Comparative Effects of Randomized Second-line Therapy for Type 2 Diabetes on a Composite Outcome Incorporating Glycemic Control, Body Weight, and Hypoglycemia: An Analysis of the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE)

Metformin is usually recommended as first-line pharmacotherapy for type 2 diabetes, with additional agents added over time if guideline-recommended glucose control targets according to glycated hemoglobin (HbA_1c) assessment are not met or for other indications such as cardiorenal protection (1,2). Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE) was a large randomized trial designed to assess the effects of each of four commonly used glucose-lowering medications (glargine, glimepiride, liraglutide, and sitagliptin), when added to metformin, on metabolic control and other outcomes (3–5). In GRADE, the cumulative incidence of HbA_1c of ≥7.0% (53 mmol/mol) (the primary GRADE metabolic outcome) differed significantly among the four groups; rates with glargine and liraglutide were similar, and lower than those with glimepiride and sitagliptin (4). Differences among groups with respect to HbA_1c >7.5% (>58.5 mmol/mol) (the secondary GRADE metabolic outcome) paralleled those of the primary outcome.

Achieving and maintaining optimal glycemic control, especially in the early years of type 2 diabetes, reduces the risk of long-term microvascular complications and perhaps cardiovascular outcomes (1). The effects of particular medication classes on reaching HbA_1c targets are important factors in decision-making according to patients (6,7) and clinicians (8). However, additional factors including side effects of medications, particularly weight gain (7–9) and/or hypoglycemia (6–8), are important determinants of drug selection and adherence in diabetes management.

Composite outcomes allow for comparison of the effects of different medications or therapeutic strategies on multiple potential outcomes of interest (10). Consideration of multiple end points to concurrently evaluate clinical benefits (glycemic control) along with the treatment‐related risk (weight gain and hypoglycemia) has become a patient‐centered approach in managing type 2 diabetes, and this approach has been incorporated in standards of clinical care for people with diabetes (1,2). In this analysis of GRADE data, we compared the effects of the four randomized medications, added to metformin, on a composite outcome incorporating deterioration of glycemic control, weight gain, and patient-reported hypoglycemia—events that might trigger modification of the treatment regimen in clinical practice. We also compared drug effects on individual components of the composite outcome, examined effects by prespecified subgroups and potential mediators, and tested whether treatment satisfaction varied among treatment groups or between those who did and those who did not reach the composite outcome. As exploratory analyses, we also looked at the effect of the addition of “rescue” insulin to randomized therapies on the remaining components of the composite outcome.

Eligibility criteria, design, randomized interventions, baseline characteristics, and major outcomes of GRADE have previously been described (3–5). Briefly, participants had type 2 diabetes of <10 years duration, were treated only with metformin, and had HbA_1c of 6.8%–8.5% (51–69 mmol/mol) at the end of a run-in period during which metformin dose was maximized to 2,000 mg, as tolerated. Eligible participants (N = 5,047) were randomized to add one of four U.S. Food and Drug Administration–approved medications to metformin (insulin glargine U-100, the sulfonylurea glimepiride, the glucagon-like peptide 1 [GLP-1] receptor agonist liraglutide, or the dipeptidyl peptidase 4 [DPP-4] inhibitor sitagliptin), with protocol-directed titration of medications based (depending on the drug) on glycemic response, tolerability, or kidney function, and followed for 5 years. All participants provided written informed consent. The study was approved by each center’s institutional review board.

Evaluations for study participants were performed quarterly, with HbA_1c measurements performed in the GRADE Central Biochemical Laboratory. At each study visit, hypoglycemic events (severe and nonsevere) during the prior 30 days were reported by participants on a self-administered questionnaire that asked about any episodes of symptomatic hypoglycemia that resolved with glucose or food, as well as about severe hypoglycemia (SH). SH was defined as episodes requiring third-party assistance to treat. Additionally, study staff queried participants at each visit about adverse events since the last study visit, which included SH. Reports of possible SH were described in detail on a specific adjudication form, followed by central review and adjudication by the outcomes subcommittee, masked to treatment. Weight was measured at baseline and at all in-person follow-up visits with standardized procedures. Measurements were collected in duplicate in kilograms to 0.1 kg.

Adherence to study metformin and randomized medication was assessed quarterly by self-report (yes/no). A longitudinal measure of adherence was calculated as the percentage of visits at which participants reported they were taking their study medications. Diabetes treatment satisfaction was assessed at baseline and at 6 months and 1 year postrandomization with the Diabetes Treatment Satisfaction Questionnaire (DTSQ) (11). A summary score is calculated from six of the eight items on the questionnaire that relate to diabetes treatment satisfaction (the remaining two items relate to perceived hypo- and hyperglycemic severity). The range of the score is 0–36, with lower values indicating lower treatment satisfaction.

The main outcome for this analysis is a composite outcome defined as the time to the first occurrence of any of the three individual components: HbA_1c >7.5% (58 mmol/mol) confirmed at next visit (the point at which, per protocol, “rescue” insulin was to be added), ≥5% weight gain from baseline, or hypoglycemia (any adjudicated SH, and/or two or more nonsevere symptomatic events in the 30 days preceding a GRADE visit). The three individual outcomes were then considered separately as component outcomes. We also conducted analyses of treatment effects in prespecified subgroups and evaluated potential mediators of the observed treatment group effects. Exploratory analyses, pertinent given the GRADE design, included the effects of addition of “rescue” insulin on the weight and hypoglycemia components of the composite outcome, and the association of the composite outcome with treatment satisfaction. Rescue insulin included the addition of glargine insulin for participants randomized to glimepiride, liraglutide, or sitagliptin and addition of prandial rapid-acting insulin for those randomized to glargine.

Analyses, other than the exploratory analyses pertaining to rescue insulin addition, were conducted under the intent-to-treat principle with use of all data from all participants by randomized treatment group assignment. Baseline characteristics of the participants were described with the median (1st and 3rd quartiles) for quantitative variables (e.g., HbA_1c) and N (%) for discrete variables (e.g., sex).

For each outcome, the cumulative incidence over time was estimated with the Kaplan-Meier estimator. Crude event rates were calculated as the number of events divided by the total length of follow-up time at risk and reported per 100 individuals at risk for 1 year. Cox proportional hazards (PH) models were used to assess group differences with respect to the risk of an outcome. For each individual outcome, the closed testing procedure was used to control the overall type I error for the six pairwise comparisons (12). A sensitivity analysis for the composite outcome was conducted with use of the Wei-Lachin test, which consists of combining (i.e., averaging) the treatment effects (i.e., the hazard ratios [HRs]) from separate Cox PH models for each of the individual outcomes. The calculation of the SE of the combined (average) effect accounted for the within-subject correlation with use of a sandwich-type estimator of the variance-covariance matrix of the HRs in the three Cox PH models (one for each individual outcome) (13). Prespecified subgroup analyses were conducted for sex (male, female), race (White, Black, other), ethnicity (Hispanic, non-Hispanic), and education (<high school [HS], HS graduate, some college, college degree or higher) and for baseline tertiles of age (<45, 45–59, ≥60 years), HbA_1c (≤7.2%, 7.3–7.7%, ≥7.8% [≤55, 56–61, ≥62 mmol/mol]), BMI (≤30.7, 30.8–36.2, ≥36.3 kg/m²), and diabetes duration (≤2.3, 2.4–5.2, ≥5.3 years). For each variable, to control the overall type I error for subgroup analyses, initially a global homogeneity test was conducted, and pairwise treatment comparisons were conducted with use of the Holm procedure only if the global homogeneity test was significant at level 0.05.

Smoking, alcohol use, adherence to metformin, and adherence to randomized medication were evaluated as potential mediators of observed treatment group differences through comparisons of the treatment group differences (i.e., the HRs) with versus without adjustment for these potential mediators. Smoking status was captured with use of a class variable with three levels: never smoker (defined as <100 cigarettes, cigars, or pipes in lifetime), past smoker (≥100 cigarettes, cigars, or pipes in lifetime but not a current smoker), or current smoker (≥100 cigarettes, cigars, or pipes in lifetime and is a current smoker or smoked in the last 30 days). Alcohol use was defined according to frequency of drinking with four categories: never, occasionally, weekly, or daily.

The rates of the weight gain and hypoglycemia components of the composite outcome were computed separately before and after protocol-advised addition of rescue insulin (addition of insulin glargine for participants randomized to glimepiride, liraglutide, or sitagliptin and addition of mealtime rapid-acting insulin for participants randomized to glargine) for participants who ever initiated rescue insulin (N = 1,577). Relative rates were assessed with Poisson models with robust (sandwich-type) SEs (14). As rescue insulin was added when participants reached the glycemic deterioration outcome (HbA_1c >7.5% [58 mmol/mol], confirmed), this component of the composite outcome was not included, although longitudinal models with generalized estimating equations were used to assess the difference in mean HbA_1c levels before versus after the rescue insulin.

The effects of treatment group assignment on treatment satisfaction were assessed with a linear model of DTSQ at 12 months as a function of treatment group with adjustment for the baseline DTSQ score. Differences in treatment satisfaction between participants who did and did not reach the composite outcome were compared with a linear model of DTSQ at 12 months as a function of an indicator (yes vs. no) of having reached the different clinical outcomes prior to 1 year with adjustment for treatment group and baseline DTSQ score.

This article is based on follow-up data and outcome assessments for the 5,047 participants enrolled into GRADE. This database will be available in the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) Central Repository in 2024.

At baseline, 36% of participants were women, 20% were Black, 65% were White, 18.4% were Hispanic, 7% had less than a HS education, and 43% had a college degree or graduate education. Participant median age was 58 years, median duration of type 2 diabetes was 4 years, median BMI was 33 kg/m², and median HbA_1c was 7.4% (57 mmol/mol) (Supplementary Table 1).

Over a median follow-up time of 5.3 years, 3,529 (71%) of the GRADE participants reached the composite outcome (28 events per 100 participants per treatment year [PTY]). Among these participants, the component outcome that was detected first was HbA_1c >7.5% (58 mmol/mol) in 1,620 (46%), weight gain ≥5% in 1,023 (29%), and hypoglycemia in 886 (25%) (n = 23 with SH and n = 863 with repeated nonsevere hypoglycemia events) participants, respectively.

The percentages of participants and risk rates for the composite outcome differed among the treatment groups. The proportion of participants who reached the composite outcome was 72% in the glargine group, 81% in the glimepiride group, 60% in the liraglutide group, and 70% in the sitagliptin group (Table 1). The cumulative incidence functions of the composite outcome and the individual outcomes are presented separately by treatment group in Fig. 1.

The risk of reaching the composite outcome was lowest for the participants randomized to liraglutide (19 events per 100 PTY), followed by those randomized to sitagliptin (26 events per 100 PTY), glargine (29 events per 100 PTY), and glimepiride (40 events per 100 PTY) (Table 1). In pairwise comparisons, participants in the glargine group had 26% lower risk of the composite outcome in comparison with the participants in the glimepiride group (i.e., HR 0.74) and 49% and 11% higher risk than the participants in the liraglutide group and the sitagliptin group, respectively. The risk for the glimepiride group was 100% and 50% higher compared with that for the liraglutide and the sitagliptin groups, while the risk for participants in the liraglutide group was 25% lower compared with the risk in the sitagliptin group. The Wei-Lachin sensitivity analyses yielded similar results (Table 1 and Fig. 1).

The pattern was somewhat different for the glycemic deterioration component of the composite outcome, with participants in the glargine group having significantly lower risk than participants in the other three groups (27%, 17%, and 39% lower vs. glimepiride, liraglutide, and sitagliptin, respectively). Participants in the liraglutide group had the second-lowest risk: 12% lower than that of the glimepiride group and 26% lower versus the sitagliptin group. Participants in the glimepiride group had the third lowest: 16% lower compared with that in the sitagliptin group (Table 1).

For the weight gain and hypoglycemia components of the composite outcome, results showed effects similar to effects on the composite outcome, except for the lack of significant difference for weight gain between glargine and glimepiride (HR 1.03 [95% CI 0.91, 1.16]) and the lack of significant difference for hypoglycemia between liraglutide and sitagliptin (HR 0.97 [95% CI 0.84, 1.12]) (Table 1 and Fig. 1).

The subgroup analyses did not show evidence of heterogeneity in the risk of reaching the composite outcome across prespecified age, sex, race, ethnicity, and diabetes duration subgroups (Table 2). While the overall homogeneity test was significant (P = 0.04) for BMI, no pairwise treatment comparisons showed significant heterogeneity in the risk of the composite outcome across the BMI tertiles. The overall homogeneity test was significant (P < 0.001) for baseline HbA_1c tertiles, and there was significant heterogeneity in the risk of the composite outcome in comparisons of sitagliptin versus glargine (P < 0.05) and glimepiride (P < 0.001). More specifically, the risk of the composite outcome was lower in the sitagliptin treatment group than in the glargine and glimepiride treatment groups for participants in the lower two HbA_1c tertiles (6.8%–7.2% and 7.3%–7.7% [51–55 and 56–61 mmol/mol]) but not significantly different for participants in highest tertile (7.8%–8.5% [62–69 mmol/mol]).

In comparing the risk of the composite outcome before versus after adjustment for the potential mediators, there was no significant mediation of relative treatment effects by any of the mediators (Supplementary Table 2).

In GRADE, 2,411 participants reached the glycemic deterioration component of the composite outcome (confirmed HbA_1c >7.5% or 58 mmol/mol) and, as dictated by the main study protocol, were to have rescue insulin added to their randomized therapy. Rescue insulin was actually added by 1,577 participants, including 1,549 of the 2,411 who met the glycemic deterioration outcome and 28 who added rescue insulin without having met the glycemic deterioration outcome. These participants form the basis for these exploratory analyses. Of the 1,577 participants who added rescue insulin, 222 (14%) were in the glargine group, 411 (26%) in the glimepiride group, 399 (25%) in the liraglutide group, and 545 (35%) in the sitagliptin group.

The risks of reaching each of the weight gain and hypoglycemia component outcomes following the addition of rescue insulin are shown in Supplementary Table 3. The event rates for reaching the weight gain component of the composite outcome were significantly lower before versus after the addition of rescue insulin in the liraglutide and sitagliptin groups and higher before versus after in the glargine and glimepiride groups. These differences disappeared after the addition of rescue insulin. In contrast, event rates for the hypoglycemia component were significantly lower before versus after addition of rescue insulin in all groups. There were no significant differences between the glargine group and each of the other three groups in the relative risk (RR) of reaching the hypoglycemia outcome after addition of rescue insulin; however, those on glimepiride had significantly increased risk of reaching the hypoglycemia outcome versus liraglutide (RR 1.79 [95% CI 1.23, 2.62]) or sitagliptin (RR 1.79 [95% CI 1.28, 2.52]).

The glycemic deterioration outcome of the composite outcome (i.e., confirmed HbA_1c >7.5% or 58 mmol/mol) had been reached when rescue insulin was added. After the addition of rescue insulin, HbA_1c levels were higher for the participants in the glargine group than for the participants in the other three groups, while HbA_1c levels were higher for participants in the glimepiride group than for the participants in the liraglutide group and those in the sitagliptin group (Supplementary Table 3).

Pairwise treatment group testing showed a small but statistically significant lower mean DTSQ score (less satisfaction) for the glargine group compared with the liraglutide group (difference −0.46 [95% CI −0.86, −0.06], P < 0.05) (Supplementary Table 4). In models with comparison of mean DTSQ scores at 1 year for participants who did and those who did not reach the outcomes (composite outcome and its components), mean DTSQ score was significantly higher (i.e., greater satisfaction) for those who did not reach the composite outcome (0.59; P < 0.001) and those who did not reach the glycemic deterioration outcome (1.12; P < 0.001). Mean DTSQ scores at 1 year did not differ between those who did and did not reach the weight gain status or hypoglycemia component outcomes (Supplementary Table 5).

These secondary analyses of GRADE provide information on a composite of outcomes of importance to people with diabetes: increase in HbA_1c that triggers additional therapy, significant weight gain, and symptomatic hypoglycemia or SH. Over the 5-year mean study follow-up, a majority of participants (∼75%) reached the composite outcome, with contributions from all three components. Our data highlight the challenges associated with reaching and maintaining evidence-based glycemic targets, owing to the progressive nature of type 2 diabetes and to important side effects, such as weight gain and risk of hypoglycemia, with any of four widely used classes of medications for type 2 diabetes. However, the risk for the composite outcome differed among treatment groups: lowest among those participants randomized to liraglutide and increasingly higher for those randomized to sitagliptin, glargine, and glimepiride, respectively. None of the factors considered significantly mediated (explained) the treatment group differences observed. Additionally, there were no significant differences in any of the prespecified subgroup analyses except for observed heterogeneity in comparisons of sitagliptin versus glargine and glimepiride by baseline HbA_1c tertile.

These findings on both the combined outcome and the individual components have relevance in considering how to decide among second-line therapies for people with type 2 diabetes of relatively short duration on metformin monotherapy, suggesting that consideration of glycemic goals be weighed with risks of weight gain and of hypoglycemia. As we previously reported, all four medications, when added to metformin, decreased HbA_1c levels. Glargine and liraglutide were significantly more effective in achieving and maintaining HbA_1c target levels, while liraglutide therapy was associated with the lowest risk of weight gain (4). However, our current analyses show that differences in the risk of weight gain disappear after rescue insulin is added, which highlights that once insulin therapy is initiated, the protection against weight gain is lost even for regimens that include medications with proven weight loss evidence such as liraglutide.

There were also clear differences in the risk of developing severe or recurrent hypoglycemia, with higher risks associated with glimepiride and glargine compared with liraglutide and sitagliptin. These differences persisted after rescue insulin was initiated. Hypoglycemia is a marker for adverse outcomes, including falls, fractures, hospitalizations, arrhythmias, and death (15,16). Hypoglycemia directly impacts adherence to recommended therapies and quality of life and may promote additional weight gain (17,18).

In recent years, guidelines for the treatment of type 2 diabetes have favored adding GLP-1 receptor agonists, rather than insulin, to oral medications in patients not meeting glycemic goals (2). Despite these guidelines, insulin use in type 2 diabetes increased and then plateaued in the past two decades in the U.S. (19), suggesting that insulin remains a common second- or third-line therapy among primary care providers. Our analyses of the effects of adding rescue insulin in GRADE suggest that the addition of insulin uniformly increases rates of weight gain, even when weight-neutral or weight loss–promoting medications are continued. This suggests that if insulin is chosen as third-line therapy, people need education and support for lifestyle efforts to avoid weight gain, regardless of additional therapies. Our findings regarding hypoglycemia suggest discontinuing a sulfonylurea, if used, if basal insulin is initiated. The addition of rapid-acting insulin to glargine led to increased risk of hypoglycemia and relatively little lowering of HbA_1c. This provides further evidence of the adverse risk-to-benefit ratio of “basal plus” therapy in people with type 2 diabetes and modestly elevated HbA_1c and supports present-day treatment guidelines (2).

Treatment satisfaction in GRADE was relatively high and similar among groups, other than a small but statistically significant lower mean score for the glargine group compared with the liraglutide group. Study factors (such as participants receiving medications, supplies, and ongoing diabetes follow-up free of charge) may have contributed to high treatment satisfaction independent of randomized therapy. However, we did find that treatment satisfaction was statistically significantly, though modestly, lower for those who reached the composite outcome, suggesting that this outcome has meaning for participants.

There are several strengths of these analyses. The GRADE study cohort is representative of the contemporary U.S. population with type 2 diabetes of <10 years’ duration, as currently seen in primary care (20). The large sample size, rigorous trial design, prespecified outcomes assessments, long follow-up, and high rates of participant retention allow for robust data analyses, including relevant subgroups and mediators that could be examined. GRADE included long follow-up with detailed assessment of multiple outcomes, including patient-reported outcomes such as treatment satisfaction and mild hypoglycemia. Our composite outcome incorporates characteristics shown to be important to patients and clinicians and is associated with treatment satisfaction.

GRADE and our secondary analyses have several potential limitations. This was a randomized trial with medications and care provided free of charge and including counseling on weight management and hypoglycemia prevention. Hence, our findings may differ from the real-world incidence of the outcomes and comparative effectiveness of these drugs. The GRADE protocol was designed in 2012 and implemented in 2013, prior to sodium–glucose cotransporter 2 inhibitors and weekly GLP-1 receptor agonists coming to market. Hence, the protocol for glucose-lowering therapy in the study diverged somewhat from guidelines that evolved over time (2). The protocol was amended when cardiovascular outcome trials emerged, such that the primary care providers of those with known cardiovascular disease were advised by GRADE physicians to add cardioprotective medications to their therapy. However, the GRADE cohort had a low prevalence at baseline (and incidence during the study) of cardiovascular events (5). Our observations on the effects of adding basal or prandial insulin to other classes of medications, still a common practice (19), support guidelines favoring GLP-1 receptor agonists as the first injectable agent rather than insulin (2).

Our composite outcome, defined a priori, included three “adverse” components, and we assessed time to achieving any of the components. Although there are no standards for diabetes composite outcomes (10,21), our approach differs from some studies with composite outcomes that incorporate several “positive” events and assess the proportion of participants achieving all components. Additionally, all of these components have been shown to be clinically relevant and of importance to individuals with diabetes in prior studies, and any of the components likely would also trigger changes in therapy. Our outcomes and approach mirror the primary GRADE outcomes (time to reaching HbA_1c above certain targets), and each of the components (glycemic deterioration, severe or repeated symptomatic hypoglycemia, significant weight gain) is one that might trigger a change in therapy in clinical practice. Although episodes of SH were rigorously adjudicated, our measures of nonsevere hypoglycemia relied exclusively on participant recollection of symptoms. In the study, blood glucose monitors were provided differentially among groups, with routine provision only for those randomized to glimepiride or glargine (and those in whom rescue insulin was initiated). To avoid biased assessment, we did not include patient reports of asymptomatic low readings on blood glucose meters. Nonsevere hypoglycemia might be better assessed with objective measures such as blinded continuous glucose monitoring.

In summary, these analyses of GRADE demonstrate differential effects of common diabetes drugs, when added to metformin, on a composite outcome incorporating glycemic deterioration, weight gain, and hypoglycemia. We also present evidence for differential effects on each component of the outcome, which may enhance clinicians’ ability to incorporate patient preferences regarding outcomes into shared decision-making. In a subset of participants who added rescue insulin to metformin and randomized study drug, the risk of weight gain was no longer different by treatment group, but similar differential treatment effects persisted for the risk of hypoglycemia. Our results add evidence to support consensus guidelines that advise thinking beyond just glycemic control in considering second- and third-line therapy for patients with type 2 diabetes.

Acknowledgements. The GRADE Research Group is deeply grateful for the participants, whose loyal dedication made GRADE possible.

Funding. GRADE was supported by a grant from the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) of the National Institutes of Health (NIH) 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 NIH 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. W.T.G. has served as a consultant on advisory boards for Boehringer Ingelheim, Eli Lilly, Novo Nordisk, Pfizer, Fractyl Health, Alnylam Pharmaceuticals, Inogen, and Merck and as a site principal investigator for multicentered clinical trials sponsored by his university and funded by Eli Lilly, Novo Nordisk, Epitomee Medical, and Pfizer. R.P.-B. received grant support to the University of Michigan from Novo Nordisk, Medtronic, and Dexcom and served as a consultant on advisory boards for Averitas Pharma, Boehringer Ingelheim, Lexicon, Nevro, Novo Nordisk, Procter & Gamble, and Roche. M.S. has received consulting fees from Eli Lilly and Neurocrine Biosciences and receives grant funding to his institution from Neurocrine Biosciences. 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 (ICMJE) authorship criteria. All named authors were involved in the conception, design, and conduct of the study and interpretation of the results. M.T., H.K.-S., and I.B. carried out the analyses. M.S.K. and R.P.-B. wrote the first draft of the manuscript, and all authors edited, reviewed, and approved the final version of the manuscript. I.B. 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.

M.S.K. and R.P.-B. are editors of Diabetes Care but were not involved in any of the decisions regarding review of the manuscript or its acceptance.