Individualizing Treatment of Type 2 Diabetes After Metformin: More Insights From GRADE

The GRADE investigators published two reports of the main results in 2022 (3,4). The mean duration of observation was 5.0 years, and retention in the study was over 90%. HbA_1c values declined initially with all treatments but then rose steadily; 71% of participants reached 7.0% or higher by the end of observation. Maintenance of glycemic control was better with glargine and liraglutide than with the two oral agents, but the differences were small. In comparison with all other treatments, liraglutide was associated with a 16% lower and glargine a 13% lower risk of HbA_1c rising to 7% or above. Even less contrast between treatments was reported for complications of diabetes. The authors concluded that “the incidences of microvascular complications and death were not materially different.” Cardiovascular events and deaths were relatively few. Only when these were combined as “any cardiovascular” event was a “possible difference” favoring better outcomes with liraglutide suggested.

These findings are reassuring. The well-known progressive nature of T2D was evident, but all of the commonly used agents controlled glucose levels for a substantial period of time. No new adverse effects were seen, and previously recognized side effects—notably hypoglycemia with glargine and glimepiride and gastrointestinal symptoms with liraglutide—seldom impeded their use. Clinical providers may conclude that any of these agents can be recommended for a broad population of patients whose glucose levels are no longer controlled with metformin.

However, the question posed by the GRADE investigators has not been fully addressed. That is: how can we best match each treatment option to the needs of each person? This issue of Diabetes Care contains a collection of 10 additional articles prepared by the GRADE investigators that may aid decisions in clinical practice. Some of their features are summarized in Table 1. Three of them concern clinical and physiologic features that are associated with success in maintaining glycemic control, both overall and with specific forms of treatment (5–7). Four others compare the several treatments in their association with end points other than glycemic control. The end points are all-cause mortality (8), a composite of glycemic control without weight gain or hypoglycemia (9), quality of life (10), and depression or diabetes distress (11). Two articles describe associations of depression or diabetes distress at entry with long-term outcomes with the different treatments (12,13). A final analysis describes the use of rescue therapy with insulin when the originally assigned second-line treatment no longer keeps HbA_1c below 7.5% (14). Each of these articles deserves a few comments.

Garvey et al. (5) report associations between easily obtained clinical information and success in maintaining HbA_1c below 7% in the whole study population and with each treatment option. Participants who were older than 60 years had better glycemic responses than those who were younger. Understandably, lower HbA_1c prior to second-line treatment predicted better maintenance of control during the study with all treatments. A multivariable model showed that, independent of other factors, glargine and liraglutide maintained HbA_1c below 7% slightly longer than glimepiride and sitagliptin, consistent with the earlier report (3). There also was a new observation: sitagliptin was particularly ineffective in improving glycemic control in the first year of treatment. However, none of the baseline characteristics emerged as a strong predictor of between-treatment differences in long-term glycemic control.

Utzschneider et al. (6) extend these findings by examining some physiologic characteristics of the participants. They assessed associations of glycemic outcomes with insulin sensitivity of tissues and insulin secretion after an oral glucose load. Serial oral glucose tolerance tests were performed at the beginning of the study and at intervals during follow-up. Established modeling procedures were used to derive information about insulin sensitivity and secretion from these tests. The models could not reliably use data from the glargine arm of the study, so the analysis focused on glimepiride, liraglutide, and sitagliptin. Both greater insulin sensitivity and greater insulin secretion were associated with better glycemic control during treatment with all three agents. However, participants in the lowest third of insulin sensitivity had especially poor glycemic control in the first year of treatment with sitagliptin.

Further physiologic studies are reported by Rasouli et al. (7). They modeled data from oral glucose tolerance tests for participants in all treatment arms and showed differences between the four agents in their effects on insulin sensitivity and secretion during the course of the study. Insulin sensitivity remained constant in all treatment arms after the first year. In the first year of treatment, an apparent improvement of sensitivity was observed with glargine. This was likely due to an artifact in the analytic method due to the presence of glargine and its metabolites, but the investigators rightly suggest that a favorable effect of glargine on insulin sensitivity cannot be excluded. Glargine had little effect on insulin secretion, but the other three agents all enhanced β-cell function in the first year. However, this benefit declined over time, especially with liraglutide. Moreover, the specific β-cell effects differed between agents. Liraglutide potentiated insulin secretion to glucose both during fasting and after a glucose load. Sitagliptin did not potentiate insulin secretion during fasting but enhanced it during the glucose tolerance test, although less so than liraglutide. Glimepiride potentiated secretion during fasting but had little effect after oral glucose. These differences in mode of action may be relevant to clinical decisions, as will be discussed later.

All-cause mortality is examined by Banerji et al. (8). The earlier report of medical outcomes in GRADE noted no clear between-treatment difference in mortality, but the cumulative incidence curve for liraglutide seemed to rise less rapidly than that for other treatments near the end of observation. The present article presents mortality data more fully. A total of 153 deaths were reported, with 0.59 deaths per 100 participant-years. This low rate was expected because the group had a relatively short duration of diabetes and few established illnesses, which limits the statistical power of the analysis. However, the glargine, glimepiride, and sitagliptin groups had similar numbers of deaths (42, 43, and 41, respectively), while the group assigned to liraglutide had 27 deaths. Equal mortality risk was reported for glimepiride compared with linagliptin in the Cardiovascular Outcome Study of Linagliptin Versus Glimepiride in Type 2 Diabetes (CAROLINA) (15) and for glargine compared with oral therapies in the Outcome Reduction With Initial Glargine Intervention (ORIGIN) trial (16), studies that were well powered to identify differences in mortality. The lower absolute number of deaths with liraglutide in GRADE is thus an outlier that calls for more attention. Even so, the numbers are too low to draw strong conclusions. Moreover, at 10%, the fraction of deaths in GRADE for which the cause is unknown is larger than that in most trials assessing mortality, limiting comparison of treatment groups by cause of death.

The article by Kirkman et al. (9) explores use of a composite end point in GRADE. Their three-part composite consisted of HbA_1c higher than 7.5%, a weight increase of 5% or more, or hypoglycemia, defined as any event requiring assistance or more than one episode of nonsevere hypoglycemia. As expected, liraglutide outperformed the other agents by this measure, maintaining glycemic control without these undesired effects. This observation is consistent with other analyses using composites of HbA_1c, weight change, and hypoglycemia that have compared liraglutide favorably with other treatments (17). However, the analysis from GRADE has some limitations. There is no consensus on how end points for weight change or hypoglycemia should be defined and weighted regarding their likely relation to the medical status of study participants (18). In this case, most hypoglycemia events were nonsevere, self-reported by participants, and not objectively confirmed. Whether such events are linked to harm in this relatively healthy population is not clear. Also, the composite did not include gastrointestinal symptoms, the leading side effect of liraglutide. Gastrointestinal symptoms were listed in the main GRADE report along with severe hypoglycemia as “serious and targeted” adverse events and were reported by 44% of participants assigned to liraglutide (3). Finally, the GRADE analysis did not identify any baseline characteristics that were clearly predictive of the composite outcome. Although reassuringly confirmatory of prior reports, the present analysis does not add much new insight regarding individualizing second-line treatment in this population.

Findings regarding health-related quality of life (HRQoL) as an outcome of therapies are reported by Cherrington et al. (10). The investigators used three validated questionnaires to assess HRQoL: the Quality of Well-Being Self-Administered Scale (QWB-SA) and the 36-Item Short-Form Health Survey mental and physical component scales (MCS and PCS). None of these measures worsened during the study with any of the treatments. The main between-treatment difference was a greater improvement of PCS score with liraglutide in the first year that was not sustained later in the study. Mediation analyses suggested the initial benefit with liraglutide was related to weight loss and was most evident among participants with higher BMI.

Gonzalez et al. (11) evaluate possible effects of the several treatments on depression or diabetes distress during the study. They used validated questionnaires to collect data from participants for both analyses. The stated goal of the analysis was to determine whether, as suggested in earlier literature, emotional distress is provoked by use of insulin. Consistent with prior evidence for favorable effects upon entering clinical studies (19), depression and distress scores in the first year improved with all treatments used in GRADE. However, both glargine and liraglutide were associated with larger improvements in the first year compared with the two oral agents. Scores were stable later in the study, with little difference between treatments. The investigators conclude that they found “no evidence for a deleterious effect of basal insulin on emotional distress.”

Two articles assess the effects of depression or diabetes distress at entry to the study on the main glycemic outcome and on adherence to treatment with the assigned study drugs. Cherrington et al. (12) report that, after adjustment for other characteristics at baseline, neither depression nor diabetes distress was associated with inability to maintain HbA_1c below 7.0%. The companion article by Hoogendoorn et al. (13) reports that both depression and diabetes distress were associated with lower self-reported adherence to the assigned study medications. This association was apparent for depression or distress at baseline as well as during the course of the study, and it was not affected by treatment assignment. No clear explanation for discordance between glycemic outcomes and medication adherence is provided. One possibility is that self-report of adherence was not always accurate. Another is that the agents used in the study, all having a duration of action longer than a day, maintained their therapeutic effects even when daily doses were occasionally missed. In any case, the findings do not support emotion-related nonadherence to medications as an important contributor to between-treatment differences in glycemic control in this population.

Finally, Hollander et al. (14) describe experience with rescue insulin in GRADE. This analysis was not specified in the original protocol. It was initiated about halfway through the study to determine why insulin was not consistently added within 6 weeks after HbA_1c was verified as higher than 7.5%, as required by protocol. Analysis of a subgroup meeting criteria for rescue with basal insulin showed that glargine was added to glimepiride, liraglutide, or sitagliptin within 6 weeks in only 39% of participants and at any time in 69%. Timely addition of one or more injections of aspart as prandial therapy for participants in the glargine arm, as required by protocol, occurred even less frequently. Site investigators completed a questionnaire for each participant in this subgroup. The most common reasons given for nonadherence were that the participant wanted to intensify lifestyle efforts or medication adherence or was reluctant to use insulin itself. Lack of adherence to the protocol could have affected planned analyses of all treatment groups after the primary end point was reached.

The GRADE investigators have energetically sought new information from these secondary and post hoc analyses, and they give us much to consider. With all the treatments, glycemic control was best maintained for participants older than 60 years and for those without markedly elevated HbA_1c at baseline. There was no evidence of excessive side effects. Thus, there seems to be no reason to delay adding a second agent once metformin is no longer successful alone or to withhold treatment for healthy older people.

Because between-treatment differences in the main end point and many secondary end points were small, the GRADE results suggest there is no clear “best” choice for second-line treatment in the broad population enrolled in this study. Instead, these results call for greater effort to individualize treatments to each person’s clinical characteristics and preferences. The analyses reviewed here suggest some ways each of the randomized treatments might be considered and prescribed.

Sitagliptin fared least well in overall glycemic effectiveness. Its limitations were often apparent immediately and were most evident when the starting HbA_1c was high and insulin sensitivity low. Its effect on β-cell function was limited to improving insulin secretion after an oral challenge. On the positive side, side effects were rare. These observations suggest an optimal clinical profile for its use. Sitagliptin might be considered for older people or others for whom side effects are especially worrisome, when hyperglycemia is mild and mainly postprandial, and when insulin sensitivity is high. These characteristics are common in Asian populations where dipeptidyl peptidase 4 (DPP-4) inhibitors are widely used. It is noteworthy that this class of drugs has been reported to provide better glycemic control for Asian populations than for other populations (20).

Glimepiride was only slightly less effective in maintaining glycemic control than liraglutide and glargine. Severe hypoglycemia accompanying its use was uncommon, reported in 2.2% of participants over 5 years of observation. This low risk might be further reduced by simple precautions. By protocol the starting dose was at least 1 mg, which immediately provides about half-maximal glucose lowering. Daily dose was titrated within a year to a mean of more than 4 mg, the usual maximally effective dose. About a third of participants in GRADE had HbA_1c 7.2% or less at entry; for them this titration scheme may have been too aggressive. Perhaps glimepiride is best suited to individuals who have HbA_1c 7.5% or higher initially, and in some cases should be started at a 0.5-mg dose with slow titration. Also, the observation that glimepiride’s physiologic effect on insulin secretion is mainly during fasting suggests it may be most effective when fasting hyperglycemia is prominent.

Liraglutide performed well in GRADE, but not without some concerns. It maintained glycemic control as well as glargine did, caused significant weight loss, and improved measures of HRQoL and diabetes-related distress in the first year. Cardiovascular end points showed promising—but not statistically reliable—trends during the study. On the other hand, the strong improvement of β-cell function with liraglutide declined after the first year, at which time weight loss leveled and patient-reported outcomes merged with those for other arms of the study. Forty-four percent of participants assigned to liraglutide reported gastrointestinal symptoms, and its use was discontinued by 23%. Because of the benefits of this class of drug for cardiovascular and renal disease, liraglutide is a very appropriate choice for individuals needing second-line therapy who have these conditions. However, GRADE included few participants with known myocardial infarction or stroke (6.5%) or estimated glomerular filtration rates below 60 (2.4%). The GRADE subgroup for which liraglutide seems most clearly indicated may be those with significant obesity, on the basis of predictable weight loss that is associated with better quality of life and perhaps lower cardiovascular risk.

As a second-line intervention, glargine was as effective as liraglutide with respect to the primary glycemic end point. It was also well accepted. Nearly all participants randomized to glargine started it, and only 14% of them discontinued during the study. The hypothesis that starting glargine might increase diabetes-related distress was rejected; a measure of distress in the first year was lowest with glargine. Severe hypoglycemia occurred in 1.3% of participants who were assigned glargine, fewer than with glimepiride (2.2%) and similar to the incidence with liraglutide (1.0%). These findings argue that glargine, used as basal insulin without mealtime boluses of rapid-acting insulin, is well tolerated as second-line therapy.

At the same time, use of glargine as rescue therapy when HbA_1c rose above 7.0% with the other treatments was unexpectedly erratic. Glargine was added to prior treatment less than half the time the protocol called for it. This observation raises new questions, which are discussed below.

Like all studies, GRADE has important limitations. The population was similar but not identical to a general population of people needing second-line therapy. Newer therapies already proven to be helpful for treating T2D—notably the sodium–glucose cotransporter 2 (SGLT2) blockers and a new class of gut-peptide agonists represented by tirzepatide—could not have been included. For these reasons, the findings of the study cannot fully be extrapolated to clinical practice today. Some key lessons, however, are apparent.

If several options for second-line therapy are nearly equal in glycemic effectiveness, decisions must be based on other distinctive features. Some insights regarding optimal use of the specific agents used in GRADE have just been reviewed. Beyond characteristics of the drugs themselves, features of a person who might use them must be considered. Physiologic aspects of diabetes in each case are relevant. Is the person insulin sensitive or resistant? Does hyperglycemia occur mainly after meals, or are fasting glucose levels always high? Lifestyle preferences differ widely, as do physical capabilities and willingness to take injected therapies. Tolerance of various side effects differs as well. Now that drugs with favorable nonglycemic effects on cardiovascular and renal disease are available, these conditions must be identified. Assessing all these factors is not easy; it requires expertise and time. How can an algorithm replace this highly personal process?

In GRADE, an effort was made to commit all therapeutic choices after randomization to a second-line treatment to a specific protocol. This was successful up to the time of reaching the primary end point, but not after. The insulin rescue plan was not followed. To a trialist, failing to adhere to protocol is deeply troubling. Nonadherence to the rescue protocol surely challenged the planned analyses after the primary end point.

However, there is a lesson here. Some of the end points may have been compromised, but the GRADE participants fared well. Most remained active in the study, patient-reported measures were generally favorable throughout, and glycemic control remained good. Near the end of observation, when duration of T2D was about 8 years, mean HbA_1c was ∼7.1% in all four randomized arms (3). The participants and site investigators, together, had individualized treatment in a variety of ways, off-protocol but nonetheless effective.

Is this not a kind of success? It calls to mind the experience in ORIGIN (21), a larger, longer comparative effectiveness and safety trial that was conducted at diabetes centers worldwide. It enrolled participants with dysglycemia or early T2D, together with evidence of high cardiovascular risk, who were using no more than one oral agent. Those with diagnosed T2D (about 11,000 participants) had, at entry, an average disease duration of 5 years and mean HbA_1c 6.5%. They were randomized to add glargine or conventional step therapy with oral agents. Whereas the GRADE protocol called for a specific rescue plan, that in ORIGIN did not. When a fasting glycemic target was exceeded, the investigators were to advance therapy according to local guidelines and their judgement of the participant’s needs, except for not using glargine in the conventional care arm. As in GRADE, good glycemic control was maintained; the mean HbA_1c after more than 5 years was still close to 6.5% in both arms. Also as in GRADE, the randomized arms had equivalent medical outcomes despite more hypoglycemia with glargine. A lesson from ORIGIN, as from GRADE, was that investigators who are experienced in managing diabetes can effectively individualize long-term treatment for many patients who come to clinic every 3 (in GRADE) or 4 (in ORIGIN) months.

Clinical studies assess the safety and efficacy of treatments or, as in GRADE, compare different agents in a selected population. However, it is said that structured studies allow better management than is possible in a real-world clinical setting. Primary care providers are expected, with the help of treatment algorithms, to individualize as best they can, but under less favorable conditions. Some people in the general population have more challenging medical conditions and living circumstances than those in studies. Newer therapies can be costly, and formulary access varies. Primary providers have limited time at office visits, and they cannot keep up with our growing collection of drugs—some with multiple formulations and names—and advancing knowledge of subgroups of patients. For all these reasons, glycemic control in a community setting often deteriorates after the first few years of T2D.

The GRADE experience provides important insights about the safety and effectiveness of the treatments compared and suggests new ways to deploy them. Just as important, it suggests we rethink the idea that T2D must be managed in a primary care setting. There are many examples of longitudinal specialty care alongside primary medical practice. Dentistry, reproductive medicine, ophthalmology, psychiatry, and oncology are a few examples. Why not diabetology? It is claimed there are too few specialists to take on this challenge, but that could be corrected. Care for type 1 diabetes is acknowledged to require specialized support, but not care for T2D, despite its complexities and hazards. Expert groups like the GRADE investigators could be built into large health systems to individualize second- and third-line care of T2D after lifestyle and metformin. Referral could be expedited and time for informed decisions assured. It is worth a thought. We have many effective new treatments; now we must use them wisely.

We are deeply indebted to the GRADE investigators for planning and completing this complex study and now giving us a collection of new analyses to move our thinking forward.

Funding. This work was supported in part by the Rose Hastings and Russell Standley Memorial Trusts for Diabetes Research.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

M.C.R. is an ad hoc editor of Diabetes Care but was not involved in any of the decisions regarding review of the manuscript or its acceptance.