Differential Treatment Effects on β-Cell Function Using Model-Based Parameters in Type 2 Diabetes: Results From the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE) Free

Over time, individuals with type 2 diabetes (T2D) experience a gradual decline in β-cell function, leading to worsening glycemic control and increased reliance on glucose-lowering medications. This decline, estimated at 5%–10% annually (1), poses a significant challenge in maintaining long-term glycemic control (2). In our previous analysis of data from the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE), we investigated the long-term effects of four glucose-lowering medications (insulin glargine U-100, glimepiride, liraglutide, and sitagliptin) when added to metformin on insulin sensitivity and β-cell function using calculated C-peptide indices over time (3) and correlated these with loss of glycemic control (4). After 1 year, noticeable improvements in C-peptide responses were observed across all treatment groups, though the degree of improvement varied. However, these improvements declined gradually over time across all groups, while insulin sensitivity remained stable (3). Understanding of how these medications differentially affect distinct aspects of β-cell function and how these in turn influence glycemic control may guide optimal medication use in clinical practice or spur development of new targeted therapies.

Glucose-lowering medications act via different mechanisms, often by modulating insulin secretion, which is a complex process involving different pools of insulin secretory granules and various cell-surface receptors that augment insulin secretory pathways. Mathematical modeling of glucose and C-peptide data allows quantification of several different aspects of β-cell function that go beyond simple calculation of changes in insulin or C-peptide concentrations during an oral glucose tolerance test (OGTT). Deconvolution of C-peptide concentrations allows for calculation of insulin secretion rates (ISRs) over time. Modeling of changes in insulin secretion with respect to glucose can indicate how responsive the β-cell is to early changes in glucose (rate sensitivity) and how sensitive the β-cell is in secreting insulin in response to varying glucose concentrations (dose response or glucose sensitivity). Additionally, the Mari model (5,6) incorporates a measure to assess the impact of potentiating factors, such as incretin hormones and glucose, that augment insulin secretion. In the present investigation, our goal was to use modeling to better understand the physiology underlying the varying effects of GRADE medications on β-cell function. While in previous research modeling has been used to explore the effects of various glucose-lowering medications on insulin secretion, these studies were limited by small sample size, predominantly single-medication interventions, and relatively short follow-up periods.

In addition to comparing the effects of GRADE treatments on β-cell parameters, we determined whether model parameters could be used to identify individuals at higher risk of losing glycemic control. Earlier GRADE cohort analysis indicated that higher C-peptide response measures correlated with a reduced rate of glycemic failure, regardless of treatment (4). Building on prior studies, we hypothesized that model-based β-cell function parameters would vary by treatment and be associated with the risk of glycemic deterioration.

In this secondary analysis of the GRADE data, we used mathematical modeling of OGTT glucose and C-peptide data to assess the impact of four glucose-lowering medications on insulin secretion and parameters reflecting different elements of β-cell function. The GRADE methods (7) and primary outcome results (8) have previously been published. In brief, GRADE was a randomized, unmasked, prospective clinical trial in adults with T2D for <10 years, with baseline A1C 6.8%–8.5% (51–69 mmol/mol), who were on metformin alone. Effectiveness for glycemic control was compared between the addition of basal insulin (glargine), a sulfonylurea (glimepiride), a glucagon-like peptide 1 (GLP-1) receptor agonist (liraglutide), or a dipeptidyl peptidase 4 (DPP-4) inhibitor (sitagliptin) with quarterly study visits for A1C measurement over a mean 5.0 years of follow-up. Per protocol, rescue insulin was initiated when A1C reached >7.5%, confirmed. Rescue insulin consisted of glargine if the participant was randomized to noninsulin treatment and aspart insulin if randomized to glargine. Institutional review board approval was obtained at all sites, and participants provided written consent. The trial was overseen by an independent data and safety monitoring board and registered on ClinicalTrials.gov (NCT01794143). The full protocol is available at https://grade.bsc.gwu.edu.

At baseline and at 1, 3, and 5 years of follow-up a standard 75-g OGTT was performed after an 8-h overnight fast. Metformin and study medications were held on the morning of the OGTT, but insulin glargine was not withheld the night before. Samples were collected in EDTA at 0, 15, 30, 60, 90, and 120 min; processed; frozen; and shipped to the central laboratory (Advanced Research and Diagnostic Laboratory, University of Minnesota) for glucose and C-peptide assays as previously described (3).

Using MATLAB, version R2023a (MathWorks, Natick, MA), we modeled β-cell function parameters from OGTT glucose and C-peptide concentrations using C-peptide deconvolution (9) as previously described (6). Only OGTTs that included fasting measurements and at least four additional time points were modeled, blinded to treatment assignment.

The model describes the dependence of ISR on glucose concentration (dose response) and the rate of early glucose change (rate sensitivity) (Fig. 1 and Supplementary Fig. 1). The model includes a time-dependent potentiation factor to account for unexplained insulin secretion (5,6). β-Cell glucose sensitivity (dose-response slope), the primary parameter characterizing β-cell function, reflects sensitivity in ISR to the glucose concentration and deteriorates during glycemic progression (10). Rate sensitivity, derived from the glucose time derivative early in the OGTT, is related to early insulin release (11). Potentiation was quantified as the potentiation factor ratio (PFR) of 100–120 min divided by values at 0–20 min (5,6).

A standardized ISR at a glucose of 8 mmol/L (ISR@8mM glucose) was calculated from the ISR-glucose relationship multiplied by the pre-OGTT potentiation factor value to represent fasting conditions (i.e., before potentiation phenomena occur). Total insulin secretion (tISR) was determined as the integral of the ISR from 0 to 120 min. All model parameters were normalized for body surface area.

HOMA of insulin sensitivity (HOMA2-%S) and HOMA of β-cell function (HOMA2-%B) were calculated from fasting glucose and C-peptide concentrations with the HOMA2 Calculator, version 2.2.3 (Diabetes Trials Unit, University of Oxford, Oxford, U.K.) (12,13). Mean OGTT glucose was calculated as the glucose area under the curve from 0 to 120 min divided by 120.

Analyses included all randomized participants with modeling data at baseline (N = 4,712) (Supplementary Fig. 2). Participants with modeling data at one or more annual follow-up visits were included in longitudinal analyses. Treatment group differences based on baseline characteristics were tested for with a Wilcoxon rank sum test for continuous variables and a χ² test for categorical variables. Extreme values of the measures of β-cell function were Winsorized and recoded to ±6 SDs from the mean (14).

The primary analyses, designated as “on treatment” (OT), included only data collected while a participant was on metformin and the randomized medication. Data collected after discontinuation of either metformin or the randomized medication, or after addition of a nonstudy glucose-lowering medication or rescue insulin, were excluded. Secondary analyses, designated as intention to treat (ITT), included all data for participants regardless of medication changes (see Consolidated Standards of Reporting Trials [CONSORT] diagram [Supplementary Fig. 2]).

We compared changes in model parameters over time among treatment groups using linear mixed-effects models. The longitudinal outcome was defined as the change in the OGTT parameter from baseline. Each statistical model included visit, treatment, and a treatment-by-visit interaction term. Statistical models were adjusted for baseline age, sex, duration of diabetes, race, BMI, and the baseline value of the parameter analyzed. Treatment group differences in the parameter change estimates were tested for at each visit, with P values adjusted for the six pairwise treatment comparisons using the false discovery rate (FDR) method (15). Similarly, longitudinal differences (i.e., year [Y]3 − Y1 and Y5 − Y3) of parameter change estimates both within and between treatment groups were tested for, with adjustment for multiple comparisons using FDR. All pairwise tests assumed asymptotic normality of the estimated means in the different treatment groups.

A Cox proportional hazards (PH) analysis of association with time to reach the outcome A1C >7.5% (58.5 mmol/mol) was performed for each of the OGTT model parameters as time-varying covariates, first as a univariate analysis and again with adjustment for HOMA2-%S as a time-varying covariate. A final multivariate model was also used that included all OGTT model parameters and HOMA2-%S. All covariates, including the HOMA2-%S covariate, were treated as time varying; i.e., the hazard at time t was assumed to be proportional to the most recent covariate prior to time t (16). Statistical models were adjusted for baseline age, sex, duration of diabetes, race, and BMI.

We used a Classification And Regression Tree (CART) approach to identify the key OGTT parameters and traditional risk factors distinguishing subsets of risk associated with time-to-event A1C >7.5% (58.5 mmol/mol) (17). The CART model included treatment group, all OGTT model parameters (time varying), HOMA2-%S (time varying), and baseline covariates (age, sex, duration of diabetes, race, BMI). We used the CART models to verify whether the associations identified aligned with the findings from our primary time-to-event analyses.

GRADE is funded by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK). This manuscript is based on follow-up data and outcome assessments from the 5,047 participants enrolled into the study. This database will be available in the NIDDK Central Repository by 2025.

Baseline demographics, glycemic status, insulin sensitivity, and β-cell function parameters did not differ across the randomized treatment groups (Supplementary Table 1). Treatment was initiated after the baseline visit and titrated per protocol. Mean daily doses of study medications at 1, 3, and 5 years were as follows: metformin, 1,907, 1,880, and 1,861 mg/day; insulin glargine, 0.33, 0.38, and 0.37 units/kg/day; glimepiride, 3.66, 3.86, and 3.89 mg/day; liraglutide, 1.58, 1.60, and 1.63 mg/day; and sitagliptin, 99, 97, and 95 mg/day. The numbers (%) excluded for the OT analyses due to initiation of rescue insulin are summarized in the CONSORT diagram (Supplementary Fig. 2).

OGTT glucose curves (Fig. 2, row 1) and changes from baseline in glucose and model parameters by treatment group (Fig. 3) illustrate differences over time. Supplementary Table 2a and b (OT) and Supplementary Table 3a and b (ITT) provide the data for the comparisons between treatment groups and over time depicted in Fig. 3 (OT) and Supplementary Fig. 3 (ITT).

Mean OGTT glucose decreased most with liraglutide, followed by insulin glargine, sitagliptin, and glimepiride, at year 1 (Fig. 3A). From baseline to years 3 and 5, similar changes were seen with liraglutide, insulin glargine, and sitagliptin, while the least improvement continued to be seen with glimepiride. Findings were similar for the ITT analysis, except for OGTT mean glucose, with the smallest change over time with sitagliptin (Supplementary Fig. 3).

ISR (Fig. 2, row 2) was higher with liraglutide throughout the study, especially late in the OGTT. The dose-response slope (Fig. 2, row 3) and potentiation (Fig. 2, row 4) were also steeper and higher, respectively, with liraglutide compared with the other three medications. The dose-response slope was steeper for sitagliptin compared with glimepiride and insulin glargine. These patterns were maintained over time.

ISR@8mM glucose increased across all treatment groups at year 1, with the largest increase observed with liraglutide, followed by glimepiride and sitagliptin (Fig. 3B). After year 1 there was a progressive decline in ISR@8mM glucose in all treatment groups. The decline was greater from year 1 to year 3 with liraglutide compared with that for insulin glargine and sitagliptin, but there were no treatment group differences from year 3 to year 5. The ITT analysis showed similar findings (Supplementary Fig. 3B).

At year 1, tISR increased in all treatment groups except insulin glargine, with which tISR decreased (Fig. 3C). The largest increase in tISR was observed with liraglutide, followed by smaller increases with sitagliptin and glimepiride. After year 1 there was a progressive decline in tISR in all treatment groups, more pronounced with liraglutide from year 1 to year 3, but no significant differences across treatments were observed from year 3 to year 5. Similar trends were observed with the ITT analysis (Supplementary Fig. 3C). By year 3, tISR had either returned to baseline levels (sitagliptin ITT) or decreased below baseline (glimepiride in both OT and ITT analyses). In the liraglutide group, tISR remained above baseline at both years 3 and 5.

Glucose sensitivity increased in all treatment groups at year 1 but then progressively declined over time (Fig. 3D). Glucose sensitivity increased the most with liraglutide followed by sitagliptin; both remained above baseline at years 3 and 5. Glucose sensitivity increased the least with glimepiride and insulin glargine at year 1 and fell below baseline thereafter. The decline in glucose sensitivity was greatest with liraglutide from year 1 to year 3 but did not differ between treatment groups from year 3 to year 5. Similar results were observed with the ITT analysis (Supplementary Fig. 3D).

Rate sensitivity increased in all treatment groups at year 1 except liraglutide (no change), with the greatest increase occurring with insulin glargine (Fig. 3E). Between years 1 and 3, rate sensitivity did not change in the glimepiride and sitagliptin groups but decreased with both insulin glargine and liraglutide. Rate sensitivity did not change from year 3 to year 5 in any treatment group. The ITT analysis differed from the OT analysis with rate sensitivity decreasing in the glimepiride and sitagliptin groups from year 1 to year 3 (Supplementary Fig. 3E).

The PFR increased in all treatment groups at year 1, with the greatest increase with liraglutide (Fig. 3F). As with the other model parameters, the PFR declined over time in all treatment groups, decreasing more with liraglutide compared with glimepiride and sitagliptin from year 1 to year 3. There were no changes in PFR from year 3 to year 5 in any of the treatment groups. Similar results were observed with the ITT analysis (Supplementary Fig. 3F).

In univariate analysis, all model parameters were inversely associated with glycemic failure. Higher model-derived β-cell parameters (with and without adjustment for HOMA2-%S) were associated with lower risks of reaching the A1C outcome >7.5% (58.5 mmol/mol) in both the OT and ITT analyses (Table 1). All model parameters except for tISR remained independently associated with glycemic outcome in the multivariate model. There was no significant effect of treatment on the relationship between model parameters and risk of A1C >7.5% (58.5 mmol/mol). Similar results were obtained for risk of reaching A1C ≥7% (53.0 mmol/mol) (data not shown).

In the OT analysis, the first split variable separates participants by age, followed closely by glucose sensitivity (Supplementary Fig. 4). Additional split variables include HOMA2-%S, ISR@8mM glucose, tISR, and PFR. Treatment was not associated with glycemic failure. Overall, lower glucose sensitivity or combinations of lower HOMA2-%S, ISR, and PFR were associated with more rapid glycemic failure.

In the ITT analysis, the first split variable separates participants by glucose sensitivity, followed closely by HOMA2-%S (Supplementary Fig. 5). Additional split variables include age, ISR@8mM glucose, PFR, and treatment group. The lowest risk was in older individuals with higher HOMA2-%S, glucose sensitivity, PFR, and/or secretion parameters. Younger age and lower insulin sensitivity were associated with glycemic failure in the sitagliptin group (node 2).

The three covariates most strongly associated with glycemic failure were the same for both the OT and ITT analyses (age, glucose sensitivity, and HOMA2-%S). Although the order differed, this is likely random, as the P values for selecting the variables were very similar in both analyses, with changes in order driven by minimal differences in test statistics. In both models, glucose sensitivity was identified to have a strong association with risk, consistent with the time-to-event analyses.

GRADE is the first large-scale clinical trial to compare OGTT model-derived parameters of β-cell function in response to multiple glucose-lowering medications. An advantage of the OGTT is that it incorporates the physiologic contributions of the gut-pancreas axis in the measure of β-cell responses. Therefore, the unique examination of model parameters over 5 years of follow-up highlights treatment effects on different aspects of β-cell function, the global nature of the pathology affecting the β-cell, and its role in predicting progressive loss of glycemic control.

Key findings include differential treatment effects on model parameters (Supplementary Fig. 6 [heat map with relative treatment effects]) and rates of progression. Model parameters improved with all treatments in the short-term (year 1) and then declined. Liraglutide showed the largest increase in the first year but also a greater decrease between years 1 and 3. Despite this, liraglutide’s effects remained the highest among all treatment groups. Importantly, all parameters reflecting β-cell function, except total insulin secretion, were independently associated with the glycemic outcome, indicating that they are measurements of different aspects of the secretory response. Thus, each independently contributes to β-cell decompensation and glycemic failure.

Liraglutide primarily increased ISR through enhanced glucose sensitivity and potentiation, consistent with results of smaller studies involving liraglutide (18) and exenatide (19). In contrast, sitagliptin had a lesser impact on insulin secretion and glucose sensitivity but notably increased rate sensitivity, indicating a beneficial effect on early insulin secretion. Similar findings were seen in previous smaller studies with DPP-4 inhibitors (20). These findings are also consistent with our previous findings that early C-peptide secretion did not differ between sitagliptin and liraglutide at year 1, but late β-cell responses were higher with liraglutide, resulting in higher total C-peptide responses (3).

The differential impact of GLP-1 receptor agonists and DPP-4 inhibitors is likely related to greater potency of the GLP-1 receptor agonists. DPP-4 inhibitors modestly raise active GIP and GLP-1 levels, while GLP-1 receptor agonists deliver pharmacologically greater GLP-1 effects. The ability of sitagliptin to increase rate sensitivity may be due to the increase in active GIP as well as GLP-1. The response to GIP is diminished in T2D but not absent. The complementary effects of GIP and GLP-1 in stimulating insulin secretion in healthy men (21) and the marked enhancement of insulin secretion in T2D with newer GIP/GLP-1 dual agonists (22) suggest potential interactive effects (23). Lack of change in rate sensitivity with GLP-1 agonist therapy was reported in previous studies with liraglutide (18) and exenatide (19). Rate sensitivity occurs in the early part of the OGTT, which encompasses both first- and second-phase responses. The ability of GLP-1 receptor agonists to enhance glucose sensitivity, the primary driver of insulin secretion, might limit the model’s ability to differentiate dynamic and static insulin responses.

Treatment with glimepiride increased insulin secretion at a fixed glucose level of 8 mmol/L, the mean fasting glucose level within the GRADE cohort, and also increased rate sensitivity after 1 year, consistent with previous results with nateglinide (24). This likely reflects the mechanism of action of sulfonylureas to stimulate insulin secretion regardless of the ambient glucose concentration. Despite model parameters declining at a relatively similar rate after year 1, fasting and OGTT mean glucose increased faster with glimepiride than with sitagliptin. Both liraglutide and sitagliptin maintained lower OGTT mean glucose and higher glucose sensitivity in comparison with glimepiride.

Treatment with insulin glargine led to small increases in glucose sensitivity and potentiation but a more pronounced increase in rate sensitivity, suggesting improvement in the early insulin response. The improvement in rate sensitivity persisted throughout the 5 years of follow-up. Investigators of one other study reported similar findings after 1 year that were not sustained after 3 years (19). We postulate that the greater glucose-lowering effect of insulin, particularly as it was titrated with the aim of a fasting glucose of 5.6 mmol/L/L (100 mg/dL), improved the early insulin response (rate sensitivity). It was reported that elevated fasting glucose can abolish the first-phase response to an intravenous glucose bolus (25), while intensive insulin treatment restored the first-phase insulin response in newly diagnosed T2D (26) and near normalization of glucose improved the insulin response to GLP-1 and GIP in individuals with T2D (27). Despite lowering of glucose with insulin glargine, the decrease in β-cell parameters over time was similar across all treatment groups between years 3 and 5. Thus, β-cell rest did not appear to slow progression of the underlying pathophysiology.

In the multivariate analysis, all parameters, except tISR, were independently associated with the glycemic outcome; however, higher glucose sensitivity was selected as the most strongly associated OGTT parameter in the CART analysis. While it is well established that β-cell dysfunction is central to T2D progression, our study is the first large clinical trial to evaluate model-based parameters in the context of common glucose-lowering medications. Our findings complement those of the Innovative Medicines Initiative DIabetes REsearCh on patient straTification (IMI - DIRECT) study, which focused on adults with recently diagnosed T2D, with identification of deteriorating insulin sensitivity and glucose sensitivity as critical factors in glycemic progression over 3 years (10).

While β-cell function parameters were inversely associated with glycemic progression, there was no significant treatment heterogeneity in the relationship between model parameters and risk of A1C >7.5% (58.5 mmol/mol) in the Cox PH analysis, and treatment was not selected as a major determinant of glycemic outcome in the CART analysis. This can be explained by the use of time-varying β-cell model parameters in the statistical models, as it is the β-cell model parameters themselves (downstream effect of treatment) that directly impact glycemic control. Only in the ITT CART analysis did treatment with sitagliptin appear to be associated, in a subgroup who were younger, were more insulin resistant, and had lower β-cell function. The greater improvement in model parameters with liraglutide is consistent with the study’s main time-to-glycemic-failure outcome results (8) showing a slower rate of reaching these glycemic outcomes in the liraglutide group. Despite improvement in β-cell model parameters with sitagliptin, the smallest decrease in A1C was seen with sitagliptin, as was the fastest rate of reaching the glycemic outcomes (8). Other treatment effects independent from direct impact on β-cell function may contribute to lowering A1C, such as exogenous insulin directly lowering glucose or GLP-1 agonists decreasing appetite and reducing food intake.

To the extent that the model parameters reflect components of β-cell function, these data support a role for decline in β-cell function over time in the pathophysiology of glycemic deterioration in patients with T2D. That these parameters are differentially affected by glucose-lowering medications commonly used in clinical practice is of relevance to all physicians who treat patients with T2D and provides a rationale to support some of the conclusions of the main GRADE study, such as the superior performance of liraglutide in relation to duration of glycemic control. Our findings suggest that while different glucose-lowering medications variably affect β-cell function, improvements do not halt disease progression. This highlights a critical need for research targeting the underlying mechanisms of β-cell dysfunction in T2D, potentially through new pharmacologic approaches. While performance of OGTTs and modeling is not clinically practical, use of mathematical modeling of β-cell parameters, as demonstrated in GRADE, holds promise for future research to help identify clinically relevant surrogate markers and develop prediction models to aid in treatment selection, i.e., personalized medicine.

Strengths include the use of repeat OGTTs and advanced modeling of β-cell function in a large, randomized trial with 5-year follow-up, which provided for a detailed evaluation of treatment effects. Limitations include a predominantly White participant cohort with more men than women, which limits generalizability, and reliance on HOMA2-%S as a surrogate for insulin sensitivity, as exogenous insulin can result in overestimation of insulin sensitivity. There is also the possibility of confounding in the glargine treatment group due to the effects of exogenous insulin on C-peptide. The OT analyses were performed to avoid these confounding effects of rescue insulin in the nonglargine treatment groups. While removal of participants reaching the glycemic outcome could cause a survivor bias, similar results with use of an ITT analysis suggest that a strong bias was not present. An additional limitation is that while rate sensitivity parallels the loss of the first-phase response (28), its precision is not comparable, which limits the ability to detect differences. Finally, the study did not incorporate newer agents such as sodium–glucose cotransporter 2 inhibitors or dual receptor agonists, which are now common in clinical practice, or thiazolidinediones.

In conclusion, glucose-lowering medications added to metformin differentially impact β-cell function parameters reflecting their mechanisms of action. Each parameter independently contributes to β-cell decompensation and glycemic failure. Despite initial improvements in β-cell function and glucose lowering, the treatments studied in GRADE were unable to reverse or prevent progression of the disease process. Research directed at physiologic parameters of β-cell function should help to identify and potentially reverse the root cause(s) of the underlying β-cell dysfunction in T2D.

Acknowledgments. The GRADE Research Group is deeply grateful to the participants whose loyal dedication made GRADE possible.

Funding. GRADE was supported by a grant from the NIDDK of the National Institutes of Health under award no. U01DK098246. The planning of GRADE was supported by a U34 planning grant from the NIDDK (U34-DK-088043). The American Diabetes Association supported the initial planning meeting for the U34 proposal. The National Heart, Lung, and Blood Institute and the Centers for Disease Control and Prevention also provided funding support. The Department of Veterans Affairs provided resources and facilities. Additional support was provided by the National Institutes of Health (grants P30 DK017047, P30 DK020541, P30 DK020572, P30 DK072476, P30 DK079626, P30 DK092926, U54 GM104940, UL1 TR000170, UL1 TR000439, UL1 TR000445, UL1 TR001102, UL1 TR001108, UL1 TR001409, 2UL1TR001425, UL1 TR001449, UL1 TR002243, UL1 TR002345, UL1 TR002378, UL1 TR002489, UL1 TR002529, UL1 TR002535, UL1 TR002537, UL1 TR002541, and UL1 TR002548). Educational materials have been provided by the National Diabetes Education Program. Material support in the form of donated medications and supplies has been provided by Becton, Dickinson and Company; Bristol-Myers Squibb; Merck & Co.; Novo Nordisk; Roche Diagnostics; and Sanofi.

The content of this manuscript is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Duality of Interest. K.M.U. reports grants or contracts from AVID Radiopharmaceuticals and Lilly and consulting fees and travel support from Nevro outside the submitted work. M.A.B. reports grants from Eli Lilly for clinical trials made to her institution; payment for presenting grand rounds on diabetes; and participation on a data safety monitoring board for Oramed Pharmaceuticals outside the submitted work. RM.B. has received research support, has acted as a consultant, or has been on the scientific advisory board for Abbott Diabetes Care, Ascensia Diabetes Care, CeQur, Dexcom, Eli Lilly, Embecta, Hygieia, Insulet, Medtronic, Novo Nordisk, Onduo, Roche Diabetes Care, Tandem Diabetes Care, Sanofi, United Healthcare, Vertex Pharmaceuticals, and Zealand Pharma. The employer of R.M.B., nonprofit HealthPartners Institute, contracts for his services, and he receives no personal income from these activities. A.L.C. reports consulting or advisory board work with Novo Nordisk, MannKind, and Zealand Pharma outside the submitted work. HealthPartners Institute/International Diabetes Center at Park Nicollet employs A.L.C. and contracts with the following companies for his services as a clinical research investigator or consultant outside the submitted work: Novo Nordisk, Medtronic/Companion Medical, Insulet, Sanofi, Dexcom, Abbott, Eli Lilly, United Health, and Tandem Diabetes Care (no personal income from any of these services goes to A.C.). R.A.D. reports grants from Boehringer Ingelheim, AstraZeneca, and 89bio; payment or honoraria from AstraZeneca, Corcept Therapeutics, and Renalytix; and participation on a data safety monitoring or advisory board for AstraZeneca, Novo Nordisk, Corcept Therapeutics, and Boehringer Ingelheim outside the submitted work. N.R. reports grants from Novo Nordisk, consulting fees from Novo Nordisk and Eli Lilly, and receipt of equipment, materials, drugs, medical writing gifts, or other services from Novo Nordisk outside the submitted work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. All authors affirmed that authorship is merited based on the International Committee of Medical Journal Editors authorship criteria. K.M.U., M.T., A.B., and N.R. contributed to the conception and design of the research. K.M.U., M.A.B., A.L.C., A.K., M.W.S., and N.R. contributed to the acquisition of data. M.T. and S.P.R. conducted the statistical analysis of data. K.M.U., M.T., N.M.B., A.M., S.P.R., A.L.C., R.A.D., M.R.G., T.H., W.I.S., A.B., and N.R. contributed to the interpretation of data and results. K.M.U., R.A.D., M.R.G., A.K., W.I.S., and N.R. contributed to the supervision and management of research. M.W.S. contributed to the drafting of the manuscript. K.M.U. drafted the initial manuscript and subsequent revisions. M.T. drafted the statistical analysis methods. A.B. drafted parts of the discussion and abstract. N.R. drafted the introduction. All authors critically reviewed the manuscript for important intellectual content. K.M.U. and M.T. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Handling Editors. The journal editor responsible for overseeing the review of the manuscript was Matthew C. Riddle.