OBJECTIVE

To describe the individual and joint associations of baseline factors with glycemia, and also with differential effectiveness of medications added to metformin.

RESEARCH DESIGN AND METHODS

Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE) participants (with type 2 diabetes diagnosed for <10 years, on metformin, and with HbA1c 6.8–8.5%; N = 5,047) were randomly assigned to a basal insulin (glargine), sulfonylurea (glimepiride), glucagon-like peptide 1 agonist (liraglutide), or dipeptidyl peptidase 4 inhibitor (sitagliptin). The glycemic outcome was HbA1c ≥7.0%, subsequently confirmed. Univariate and multivariate regression and classification and regression tree (CART) analyses were used to assess the association of baseline factors with the glycemic outcome at years 1 and 4.

RESULTS

In univariate analyses at baseline, younger age (<58 years), Hispanic ethnicity, higher HbA1c, fasting glucose, and triglyceride levels, lower insulin secretion, and relatively greater insulin resistance were associated with the glycemic outcome at years 1 and/or 4. No factors were associated with differential effectiveness of the medications by year 4. In multivariate analyses, treatment group, younger age, and higher baseline HbA1c and fasting glucose were jointly associated with the glycemic outcome by year 4. The superiority of glargine and liraglutide at year 4 persisted after multiple baseline factors were controlled for. CART analyses indicated that failure to maintain HbA1c <7% by year 4 was more likely for younger participants and those with baseline HbA1c ≥7.4%.

CONCLUSIONS

Several baseline factors were associated with the glycemic outcome but not with differential effectiveness of the four medications. Failure to maintain HbA1c <7% was largely driven by younger age and higher HbA1c at baseline. Factors that predict earlier glycemic deterioration could help in targeting patients for more aggressive management.

Type 2 diabetes (T2D) is a chronic disease characterized by progressive worsening of glycemia and complications that confer morbidity and mortality (1). To address deterioration in glycemic regulation, there is a predictable need to escalate and intensify therapy by using different classes of glucose-lowering drugs in combination. While metformin has been recommended as a first-line medication (2,3), there remains no clear consensus on the optimal add-on medication to achieve acceptable glycemic control.

Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE) tested the durability of glycemic control of four glucose-lowering agents added to metformin in a randomized comparative effectiveness clinical trial (4). GRADE compared the effectiveness of a basal insulin (glargine U-100), sulfonylurea (glimepiride), glucagon-like peptide 1 (GLP-1) receptor agonist (liraglutide), and dipeptidyl peptidase 4 inhibitor (sitagliptin) in maintaining HbA1c at target values over time. Participants exhibited a high degree of individual variability in the ability of these medications to maintain glycemia at target levels (5). Identification of baseline characteristics associated with differences in glycemic responses to the four medications is necessary to derive a strategy for optimal selection of second-line agents. These data could provide a rationale for more personalized and effective approaches to treating patients with T2D.

A broad range of baseline factors were assessed at the time of enrollment into GRADE, including demographics, HbA1c and diabetes duration, socioeconomic factors, cognitive ability, metabolic syndrome traits, and indices of insulin sensitivity and secretion (46). A limited number of baseline variables were previously assessed in GRADE, with higher baseline HbA1c values previously found to be associated with a shorter time to glycemic failure (5). We have now conducted a comprehensive study of many baseline variables to identify factors, individually and jointly, associated with glycemic failure by years 1 and 4 of study follow-up. To identify a priori predictors of efficacy, we assessed each baseline characteristic separately against a glycemic outcome of HbA1c ≥7.0% (53 mmol/mol) with all participants combined regardless of treatment modality. In addition, similar analyses were performed within each randomized treatment group for assessment of whether associations with glycemic outcomes differed among treatment groups.

GRADE was launched in 2013 as a parallel-group, randomized comparative effectiveness trial across 36 clinical centers in the U.S. to compare the effectiveness of four glucose-lowering medications added to metformin in people with T2D (4). Eligible participants had T2D diagnosed for <10 years, had HbA1c 6.8–8.5% (51–69 mmol/mol) at the time of randomization, and were taking at least 1,000 mg/day metformin, up to a maximum of 2,000 mg daily, as tolerated. Eligible participants were randomly assigned to one of four medications (glargine, glimepiride, liraglutide, or sitagliptin) and the 5,047 participants were followed for an average of 5 years (6). Written informed consent was provided by all participants at the screening visit and again prior to randomization.

Baseline Factors

The following categories of baseline factors were studied: demographic and socioeconomic factors, clinical measures, psychosocial factors, cognitive ability, and metabolic variables (4,6).

Demographic and socioeconomic factors were assessed by self-report and included sex (male, female), race (Black, White, others), ethnicity (Hispanic, non-Hispanic), education (less than high school, high school/GED, some college, college graduate and above), income (USD <10,000, 10,000–15,000, 15,000–20,000, 20,000–25,000, 35,000–50,000, 50,000–75,000, ≥75,000), and marital status (married/living together, other) (6). Age was grouped as <45, 45–59, and ≥60 years and divided at the median and denoted “younger” and “older” as <58 and as ≥58 years, respectively.

Clinical measures were assessed at run-in or at baseline by centrally trained clinical research staff (6) and included waist-to-hip ratio, systolic and diastolic blood pressure measurements (mmHg), BMI (weight in kilograms divided by the square of height in meters), distal symmetrical polyneuropathy (DSPN) assessed with the Michigan Neuropathy Screening Instrument, and hypertension (blood pressure ≥130/80 mmHg or treatment with blood pressure–lowering agents). Diabetes duration (years), alcohol use (never, occasionally, weekly, daily), smoking status (never, past, current), and depression (history of diagnosis or treatment) were assessed by self-report.

All laboratory tests were measured by the Central Biochemistry Laboratory (Department of Laboratory Medicine and Pathology, University of Minnesota) (5,6). The insulinogenic index (IGI) was used as a measure of insulin secretion (7) and the Matsuda index as a measure of whole body insulin sensitivity, the reciprocal of which is a measure of insulin resistance (8). The Chronic Kidney Disease Epidemiology Collaboration creatinine equation was used for estimated glomerular filtration rate (eGFR) calculations (9). Baseline HbA1c, measured in EDTA whole blood on the Tosoh HPLC Glycohemoglobin Analyzer and standardized per NGSP protocol, was assessed at the final run-in visit (10).

Cognitive ability (11) was assessed with the Spanish-English Verbal Learning Test (SEVLT), Digit Symbol Substitution Test (DSST), letter fluency test (Word Fluency Test [WFT] Letter), and animal fluency test (WFT Animal).

Glycemic Outcome

The primary glycemic outcome was HbA1c ≥7.0% (53 mmol/mol), confirmed at the subsequent follow-up visit (4,5), and was assessed separately at 1 and 4 years following randomization.

Statistical Analyses

Logistic regression analyses were used to assess the univariate association (odds ratio [OR]) of each baseline factor separately with the glycemic outcome at years 1 and 4 in the full cohort. The proportions reaching each outcome within categories of a given factor were estimated from the models. For each quantitative factor, a model-free (regression spline) analysis was used to assess the risk gradient for that factor with each of the outcomes; for ease of interpretation, associations with quantitative factors are described with use of categories (e.g., tertiles) of the distribution of that factor.

We also addressed whether there were interactions involving the four treatment groups with each baseline factor, with significant interactions indicating that the association of the factor with the outcome differed among treatment groups. Benjamini-Hochberg false discovery rate (FDR)–adjusted P values (12) with a 1% FDR were used for evaluation of differential univariate associations with the glycemic outcome as a function of treatment group. For each factor demonstrating a significant interaction with treatment group, we further compared the four groups within each of the categories (strata) of the factor, and in the case of nominal significance, the six pairwise group comparisons were tested with the Holm procedure (13) applied to adjust for multiple comparisons.

Multivariate logistic models of the outcome at years 1 and 4 were used including the subset of baseline factors that were significant in the univariate analyses. Factors with significantly differential associations were included in the multivariate model with an interaction term with treatment group and a main effect for the factor. If the univariate analyses showed a nominally significant marginal association of a baseline factor with the outcome at the 0.01 significance level, then the multivariate model for that outcome only included a main effect for the factor.

In addition, for the outcomes at 1 and 4 years, classification and regression tree (CART) models involving all baseline factors and the four treatment groups empirically generated the main drivers of glycemic failure with use of simple and limited sets of variables (14). The tree was used to classify participants into specific risk categories that optimally predicted the outcome, operationalized as maximization of Gini diversity index (15). The CART model identified factors, including interaction terms, that were jointly predictive of the given outcome; for each quantitative factor, optimal cut points were determined. For CART, the predictor variable interactions and categorizations included in the model are identified empirically from the data, instead of prespecified by the analyst as for a traditional multivariate regression analysis. The classification tree was “pruned” with cross validation to prevent overfitting the model to the data, and the resultant tree was used to predict the outcome for each risk category.

Data and Resource Availability

This article is based on data collected at baseline and follow-up 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.

The baseline characteristics for the 5,047 participants in these analyses are presented overall and by individual treatment groups in Supplementary Table 1, and some have previously been published (5,6,11,16). There were no statistically significant differences among the treatment groups except that the IGI was slightly higher and the Matsuda index slightly lower in the group randomized to glargine.

Univariate Analyses: Factors Associated With the Glycemic Outcome in the Overall Cohort

The following baseline factors were associated with the glycemic outcome (confirmed HbA1c ≥7.0% [53 mmol/mol]) in univariate analyses at years 1 and 4: treatment group, younger age, age-race jointly, high HbA1c and fasting glucose, absence of hypertension, high eGFR, never smoking, and lower insulin secretion (as assessed by IGI) (Table 1 and Supplementary Fig. 1). As previously reported, assigned treatment had a significant effect on glycemic outcomes. Sitagliptin performed more poorly than the other medications with respect to the outcome at year 1, and liraglutide and glargine performed better than sitagliptin and glimepiride at year 4 (Table 1). Significant associations only at year 1 included Hispanic ethnicity, high triglycerides, and whole-body insulin sensitivity (as assessed with Matsuda index); at year 4, there was an association with absence of depression (Table 1).

Supplementary Fig. 1 also illustrates the probabilities of the glycemic outcome at years 1 and 4 as a continuous function for the designated quantitative baseline factors in Table 1. (These quantitative factors were analyzed by tertiles in Table 1.) Associations involving quantitative factors are shown as smoothed (model-free) functions of the outcome probability over the approximate range of the factor, and for most factors the association is approximately linear. In general, the numerical probabilities, directionality, and linearity of relationships between the baseline factors and the glycemic outcome at year 1 were similar to those at year 4 (Supplementary Fig. 1).

Univariate Analyses: Factors Differentially Associated With Glycemic Response by Treatment Group

Five baseline characteristics were selectively associated at year 1 with response to therapy in one or more of the four treatment groups; i.e., there was a factor–by–treatment group interaction (Supplementary Fig. 1). No baseline characteristics were selectively associated with glycemic failure within individual treatment groups at year 4. These interactions at year 1 included treatment group with sex, baseline HbA1c, fasting glucose, insulin sensitivity (Matsuda index), and the joint categories of age-race. For baseline HbA1c, fasting glucose, and Matsuda Index, a model-free figure shows the probability of failure among groups over the range of the factor (Supplementary Fig. 1). The probability of glycemic failure increased progressively in all treatment groups with increases in baseline HbA1c and fasting glucose. As the Matsuda index values increased (i.e., greater insulin sensitivity), there was a tendency toward lower probabilities of failure for all four medications (Supplementary Fig. 1). Compared with the other medications, however, liraglutide was somewhat more effective at lower values on the Matsuda index (indicating an individual is relatively more insulin resistant), while the performance of all four medications was similar at higher values. Interactions with treatment group only remained significant for sex and age-race in a multivariate model (described below).

Multivariate Analyses Involving Baseline Factors and Treatment Group

For assessment of the joint effects of baseline factors on glycemic outcomes, multivariate logistic regression models were developed separately for the glycemic outcome at year 1 (Table 2) and year 4 (Table 3).

In the multivariate model at year 1, treatment group and interactions of treatment with sex and treatment with age-race remained significant (Table 2), with no other significant associations between factors and the year 1 outcome. At year 1, within both sexes and most age-race strata the probability of the glycemic outcome for those on sitagliptin was higher than for those on glargine, glimepiride, or liraglutide, except among older participants in the Black and others race categories (Table 2). In the glimepiride group, there was lower risk for male versus female participants, with an OR of 0.74, whereas the risk was higher among male participants in the liraglutide group (OR 1.63). The OR for the glycemic outcome was greater for younger Black versus younger White participants for glargine (OR 2.44) and was greater for older Black versus older White participants for glimepiride (OR 2.68).

For the glycemic outcome at year 4 (Table 3), treatment group, younger age, and higher HbA1c and fasting glucose were the only statistically significant factors in the multivariate model. Participants continued to have lower likelihood of the outcome on glargine and liraglutide despite the adjustment for covariates.

CART Analysis

The regression analyses describe numerical associations of covariates with the risk of an outcome. In complimentary analyses with CART, the cohort was stratified based on patient characteristics according to the risk of the glycemic outcome. The only baseline factors identified as jointly affecting probability of the glycemic outcome were treatment group and HbA1c at year 1 and age and HbA1c at year 4 (Fig. 1). Among the 4,982 evaluated at year 1, the probability (risk) for the outcome was decreased (probability 0.23) when baseline HbA1c was <7.5% (58 mmol/mol) (n = 2,631 [53% of the cohort]). When baseline HbA1c was ≥7.5% (58 mmol/mol) the risk differed by treatment group. The risk among the 594 (12%) in the sitagliptin group was increased (0.64), whereas that among the 1,757 (35%) in the other groups was intermediate (0.42). Among the 4,561 at year 4, the probability of the outcome was greater for baseline HbA1c ≥7.4% (57 mmol/mol) (probability 0.73; n = 2,429 [53%]) and for younger participants (age <58 years) when baseline HbA1c was <7.4% (57 mmol/mol) (probability 0.62; n = 990 [22%]). Among older participants (age ≥58 years) with HbA1c <7.4% (57 mmol/mol), the risk was intermediate (probability 0.45; n = 1,142 [25%]).

The common failure of initial therapy with metformin to sustain glycemic targets (17) and the need for add-on medications present clinicians with three critical questions. First, which add-on medication(s) is most likely to maintain target glycemia? Second, are there factors that portend more rapid deterioration in glycemic control that could be used to target individual patients for more aggressive therapy? Finally, can baseline factors be used to identify the optimal medication as an add-on to metformin for achieving and maintaining HbA1c at target and for the longest duration?

GRADE previously answered the first question (5), and we address the other two issues herein. In the entire cohort, treatment group, younger age, Hispanic ethnicity, high values of fasting glucose and HbA1c, the absence of hypertension, high triglycerides, high eGFR, never smoking, absence of depression, lower insulin secretion, and relative insulin resistance were all key factors associated with HbA1c ≥7% at years 1 and/or 4 in univariate analyses (results with triglyceride–to–HDL cholesterol ratio were similar). Notably, several other factors were not associated with the glycemic outcome, including BMI, waist-to-hip ratio, diabetes duration, blood pressure measurements, neuropathy (DSPN), albumin-to-creatinine ratio (ACR) >30 mg/g, socioeconomic factors, alcohol use, and measures of cognition.

A limited number of baseline characteristics were selectively associated with differential response to assigned medication but only at year 1. Liraglutide performed better than other medications in women and younger White participants, while sitagliptin was associated with higher risk of HbA1c ≥7% in White but not in Black participants. The probability of glycemic failure at year 1 rose progressively with increases in baseline HbA1c in all four treatment groups. However, participants treated with sitagliptin were most likely to experience the glycemic outcome as a function of rising baseline HbA1c with no differences among glargine, glimepiride, and liraglutide. The probabilities of treatment failure increased for all four medications with progressive worsening of baseline insulin resistance, with liraglutide relatively more effective (i.e., lower probability of the outcome) than the other medications at 1 year in participants who were more insulin resistant. Although previous reports have linked markers of β-cell failure with early poor glycemic response to GLP-1 receptor agonists (1820), this finding was not confirmed in the current study with longer-term follow-up (P = NS [data not shown]). Importantly, no factors were associated with differential effectiveness of the medications at year 4.

Multivariate regression analyses were performed to identify which baseline factors were independently associated with glycemic outcomes. At year 1, the only significant factors in the multivariate model were treatment group, the interaction of treatment and sex (the glycemic outcome on glimepiride was less likely in male compared with female participants and less likely in female participants on liraglutide), and the interaction of treatment and age-race (outcome was more likely to occur in younger Black participants on glargine and in older Black participants on glimepiride compared with their White counterparts). At year 4, the multivariate model revealed only treatment group, younger age, and higher HbA1c and fasting glucose as significantly associated independent variables. Thus, multiple factors found to be associated with glycemic outcomes in univariate analyses were likely related to other predictors of the glycemic outcomes and so did not meet criteria for statistical significance in the multivariate models.

Differences among the four add-on medications in preventing the glycemic outcome were similar in both univariate and multivariate analyses. At year 1, glycemic failure was more likely to occur with sitagliptin (with no differences among glargine, glimepiride, and liraglutide), and at year 4 both liraglutide and glargine were modestly more effective than sitagliptin and glimepiride. This parallels the previous report of GRADE findings (5) and confirms that these differences in treatment efficacy persist even in multivariate models that control for multiple baseline factors. When we examined longer-term changes, we did not see persistence of the effects, particularly for heterogeneity in interactions with treatment effects, as seen at 1 year. The short-term treatment effect differed by sex, age, and race, but the long-term treatment effects were generally homogenous across different demographic characteristics.

The CART models empirically demonstrated how baseline factors could be used to stratify the cohort according to different risks of treatment failure. Only younger age, higher HbA1c, and treatment group were found to affect the risk of the glycemic outcome. Treatment group was only identified by the CART model in year 1; when the baseline HbA1c was ≥7.5% (58 mmol/mol), sitagliptin performed worse than the other medications. At year 4, only HbA1c and age interacted to predict probability of treatment failure; the outcome was more likely when baseline HbA1c was ≥7.4% (57 mmol/mol) and for younger participants when the HbA1c was <7.4% (57 mmol/mol).

The results pertaining to HbA1c are consistent with previous data showing that higher HbA1c values increase the probability of a suboptimal response to therapy (5,21). Hispanic patients have been characterized as being more likely to have poor glycemic control, which has been attributed to factors such as barriers to health care, poor adherence to prescribed medications, health illiteracy, language difficulties, and lack of health insurance (2224). In GRADE, Hispanic ethnicity was associated with an increased probability of poor glycemic outcomes, though this could not be attributed to education level, marital status, cognitive ability, differences in baseline HbA1c, or health insurance, as all diabetes care was provided free of charge.

Other observations would not be expected or are somewhat novel. The finding that baseline cognitive ability did not influence glycemic outcomes was surprising, as poor cognitive function was associated with higher HbA1c values in the Memory in Diabetes (MIND) substudy of the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial (ACCORD-MIND) (25). This discordance may be explained by the younger mean age of the GRADE participant cohort (57 years) compared with ACCORD (62 years), a shorter duration of diabetes (<10 years), and that GRADE participants were largely cognitively intact. A novel observation is that younger participants were more likely to experience treatment failure than older participants. A cross-sectional analysis of data from the 2005–2010 National Health and Nutrition Examination Survey (NHANES) showed that a higher proportion of adults age <65 years at the time of diabetes diagnosis had higher HbA1c values than older patients (≥65 years), even after adjustment for multiple factors (26). Patients with younger-onset diabetes are more likely to be obese, Hispanic, or non-Hispanic Black and have a longer duration of diabetes (26). Moreover, earlier onset of T2D in adults has been associated with higher genetic risk scores and unique polymorphisms compared with later-onset disease (2730). Finally, adolescents and younger adults diagnosed with T2D may have a higher degree of insulin resistance and more rapidly increasing glucose levels (31,32). Thus, previous reports of worse glycemic control in younger patients with T2D are consistent with the current finding that these patients are also more likely to experience early treatment failure. These observations suggest that more aggressive treatment is justified in these patients to achieve and maintain better glycemic control and reduce the risk of future complications.

While only age and baseline HbA1c jointly predicted the glycemic outcome at year 4 in multivariate and CART analyses, the univariate associations of individual factors with glycemic outcomes are still relevant in identifying those patients at greater risk of glycemic failure. The association of these factors with early treatment failure could justify more aggressive treatment following diagnosis. For example, patients with one or more risk factors for early treatment failure could be targeted for initial combination therapy (33,34), weight loss therapy (35), or other more intensive approaches. Such an approach to risk stratification and individualized prescriptions for more aggressive interventions is deserving of additional research with assessment of long-term outcomes.

We also addressed whether baseline factors could be used to select an optimal add-on medication to metformin. Though numerous factors related to poor glycemic outcomes in the overall cohort, only a limited number of factors were found to predict differential response for any one medication as compared with the others, and only for the glycemic outcome (HbA1c ≥7.0%) at year 1. By year 4, there were no significant associations of any baseline factor for prediction of differential responses to any of the medications. Thus, baseline factors may be of limited value in identifying individuals who would respond more effectively to a specific medication.

The study has several limitations. Several baseline factors (e.g., depression) were assessed only by self-report. Cognitive function was assessed with validated instruments, but the sample population on the whole was relatively young and cognitively intact. Given the numerous baseline factors, there were often high degrees of colinearity. For example, univariate analysis produced an unexpected association between high eGFR and glycemic outcomes, which was likely due to a confounding relationship with HbA1c in our sample, and which was lost in multivariate analyses. Likewise, the demonstrated associations of the absence of hypertension and the absence of smoking (i.e., status “never smoked”) with the glycemic outcome in univariate analysis strike us as anomalous but may be the result of confounding in our sample. The definition of baseline hypertension in this analysis included those with a preestablished diagnosis, on antihypertensive medications, or with elevated recorded blood pressure on the day of the visit. Those with no clinical diagnosis of hypertension who were on angtensin-converting enzyme (ACE) inhibitors or angiotensin receptor blockers for “renal protection,” for example, may have yielded a misclassification. GRADE did not include thiazolidinediones, sodium–glucose cotransporter 2 inhibitors, or newer, more potent GLP-1, GLP-1/GIP, or GLP-1/GIP/glucagon receptor agonists as treatments, and therefore we cannot assess the interaction of baseline factors with these more recent agents. A comparison of the GRADE study population with the general population as represented in NHANES has previously been reported (6). However, underrepresentation of lower socioeconomic groups and lower educational attainment (Supplementary Table 1) than in the general U.S. population may have limited the power to detect relationships with certain baseline factors. On the other hand, GRADE was well designed to identify factors predictive of glycemic outcomes. The study enrolled a large, diverse cohort followed long-term and with minimal dropout, and participants were randomized to commonly prescribed medications and managed in a manner comparable with real-world care.

In summary, younger age, self-identified Hispanic ethnicity, high values of fasting glucose and HbA1c, high triglycerides, poor insulin secretion, and relatively greater insulin resistance at baseline were associated with treatment failure in univariate analyses and could justify more intensive monitoring and treatment in such patients. These baseline factors, however, were not associated with differential effectiveness of the glucose-lowering medications at year 4 and cannot be used in a personalized approach to select medications providing longer-term glycemic control. CART and multivariate analyses confirmed more likely treatment failure at year 4 regardless of treatment group in participants who were younger and with higher HbA1c at baseline. Such patients with diabetes should be considered for more aggressive treatment early in the course of the disease.

Clinical trial reg. no. NCT01794143, clinicaltrials.gov

This article contains supplementary material online at https://doi.org/10.2337/figshare.24543754.

*

A complete list of members of the GRADE Research Group can be found in the supplementary material online.

This article is featured in podcasts available at diabetesjournals.org/care/pages/diabetes_care_on_air.

Funding. GRADE was supported by a grant from the 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 DK020541-44, P30 DK020572, P30 DK072476, P30 DK079626, P30 DK092926, U54 GM104940, UL1 TR000439, UL1 TR000445, UL1 TR001108, UL1 TR001409, UL1 TR001449, UL1 TR002243, UL1 TR002345, UL1 TR002378, UL1 TR002489, UL1 TR002529, UL1 TR002535, UL1 TR002537, and UL1 TR002548. Educational materials were provided by the National Diabetes Education Program. Material support in the form of donated medications and supplies were provided by Becton, Dickinson and Company, Bristol-Myers Squibb, Merck, 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. reports serving 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, Neurovalens, and Pfizer outside the submitted work. E.S. reports serving on advisory boards for Eli Lilly, Novo Nordisk, and Zucara Therapeutics and as a site principal investigator for multicentered clinical trials sponsored by her university and funded by Eli Lilly outside the submitted work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. All authors affirmed that authorship was merited based on the International Committee of Medical Journal Editors (ICMJE) authorship criteria. W.T.G., R.M.C., and J.M.L. contributed to the conception and/or design of the research. W.T.G., R.M.C., A.A., J.K., C.L.M., E.S., M.W.S., and J.M.L. contributed to acquisition of data. N.M.B., E.J.K., N.Y., S.P.R., and J.M.L. contributed to statistical analysis of data. W.T.G., R.M.C., N.M.B., E.J.K., N.Y., C.E.S., P.A.H., J.K., C.L.M., E.S., M.W.S., and J.M.L. contributed to interpretation of data and results. W.T.G. and J.M.L. contributed to acquisition of funding. C.E.S., A.A., J.K., E.S., and J.M.L. contributed to the supervision and management of the research. W.T.G., R.M.C., N.M.B., E.J.K., and J.M.L. contributed to the drafting of the manuscript. W.T.G., R.M.C., N.M.B., E.J.K., N.Y., S.P.R., C.E.S., A.A., P.A.H., J.K., C.L.M., E.S., M.W.S., and J.M.L. contributed to the critical review and revision of the manuscript. W.T.G. and N.Y. 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.

1.
Garvey
WT, Robertson RP.
Type 2 diabetes mellitus: etiology, pathogenesis, and natural history. In
DeGroot’s Endocrinology.
8th ed.
Robertson
RP
, Ed.
Elsevier
,
2023
, pp. 557–588
2.
ElSayed
NA
,
Aleppo
G
,
Aroda
VR
, et al;
American Diabetes Association
.
9. Pharmacologic approaches to glycemic treatment: Standards of Care in Diabetes—2023
.
Diabetes Care
2023
;
46
(
Suppl. 1
):
S140
S157
3.
Blonde
L
,
Umpierrez
GE
,
Reddy
SS
, et al
.
American Association of Clinical Endocrinology clinical practice guideline: developing a diabetes mellitus comprehensive care plan-2022 update
.
Endocr Pract
2022
;
28
:
923
1049
4.
Nathan
DM
,
Buse
JB
,
Kahn
SE
, et al;
GRADE Study Research Group
.
Rationale and design of the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE)
.
Diabetes Care
2013
;
36
:
2254
2261
5.
Nathan DM, Lachin JM, Balasubramanyam A, et al.; GRADE Study Research Group. Glycemia reduction in type 2 diabetes - glycemic outcomes
.
N Engl J Med
2022
;
387
:
1063
1074
6.
Wexler
DJ
,
Krause-Steinrauf
H
,
Crandall
JP
, et al;
GRADE Research Group
.
Baseline characteristics of randomized participants in the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE)
.
Diabetes Care
2019
;
42
:
2098
2107
7.
Seltzer
HS
,
Allen
EW
,
Herron
AL
Jr
,
Brennan
MT.
.
Insulin secretion in response to glycemic stimulus: relation of delayed initial release to carbohydrate intolerance in mild diabetes mellitus
.
J Clin Invest
1967
;
46
:
323
335
8.
Matsuda
M
,
DeFronzo
RA.
.
Insulin sensitivity indices obtained from oral glucose tolerance testing: comparison with the euglycemic insulin clamp
.
Diabetes Care
1999
;
22
:
1462
1470
9.
Levey
AS
,
Stevens
LA.
.
Estimating GFR using the CKD Epidemiology Collaboration (CKD-EPI) creatinine equation: more accurate GFR estimates, lower CKD prevalence estimates, and better risk predictions
.
Am J Kidney Dis
2010
;
55
:
622
627
10.
Nathan
DM
,
Krause-Steinrauf
H
,
Braffett
BH
, et al;
GRADE Research Group
;
DCCT/EDIC Research Group
.
Comparison of central laboratory HbA1c measurements obtained from a capillary collection versus a standard venous whole blood collection in the GRADE and EDIC studies
.
PLoS One
2021
;
16
:
e0257154
11.
Luchsinger
JA
,
Younes
N
,
Manly
JJ
, et al;
GRADE Research Group
;
GRADE Research Group Investigators
.
Association of glycemia, lipids, and blood pressure with cognitive performance in people with type 2 diabetes in the Glycemia Reduction Approaches in Diabetes: A Comparative Effectiveness Study (GRADE)
.
Diabetes Care
2021
;
44
:
2286
2292
12.
Benjamini
Y
,
Hochberg
Y.
.
Controlling the false discovery rate: a practical and powerful approach tomultiple testing
. J R Stat Soc Series B StatMethodol
1995
;
57
:
289
300
13.
Holm
S.
.
A simple
sequentially rejective multiple
test procedure
.
Scand J Stat
1979
;
6
:
65
70
14.
Breiman
L.
Classification and Regression Trees.
Belmont, CA
,
Wadsworth International Group
,
1984
15.
Therneau
TM
,
Atkinson
EJ.
An Introduction to Recursive Partitioning Using the RPART Routines.
Mayo Foundation,
1997
16.
Nathan DM, Lachin JM, Bebu I, et al.; GRADE Study Research Group. Glycemia reduction in type 2 diabetes - microvascular and cardiovascular outcomes
.
N Engl J Med
2022
;
387
:
1075
1088
17.
Brown
JB
,
Conner
C
,
Nichols
GA.
.
Secondary failure of metformin monotherapy in clinical practice
.
Diabetes Care
2010
;
33
:
501
506
18.
Jones
AG
,
McDonald
TJ
,
Shields
BM
, et al;
PRIBA Study Group
.
Markers of β-cell failure predict poor glycemic response to GLP-1 receptor agonist therapy in type 2 diabetes
.
Diabetes Care
2016
;
39
:
250
257
19.
Kondo
Y
,
Satoh
S
,
Nagakura
J
,
Kimura
M
,
Nezu
U
,
Terauchi
Y.
.
Defining criteria for the introduction of liraglutide using the glucagon stimulation test in patients with type 2 diabetes
.
J Diabetes Investig
2013
;
4
:
571
575
20.
Kozawa
J
,
Inoue
K
,
Iwamoto
R
, et al
.
Liraglutide is effective in type 2 diabetic patients with sustained endogenous insulin-secreting capacity
.
J Diabetes Investig
2012
;
3
:
294
297
21.
Jones
AG
,
Lonergan
M
,
Henley
WE
,
Pearson
ER
,
Hattersley
AT
,
Shields
BM.
.
Should studies of diabetes treatment stratification correct for baseline HbA1c
?
PLoS One
2016
;
11
:
e0152428
22.
Banuelos Mota
A
,
Feliz Sala
EE
,
Perdomo
JM
, et al
.
Assessing barriers to medication adherence among Latinos with diabetes: a cross-sectional study
.
J Gen Intern Med
2020
;
35
:
603
605
23.
Fernández
A
,
Quan
J
,
Moffet
H
,
Parker
MM
,
Schillinger
D
,
Karter
AJ.
.
Adherence to newly prescribed diabetes medications among insured Latino and white patients with diabetes
.
JAMA Intern Med
2017
;
177
:
371
379
24.
Fortmann
AL
,
Gallo
LC
,
Philis-Tsimikas
A.
.
Glycemic control among Latinos with type 2 diabetes: the role of social-environmental support resources
.
Health Psychol
2011
;
30
:
251
258
25.
Cukierman-Yaffe
T
,
Gerstein
HC
,
Williamson
JD
, et al;
Action to Control Cardiovascular Risk in Diabetes-Memory in Diabetes (ACCORD-MIND) Investigators
.
Relationship between baseline glycemic control and cognitive function in individuals with type 2 diabetes and other cardiovascular risk factors: the Action to Control Cardiovascular Risk in Diabetes-Memory in Diabetes (ACCORD-MIND) trial
.
Diabetes Care
2009
;
32
:
221
226
26.
Berkowitz
SA
,
Meigs
JB
,
Wexler
DJ.
.
Age at type 2 diabetes onset and glycaemic control: results from the National Health and Nutrition Examination Survey (NHANES) 2005-2010
.
Diabetologia
2013
;
56
:
2593
2600
27.
Mars
N
,
Koskela
JT
,
Ripatti
P
, et al;
FinnGen
.
Polygenic and clinical risk scores and their impact on age at onset and prediction of cardiometabolic diseases and common cancers
.
Nat Med
2020
;
26
:
549
557
28.
Piko
P
,
Werissa
NA
,
Fiatal
S
,
Sandor
J
,
Adany
R.
.
Impact of genetic factors on the age of onset for type 2 diabetes mellitus in addition to the conventional risk factors
.
J Pers Med
2020
;
11
:
6
29.
Sale
MM
,
Lu
L
,
Spruill
IJ
, et al
.
Genome-wide linkage scan in Gullah-speaking African American families with type 2 diabetes: the Sea Islands Genetic African American Registry (Project SuGAR)
.
Diabetes
2009
;
58
:
260
267
30.
Haupt
A
,
Staiger
H
,
Schäfer
SA
, et al
.
The risk allele load accelerates the age-dependent decline in beta cell function
.
Diabetologia
2009
;
52
:
457
462
31.
D’Adamo
E
,
Caprio
S.
.
Type 2 diabetes in youth: epidemiology and pathophysiology
.
Diabetes Care
2011
;
34
(
Suppl. 2
):
S161
S165
32.
Gunasekaran
U
,
Gannon
M.
.
Type 2 diabetes and the aging pancreatic beta cell
.
Aging (Albany NY)
2011
;
3
:
565
575
33.
DeFronzo
RA
,
Eldor
R
,
Abdul-Ghani
M.
.
Pathophysiologic approach to therapy in patients with newly diagnosed type 2 diabetes
.
Diabetes Care
2013
;
36
(
Suppl. 2
):
S127
S138
34.
Phillips
LS
,
Ratner
RE
,
Buse
JB
,
Kahn
SE.
.
We can change the natural history of type 2 diabetes
.
Diabetes Care
2014
;
37
:
2668
2676
35.
Garvey
WT.
.
New horizons. A new paradigm for treating to target with second-generation obesity medications
.
J Clin Endocrinol Metab
2022
;
107
:
e1339
e1347
Readers may use this article as long as the work is properly cited, the use is educational and not for profit, and the work is not altered. More information is available at https://www.diabetesjournals.org/journals/pages/license.