Association of Metabolic Syndrome With Neuroimaging and Cognitive Outcomes in the UK Biobank

Metabolic syndrome (MetS) has become a serious public health concern, affecting approximately 25% of adults globally (1). MetS is characterized by a cluster of cardiometabolic and vascular risk factors that tend to co-occur in individuals (1). Generally, a MetS diagnosis is defined as the presence of at least three of the following traits: larger waist circumference; elevated levels of triglycerides, blood glucose, and blood pressure; or reduced HDL cholesterol (1). MetS has been identified as a potentially modifiable risk factor for cardiovascular and cerebrovascular diseases (2), making it an important target for disease prevention strategies.

Global dementia incidence has also been rapidly increasing and is expected to triple in the next 30 years (3). There is currently no definitive cure for dementia; thus, there is an urgent need to understand and identify which risk factors can be effectively targeted to help prevent or delay its progression. Recent research has linked MetS to an increased risk of dementia (4,5). Building on this work by exploring links between MetS and preclinical and intermediate markers of dementia, such as early changes in neuroimaging and cognition, can enhance our understanding of the relationship between MetS and dementia risk. Maintaining both brain structure and cognition (hereafter, collectively referred to as brain health) is crucial for healthy aging because loss of both are key indicators of progression toward dementia (6). Evaluating neuroimaging and cognitive measures can provide distinct insights regarding which aspects of brain health may be affected, such as atrophy, vascular pathology, and global or region- or domain-specific brain and cognitive changes. Different health conditions may influence these markers of brain health in distinct ways; some may exert a widespread effect, affecting multiple aspects of brain health, whereas others may exert a localized effect (7). Traditional cardiometabolic and vascular risk factors (e.g., diabetes, obesity, hypertension, hyperlipidemia) have been individually associated with poorer brain health (8). However, it is well known that these risk factors are highly correlated with each other, often making it difficult to isolate their individual effects on overall brain health (8). These relationships remain to be investigated in MetS, which is particularly pertinent because the clustering of conditions inherent in MetS are more closely in line with real-world representations of known individual risk factors.

In a 2021 meta-analysis of 12 studies (sample sizes ranging from 50 to 7,000 participants), MetS was consistently associated with lower scores on global tests of cognition; however, its impact on individual cognitive domains was mixed (9). Moreover, although existing research has examined MetS in relation to either structural neuroimaging outcomes or cognition, few studies have examined this together (7–10), with most research focused on specific populations (e.g., MetS in schizophrenia) (11–13).

Given these knowledge gaps, we sought to investigate the association of MetS with several structural neuroimaging outcomes and cognitive domains in more than 37,000 participants from the UK Biobank. We also explored the impact of an increasing number of MetS components on these outcomes. To our knowledge, this is the largest and most comprehensive study of its kind which incorporates detailed phenotypic data and specific genotypic information on apolipoprotein (APOE)-ε4 carrier status, enabling a robust assessment of these associations.

In this study, we used data from the UK Biobank, a population-based prospective cohort of 0.5 million participants aged 40–69 years recruited between 2006 and 2010 in England, Scotland, and Wales (14). At baseline (2006–2010), participants provided information on sociodemographic, lifestyle, environmental, and health-related factors via touch-screen questionnaires and a nurse-led verbal interview; underwent physical examinations; and provided blood samples to measure key biomarkers. Since 2014, participants have been reinvited to attend a follow-up imaging assessment in order to undergo brain, cardiac, and abdominal MRI, DEXA, and carotid ultrasound. During the imaging assessment, participants also underwent detailed tests of various cognitive domains, and provided repeated measures of baseline assessments (with the exception of biomarker data). The imaging assessments were performed 9.27 (SD ±2.01) mean years after baseline. Using a standardized imaging protocol, participants were scanned across four imaging centers using Siemens Skyra 3T scanners (software VD13) with a standard 32-channel head coil.

At the time of data analysis (June 2023 to February 2024), approximately 50,000 participants had undergone an imaging assessment. We excluded participants who were missing data on imaging-related confounders (n = 22), and those with prevalent self-reported or diagnosed dementia, Parkinson’s disease, or any other chronic neurodegenerative condition (e.g., motor neuron disease), brain cancer, brain hemorrhage, brain abscess, aneurysm, cerebral palsy, encephalitis, head injury, nervous system infection, traumatic brain injury, or stroke (n = 93). We further excluded those with one or more missing values in any of the five MetS components (n = 8,888) for the main analyses and imputed missing MetS components and covariates in sensitivity analyses. See Supplementary Fig. 1 for the study cohort flow diagram.

Participants provided electronically signed consent for their data to be used in health-related research. The UK Biobank received ethical approval from the National Health Service North West Centre for Research Ethics Committee (reference no. 11/NW/0382).

MetS was defined at baseline using the 2009 Harmonized Criteria (1). The presence of at least three of the following components constituted a MetS diagnosis: 1) abdominal obesity (larger waist circumference: ≥102 cm in males; ≥88 cm in females); 2) elevated triglyceride levels (≥150 mg/dL [1.7 mmol/L]) or lipid-modifying medications; 3) elevated blood pressure (≥130 mmHg systolic blood pressure and/or ≥85 mmHg diastolic blood pressure) or antihypertensive medication use; 4) elevated fasting blood glucose level (≥100 mg/dL or ≥5.6 mmol/L) or drug treatment for elevated blood glucose; and 5) reduced HDL cholesterol level (<40 mg/dL [1.0 mmol/L] in males; <50 mg/dL [1.3 mmol/L] in females). Data on circulating glucose levels were obtained predominantly from nonfasting blood samples, which are more likely to be affected by recent food intake (as compared with fasting samples), potentially leading to high variability in glucose measurements. Therefore, we used glycated hemoglobin (HbA_1c) as a proxy measure based on recommendations from the American Diabetes Association, with a cut point of HbA_1c ≥5.7% (39 mmol/mol) to indicate hyperglycemia (15). Medication use was captured using Anatomical Therapeutic Chemical codes, informed by a thorough literature review of previous studies defining MetS using those codes, and expert clinical input (5).

Participants were categorized into two groups: no MetS (reference) and MetS. Supplementary Tables 1 and 2 contain details regarding variables and medication codes used to define MetS components.

Imaging derived phenotypes (IDPs) were derived from T1- and T2-weighted brain MRI scans using an image-processing automated pipeline developed by an external group in collaboration with the UK Biobank (16). IDPs were selected on the basis of their relevance to markers of brain health in the context of cognitive decline and/or dementia (17). These included global gray matter, white matter, and total brain (sum of gray and white matter) volumes, which were segmented using the Functional Magnetic Resonance Imaging of the Brain (FMRIB) Analysis Group’s Automated Segmentation Tool (18); hippocampal volume (left, right) using FMRIB’s Integrated Registration and Segmentation Tool (19), and white matter hyperintensity (WMH) volume from the Brain Intensity Abnormality Classification Algorithm (20), which uses both T1-weighted and T2-weighted data.

In this study, white matter volume refers to the total volume of white matter in the brain. In contrast, WMH refers to areas of white matter that appear hyperintense (i.e., brighter) on brain MRI; these are often linked to aging, cognitive decline, and dementia (17). Outliers were defined as having a value >8 times the median absolute deviation from the median for any given IDP (n = 1,008) and were removed per previous methods (16). Full details on the neuroimaging variables are listed in Supplementary Table 3.

We chose seven cognitive tests administered at the UK Biobank imaging assessment, which assessed different cognitive domains (21): trail-making tests (TMTs) A and B (executive function), fluid intelligence test (verbal and numerical reasoning), backward digit span task (working memory), symbol-digit substitution test (complex processing speed), paired associate learning task (verbal declarative memory), and the matrix pattern completion test (nonverbal reasoning). Higher scores indicate poorer cognitive performance for TMTs (i.e., time taken to complete test), whereas lower scores reflect poorer performance for all other tests. Individuals with TMT scores of ≤0 were removed from the analyses. Detailed descriptions regarding the UK Biobank cognitive tests have been published elsewhere (21). Full details regarding cognition variables in this study are given in Supplementary Table 3.

Covariates included factors considered potential confounders in the relationship between the exposure (MetS) and outcomes (neuroimaging and cognition) (3,9,17,22). Age in years was calculated as the interval between date of birth and the date of attending an imaging assessment center. Location of imaging assessment center (namely, Cheadle, Reading, Newcastle, and Bristol) was defined as attendance at the imaging site. The Townsend deprivation index was used as a proxy for socioeconomic deprivation and was assigned to each study participant using their residential postal code at baseline, and was categorized into fifths (1, least deprived; to 5, most deprived) (23). Sex (male, female) was defined using National Health Service records and/or self-reported data. Self-reported ethnicity (White, non-White), education level (primary, secondary, postsecondary nontertiary, tertiary), household income in British pound sterling (<18,000, 18,000–30,999, 31,000–51,999, 52,000–100,000, >100,000), smoking status (never, previous, current), and alcohol intake (never, former drinker, infrequent [special occasions only; one to three times per month], once or twice per week, three or four times per week, daily or almost daily) were measured via touchscreen questionnaires. Physical activity level (low: MET minutes ≤1,200; high: MET >1,200) was derived from touchscreen questionnaire items that were adapted to the validated short International Physical Activity Questionnaire (24); time spent conducting vigorous, moderate, and walking activity was weighted by the amount of energy expended, which allowed us to derive total MET minutes/week. APOE-ε4 carrier status (ε4 noncarrier, ε4 carrier) was derived using rs429358 and rs7412 single nucleotide polymorphisms, which were directly genotyped (25). Participants with any missing covariate data, or who responded with “prefer not to answer/do not know,” were assigned as a separate category for each categorical variable.

We also incorporated several imaging-related confounders: 1) head size (i.e., intracranial volume); 2) head position (using x-, y-, and z-axis position coordinates); and 3) head motion (16).

Descriptive statistics were used to compare baseline characteristics between participants with and without MetS; mean and SD values were calculated for normally distributed variables, and median and interquartile range were calculated for skewed variables. Skewness was assessed for outcomes by visually inspecting their distributions through histograms. Skewed variables (i.e., TMT A/B, and WMH volume) were normalized using natural logarithm transformations (21). All outcomes were then standardized (mean = 0; SD = 1) to permit comparison of effect size across outcomes. Therefore, we report standardized β coefficients and 95% CIs throughout the results.

Separate multivariable linear regression models were used to estimate the association of MetS with neuroimaging and cognitive outcomes. Models were adjusted for age, sex, ethnicity, education, Townsend deprivation index, household income, smoking status, alcohol intake, physical activity level, and APOE-ε4 carrier status. To correct for interactions between age and sex, analyses were also adjusted for age² (nonlinear effects), sex × age, and sex × age². Per previous recommendations, we included quadratic age terms in our models (age²) to account for nonlinear age effects across all outcomes (26). This decision was further supported by ANOVA tests comparing models with and without the age² term, which demonstrated that its inclusion significantly improved model fit. Similarly, age–sex interaction terms were included to comprehensively account for their potential joint effects on brain health. Age and sex are known to interact in a nonadditive manner in the context of brain health outcomes; thus, the product of these two confounders must be considered (26). For neuroimaging analyses only, we further adjusted for the following imaging-related confounders (i.e., precision variables): head size, head position, and head motion (26).

In secondary analyses, we examined associations of individual MetS components and the categorical number of MetS components present (0–5) with each neuroimaging and cognitive outcome. In a sensitivity analysis, we repeated the main analysis using multiple imputation to investigate the impact of missing data for the exposure and covariates. We also examined the interaction between MetS and age (<65 years, ≥65 years) and between MetS and sex, due to prior evidence indicating differences in structural neuroimaging and cognition based on these factors (17,27–29).

All P values were two sided, with statistical significance set at P < 0.05. All analyses were performed using RStudio, version 4.2.2.

The UK Biobank Resource (application no. 33952) holds the data used in this article. Data can be requested at www.ukbiobank.ac.uk/register-apply.

We identified 37,395 participants (mean age, 55.01 ± 7.55 years), among whom 7,945 (21.2%) had MetS. The average time between baseline and follow-up imaging assessment was similar between participants with and without MetS (MetS: 9.21 ± 2.04 years; no MetS: 9.28 ± 2.0 years). Compared with no MetS, those with MetS were more likely to be older, male, of non-White ethnicity, have less education, have lower household income, reside in more socioeconomically deprived areas, be previous smokers, be less physically active, and APOE-ε4 carriers (Table 1). Among those with MetS, 65.7%, 27.4%, and 6.9% had three, four, and five MetS components, respectively. Among MetS participants, the most prevalent component was elevated blood pressure (93.2%), followed by elevated triglyceride levels (83.7%), larger waist circumference (72.0%), reduced HDL cholesterol levels (54.6%), and elevated HbA_1c (37.7%).

Compared with no MetS, the presence of MetS was associated with lower total brain (β: −0.06; 95% CI −0.08, −0.04) and gray matter volume (β: −0.10; 95% CI −0.12, −0.08), as well as increased WMH volume (β: 0.08; 95% CI 0.06, 0.11) (Table 2). MetS was also associated with lower left (β: −0.03; 95% CI −0.05, −0.01) and right (β: −0.04; 95% CI −0.07, −0.02) hippocampal volumes. No significant association was observed between MetS and white matter volume. Results remained similar after performing multiple imputation for missing exposure and covariate data (Supplementary Table 4).

A dose-response relationship was observed between the number of MetS components present and several neuroimaging measures (Fig. 1, Supplementary Table 5). Specifically, an increasing number of MetS components were associated with lower total brain (β range: −0.02 to −0.21; P for trend < 0.001) and gray matter (β range: −0.03 to −0.31; P for trend < 0.001) volume, and greater WMH volume (β range: 0.13 to 0.30; P for trend < 0.001). An increasing number of MetS components also showed a dose-response association with lower left (β range: 0.00 to −0.11; P for trend = 0.003) and right (β range: 0.00 to −0.09; P for trend = 0.002) hippocampal volume. No association was observed with white matter volume.

In analysis of individual components, larger waist circumference and higher HbA_1c were significantly associated with lower total brain and gray matter volumes, and having elevated blood pressure showed the strongest association with greater WMH volume (Supplementary Table 6).

Associations between MetS and neuroimaging outcomes were consistent among younger (<65 years) and older (≥65 years) participants. However, there was a significant interaction (P < 0.001) of age with MetS and WMH volume, with the strength of the association being greater among younger participants (<65 years: β: 0.09, 95% CI 0.07, 0.11; ≥65 years: β: 0.04, 95% CI −0.02, 0.11) (Supplementary Table 7). Moreover, there was a significant interaction of sex with MetS and total brain, gray matter, and white matter volumes (P < 0.001), with the strength of these associations being greater in male participants. Associations among other neuroimaging outcomes remained similar (Supplementary Table 8).

MetS participants performed significantly poorer on tests measuring several different cognitive domains, including working memory (β: −0.10; 95% CI −0.13, −0.07), verbal declarative memory (β: −0.08; 95% CI −0.11, −0.05), processing speed (β: −0.06; 95% CI −0.09, −0.04), verbal and numerical reasoning (β: −0.07; 95% CI −0.09, −0.04), nonverbal reasoning (β: −0.03, 95% CI −0.05, −0.01), and executive function (for TMT B, β: 0.05, 95% CI 0.03, 0.08; for TMT A: β: 0.03, 95% CI 0.01, 0.06) (Table 2). Results remained similar after performing multiple imputation for missing exposure and covariate data (Supplementary Table 4).

We also observed a dose-response relationship between the number of MetS components present and cognitive performance across several domains (Fig. 2, Supplementary Table 9). An increasing number of MetS components were associated with poorer cognitive performance on tests of working memory (β range: −0.05 to −0.27; P for trend < 0.001), verbal declarative memory (β range: −0.05 to −0.22; P for trend < 0.001), processing speed (β range: −0.02 to −0.15; P for trend < 0.001), executive function (TMT A or TMT B: β range: 0.01 to 0.16; P for trend < 0.001), and verbal and numerical reasoning (β range: −0.04 to −0.13; P for trend < 0.001). Increasingly poorer performance was also noted for nonverbal reasoning, although this was less pronounced (β range: −0.03 to −0.13; P for trend = 0.016).

Regarding individual components, elevated blood pressure was consistently associated with significantly poorer performance across all cognitive domains (Supplementary Table 6). Associations of MetS with poor cognitive performance were generally consistent among younger (<65 years) and older (≥65 years) participants. However, no significant interaction of age with MetS and cognition was observed for any of the cognitive domains (Supplementary Table 10). In contrast, there was a significant interaction of sex with MetS and verbal and numerical reasoning (P = 0.023), where the strength of the association was greater among males. Associations among other cognitive outcomes remained similar when comparing by sex (Supplementary Table 11).

In this large, population-based cohort of more than 37,000 adults without mid- to late-life dementia, MetS was associated with less brain volume, greater vascular pathology, and poorer cognition. There was a dose-response relationship between an increasing number of MetS components and smaller brain volume, vascular pathology, and poorer cognition. These findings suggest that MetS is associated with poorer global brain health rather than having region- or domain-specific effects, which might have implications for understanding previously observed relationships between MetS and dementia.

Several studies have associated MetS with several brain abnormalities, including lower total brain volume, silent brain infarcts, decreased gray matter volume in specific brain regions, and increased WMHs (7,8). However, other research studies have shown inconsistent findings. Cavalieri et al. (27) found no significant differences in total brain volume between those with and without MetS in 819 dementia-free participants. Additionally, Sala et al. (30) found no association between MetS and gray and white matter volumes among elderly participants from families with a history of longevity. Tiehuis et al. (13) found that MetS was associated with smaller brain volumes but not increased WMHs in 1,232 patients with manifest arterial disease. Similarly, evidence regarding MetS and cognitive performance has been equally varied. Although previous studies have been relatively consistent in demonstrating an association between MetS and impaired global cognition scores, the impact of this association on individual cognitive domains remains largely inconsistent (10). Inconsistencies could be due to several methodological issues, including 1) most studies comprised relatively small samples; 2) most studies focused on specific disease populations (e.g., MetS in those with mental disorders), limiting generalisability; 3) exposure misclassification resulting from heterogeneity in the components and specific cutoffs used to define MetS; 4) the vast number of cognitive tests used to assess the same cognitive domain, with each having different sensitivity that could impact the ability to detect cognitive issues; and 5) large variation in sample characteristics (i.e., age, sex, education), and covariate adjustment. (22) Findings from our study contribute to this existing body of research by providing a more robust and comprehensive evaluation of the influence of MetS on various neuroimaging and cognitive measures in a large sample.

In our study, MetS was associated with several neuroimaging and cognitive indicators of poor brain health. Specifically, we found that MetS was linked to lower total brain and gray matter volumes, including hippocampal volume, as well as increased WMHs, which were paralleled by cognitive deficits across several domains, including memory, reasoning, processing speed, and executive function. Collectively, these findings suggest that rather than having a localized effect, MetS appears to have a more widespread effect on brain health. These findings align with and build upon our previous study using the UK Biobank data, as well as other dementia research studies from Europe and Asia, which demonstrate consistent associations of MetS with increased dementia risk, irrespective of the dementia subtype (5,31–34).

We also found that an increasing number of MetS components were associated with greater reductions across several measures of brain volume and poorer performance on various cognitive domains. Although several studies have reported similar findings in relation to the number of MetS components and cognitive outcomes (9,10), few comparable studies have explored these relationships with structural neuroimaging outcomes, yielding inconsistent findings (30,35,36). Our study also provides a detailed evaluation of various cognitive domains in relation to the number of MetS components present, showing that the presence of a greater number of components incrementally increases cognitive deficits in areas of memory, processing, and executive function. Taken together, these results indicate that the impact of MetS on brain health is characterized by a dose-response relationship. Therefore, considering the overall number of MetS components present, rather than focusing solely on its threshold effect (i.e., presence of at least three components) may prove useful for devising prevention strategies to mitigate cognitive decline and dementia.

Interventional studies show promising results in managing MetS components to improve brain health. The SPRINT-MIND (Systolic Blood Pressure Intervention Trial Memory and Cognition in Decreased Hypertension - Memory in Diabetes) trial demonstrated that aggressive blood pressure control reduced WMH progression (37,38). The Sydney Memory and Aging Study observed that people with diabetes who used metformin experienced less cognitive decline and dementia than those not treated (38). Moreover, physical activity was shown to improve cognition in people with diabetes (37). The ACCORDION-MIND (an extension of the MIND substudy of the Action to Control Cardiovascular Risk in Diabetes) trial noted short-term increases in brain volume from intensive hyperglycemia management, though benefits did not persist in the long term (37). Conversely, trials targeting hypercholesterolemia did not find benefits for brain health (37). Perhaps the most promising findings come from the Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability (FINGER) trial, a multidomain intervention involving a diet, exercise, cognitive training, and vascular risk monitoring program in older adults with high rates of MetS-related components (37,38). Results from the FINGER trial indicated that the intervention group had significantly less cognitive decline than controls (although there were no significant changes in brain volume or WMHs) (37,38). Further large-scale intervention studies with longer follow-up are necessary to understand how reversal of MetS as a whole, or its components, might affect overall brain health and dementia risk.

Key strengths of this study include a comprehensive evaluation of the relationship of MetS with several neuroimaging measures and validated domain-specific cognitive outcomes (21). To our knowledge, this is also the largest study of its kind, incorporating extensive phenotypic data and specific genotypic information on APOE-ε4 carrier status, allowing for a robust assessment of these associations. This study also has several limitations. First, MetS was defined using baseline measurements (due to lack of biomarker data at the follow-up assessment), with neuroimaging and cognitive outcomes assessed several years later, limiting our ability to account for variability in MetS status over time. However, the validity of incorporating baseline biomarker data is supported by previous work demonstrating the high stability of these data over time (39). Second, we used HbA_1c as a proxy for fasting glucose level (given the low number of fasting samples), which varies from the Harmonized Criteria for MetS. However, previous recommendations from the American Diabetes Association support use of this measure as a suitable proxy for glucose (15). Third, our study was limited by lack of information on the severity and duration of MetS and its individual components. Fourth, although we excluded prevalent dementia, some cases may have gone undetected; this could introduce selection bias, particularly if the prevalence of undetected cases varies between those with and without MetS. Fifth, the design of this work makes our findings susceptible to residual confounding from unmeasured factors, which is an inherent limitation of all observational studies. Moreover, we did not adjust for multiple comparisons, increasing the potential for type I error. Additionally, UK Biobank participants are known to be healthier than the general population, especially among those who attended the follow-up imaging assessment (i.e., healthy-volunteer bias). Thus, it is possible that our findings may underestimate the true magnitude of the association between MetS and brain health (40). Moreover, findings from the UK Biobank may not be broadly generalizable. Finally, due to the nature of data availability, we were unable to assess how MetS influences changes in neuroimaging and cognition over time.

In conclusion, we found that MetS, a cluster of risk factors associated with cognitive decline and dementia, was linked to significantly smaller brain volumes and greater vascular pathology, which were paralleled by impaired performance across a range of cognitive domains. These findings demonstrate that MetS is related to poorer global brain health, aligning with previously observed associations between MetS and increased risk of several dementia subtypes (5,31–34). Further research is necessary to understand whether reversal or improvement of MetS (and its components) can improve brain health and reduce risk for dementia.

Acknowledgments. The authors are grateful to the participants for generously dedicating their time to take part in the UK Biobank study. This research has been conducted using the UK Biobank Resource under application number 33952. This work uses data provided by patients and collected by the NHS as part of their care and support. Copyright (2024), NHS England. Reused with permission of the NHS England and UK Biobank. All rights reserved. This research also 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 (grant reference: MC_PC_20058).

The authors thank Dr. Ben Lacey and Dr. Robert Clarke (from the Nuffield Department of Population Health, University of Oxford) for their valuable guidance on the selection and categorization of medications used to define the exposure in this study. Dr. Lacey and Dr. Clarke received no financial support for their participation.

Funding. This work was supported by a doctoral research grant from the Canadian Institutes of Health Research Institute of Aging (Priority Announcement – Aging; Funding reference no. DAA–181627), and a scholarship offered by the Nuffield Department of Population Health at the University of Oxford (to D.Q.). This work was also supported by the Nicolaus and Margrit Langbehn Foundation (a grant to E.K.). For the purpose of open access, the authors have applied a Creative Commons Attribution license to any author accepted manuscript version arising.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. D.Q. contributed to the conception and design of the study as well as the data analysis and interpretation, and drafted the manuscript. T.J.L. and N.E.A. contributed to the conception and design of the study and data acquisition, and they critically reviewed and edited the manuscript. A.T., E.K., and S.U.A. contributed to the conception and design of the study and critically reviewed and edited the manuscript. All authors approved the final version of the manuscript. D.Q. 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.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were John B. Buse and Jonathan E. Shaw.