Relationship Among Diabetes, Obesity, and Cardiovascular Disease Phenotypes: A UK Biobank Cohort Study

Diabetes and obesity are both associated with increased cardiovascular morbidity and mortality (1). Obesity is a complex condition of increased subcutaneous and visceral adiposity often associated with adipose dysfunction and insulin resistance, which increase the risk of diabetes and cardiovascular disease (2). Most people with diabetes are overweight or obese, except for the minority with autoimmune or genetic forms of diabetes, and for each unit increase in BMI, the likelihood of diabetes increases exponentially. Moreover, diabetes and obesity are associated with increased vascular stiffness and accelerated atherosclerosis, processes that lead to premature cardiovascular disease and death (3).

Large population studies attempting to discern the independent cardiovascular risk conferred by diabetes suggest that adjusting for BMI does not substantially diminish the association between diabetes and cardiovascular mortality (4). However, the relationship between BMI and cardiovascular disease is potentially complex, with BMI above or below the normal range being associated with a higher risk of cardiovascular (and all-cause) mortality (5,6). Such data may reflect residual confounding factors and suggest cautious interpretation of epidemiological data in isolation. Cardiovascular imaging studies, while much smaller, offer an alternative approach to characterize overt and subclinical cardiovascular disease. These support the notion that diabetes in the context of normal BMI is still associated with important cardiovascular abnormalities (7,8).

The literature defining the complex relationship among diabetes, BMI, and cardiovascular disease lacks data leveraging multimodality cardiovascular imaging and hard outcomes within a single cohort powered to study people with normal BMI and diabetes. To address this lack of data, we used the UK Biobank (UKB) cohort study. We hypothesized that diabetes with normal BMI is associated with a cardiovascular phenotype and event rate comparable to obesity without diabetes.

UKB is a prospective, observational, cohort study of 502,462 participants aged 37–73 years recruited from 22 assessment centers across the U.K. between 2006 and 2010. It is an open access resource developed using U.K. government and biomedical research charity funding, which links wide-ranging phenotypic and health care record data. The UKB resource is open to all bona fide researchers. Full details of its design and conduct are available online (https://www.ukbiobank.ac.uk). UKB received ethical approval from the National Health Service (NHS) Research Ethics Service (11/NW/0382); we conducted this analysis under application number 59585. All participants provided written informed consent, and the research was conducted in line with the Declaration of Helsinki. The study was reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement.

Baseline sociodemographic characteristics, comorbidities, and medications were recorded by participants completing a touchscreen and nurse-led interview at study recruitment, as previously described (9). Data from the face-to-face nurse-led interview were used to ascertain baseline comorbidities and medications. Diabetes was classified as any diabetes (UKB field identifier 1220), type 1 diabetes mellitus (1222), type 2 diabetes mellitus (1223), diabetic eye disease (1276), diabetic neuropathy/ulcers (1468), and diabetic nephropathy (1607). Duration of diabetes was defined as the time between self-reported diagnosis and study recruitment. Triglyceride-to-HDL cholesterol ratio was used as a proxy of insulin resistance, as previously described (10).

BMI was assessed using standing height and weight data collected by the UKB at study recruitment. BMI category was adjusted for ethnicity in accordance with World Health Organization (WHO) ethnicity-specific threshold recommendations: normal, BMI ≥18.5 kg/m² to <25 kg/m² or ≥18.5 kg/m² to <23 kg/m² if South Asian ethnicity; overweight, BMI ≥25 kg/m² to <30 kg/m² or ≥23 kg/m² to <27.5 kg/m² if South Asian ethnicity; and obese, BMI ≥30 kg/m² or ≥27.5 kg/m² if South Asian ethnicity (11). Participants with below-normal BMI were excluded from this analysis (n = 2,316) because of an insufficient sample size to study these individuals with diabetes; we did not apply an upper BMI limit for inclusion in the analysis. Definitions of other comorbidities at recruitment has previously been described (12). We excluded participants with missing data related to BMI (n = 10,135), loss to follow-up or withdrawal of consent (n = 1,298), or confounding factors, including smoking status (n = 2,949), ethnicity (n = 2,777), socioeconomic status (n = 624), and systolic blood pressure (SBP) (n = 34,439).

From 2014, all surviving participants were invited by e-mail and then postal mail to take part in multimodality imaging assessment. The imaging included cardiac MRI (cMRI), carotid artery ultrasound, photoplethysmography-derived arterial stiffness index (PASI), and abdominal MRI. Anthropometric measurements of body composition and measurements of serum lipids and biochemistry were collected at baseline. Responding participants were screened for eligibility for inclusion based on safety and tolerability criteria. All participants with metal implants in their body were excluded for safety and image quality concerns (13). Further details regarding cardiometabolic phenotyping are found in the Supplementary Materials, and full details of all protocols have been previously published (14–19).

Our primary end point was cardiovascular mortality, and our secondary end points were nonfatal myocardial infarction (MI), nonfatal stroke, and all-cause mortality; all other analyses were exploratory. UKB mortality outcomes are obtained from the official U.K. national death registry from NHS Digital for participants in England and Wales and from the NHS Central Register for participants in Scotland. We censored outcomes on 23 March 2021. Cardiovascular mortality was defined according to ICD-10 codes as previously described (20). In brief, this definition included all cardiovascular ICD-10 codes from I00 to I99, excluding codes related to infection mortality. Incident nonfatal MI or nonfatal ischemic stroke were defined using a UKB algorithm (21,22). We only included events ascertained from hospital admission data (excluding self-reported outcomes) and where nonfatal MI or nonfatal ischemic stroke was the primary diagnosis.

All analyses were performed using Stata/MP statistical software. All statistical tests were two-sided, and statistical significance was defined as P < 0.05. However, in tables presenting multiple exploratory statistical tests, this threshold for significance was reduced to P < 0.005. Missing data were not imputed. Continuous data are presented as medians with 25th–75th centiles. Categorical data are presented as counts with percentages. Normality of distribution was checked using skewness and kurtosis tests; all continuous variables were found to be nonnormally distributed. Differences between BMI categories within the diabetes or nondiabetes groups were assessed using Kruskal-Wallis H tests or χ² test for continuous and categorical variables, respectively. Differences between the diabetes and nondiabetes groups within each BMI category were assessed using Mann-Whitney U tests. Where appropriate, some analyses were repeated after stratification by sex. We used correlation matrixes of Pearson model coefficients to assess correlations between covariates; no correlation coefficients >0.3 or <−0.3 were observed.

Unadjusted and adjusted incident rate ratios (IRRs) and their 95% CIs were estimated for all-cause mortality, cardiovascular mortality, nonfatal MI, and ischemic stroke using Poisson regression models; exposure time was modeled, but time-varying covariates were not used. A dummy variable was used to compare outcomes of diabetes-BMI groups with reference to the normal BMI-nondiabetes group. Where indicated, models were adjusted for covariates indicated in the accompanying table or figure legend; this included interaction terms between diabetes and BMI categories to identify differential associations between BMI category and outcome in people without or with diabetes. Crude mortality rates were calculated per 1,000 person-years of follow-up for all-cause mortality, cardiovascular mortality, nonfatal MI, and ischemic stroke by BMI category for the total population and then stratified by diabetes status. Kaplan-Meier curves were used to illustrate unadjusted event rates among participants grouped by diabetes status and BMI category. Where specified, we separately modeled BMI as a continuous variable using restricted cubic splines with four knots for all clinical outcomes, as this provided the best fit as assessed by minimizing Akaike and Bayesian criteria (models including categorical, linear, or cubic splines with three, four, and five knots and first-degree and second-degree fractional polynomials were compared). Models were constructed independently for participants with and without diabetes. The reference knot was set at the median BMI of the whole cohort, and spline curves were truncated at the 1st and 99th centiles.

A sensitivity analysis was performed to define cardiovascular events in participants with diabetes after exclusion of those diagnosed before the age of 40 years who were also receiving insulin treatment at the time of recruitment to UKB. A further sensitivity analysis stratified participants with and without diabetes based on central obesity, defined using a waist-to-hip ratio (WHR) ≥0.85 if female and ≥0.90 if male, in lieu of obesity defined with BMI (23).

We included 451,355 participants of whom 22,451 (4.9%) had diabetes at recruitment. Among participants with diabetes, 2,290 (10.1%) were normal weight, 7,732 (34.4%) overweight, and 12,429 (55.3%) obese. In participants without diabetes, 143,557 (33.5%) were normal weight, 185,758 (43.3%) overweight, and 99,589 (23.2%) obese. Participants with diabetes were older, more often male, less physically active, less often of non-White ethnicity, and more socioeconomically deprived (Table 1). There was greater prevalence of all studied cardiometabolic comorbidities at baseline among participants with diabetes, and the prevalence of cardiometabolic comorbidities (except for peripheral vascular disease) increased with BMI, irrespective of diabetes status. Antihypertensive, statin, aspirin, and diuretic use was higher among participants with diabetes at enrollment, and usage increased with increasing BMI category, regardless of diabetes status. Most participants with diabetes were taking metformin, and approximately one-fifth were receiving insulin (Supplementary Table 1). Diabetes medication use rose with increasing BMI.

During 5,323,190 person-years of follow-up (median 12.0 years), 29,931 participants died (6.6%) of whom 5,831 (1.3%) died from cardiovascular causes. A total of 7,179 participants (1.6%) had nonfatal MI, and 3,469 (0.8%) had nonfatal ischemic stroke during follow-up. Kaplan-Meier curves illustrating cardiovascular and all-cause mortality during follow-up are shown in Supplementary Fig. 1. Broadly, modestly rising mortality was found across BMI categories, with much greater mortality in groups with diabetes. Indeed, absolute unadjusted rates of all-cause mortality, cardiovascular mortality, nonfatal MI, and nonfatal ischemic stroke climbed modestly with increasing BMI category, with much greater mortality in participants with diabetes (Supplementary Table 2). Unadjusted and adjusted IRRs for all-cause and cardiovascular mortality are shown in Table 2. Among participants without diabetes, the risk of cardiovascular death was comparable among overweight participants (adjusted IRR 1.00 [95% CI 0.93–1.08]) and increased in obese participants (adjusted IRR 1.27 [95% CI 1.17–1.37]) compared with those with normal BMI. Participants with diabetes and normal BMI experienced a nominally larger risk of cardiovascular death (adjusted IRR 1.55 [95% CI 1.21–1.98]), with further increases in the overweight with diabetes group (adjusted IRR 1.71 [95% CI 1.47–1.98]) and obese with diabetes group (adjusted IRR 1.96 [95% CI 1.71–2.25]). There was no significant interaction between BMI category and diabetes associated with cardiovascular mortality; this implies that rising BMI has a similar association with cardiovascular mortality irrespective of diabetes status. When directly comparing obese participants without diabetes with those with normal BMI and diabetes, their risk of cardiovascular death was not statistically different despite being nominally lower (adjusted IRR 0.82 [95% CI 0.64–1.04]; P = 0.1) (Supplementary Table 3). Similar patterns were observed in risk of nonfatal MI or nonfatal ischemic stroke (Table 2 and Supplementary Table 3).

A sensitivity analysis excluding participants with diabetes diagnosed before the age of 40 years who were also receiving insulin treatment at the time of recruitment yielded similar findings (Supplementary Table 4), which suggests that our findings are unlikely to be driven by the inclusion of participants with type 1 diabetes. Another sensitivity analysis replacing BMI categories with central adiposity defined using WHO sex-specific WHR thresholds also yielded similar findings (Supplementary Table 5 and Supplementary Fig. 2). This finding showed that central adiposity is associated with cardiovascular death, nonfatal MI, and nonfatal stroke in participants without diabetes; however, the magnitude of this risk was larger for participants with diabetes and no central adiposity. The addition of central adiposity to diabetes did not substantially increase the adjusted risk of cardiovascular events beyond diabetes alone. Indeed, there was a significant interaction between diabetes and central obesity, suggesting differential association between central obesity and cardiovascular outcomes in participants without and with diabetes. When modeled as a continuous variable, rising BMI had a steeper relationship with cardiovascular mortality in participants without diabetes, although with overlapping 95% CIs (Supplementary Fig. 3).

Participants with diabetes had elevated BMI, WHR, body fat percentage, and reduced whole-body impedance compared with those without diabetes within any given BMI category (Table 3). Increasing BMI category was associated with significantly higher serum triglycerides and LDL cholesterol and lower HDL cholesterol, irrespective of diabetes status. Total cholesterol, LDL cholesterol, HDL cholesterol, and triglycerides were lower in participants with diabetes compared with those without within any given BMI category. Participants with diabetes had higher serum creatinine, serum cystatin C, urinary microalbumin, and serum ALT compared with those without diabetes, which rose with increasing BMI category. Higher BMI categories were also associated with increased serum C-reactive protein and HbA_1c, regardless of diabetes status. Total abdominal adipose tissue index and abdominal fat ratio increased in higher BMI groups but did not differ according to diabetes status. Given the sexual dimorphism in body composition, we also performed stratified analyses that showed similar patterns in relation to diabetes and BMI category in both sexes (Supplementary Table 6). Collectively, these data illustrate important differences in metabolic parameters associated with increasing BMI, irrespective of diabetes status; however, many of these are also abnormal in people with diabetes and normal BMI versus those without diabetes.

Resting heart rate, SBP, and diastolic blood pressure (DBP) increased across rising BMI categories. Participants with diabetes also had higher resting heart rate and SBP but lower DBP than participants without diabetes within each BMI category (Table 4). Participants with diabetes had higher CIMT and PASI than those without diabetes within each BMI category; both measures increased with rising BMI in the nondiabetes group, but only PASI (not CIMT) increased with BMI in the diabetes group. Among participants who were overweight or obese, LVEF was lower in those with diabetes compared with those without. Furthermore, LVEF declined with rising BMI category, irrespective of diabetes status. Among participants without diabetes, rising BMI was associated with lower left ventricular stroke volume (LVSV), left ventricular end-diastolic volume, and cardiac index; however, cardiac contractility index (CCI) increased with rising BMI. In participants with diabetes, rising BMI was only significantly associated with lower LVSV and not CCI. Within obese participants, diabetes was associated with elevated CCI. Direct comparison of cardiovascular imaging phenotypes between obese participants without diabetes and participants with normal BMI and diabetes revealed statistically greater CIMT and lower PASI but similar LVEF and CCI in the normal BMI and diabetes group (Supplementary Table 7). A sensitivity analysis excluding participants with diabetes diagnosed before the age of 40 years who were also receiving insulin treatment at the time of recruitment did not substantially alter the conclusions drawn from cardiovascular imaging phenotypes (Supplementary Table 8). Given the potential sexual dimorphism in these measures, we also performed stratified analyses, which showed similar patterns in relation to diabetes and BMI category in both sexes (Supplementary Table 9).

Collectively, these data illustrate an adverse cardiovascular phenotype associated with rising BMI in participants without diabetes, although this relationship is less clear in participants with diabetes who exhibit marked abnormalities, even in the normal BMI group. A sensitivity analysis replacing BMI categories with central adiposity defined using WHO sex-specific WHR thresholds also yielded similar findings, with adverse associations between central adiposity and all cardiovascular imaging phenotypes in participants without diabetes (Supplementary Table 10). In participants with diabetes, there was also an adverse association with central adiposity for CIMT, LVEF, and PASI but not CCI (Supplementary Table 10). Next, we explored the association between metabolic parameters and cardiovascular imaging phenotypes. Pairwise correlation analysis demonstrated that duration of diabetes weakly correlated with CIMT and LVEF but not PASI or CCI (Supplementary Fig. 4); no correlations were noted with HbA_1c in the diabetes group, but in participants without diabetes, HbA_1c correlated with PASI, CIMT, LVEF, and CCI. The strongest correlates of cardiovascular parameters were markers of adiposity; this was apparent in the diabetes and nondiabetes groups, although these remained modest, with Pearson correlation coefficients of no more than 0.25. Importantly, all four cardiovascular imaging phenotypes were associated with our primary end point of cardiovascular mortality, and this association was comparable in participants with and without diabetes (Supplementary Table 11).

We present a detailed analysis of cardiovascular phenotypes and outcomes in relation to metabolic parameters in a large cohort stratified by baseline diabetes status and BMI category. Cardiovascular mortality and nonfatal events were more common as BMI rose, but in the presence of diabetes, event rates were much greater. Indeed, UKB participants with diabetes and normal BMI experienced nominally higher adjusted event rates than obese participants without diabetes. However, the adverse association of obesity with cardiovascular mortality and events was comparable in participants with and without diabetes, indicating that they combined additively rather than synergistically. In contrast, while our sensitivity analysis revealed that central obesity was associated with increased adjusted risk of cardiovascular death in participants without diabetes, this was not apparent in those with diabetes. These data emphasize the complex interplay between diabetes and obesity in the modulation of cardiovascular disease.

Imaging data corroborated progressive cardiovascular abnormalities with rising BMI in participants without diabetes. Notably, the cardiovascular imaging phenotype of diabetes with normal BMI was broadly comparable to obesity without diabetes, with only a modest progression of these cardiovascular abnormalities with rising BMI in the diabetes group. Furthermore, duration of diabetes and HbA_1c correlated poorly with cardiovascular phenotype in participants with diabetes, with metrics of adiposity demonstrating the strongest (albeit modest) correlation. This finding raises the possibility that the ideal target range for metrics of adiposity may be lower than currently proposed in people with diabetes. Our data also reveal scope to improve adherence to existing targets around modifiable cardiovascular risk factors (e.g., smoking cessation) in people with diabetes and/or elevated BMI.

To our knowledge, our study is the largest investigation of interactions between diabetes and obesity in terms of cardiovascular phenotype and long-term outcomes. However, other epidemiological studies have provided important context. In >10 million participants, Di Angelantonio et al. (6) showed being overweight or obese was associated with higher all-cause and cardiovascular mortality versus normal weight. However, they did not directly compare cardiovascular mortality stratified by diabetes status. In a prospective cohort of 10,568 people with type 2 diabetes, Costanzo and colleagues (24,25) showed that overweight and obese people were more likely to be hospitalized for MI or stroke, although cardiovascular mortality was not reported. In a meta-analysis of 16 cohort studies including 445,125 people with type 2 diabetes, Kwon et al. (26) reported a U-shaped relationship between BMI and all-cause or cardiovascular mortality; no data were included from people without diabetes. Notably, their analysis found a nadir of cardiovascular mortality at a BMI of 29–31 kg/m², highlighting an obesity paradox. This finding contrasts with our analysis and that of Costanzo et al., possibly reflecting cohort differences in important confounding factors, such as comorbidity and smoking status.

We found that participants with diabetes had more advanced atherosclerosis and greater arterial stiffness than participants without diabetes within all BMI categories. Moreover, participants with normal BMI and diabetes had a nominally higher CIMT than any other diabetes-BMI category, emphasizing their substantial burden of arterial disease. Participants with diabetes also exhibited supraphysiological cardiac contractility without differences in cardiac output. This, combined with associated increased arterial stiffness, suggests chronically elevated left ventricular afterload, a risk factor for incident heart failure and atrial fibrillation, among others (27). Indeed, serial cMRI studies in people with uncomplicated type 2 diabetes have revealed important reductions in LVEF over a 6-year period (28). While we found statistically lower LVEF in participants with diabetes, these were smaller than the margin of error with any cardiac imaging modality. Given that we observed much larger differences in CCI, this may be a better biomarker of early diabetic heart disease and warrants assessment in future studies.

We showed that UKB participants with diabetes and normal BMI had significantly elevated WHR and a numerically greater abdominal fat ratio and total abdominal adipose tissue index compared with participants with normal BMI without diabetes. These measures suggest unfavorable body composition and greater visceral versus subcutaneous adipose deposition. Chowdary et al. (8) recently described that people with diabetes have greater visceral adiposity than those without diabetes, even when of normal weight. Visceral adiposity is independently associated with higher 10-year cardiovascular disease risk (29), which is supported by our sensitivity analyses of central obesity defined by WHR (Supplementary Table 5). However, this sensitivity analysis did not find additive effects of this measure of central obesity with diabetes, highlighting complex interactions between diabetes and obesity. Notably, indices of adiposity were the strongest (albeit modest) correlates of abnormal cardiovascular phenotypes in people with and without diabetes in our analysis. Hence, better assessment of visceral adiposity in routine practice may allow clinicians to define high-risk groups, and our data raise the question of whether lower BMI targets, or possibly alternate adiposity metrics, should guide use of existing and novel therapies.

Increased risk of cardiovascular disease and death among people with diabetes and normal BMI is also likely to relate to suboptimal control of traditional cardiovascular risk factors. In our study, while this group had relatively good glycemic control, they had low physical activity, higher than ideal SBP, and suboptimal LDL cholesterol, and approximately one in seven currently smoked. These data emphasize the potential benefits of more effective application of existing cardiovascular risk modification guidelines. Nevertheless, all groups in our analysis had suboptimal cardiovascular risk factor profiles, emphasizing the challenges of preventive medicine. Interestingly, systemic inflammation (defined using serum C-reactive protein) was similar in participants with and without diabetes of normal BMI, conflicting with some literature. However, systemic inflammation increased with BMI irrespective of diabetes status, as expected (30). The modest correlation of included cardiovascular risk factors with cardiovascular imaging phenotypes in participants with diabetes suggests the need for better routine clinical biomarkers of cardiovascular disease in people with diabetes.

The strengths of our study include detailed cardiometabolic phenotyping with multimodality assessment, high-quality long-term outcome data, and statistical power to study diabetes with normal BMI. However, we must also acknowledge limitations. First, phenotypic measures of cardiometabolic disease were not assessed in every participant because of the design of UKB, which only performed more complex assessments in a subset. This introduces survivor bias in participants who underwent detailed imaging assessment, which commenced in 2014 (13). We also lacked the statistical power to conduct analyses restricted to participants with more pronounced obesity. Second, we did not stratify by type of diabetes since only 404 participants had self-reported type 1 diabetes. However, a sensitivity analysis excluding 1,756 participants with diabetes diagnosed before the age of 40 years who also received insulin at the time of recruitment reached similar conclusions, suggesting that our findings are unlikely to be substantially driven by the inclusion of people with type 1 diabetes. Third, our work is observational, so causality cannot be inferred. Fourth, participants were recruited before the use of modern diabetes therapeutics (e.g., sodium–glucose cotransporter 2 inhibitors), which improve cardiovascular outcomes (31). Therefore, our observed event rates in participants with diabetes may be higher than contemporary rates. Fifth, UKB is not representative of the U.K. population regarding socioeconomic deprivation, some noncommunicable diseases, and ethnic minority groups (32). While this means event rates should be cautiously extrapolated to the U.K. population, UKB remains a robust resource to define exposure-disease relationships (32). Sixth, our adjusted analyses do not account for dietary factors, which may mediate some of the adverse cardiovascular outcomes associated with diabetes and obesity and are an essential target in disease prevention. Finally, we present many statistical analyses beyond our primary and secondary outcome measures, and while these data provide valuable context, these exploratory analyses must be interpreted cautiously.

In conclusion, both obesity and diabetes are independently and additively associated with more advanced cardiovascular disease and more frequent major adverse cardiovascular events. While adiposity metrics are more strongly correlated with cardiovascular biomarkers than diabetes-oriented metrics, both correlate weakly, suggesting that other factors underpin the high cardiovascular risk of normal-weight diabetes. While there is clear scope for better use of existing screening and preventive approaches in people with diabetes, our data suggest that more refined risk assessment aligned with targeted preventive interventions are needed to improve outcomes.

Acknowledgments. This work uses data provided by patients and collected by the NHS as part of their care and support (from NHS Digital, reused with the permission of UK Biobank). This research has been conducted using UK Biobank resource application number 59585.

Funding. This research was funded by the British Heart Foundation (RG/F/22/110076). This research used data assets made available by National Safe Haven as part of the Data and Connectivity National Core Study, led by Health Data Research UK in partnership with the Office for National Statistics and funded by UK Research and Innovation (research commenced between 1 October 2020 and 31 March 2021, grant MC_PC_20029, and 1 April 2021 and 30 September 2022, grant MC_PC_20058). M.D. and S.S. were supported by a British Heart Foundation Clinical Research Training Fellowship. E.L. is funded by Wellcome Trust Clinical Career Development Fellowship award 221690/Z/20/Z and receives support from the National Institute for Health and Care Research (NIHR) Leeds Biomedical Research Centre. L.D.R. was supported by Diabetes UK RD Lawrence Fellowship award 16/0005382. K.J.G. is an NIHR Academic Clinical Lecturer. M.A.B. is supported by British Heart Foundation Intermediate Clinical Research Fellowship award FS/18/12/33270. M.T.K. is a British Heart Foundation professor. R.M.C. was supported by a British Heart Foundation Intermediate Clinical Research Fellowship.

At no time did any authors or their institutions receive other payment or services from a third party for any aspect of the submitted work. The views expressed are those of the authors and not necessarily those of the NHS, NIHR, or the Department of Health and Social Care.

Duality of Interest. S.S. has received speaker fees, honoraria, and nonfinancial support from AstraZeneca. R.A. has received institutional research grants from Abbott Diabetes Care, Bayer, Eli Lilly, Novo Nordisk, Roche, and Takeda and honoraria and consultant fees from Abbott Diabetes Care, AstraZeneca, Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, Eli Lilly, GlaxoSmithKline, Menarini, Merck Sharp & Dohme, Novo Nordisk, and Takeda. M.A.B. has received speaker fees from Medicon and Amgen. M.T.K. has received honoraria from AstraZeneca, speaker fees from Merck and Novo Nordisk, and unrestricted research awards from AstraZeneca and Medtronic. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. O.I.B. and M.D. were responsible for the data analysis. O.I.B., M.D., H.M., M.G., M.C.-R., J.G., S.S., S.B.W., K.B., L.D.R., E.L., R.A., K.J.G., M.A.B., M.T.K., and R.M.C. were responsible for study conception, critical revision of the manuscript, and approval of the final version of the manuscript. O.I.B., M.D., H.M., and R.M.C were responsible for manuscript drafting. M.D. and R.M.C. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. A non–peer-reviewed version of this article was submitted to the medRxiv preprint server (https://www.medrxiv.org/content/10.1101/2023.01.03.23284157v1) on 5 January 2023.