD subtypes. Specifically, in the scientific literature, data on PAD are rarely presented. PAD incidence was strongly linked to glycemic control and, importantly, intraindividual HbA_1c variability. In addition, arterial hypertension and dyslipidemia were also significant and independent risk factors. Nevertheless, PAD was also the only CVD subtype associated with SDR, confirming the association of PAD with microvascular damage in longstanding T1D, as well as in type 2 diabetes, as was recently reported (25).

The main strength of this study is that we had the possibility to provide a multifactorial view regarding the occurrence of CVD events and mortality by combining clinical data from the FinnDiane visits and data from reliable national registries. Therefore, we were able to link the dates of diabetes diagnosis with the dates of CVD diagnosis. This enabled us to present the occurrence of different CVD subtypes in a prospective manner from the time of diabetes onset. Furthermore, we were able to confirm the representativeness of our cohort with data from the national T1D 50-year cohort. Finally, we showed excess CVD risk in our T1D cohort compared with age-, sex-, and geographical regionmatched individuals without diabetes. Notwithstanding, our study has several limitations. Longitudinal association of different risk factors with CVD and mortality is suggestive of causality but is rarely conclusive in its own right. Also, although we have longitudinal data on HbA_1c values, the period covers only ∼15 years’ diabetes duration, and therefore, we could assess neither the impact of glycemic control nor the impact of other clinical CVD risk factors since diabetes onset as time-varying covariates for CVD incidence. Furthermore, we included the ratio of triglycerides to HDL cholesterol instead of the usual LDL cholesterol as a lipid variable in our models. This was due to the fact that in our previous studies, we have shown that the ratio of triglycerides to HDL cholesterol is the best predictor of CAD in T1D (15), possibly because of the interference of albuminuria with lipid status. In addition, besides LDL cholesterol, triglycerides have been shown to be causally associated with CVD (26). Finally, data on family history of CVD are missing for the majority of individuals.

In summary, individuals with T1D of >50 years duration are at high risk of CVD events and have a more than triple CVD risk compared with the matched population without diabetes, even in the absence of albuminuria. Furthermore, glycemic control remains one of the most important determinants of cardiovascular health throughout the lifespan. Together, our findings again highlight the importance of continuous and meticulous control of classical risk factors for CVD and challenge the current clinical practice recommendations for older patients with T1D.

Acknowledgments. The authors acknowledge all physicians and nurses at each FinnDiane center participating in patient recruitment and characterization. The complete list of physicians and nurses is presented in the Supplementary Data.

Funding. This research was supported by the Folkhälsan Research Foundation, Academy of Finland (grants 299200 and 316664), Wilhelm and Else Stockmann Foundation, Liv och Hälsa Society, Novo Nordisk Foundation (grant OC0013659), Finnish Foundation for Cardiovascular Research, and Finnish Diabetes Research Foundation. D.P.B. was supported by the European Association for the Study of Diabetes Albert Renold Fellowship, offered by the European Foundation for the Study of Diabetes.

Duality of Interest. D.P.B. reports lecture honoraria from AstraZeneca, Boehringer Ingelheim, Eli Lilly, Krka, Novartis, and Novo Nordisk and advisory board membership for AstraZeneca, Krka, Eli Lilly, and Novo Nordisk. P.-H.G. reports lecture honoraria from Astellas, AstraZeneca, Boehringer Ingelheim, Eli Lilly, Elo Water, Genzyme, Medscape, Merck Sharp & Dohme, Mundipharma, Novartis, Novo Nordisk, PeerVoice, Sanofi, and Sciarc and advisory board membership for AbbVie, Astellas, AstraZeneca, Bayer,Boehringer Ingelheim, Eli Lilly, Janssen, Medscape, Merck Sharp & Dohme, Mundipharma, Novartis, Novo Nordisk, and Sanofi. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. V.H. was responsible for the study design, data acquisition, statistical analyses, data interpretation, and drafting of the manuscript. D.P.B. participated in the study conception, literature search, data analysis, data interpretation, and drafting of the manuscript. D.G. participated in the design of the study, data interpretation, and critical revision of manuscript for important intellectual content. C.F. contributed to data acquisition, data interpretation, and critical revision of the manuscript. G.K. contributed to data interpretation and critical revision of the manuscript. P.-H.G. is the principal investigator of the study and participated in the study conception, data interpretation, and critical revision of manuscript for important intellectual content. P.-H.G. 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.