Diabetes Remission in the Alliance of Randomized Trials of Medicine Versus Metabolic Surgery in Type 2 Diabetes (ARMMS-T2D)

More than 34.2 million Americans, or 10.5% of the population, have type 2 diabetes (1), and ∼40% have obesity (2). Obesity and type 2 diabetes cause enormous individual health burden and societal health care costs, including in unexpected ways, such as increased morbidity and mortality from coronavirus disease 2019 (3,4). Substantial evidence from nonrandomized and smaller controlled trials indicates that metabolic/bariatric surgery (hereafter termed metabolic surgery) is a superior modality for weight loss and glycemic control. To date, however, randomized controlled trials (RCTs) have been limited by number, sample size, single site, surgery type, and follow-up duration. The lack of large and definitive randomized trials likely contributes to the extremely limited utilization of metabolic surgery to treat type 2 diabetes among qualified candidates (5).

The 1st and 2nd Diabetes Surgery Summits, comprised of international groups of experts across multiple disciplines, concluded with strong consensus that “gastrointestinal surgery may be appropriate for the treatment of type 2 diabetes” (6,7). The consensus opinion encompassed patients “who are good surgical candidates with a BMI of ≥30 kg/m² (≥27.5 kg/m² for Asians) and whose glycemia is inadequately controlled by lifestyle and medical therapy.” Despite these recommendations, however, which were adopted by the American Diabetes Association, the International Diabetes Federation, and formally ratified by 56 other leading medical, surgical, and scientific organizations worldwide (8), the role of operative interventions for type 2 diabetes treatment in patients with moderate obesity is still considered insufficiently addressed due to the absence of long-term data from large, randomized cohorts. Furthermore, many health insurance providers in the U.S. and across the globe indicate they will not provide coverage for metabolic surgery until such evidence is available.

Previously, four separate investigative groups (STAMPEDE [9], TRIABETES [10], SLIMM-T2D [11,12], and CROSSROADS [13]) initiated RCTs to evaluate the effectiveness and safety of metabolic surgery compared with multidisciplinary medical and lifestyle management of type 2 diabetes and obesity, including among individuals with class I obesity, for whom surgery is a not a widely used treatment option (14). These four RCTs were harmonized to form the Alliance of Randomized Trials of Medicine versus Metabolic Surgery in Type 2 Diabetes (ARMMS-T2D) consortium and provided pooled and longitudinal follow-up data for participants previously randomly assigned to surgical or nonsurgical diabetes treatment approaches in the parent trials. This group of participants now represents the largest cohort of patients with type 2 diabetes (approximately one-third having a BMI <35 kg/m²) to undergo randomized assignment to metabolic surgery or medical/lifestyle intervention.

The overall aim of ARMMS-T2D is to assess the durability and effectiveness of metabolic surgery compared with medical/lifestyle management in patients with type 2 diabetes and class I–III obesity. In this study, we report 3-year outcomes comparing medical/lifestyle to surgical intervention for treatment of type 2 diabetes. Specific primary and secondary outcomes include efficacy and longer-term durability of diabetes control, body weight, BMI, dyslipidemia, hypertension, number of medications required to manage diabetes and metabolic comorbidities, and longer-term complications of medical versus surgical interventions in each of the treatment groups.

ARMMS-T2D is a prospective observational study of participants with type 2 diabetes and overweight/obesity who were previously randomly assigned to undergo metabolic surgery or medical/lifestyle diabetes management approaches (ClinicalTrials.gov NCT02328599). Participants were formerly enrolled in registered RCTs at the Cleveland Clinic (NCT00432809), University of Pittsburgh (NCT00465829), Harvard Joslin Diabetes Center/Brigham and Women’s Hospital (NCT01073020), or University of Washington and Kaiser Permanente Washington (NCT01295229) (9–13). These trials included a set of common measures, and all of the trials had completed a minimum of 1 year of follow-up assessments. All eligible participants were re-enrolled into ARMMS-T2D and provided written informed consent in accordance with the guidelines for the protection of human subjects at each of the clinical sites.

Key criteria for participation in the parent trials included: candidate for general anesthesia and unsupervised exercise, aged 20–65 years, BMI 27–45 kg/m², diagnosis of type 2 diabetes confirmed by requiring diabetes medication and/or based on American Diabetes Association diagnostic criteria (15), ability and willingness to participate in the study and agreement to any of the research arms at their site, able to understand the options and comply with the requirements of each program, and negative urine pregnancy test at screening or baseline visits before intervention. Additionally, participants were required to have initiated randomized intervention and had at least one subsequent assessment.

The prespecified primary end point of ARMMS-T2D was the success rate for biochemical remission of diabetes at 3 years. Diabetes remission was defined as achieving hemoglobin A_1c (HbA_1c) levels ≤6.5% after cessation of glucose-lowering medications for at least 3 months, and this terminology and interpretation of remission is similar to the current American Diabetes Association consensus guidelines (16). Predetermined secondary end points included long-term diabetes control (mean and change from baseline in HbA_1c and fasting plasma glucose), body weight and waist circumference, number of medications required for targeting glycemic control (number and class of medications), development or progression of micro/macroalbuminuria and decline in glomerular filtration rate, hypertension, and dyslipidemia. Safety end points included death from any cause, microvascular and macrovascular complications, and adverse events (AEs).

All patients were analyzed in accordance with the intention-to-treat. Continuous variables were summarized using descriptive statistics and include mean, SD, median, and interquartile range. Categorical variables were presented as the number and percentage of participants in each category. A generalized linear model adjusted for the prespecified covariates of baseline HbA_1c value, year since randomization, diabetes duration, insulin use, and sex were used to analyze the primary end point. The secondary analyses were all unadjusted. An exploratory analysis was performed, including clinic site in the model, which potentially overestimated effect magnitudes, so site was conservatively not included in the final generalized linear model. The inverse-link transformation was used to obtain event probabilities. Continuous outcomes for body measurements, blood pressure, and laboratory markers are reported using observed values at baseline and 3 years. Change from baseline was calculated directly. For graphical purposes, least-square means from a mixed model of repeated measurements for HbA_1c and body weight were plotted over time. The Wilcoxon rank sum test was used in cases in which normality assumptions were violated. Unless otherwise specified, variables were analyzed in their original units. Analyses were performed using SAS, version 9.4 (SAS Institute, Cary, NC).

Of the 355 participants randomly assigned in the original parent trials, 39 did not undergo intervention; thus, a total of 316 patients previously randomly assigned to either surgical (N = 195) or medical/lifestyle (N = 121) treatment of diabetes were considered for participation in ARMMS-T2D (Fig. 1). All randomized trials were balanced at entry, but the merging of four trials with unequal contributions in the number of patients to each group contributed to a relatively small but statistically significant imbalance between surgical and medical/lifestyle intervention groups. Five people in the medical/lifestyle group eventually sought surgical intervention within the 3-year follow-up period (four underwent Roux-en-Y gastric bypass [RYGB], and one had sleeve gastrectomy [SG]). Nevertheless, they were analyzed with the medical/lifestyle group, in accordance with the intention-to-treat principle. Of the patients randomly assigned to surgery, 55% underwent RYGB, 25% SG, and 20% adjustable gastric banding (AGB).

Baseline characteristics of patients are provided in Supplementary Table 1. Among these, sex (70 vs. 62% female), race (73 vs. 67% White), BMI (37 ± 4 vs. 37 ± 3 kg/m²), and waist circumference (115 ± 10 vs. 115 ± 10 cm) did not differ between surgical and medical groups, respectively. Surgical patients were younger (49 ± 9 vs. 52 ± 7 years) and had higher systolic blood pressure compared with the medical/lifestyle group (134 ± 18 vs. 130 ± 16 mmHg), but diastolic blood pressures were similar (81 ± 10 vs. 79 ± 10 mmHg, respectively). Surgical and medical/lifestyle patients had a comparable duration of diabetes (9 ± 6 vs. 9 ± 6 years), but a greater number of surgical patients were using insulin (52 vs. 41%). Usage of other diabetes medications was similar between groups (Supplementary Table 2). Surgical patients had higher HbA_1c (8.8 ± 1.7 vs. 8.2 ± 1.4%) and albumin-to-creatinine ratio (median [quartile (Q)1, Q3], 8 [4, 23] vs. 6 [3, 12]) and a trend for higher fasting glucose concentrations (median [Q1, Q3], 163 [119, 223] vs. 146 [122, 180] mg/dL; P = 0.06). Surgical and medical/lifestyle patients had similar LDL cholesterol (99 ± 34 vs. 96 ± 32 mg/dL), HDL (43 ± 12 vs. 44 ± 13 mg/dL), and triglycerides (median [Q1, Q3], 144 [103, 226] vs. 152 [94, 231] mg/dL). Surgical and medical/lifestyle patients were also similar for use of statins (74 vs. 76%), β-blockers (18 vs. 21%), and ACE inhibitors or angiotensin receptor blockers (66 vs. 66%), while calcium channel blockers were more widely used in the medical/lifestyle group (8 vs. 19%) (Supplementary Table 2). Demographics and baseline characteristics by surgery type are shown in Supplementary Table 3. Demographics and baseline characteristics for participants included in the 3-year analysis compared with participants not included are provided in Supplementary Table 4.

Primary and secondary end points for participants with complete HbA_1c data at 3 years are displayed in Supplementary Table 1 and Fig. 2A. Three years after randomization, substantially more surgical patients achieved diabetes remission, defined as HbA_1c ≤6.5% off diabetes medications, compared with patients in the medical/lifestyle group (60 of 160 [37.5%] vs. 2 of 76 [2.6%], respectively; P < 0.001). Achieving glycemic targets of HbA_1c ≤6.5% or ≤7.0%, with or without diabetes medications, were also greater following surgical than medical/lifestyle intervention. When adjusted for treatment allocation, annual visit, baseline HbA_1c, diabetes duration, insulin use, and sex, the predicted probability of remission at 3 years with surgery was 41.6% (95% CI 29.6–58.3) compared with 1.0% (95% CI 0.2–4.0) in the medical/lifestyle group (P < 0.001). Surgical patients had greater reductions in HbA_1c (Δ = −1.9 ± 2.0 vs. −0.1 ± 2.0%; P < 0.001) and fasting plasma glucose (Δ = −52 [−105, −5] vs. −12 [−48, 26] mg/dL; P < 0.001) than medical/lifestyle patients. Of note, RYGB and SG achieved a more effective reduction in HbA_1c (−2.1% and −2.5%, respectively) (Supplementary Table 5) than AGB (−0.9%), but all surgical procedures were more effective in improving HbA_1c than the medical/lifestyle intervention (Fig. 2A).

Surgery yielded greater reductions in BMI (Δ = −8.0 ± 3.6 vs. −1.8 ± 2.9 kg/m²; P < 0.001), body weight (Δ = −22.7 ± 10.5 vs. −5.0 ± 8.7%; P < 0.001) (Fig. 2B), and waist circumference (Δ = −17.5 ± 10.2 vs. −2.1 ± 9.6 cm; P < 0.001) compared with medical/lifestyle intervention. Systolic but not diastolic blood pressure was improved after surgery compared with the medical/lifestyle intervention (Δ = −2.7 ± 19.3 vs. 3.2 ± 19.2 mmHg; P < 0.03). Surgical patients also achieved greater increases in HDL cholesterol (Δ = 14.5 ± 11.1 vs. 2.3 ± 10.1 mg/dL; P < 0.001) and reductions in triglycerides (Δ = −48 [−110, −2] vs. −10 [−70, 15] mg/dL; P < 0.004), but similar changes in LDL cholesterol (Δ = 2.2 ± 38.5 vs. −0.2 ± 26.3 mg/dL; P = 0.62). The number of patients with metabolic syndrome was lower 3 years after the surgical intervention (56 of 144 [38.9%] vs. 46 of 67 [68.7%]; P < 0.001). Whereas the albumin-to-creatinine ratio was reduced in the surgical group, there was no change in the medical/lifestyle group (Δ = −2 [−13, 1] vs. 0 [−4, 4] µg/mg creatinine; P = 0.003). Reductions in estimated glomerular filtration rate did not differ between groups (Δ = −4.6 ± 17.3 vs. −5.6 ± 17.5 mL/min/1.73 m²; P = 0.55) (Table 1). The percentage of patients requiring medications for diabetes, dyslipidemia, and hypertension at 3 years were all reduced in the surgical compared with the medical/lifestyle group (P < 0.001 for each) (Fig. 2C and Supplementary Table 2). For those receiving diabetes medications, regimens were also simplified, with less insulin per dose and fewer individuals requiring triple therapy. Key end points for the different surgical procedures are provided in Supplementary Table 5.

Table 2 displays cumulative serious AEs (SAEs) from the start of the original intervention through 3 years of follow-up in ARMMS-T2D. There were 16 cardiovascular-related SAEs (including 1 death and 6 angioplasty/stent procedures) in the medical/lifestyle intervention and 8 in the surgical intervention (4 in RYGB, 2 in SG, and 2 in AGB). Coronary revascularization–related SAEs were more prevalent in the medical/lifestyle intervention compared with the surgical groups. There were multiple percutaneous coronary interventions in two patients; however, there were insufficient numbers of total cardiovascular events to demonstrate statistical differences among the groups. More detailed analysis by type of surgery revealed a greater frequency of gastrointestinal-related SAEs after RYGB (12.1%), SG (12.2%), and AGB (17.9%) compared with medical/lifestyle intervention (0.8%). The RYGB group had a greater frequency of nutrition/metabolic-related SAEs (mostly related to dehydration) than did the other surgical groups or medical/lifestyle intervention. There were very few cancer-related SAEs in all groups, with no differences across groups (Table 2 and Supplementary Table 6).

A greater frequency of invasive procedures occurred after AGB (20%), RYGB (15%), and medical/lifestyle intervention (11%) compared with SG (4%) (Supplementary Table 7). AEs of interest not sufficiently severe to qualify as SAEs included AGB malfunction not requiring hospitalization or surgery (n = 6), emergency room visits for abdominal pain in two patients who received AGB, marginal ulcers after RYGB (n = 5), and gastric ulcer after medical/lifestyle (n = 1).

Despite growing evidence that metabolic surgery is the most effective treatment for type 2 diabetes (17), it is estimated that <1% of eligible patients actually undergo such operations (18). This is largely attributed to concerns from referring providers and patients about long-term safety and durability (19). In this study, we report data from the largest cohort of participants prospectively randomized to either surgical or medical/lifestyle-based management of type 2 diabetes. Our study includes a substantial proportion of patients who were only overweight or had mild/moderate obesity (BMI 27–34 kg/m²), as well as those with more severe obesity (BMI ≥35 kg/m²). The inclusion of patients from four different centers across the U.S. enhances the generalizability of our data and strongly supports the efficacy, durability, and safety of metabolic surgery to treat type 2 diabetes. We examined a range of metabolic surgical procedures and are one of only a few RCTs to include data on SG, which is now the most performed procedure worldwide.

We found that metabolic surgery was superior to medical/lifestyle intervention for achieving diabetes remission as well as lowering HbA_1c, fasting glucose, BMI, and other cardiovascular risk factors using substantially fewer medications at 3 years of follow-up. Our data also demonstrate that comprehensive combined medical and lifestyle intervention, modeled after the Diabetes Prevention Program (20) and Look AHEAD trials (21), was not sufficient in most patients to induce durable diabetes remission or achieve the guidance-driven HbA_1c target of ≤7.0% with or without diabetes medications. Despite education in nutrition, exercise, and self-monitoring, as well as use of newer diabetes medications as they became available (e.g., glucagon-like peptide 1 receptor agonists and dipeptidyl peptidase 4 inhibitors), diabetes remission was achieved in only 2.6% of patients, which was substantially less than that attained with metabolic surgery (37.5%).

Multidisciplinary weight-management approaches are emerging as viable and potentially cost-effective treatment options for patients with diabetes and overweight/obesity. Such approaches, however, often fail to achieve durable weight loss of >5–10% (22), and a new report suggests that despite greater medication use, diabetes control is getting worse, not better (23). One encouraging result is the recently completed DiRECT trial, a primary care–led diet intervention in the U.K. that successfully achieved diabetes remission (HbA_1c ≤6.5% off medications) in 46% of patients with diabetes after 12 months and 36% after 24 months (24,25). It is notable, however, that participants in DiRECT had very mild, early diabetes prior to intervention. The DIADEM-I trial examined an intensive lifestyle intervention combined with usual medical care in primary and community settings among adults with overweight/obesity and short diabetes duration (<3 years). With the 12-month intervention, diabetes remission occurred in 61% of participants compared with 12% of participants in the control condition (26). Additionally, newer medications such as semaglutide, which was not available when our study participants initiated their interventions, have recently proven efficacy in lowering HbA_1c and achieving 1-year weight loss that rival the effects of metabolic surgery seen at 1 year in the ARMMS-T2D parent RCTs (9–13,27). The maximum pharmaceutical weight loss yet reported with semaglutide is 15.8% at 68 weeks in adults with overweight or obesity; of note, 38.5% of the participants lost >20% body weight, 84% of participants experienced adverse gastrointestinal effects, and none of these participants had type 2 diabetes (28). By comparison, patients who underwent metabolic surgery have maintained a mean 3-year total weight loss of 22%, with RYGB producing a 27% weight loss. Other studies demonstrate comparable magnitudes of weight loss after surgery persisting for up to 20 years (29). While improving in efficacy, antiobesity medications approved by regulatory agencies including the Food and Drug Administration, the National Institute for Health and Care Excellence, and the European Medicine Agency are rapidly gaining in efficacy and ability to produce weight loss that is nearly comparable to surgery. Whether this amount of weight loss is sustainable with medication remains to be determined. Continued follow-up in ARMMS-T2D will provide the critical data to evaluate the efficacy of newer strategies alone or in combination with metabolic surgery for comparative and synergistic efficacy to facilitate remission of type 2 diabetes.

Longer-term results from both the Diabetes Control and Complications Trial and Epidemiology of Diabetes Interventions and Complications (DCCT/EDIC) (30) (albeit type 1 diabetes) and UK Prospective Diabetes Study (UKPDS) (31) demonstrate persistent health benefits of glycemic control initiated early after diabetes diagnosis, referred to as “metabolic memory,” with emergent cardiovascular benefit over a much longer time period, even long after intensive glycemic control has ceased (32). Together, these findings support the need for improved diabetes management options and potential for long-term health impact following metabolic interventions. However, our participants had diabetes diagnosed for 8 years on average, and it remains uncertain if later initiation of glycemic control will provide the same health benefits over time as that seen with early glycemic intervention. With the exception of recent 10-year follow-up data on 60 patients from Mingrone et al. (33), there are no long-term term data on patients randomly assigned to surgical versus medical/lifestyle intervention regarding the comparative effects on diabetes complications, particularly in patients with lower BMI (34). Furthermore, the study from Mingrone et al. (33) only included patients undergoing RYGB or biliopancreatic diversion, limiting its generalizability from a safety and durability perspective.

Macrovascular disease is a leading cause of morbidity and mortality for patients with type 2 diabetes. Furthermore, the long-term sequelae of microvascular disease (including neuropathy, nephropathy, and retinopathy) reduce quality of life (35). Therefore, optimizing glycemic control, blood pressure, and serum lipid levels are essential diabetes treatment goals. In addition to glycemic differences, we observed improvements in components of the metabolic syndrome, including waist circumference, systolic blood pressure, HDL cholesterol, and triglycerides, in the surgical compared with medical/lifestyle interventions (Table 1). Although the albumin-to-creatinine ratio, a marker of glomerular damage was lower in patients undergoing metabolic surgery, this statistically significant difference is not likely to be clinically significant, as mean baseline levels were in the normal range. There were four times as many serious adverse cardiovascular events among medical/lifestyle compared with surgical patients. These included one death after coronary artery bypass graft (CABG), one resuscitated cardiac arrest, and several angioplasty/stent events. Among patients undergoing surgery, there was one stroke and one CABG. Further analysis of macrovascular disease will require longer-term follow-up of our randomized cohort. Nevertheless, there is mounting evidence from large, well-controlled, and well-conducted but nonrandomized metabolic surgery trials that show improved micro- and macrovascular disease outcomes and reduced rates of major cardiovascular AEs (36–38).

There were too few AEs in ARMMS-T2D to draw firm, long-term safety conclusions. Beyond the macrovascular events noted above, important but less severe events included lightheadedness, vertigo, and foot ulcers. However, gastrointestinal and nutrition-related AEs predominated in the surgical compared with medical/lifestyle group. A limitation of our study was the lack of measurements of serum biomarkers at all sites to detect levels of micro- and macronutrient deficiencies resulting from these gastrointestinal operations. These nutritional deficiencies, including iron, thiamine, and vitamins B12, D, and A, are well documented in the literature and require lifelong prevention, monitoring, and management (39). Subsequent invasive procedures were common after all interventions, including AGB (20%), RYGB (15%), and medical/lifestyle intervention (11%), with the exception of SG (4%). Health implications for patients with these types of SAEs by treatment approach should also be considered, and longer follow-up will be informative.

Limitations to our study include a necessarily open-label design, differences in parent trial study protocols, and lack of uniform baseline and follow-up data at all sites related to diabetes complications (retinopathy and neuropathy) to draw conclusions about the impact of metabolic surgery on these events. We also note that the study was not powered to detect a difference between surgical groups. Some may argue that our primary end point, diabetes remission (HbA_1c ≤6.5% without pharmacotherapy), is simply not possible for the medical/lifestyle intervention group, especially given the advanced duration of disease. Further, serum measures of nutritional deficiencies, quality-of-life measures, and measures of alcohol use disorder were not uniformly collected at all sites and thus not available for analysis. We also note that most participants were female, which is consistent with real-world observations that women are more inclined to undergo weight-loss surgery. This does somewhat limit generalizability of the observations; however, more than one-quarter of our cohort was non-White, which is relatively diverse compared with prior bariatric research studies.

In summary, prospective randomized interventional data from the largest cohort of patients to date demonstrates that metabolic surgery improves glycemic control, diabetes-related comorbidities, and weight loss to a greater extent than medical/lifestyle intervention for up to 3 years after treatment, with minimal and generally tolerable AEs.

Acknowledgments. The authors thank Reba Blissell (University of Washington), Emily Eagleton (University of Pittsburgh), Kathleen Foster (Joslin Diabetes Center), Chytaine Hall (Cleveland Clinic), Claire Pothier (Cleveland Clinic), Katie Wicklander (University of Washington), and Danielle Wolfs (Joslin Diabetes Center) for study coordination, Christopher Axelrod (Pennington Biomedical Research Center) for technical assistance, the investigators of the parent trials, and the study volunteers for the considerable time and effort on the trial. The authors also thank the Cleveland Clinic Clinical Research Unit and the Joslin Clinical Research Center for technical support.

Funding. This research was primarily supported by an investigator-initiated grant from Ethicon Endo-Surgery and Medtronic and by in-kind support from LifeScan and Novo Nordisk.

Duality of Interest. D.E.C. is on the Scientific Advisory Boards of GI Dynamics, Endogenex, and Gila Therapeutics. A.B.G. completed this work at the Joslin Diabetes Center and is now employed by Novartis Institutes for BioMedical Research. S.R.K. is on the Scientific Advisory Boards of Fractyl Health; serves on the Medication Monitoring Committee for GI Dynamics; and has received grant support from Janssen Pharmaceuticals, Fractyl Health, Ethicon Endo-Surgery, and Covidien. D.C.S. is a stockholder/shareholder of G.I. Windows. D.E.A. received travel funds to the World Congress on Interventional Therapy for Diabetes and the International Federation for the Surgery of Obesity Latin America Chapter meeting. J.M.J. is on the Scientific Advisory Boards for WW International, Inc. and Wondr Health (formerly Naturally Slim). M.E.P. has been a coinvestigator on a National Institutes of Health R44 grant together with Xeris Pharmaceuticals; has consulted for Eiger Biopharmaceuticals; has received investigator-initiated grant support from Janssen Pharmaceuticals, Medimmune, Sanofi, AstraZeneca, Genesis, and Nuclea Biotechnologies; has been a site investigator for XOMA and Xeris Pharmaceuticals; and acknowledges clinical trial product support from Dexcom within the past 5 years, all unrelated to the current study. P.R.S. has received research funding from Ethicon, Medtronic, and Pacira; has consultancy agreements with GI Dynamics, Keyron Ltd., Persona Mediflix, Ethicon, Medtronic, and BD Surgical; and has ownership interest in SE Healthcare LLC. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. J.P.K. was responsible for conceptualization, methodology, investigation, resources, writing of the original draft, supervision, project administration, and funding acquisition. A.P.C. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). D.E.C. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). A.B.G. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). S.R.K. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). D.C.S. was responsible for conceptualization, methodology, investigation, resources, writing (review and editing). D.E.A. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). W.F.G. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). A.H.V. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). J.M.J. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). M.E.P. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing). K.W. was responsible for conceptualization, methodology, software, validation, formal analysis, investigation, writing (review and editing), and visualization. P.R.S. was responsible for conceptualization, methodology, investigation, resources, and writing (review and editing), and funding acquisition. J.P.K. is the guarantor of this work and, as such, had full access to all of 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 at the 81st Scientific Sessions of the American Diabetes Association, 25–29 June 2021.