Association of Timing of Moderate-to-Vigorous Physical Activity With Changes in Glycemic Control Over 4 Years in Adults With Type 2 Diabetes From the Look AHEAD Trial

Regular physical activity is a central component of blood glucose management in diabetes. Elevated HbA_1c, a measure of chronic glycemia, is predictive of cardiovascular events and microvascular complications in type 2 diabetes (1). Moderate-to-vigorous physical activity (MVPA) reduces HbA_1c in a dose-dependent manner (2). MVPA enhances muscle glucose uptake and improves long-term glycemic control by increasing skeletal muscle oxidative capacity and insulin signaling (3). While many studies focus on the type and dose of MVPA for optimal health benefits, little is known whether performing MVPA at different times of the day influences long-term glycemic control.

Recent animal studies demonstrated that metabolic responses to one MVPA bout varied by the times of day, suggesting time-of-day–dependent beneficial effects of bout-related MVPA (bMVPA) (4,5). Indeed, small human studies found better improvement in glycemic control when people with type 2 diabetes or obesity/overweight exercised in the afternoon/evening compared with the morning (6–10). However, these studies assessed small groups of participants for a short duration (i.e., 1 day to 12 weeks) in the context of a supervised exercise training program. There are also cross-sectional studies reporting afternoon or evening MVPA was negatively associated with insulin resistance (11) and HbA_1c (12), and a prospective study reporting an increase in nighttime physical activity correlates with increased HbA_1c across 3 months in 207 sedentary people (13). However, whether timing of unsupervised bMVPA modulates longer-term improvements in glycemic control remains unclear.

The Look AHEAD (Action for Health in Diabetes) study (14) evaluated an intensive lifestyle intervention in people with type 2 diabetes and included accelerometry for objective physical activity assessment. Participants were randomly assigned to a 4-year control treatment or a lifestyle intervention, in which the most intensive lifestyle intervention occurred in year 1. We aimed to examine the longitudinal associations of participants’ timing of unsupervised bMVPA assessed at year 1 and year 4 with changes in glycemic measures and glucose-lowering medication use over 4 years.

Look AHEAD was a multicenter, randomized controlled trial designed to test the effects of an intensive lifestyle intervention (ILI) intended to produce 7–10% weight loss on a composite outcome of death from cardiovascular causes, nonfatal myocardial infarction, nonfatal stroke, or hospitalization for angina. Participants had been diagnosed with type 2 diabetes, had a BMI ≥25 kg/m² (or ≥27 kg/m² when taking insulin), and were 45–76 years of age (14). Eligible patients were randomly assigned to the intervention or control group. The ILI is intensive behavioral treatment aiming to increase physical activity and reduce caloric intake. Year 1 was the most intensive, as it included individual and group counseling sessions occurring weekly for the initial 6 months of the trial, followed by two group sessions and one individual session per month for the second 6 months. Years 2 to 4 included two intervention contacts per month to facilitate continued engagement in the prescribed lifestyle behaviors. Participants in the control group (diabetes support and education [DSE]) received a less intense educational intervention, including three annual brief diet and exercise educational sessions and social support during years 1 through 4 (14). Of the 5,145 participants enrolled in the Look AHEAD study, 2,627 participants were enrolled in a substudy to assess physical activity using accelerometry (15). Among those, accelerometry data were available for 1,855 participants at year 1 and for 2,200 at year 4. All participants signed a consent form approved by their local institutional review board.

Participants attended a baseline clinic visit as well as annual follow-up visits for assessments of demographic lifestyle and anthropometric factors (14,16). Participants brought all prescription medications to the clinic visits to ensure recording accuracy. Physical activity was assessed using a waist-mounted RT3 triaxial accelerometer (StayHealthy, Monrovia, CA) (17). Participants were instructed to wear this device during all waking hours for a period of 7 days at each assessment period (year 1 and year 4). Data collection mode, quality control, and data reduction criteria for the accelerometer were implemented as previously described (18,19). Bouts of MVPA were identified as activity ≥3 METs (i.e., metabolic equivalents; one MET is the amount of energy used while sitting quietly) and ≥10 min in duration, allowing for one 1–2 min interruption. The current analyses used MVPA data at years 1 and 4 to define the bMVPA timing group and to calculate weekly bMVPA volume (MET × min/week) and intensity (METs/bout). For details and assessments of other covariates, see Supplementary Methods.

The primary outcome of this study was change from the previous assessment (i.e., baseline to year 1, and year 1 to year 4) in HbA_1c (percentage points). The secondary outcome was change in fasting glucose (mg/dL). Considering change in status of whether taking any glucose-lowering medication is a potential indicator for improvement in glycemic control, we also included it as an exploratory outcome (three-level ordered ordinal response), which was reported as the between-group odds ratio for discontinuation versus continuation or initiation in glucose-lowering medications compared with the previous assessment.

We assigned bMVPA timing groups based on their timing of bMVPA at year 1, as previously described (18), recategorized the participants at year 4 with the same method, and then used them as the time-varying exposure. In short, we first examined the distribution of bMVPA from the study population at year 1 and year 4, separately, across the clock hours (Supplementary Fig. 2A). After excluding bouts in the overnight hours (0000–0500), the 5 clock hours that had the lowest numbers of bouts (<20/h), we divided the remaining clock hours into four time windows in a way that each window had a quarter of the total MVPA bouts: morning (0500–1042), midday (1043–1342), afternoon (1343–1700), and evening (1700–2400) at year 1, and morning (0500–1030), midday (1031–1318), afternoon (1319–1636), and evening (1637–2400) at year 4. As we previously described (18), this data-driven approach allows each time-specific group to have a comparable number of participants, thus avoiding misinterpretation as a result of small subgroup size. For the participants who did not have bMVPA or only had bMVPA in 1 day, we assigned them into the inactive group. For those who had bMVPA on at least 2 different days, if ≥50% of their bMVPA volume (calculated as METs × min) occurred during the same time window, as defined above, we then considered them as engaged in temporally consistent bMVPA and classified them into the corresponding timing group. Representative examples for each of the active groups are shown in Supplementary Fig. 2B. A detailed breakdown of the percentage of bMVPA in each time window across different groups is represented in Supplementary Fig. 2C. There were seven participants in each year who had ≥50% of their bMVPA amount between 0000 and 0459 and were excluded. This, in addition to the aforementioned criteria, left us with 1,755 and 2,047 participants at year 1 and 4, respectively, qualified for further analyses. See Supplementary Fig. 1 for a Consolidated Standards of Reporting Trials (CONSORT) diagram.

Descriptive statistics (number, mean [SD], median [interquartile range; IQR]) and percentages for categorical variables were calculated for participant baseline characteristics. To test differences between the six groups based on timing of bMVPA, we used the Kruskal-Wallis test for continuous variables and the χ² test for categorical variables. χ² Statistics were also calculated for each cell to determine the most contributing cell.

We used linear mixed-effect models to test bMVPA timing group with respect to changes across 4 years in HbA_1c and fasting glucose. Models included participants as random effect, and fixed effects for time since randomization (i.e., 1 or 4 years), bMVPA timing group, and time-by-group interactions. For the ordinal outcome, change in glucose-lowering medication, we used the multinomial generalized estimate equation to test the association of bMVPA timing group with odds of discontinuation versus continuation and initiation in glucose-lowering medications over time. Considering that no information was available on insulin dosing and that the case numbers of initiation or discontinuation in insulin usage were small, we focused on oral glucose-lowering medications in noninsulin users in the main analysis. We obtained the least squares mean difference or odds ratio and 95% CI compared with the inactive group. When significant bMVPA timing group differences were found, post hoc pairwise comparisons with adjustment for false discovery rate (P_FDR) were performed to determine the specific differences between individual bMVPA timing group(s) for each year.

Multivariable-adjusted analyses (model 1) included time-invariant covariates, such as sex, age, race, education, clinic site, and diabetes duration, and time-varying covariates assessed at years 1 and 4, including current smoking status, alcohol consumption, and, for the mixed model only, glucose-lowering medications (i.e., insulin, thiazolidinedione, biguanides, sulfonylureas, meglitinides, incretin mimetics, and dipeptidyl peptidase 4 inhibitor). In model 2, we further adjusted for treatment (ILI/DSE). In model 3, additional factors, specifically BMI, cumulative weekly bMVPA, and average bout intensity were included as time-varying covariates to evaluate whether the relationship between bMVPA timing groups and glycemic control was independent of BMI and the volume and intensity of bMVPA. The least squared means generated from model 3 were used to visualize the trajectory of changes over time.

In secondary analyses, we tested interactions of treatment or sex by bMVPA timing group on the outcomes; if significant, we performed stratified analysis. Given that HbA_1c varied across the bMVPA timing groups at baseline, sensitivity analyses were performed (see Supplementary Methods). Statistical analyses were conducted with PROC MIXED and PROC SAS GEE software in SAS 9.4 (SAS Institute, Cary, NC). Bonferroni correction was applied for stratified analyses; otherwise, a two-sided P < 0.05 was considered statistically significant.

The data used for analysis during the current study were housed at the data coordinating center and are not available for public distribution. Data on Look AHEAD participants have been supplied to the National Institute of Diabetes and Digestive and Kidney Diseases Central Repository and are publicly available at https://repository.niddk.nih.gov/studies/lookahead/?query5Look%20AHEAD.

Among the 2,627 participants who were enrolled in a substudy to assess physical activity using accelerometry, there were 1,755 and 2,047 participants having ≥4-day valid accelerometry data at year 1 or 4, respectively (Supplementary Fig. 1), which left an analytical sample size of 2,331. Overall, 57% of the study population were women, 72% were non-Hispanic White, and 50% were randomly assigned to the ILI. At baseline, mean age was 59.2 years, mean BMI was 36.3 kg/m², mean diabetes duration was 6.8 years, and 15% were treated with insulin (Supplementary Table 1). At both year 1 and year 4, there were significant differences in the amount and intensity of bMVPA among bMVPA timing groups (Ps < 0.0001). The inactive group, by definition, had the lowest amount and intensity of bMVPA, while the morning group at year 1 (median [IQR], 635.5 [1065.4] MET × min/week, 4.7 [1.4] METs/min) and the mixed group at year 4 (554.6 [1071.8] MET × min/week, 4.7 [1.2] METs/min) had the highest amount and intensity of bMVPA. Detailed baseline characteristics and amount and intensity of bMVPA across 4 years are shown in Table 1, Supplementary Table 2, and Supplementary Fig. 3, respectively, stratified by bMVPA timing groups at years 1 and 4.

Changes in HbA_1c from baseline to year 1 significantly varied across the bMVPA timing groups (Table 2, model 1; P = 0.02). Importantly, further adjustment for treatment arm, physical activity-related measures (weekly bMVPA volume and intensity), and BMI did not alter the significant results at year 1 (models 2 and 3, P = 0.02, respectively). This indicated that the association between bMVPA timing and changes in HbA_1c was independent of the overall volume, average intensity of bMVPA, and BMI. The afternoon group had the greatest decrease in HbA_1c at year 1 (Fig. 1A and model 3 in Table 2). Specifically, the least squared means of decrease in HbA_1c in the afternoon group was ∼50% larger than the decrease in the mixed group at year 1 (afternoon: −0.89% [−1.34 to −0.44] vs. mixed: −0.60% [−1.04 to −0.16]), with an adjusted difference of −0.20 to −0.29 HbA_1c% compared with all other groups. For all the year-4 bMVPA timing groups, there was no significant change in HbA_1c between year 1 and year 4 (Fig. 1A and Table 2).

Sensitivity analysis excluding the inactive group and using the mixed group as the reference produced largely similar results, with the afternoon group showing the largest reduction in HbA_1c (Supplementary Table 3). Using proportional changes in HbA_1c as the outcome did not change results on year 1 but revealed a trend association with bMVPA timing at year 4 (Supplementary Table 4).

Moreover, we found that the associations between bMVPA timing and changes in HbA_1c modified by treatments (P_{group×treatment} = 0.02), but not by sex (P_group×sex = 0.67) (Supplementary Table 5). In the ILI, there were significant differences by the bMVPA timing group at year 1 (P_adjusted = 0.006), independent of BMI and weekly bMVPA volume and intensity, in which the afternoon group had the greatest decrease in HbA_1c at year 1 (−0.40%, −0.63 to −0.17) (Table 3). However, no significant group differences were found in the DSE. As for changes in fasting glucose, we did not find any significant differences across the bMVPA timing groups over 4 years (all Ps > 0.10) (Supplementary Table 6 and Supplementary Fig. 4).

Among the 1,939 noninsulin users, odds for discontinuation versus continuation or initiation of glucose-lowering medications over 4 years differed by bMVPA timing (Supplementary Table 7, model 1; P = 0.04). The afternoon group had the highest odds for discontinuation of glucose-lowering medications, with an odds ratio of 2.13 (95% CI 1.29–3.52) at year 1 compared with the inactive group. The significant association by timing of bMVPA was lost after further adjustment (P ≥ 0.13). The association between bMVPA timing and change in glucose-lowering medication did not differ between treatments (P_{group×treatment} = 0.62). Least squared means of odds for discontinuation versus continuation or initiation of glucose-lowering medications over 4 years, stratified by bMVPA timing group, are shown in Fig. 1B. For the completeness of the analyses, we also compared odds for discontinuation versus continuation or initiation of any glucose-lowering medications (i.e., including insulin) in all participants. The results remained similar (Supplementary Table 8 and Supplementary Fig. 4).

In our analyses of the Look AHEAD trial, we found that changes in glycemic control varied by bMVPA timing groups at year 1, when the most intensive behavioral treatment had occurred. Participants who performed most of the bMVPA in the afternoon had the greatest improvement in HbA_1c in year 1 and were more likely to stop taking glucose-lowering medications at year 1. Importantly, the association between timing of bMVPA and changes in HbA_1c at year 1 was independent of glucose-lowering medications, weekly bMVPA volume, bout intensity, and BMI. The association with bMVPA timing was attributable to participants randomized to ILI. We did not detect any reduction or rebound in HbA_1c or odds of glucose-lowering medication use in any bMVPA timing groups from year 1 to year 4. These findings indicate that being active in the afternoon may gain the most metabolic benefits, especially during the initial 12 months when the most intensive intervention occurred. Our study not only provides insights into potential time-of-day specific effects of physical activity but also may have important clinical implications on blood glucose management by lifestyle interventions.

Several small, controlled trials in humans (6–8,20) assessed the time-of-day–dependent effects of short-term exercise training on glycemic control. Savikj et al. (8) found in a randomized crossover trial that 2 weeks of high-intensity interval training in the afternoon was more efficacious at lowering 24-h blood glucose in 11 men with type 2 diabetes than training in the morning. In a retrospective observational study, Mancilla et al. (7) reported similar results in which 32 men had greater reduction in peripheral insulin resistance and fasting glucose levels after 12 weeks of exercise training in the afternoon versus morning. In addition, Moholdt et al. (6) documented that under a high-fat diet, a significant reduction in fasting glucose/insulin levels was only found in men randomly allocated to evening exercise, but not morning exercise. In contrast, Teo et al. (20) reported that 12 weeks of either a morning or evening multimodal exercise training program resulted in similar improvement in glycemic outcomes in 20 patients with type 2 diabetes.

However, three of these four small, short-term (≤12 weeks) human studies included only men, thus limiting generalizability to women. Also, these studies compared exercise at two times during the day, which only provided a snapshot of the diurnal variations. Moreover, these prior studies were conducted using structured exercise training protocols, which represent a more intensive approach for modifying exercise behavior compared with recommendations that are typically prescribed within clinical settings and more feasible to patients with type 2 diabetes (21). Despite the differences in study design, our results point in the same direction as most of the above studies. This is very encouraging because the consistent results support the notion that the efficacy of physical activity may be optimized by changing its timing.

The current findings may have important clinical implications. The adjusted reduction in HbA_1c in the afternoon group relative to the other groups (difference of 0.20 to 0.29 HbA_1c%) was equivalent to the previously reported effect size of ILI on HbA_1c over 10 years of follow-up (−0.22%), or about 35% of the effect size (−0.64%) in year 1 (14,22). HbA_1c values reflect a 2- to 3-month average serum glucose and include fasting and postprandial blood glucose levels (23). We did not find any evidence of an association between timing of bMVPA and changes in fasting glucose levels over 4 years, which suggested that the association observed between bMVPA timing and changes in HbA_1c might be attributable to differential improvements in postprandial glucose tolerance rather than fasting glucose levels. Future studies should include assessment of glucose tolerance as an outcome to examine this hypothesis.

The fact that the association between bMVPA timing and 1-year changes in HbA_1c was strong in ILI but not in DSE may suggest that the ILI might have helped the time-of-day–dependent effects of bMVPA manifest. However, that contribution may not be solely a result of physical activity, given that the ILI also included dietary counseling to assist in reducing energy intake to facilitate weight loss. Unfortunately, data on changes in dietary intake are not available for the participants involved in the accelerometry substudy, and therefore, this factor cannot be examined. This is a limitation of this study. Moreover, given that the effect was observed at year 1 but not from year 1 to 4, this may be a result of infrequent assessment of physical activity. For example, while we assume that a 1-week measurement of physical activity at year 1 would be able to represent most of the physical activity behavior from baseline to year 1, the same measurement at year 4 may not adequately represent the timing of bMVPA of participants across the period of year 1 to year 4, but annual or more frequent assessments were not available here. The lifestyle counseling in ILI also became less frequent after year 1, which led to a large variability in adherence to weight, diet, and activity goals. Moreover, during this period, there was weight regain and many participants gradually lost metabolic benefits gained at year 1 (14). The added variability and reduced overall improvement may also account for a reduced power to detect an association between bMVPA timing and changes in HbA_1c in the ILI from year 1 to year 4. As our prior study has detected sex-specific cross-sectional association between timing of physical activity and cardiovascular risk factors (18), we tested for sex differences in the current study. However, we did not find any sex-dependent effects, which could be due to lack of power.

When we used percentage changes in HbA_1c for sensitivity analysis, the results suggested a trend association between bMVPA timing at year 4. This association was largely driven by a rebound of HbA_1c in the evening group. One possibility could be a delayed dinner due to the evening activity, as late eating is associated with impaired glucose tolerance (24,25). However, we could not test the underlying cause in the current study.

The association between timing of bMVPA and changes in HbA_1c at year 1 was independent of glucose-lowering medication use. Indeed, this made the results even more meaningful because the afternoon group had higher odds for a reduction in HbA_1c despite a decrease in glucose-lowering medications. Similarly, our results also suggested that the amount and intensity of bMVPA could not account for the significant association. In fact, the afternoon group, which had the greatest improvement in HbA_1c, had the second lowest total amount of bMVPA among all the groups, suggesting the association between timing of bMVPA and HbA_1c cannot be fully explained by bMVPA volume and intensity. This finding indicated that timing is likely an important aspect of physical activity with a unique impact on metabolic health.

The circadian system may play a role in the time-specific benefits of bMVPA. Previous animal and human experimental work demonstrated that the adaptive metabolic responses to a bout of MVPA vary depending on time of day, and are directly under the regulation of the circadian clock genes (4,5,26–29). In addition, other behavioral factors (e.g., fasting/postprandial states, sleep-wake cycles) may also contribute to the diurnal variations in the adaptation to bMVPA (30,31). For example, postmeal physical activity, which may be occurring most often after lunch in the afternoon group, is an effective strategy for managing postprandial glucose excursions in type 2 diabetes even in the face of diabetes medications (32–34). Furthermore, the MVPA may also interact with the well-known “dawn phenomenon” (i.e., an increase in blood glucose levels and decreased insulin sensitivity in the early morning fasted hours) in patients with type 2 diabetes (35), which leads to time-of-day–dependent benefits. Future human experimental studies are needed to investigate the underlying mechanisms of this time-specific association.

Although this study has many strengths, including the large sample size, objective measures of unsupervised physical activity, a well-characterized cohort, and the long follow-up, there are limitations. First, our findings should be interpreted cautiously because of the potential for unmeasured confounders and selection bias that are known to occur in secondary analyses. For example, we did not have data on sleep, dietary intake, and meal timing. All were shown to strongly influence glucose control (24,36,37). Participants who performed most bMVPA in the afternoon might have had better underlying health or health behaviors (e.g., healthy diet, early meal timing, good sleep hygiene) to facilitate glucose management. In addition, because the bMVPA timing and glycemic measures were collected at same time point, reverse causation is possible. Thus, causality cannot be inferred from our findings. Instead, they warrant closer examination of the interaction effects of bMVPA and time of day on glycemic outcomes in patients with type 2 diabetes through large, randomized control trials.

Second, we used clock time to define timing of bMVPA. Given the individual differences in circadian timing relative to clock hour, it may be more physiologically relevant to use timing of bMVPA relative to endogenous circadian timing (38) or habitual sleep timing (39), which were not available for this data set.

Third, we studied 45- to 76-year-old adults with overweight/obesity and type 2 diabetes. It is important to validate our findings in the general population and younger adults with early-onset type 2 diabetes.

Last, it is noteworthy that the percentage body fat can be a better predictor for risks of type 2 diabetes than BMI (40). However, there were not enough participants with body composition data in the current study population. Thus, it would be important for future cohorts to collect this information to allow investigation of the interaction of MVPA timing and other behavioral factors.

In conclusion, this study indicated that participants who performed more bMVPA in the afternoon had the greatest reduction in HbA_1c and highest odds of discontinuing glucose-lowering medication in year 1 when the most intense lifestyle intervention occurred. The association of timing of bMVPA with changes in HbA_1c was independent of weekly bMVPA volume, bout intensity, and BMI. Further research with in-depth information on sleep patterns and nutrient intake is needed to better delineate the relationships between timing of bMVPA and glycemic control in type 2 diabetes. Our findings, together with others (6–8), highlight an exciting interdisciplinary research frontier on lifestyle interventions and circadian biology, which holds promise for optimizing treatment efficacy.

Acknowledgments. The authors thank the other investigators, the staff, and the participants of the Look AHEAD study for valuable contributions. Some of the information contained herein was derived from data provided by the Bureau of Vital Statistics, New York City Department of Health and Mental Hygiene.

Funding. This study was funded by the National Heart, Lung, and Blood Institute (K99-HL-148500, J.Q., principal investigator and R01-HL140574 to F.A.J.L.S.), National Institute on Aging (RF1AG059867 and RF1AG064312 to K.H.), and National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) (K23-DK114550 to R.J.W.M.). The Look AHEAD study was supported by the Department of Health and Human Services through the following cooperative agreements from the National Institutes of Health, NIDDK: DK57136, DK57149, DK56990, DK57177, DK57171, DK57151, DK57182, DK57131, DK57002, DK57078, DK57154, DK57178, DK57219, DK57008, DK57135, and DK56992. Additional funding was provided by the National Heart, Lung, and Blood Institute, National Institute of Nursing Research, National Institute on Minority Health and Health Disparities, National Institutes of Health Office of Research on Women’s Health, and the Centers for Disease Control and Prevention. This research was supported in part by the NIDDK Intramural Research Program. Additional support was received from the Johns Hopkins Medical Institutions Bayview General Clinical Research Center (M01RR02719), the Massachusetts General Hospital Mallinckrodt General Clinical Research Center and the Massachusetts Institute of Technology General Clinical Research Center (M01RR01066), the University of Colorado Health Sciences Center General Clinical Research Center (M01RR00051) and Clinical Nutrition Research Unit (P30 DK48520), the University of Tennessee at Memphis General Clinical Research Center (M01RR0021140), the University of Pittsburgh General Clinical Research Center (M01RR000056), the Clinical Translational Research Center funded by the Clinical & Translational Science Award (UL1RR024153) and National Institutes of Health NIDDK grant (DK 046204), the VA Puget Sound Health Care System Medical Research Service, Department of Veterans Affairs, and the Frederic C. Bartter General Clinical Research Center (M01RR01346). The Indian Health Service (IHS) provided personnel, medical oversight, and use of facilities.

The opinions and views expressed in this article are those of the authors and do not necessarily represent the views of the Indian Health Service or other funding sources, the NIDDK, the National Institutes of Health, or the U.S. Department of Health and Human Services.

Duality of Interest. The following organizations have committed to make major contributions to Look AHEAD: Federal Express, Health Management Resources, Johnson & Johnson, LifeScan Inc., Optifast-Novartis Nutrition, Roche Pharmaceuticals, Ross Product Division of Abbott Laboratories, SlimFast Foods Company, and Unilever. J.M.J. is on the scientific advisory board for Wondr Health, Inc. F.A.J.L.S. serves on the Sleep Research Society Board of Directors and has received consulting fees from the University of Alabama at Birmingham. F.A.J.L.S. interests were reviewed and managed by Brigham and Women’s Hospital and Partners HealthCare in accordance with their conflict-of-interest policies. F.A.J.L.S. consultancies are not related to the current work. R.J.W.M. has received research funding from Novo Nordisk unrelated to this work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.Q. contributed to formal analysis. J.Q. wrote the original draft. J.Q., Q.X., J.U., J.M.J., K.H., F.A.J.L.S., and R.J.W.M. contributed to methodology and data interpretation. J.Q., F.A.J.L.S., and R.J.W.M. conceptualized the study. M.P.W. contributed to data curation. J.Q., Q.X., M.C., M.L.E., J.U., J.M.J., K.H., F.A.J.L.S., and R.J.W.M. contributed to writing, review, and editing of the manuscript. J.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.

Prior Presentation. Parts of this study were presented in abstract form at the 82nd Scientific Sessions of the American Diabetes Association, virtual and at New Orleans, LA, 3–7 June 2022.