Plain Language Summary
The high blood glucose levels that characterise type 2 diabetes mellitus (T2DM) are treated using antihyperglycaemic drugs. Patients’ responses to antihyperglycaemic drugs, which are usually evaluated by measuring glycated haemoglobin (HbA1c), are affected by several factors, including the patient’s baseline HbA1c level. Previous investigations of this topic have involved dividing the baseline HbA1c values of the patient population into a number of categories. However, this loses potentially valuable information. In this analysis, we treated baseline HbA1c as a continuous variable. Our analysis used data from three large clinical trials to investigate changes in HbA1c in patients with T2DM who received one of four different antihyperglycaemic agents: dulaglutide (0.75 or 1.5 mg once weekly), sitagliptin (100 mg once daily), liraglutide (1.8 mg once daily), or exenatide (10 μg twice daily). The patients had a range of baseline HbA1c values.
In all treatment groups, improvements in HbA1c occurred across the entire baseline HbA1c range and patients with higher baseline HbA1c values showed greater reductions. The mean HbA1c reduction was similar in patients treated with dulaglutide 1.5 mg and those receiving liraglutide 1.8 mg across the range of baseline HbA1c values. However, patients receiving dulaglutide 1.5 mg had greater reductions than patients receiving exenatide or sitagliptin. As baseline HbA1c increased, patients receiving dulaglutide had progressively greater HbA1c reductions compared with those receiving sitagliptin. These results suggest that dulaglutide is an appropriate therapy for patients with T2DM across a wide range of baseline HbA1c values, including those with poor blood glucose control.
Introduction
For patients with type 2 diabetes mellitus (T2DM), treatment with most antihyperglycaemic agents—including incretin-based therapies, such as glucagon-like peptide-1 receptor agonists (GLP-1 RA) and dipeptidyl peptidase-4 (DPP-4) inhibitors—is associated with an improvement in glycated haemoglobin (HbA
1c) levels [
1‐
4]. Baseline HbA
1c is a predictor of HbA
1c change in patients responding to treatment, with greater reductions in HbA
1c levels occurring in patients with higher baseline values [
5‐
11]. Of clinical interest is a meta-analysis involving DPP-4 inhibitors, which indicated that the baseline HbA
1c level was the strongest predictive factor for change in HbA
1c [
8]. Similar results have been found for the GLP-1 RA dulaglutide [
11].
Dulaglutide has demonstrated significant improvements in HbA
1c, and fasting and postprandial blood glucose levels in patients with T2DM, and is associated with a substantial percentage of patients achieving HbA
1c < 7% (53.0 mmol/mol) [
12‐
19]. These effects are durable up to 104 weeks [
16,
20]. This agent also demonstrates a potential for weight loss and has a safety profile similar to that of other agents in the GLP-1 RA class [
21].
A post hoc analysis of data from the AWARD trials that used gradient-boosting analysis and multivariable regression to examine baseline factors associated with glycaemic response in dulaglutide-treated patients with T2DM [
11] found that higher baseline HbA
1c was the major factor associated with a greater reduction in HbA
1c (a 1% [11 mmol/mol] increase in baseline HbA
1c was associated with a decrease in HbA
1c of 0.60% [6.54 mmol/mol;
P < 0.0001]).
When comparing antihyperglycaemic agents with each other, it is often of interest to understand whether patients experience different responses to each treatment according to baseline or disease characteristics of interest. In the case of a continuous variable, such as baseline HbA
1c, this is typically done by dividing the population into subgroups on the basis of HbA
1c level before applying the statistical model [
22‐
25]. For example, both Bain et al. [
24] and Gallwitz et al. [
25] divided their respective GLP-1 RA-treated populations into two subgroups (baseline HbA
1c < 8.5% and ≥ 8.5%). Baseline HbA
1c was thus treated as a categorical variable. Although informative, this type of analysis fails to utilise all the information provided by a continuous variable [
26]. It is also associated with reduced statistical power, increased risk of a false positive result, underestimation of the extent to which the outcome varies among groups, and potential concealment of a non-linear relationship between variable and outcome [
26]. In addition, this approach can result in the performance of multiple analyses if more than one set of cutoffs is used to generate the categories (e.g. < 8.5% vs ≥ 8.5% followed by < 7.5% vs ≥ 7.5%) and, as a consequence, it may generate conflicting results. In order to avoid these problems, in the current analysis we have treated baseline HbA
1c as a continuous variable—an approach that allows us to fully explore the relationship between baseline HbA
1c and antihyperglycaemic endpoints both within and between treatment groups. This analysis—which represents an alternative to the more typical (categorical) analysis—should provide a deeper understanding of changes in antihyperglycaemic measures associated with individual agents in patients with different baseline HbA
1c values, and of the relative changes associated with different agents in patients at each level of baseline glycaemia. To our knowledge, this type of analysis has not been previously performed on HbA
1c data from patients with T2DM who are receiving incretin therapies.
In the present post hoc analysis, we used data from three phase III randomised controlled AWARD trials to investigate the consistency of glycaemic response across a continuous range of baseline HbA
1c values, in patients treated with once-weekly dulaglutide 1.5 mg and 0.75 mg versus other incretin-based therapies: the DPP-4 inhibitor, once-daily sitagliptin 100 mg (AWARD-5 [
13]), and the GLP-1 RAs, once daily liraglutide 1.8 mg (AWARD-6 [
14]) and twice-daily exenatide 10 μg (AWARD-1 [
15]). These three AWARD trials were selected for this analysis because they involved an incretin-based comparator and enrolled patients with T2DM with a wide and continuous range of baseline HbA
1c values. Each of these trials demonstrated that dulaglutide is at least as efficacious as these other incretin-based therapies and has an acceptable tolerability and safety profile.
Methods
This exploratory post hoc analysis included data from all adult patients enrolled in the intent-to-treat (ITT) populations who received active treatment (dulaglutide or another incretin-based therapy) in AWARD-5, AWARD-6 and AWARD-1. AWARD-5 and AWARD-1 patients randomised to receive placebo were not included in the analysis. The design and results of these trials have been previously published [
13‐
15]. Each of these original studies was conducted in accordance with the Declaration of Helsinki guidelines on good clinical practice. In all cases the protocol was approved by local institutional review boards, and all patients provided written informed consent before participation in the trial [
13‐
15]. The full list of institutional review boards that approved the three studies can be found in the Supplementary Material (Table S1). A summary of each of these trials is provided below.
In AWARD-5 [
13], 921 patients with T2DM on metformin (≥ 1500 mg/day) were randomised to receive concomitant therapy with once-weekly dulaglutide 1.5 mg or 0.75 mg, or once-daily sitagliptin 100 mg. Patients were eligible for inclusion if, at screening, their HbA
1c was > 8% and ≤ 9.5% (> 63.9 and 80.3 mmol/mol), if treated with diet and exercise, or ≥ 7.0% and ≤ 9.5% (≥ 53.0 and ≤ 80.3 mmol/mol) if receiving oral antihyperglycaemic medication. Eligible patients then entered a lead-in period that lasted up to 11 weeks. During the lead-in period, patients received metformin monotherapy (minimum dose ≥ 1500 mg/day) for ≥ 6 weeks prior to randomisation; all other oral antihyperglycaemic medications were discontinued.
In AWARD-6 [
14], 599 patients with T2DM on metformin (≥ 1500 mg/day) were randomised to receive concomitant therapy with once-weekly dulaglutide 1.5 mg or once-daily liraglutide 1.8 mg. Patients treated with antihyperglycaemic drugs other than metformin were excluded. The inclusion criteria included an HbA
1c level ≥ 7.0% and ≤ 10.0% (≥ 53.0 and ≤ 85.8 mmol/mol) at screening. This study did not include a lead-in period.
In AWARD-1 [
15], 835 patients with T2DM on maximally tolerated doses of metformin (1500–3000 mg/day) and pioglitazone (30–45 mg/day) were randomised to receive concomitant therapy with once-weekly dulaglutide 1.5 mg or 0.75 mg, or exenatide 10 μg twice daily. At screening, patients were eligible for inclusion if their HbA
1c was ≥ 7.0% and ≤ 11.0% (≥ 53.0 and ≤ 96.7 mmol/mol), if on oral antihyperglycaemic medication monotherapy, or ≥ 7.0% and ≤ 10.0% (≥ 53.0 and ≤ 85.8 mmol/mol) if on combination oral antihyperglycaemic medication. After a lead-in period that lasted up to 12 weeks, patients were eligible for randomisation if their HbA
1c was > 6.5% (> 47.5 mmol/mol). Previous oral antihyperglycaemic medications other than metformin and pioglitazone were discontinued during the lead-in period.
In the current post hoc analysis, changes in HbA1c were evaluated at week 26 for each of the trials (AWARD-5, AWARD-6 and AWARD-1). Week-52 data are available for one study only (AWARD-5) and are mentioned briefly. Data were analysed by study using last observation carried forward (LOCF) analysis of covariance (ANCOVA), starting with the predefined statistical models from the respective study protocols, which assumed a linear association between baseline HbA1c and change in HbA1c.
Backward selection methods were applied to assess whether the respective predefined models were appropriately capturing this linear association or needed to be adapted. In order to produce meaningful and interpretable estimates and P values for the factor treatment, the factor for baseline HbA1c and the treatment-by-baseline HbA1c factor were centred around a baseline HbA1c of 8.0% (63.9 mmol/mol). Consistency of potential treatment effect across the range of baseline HbA1c values was investigated via the corresponding treatment-by-baseline HbA1c interaction term, with P < 0.1 interpreted as a potential differential treatment effect. Sensitivity analyses were performed as appropriate, using only data from patients treated with active therapy in the ITT population of each study who met the HbA1c inclusion criteria.
Baseline characteristics data are presented for the active treatment arms (dulaglutide and incretin-based comparators) of the ITT population of each study; full baseline characteristics for each of the respective studies have been previously published [
13‐
15].
Discussion
By investigating HbA
1c reduction across the continuum of baseline HbA
1c values, this exploratory post hoc analysis of data from the AWARD-5, AWARD-6 and AWARD-1 studies provides an in-depth evaluation of the primary results of these studies. The results of the current analysis indicate that incretin-based therapies effectively lowered HbA
1c irrespective of the HbA
1c value at baseline. Across the range of baseline HbA
1c, patients treated with dulaglutide 1.5 mg achieved a similar mean reduction in HbA
1c to patients treated with liraglutide 1.8 mg (AWARD-6: Fig.
1b), and a greater HbA
1c mean reduction than patients treated with exenatide twice daily (AWARD-1; Fig.
1c) or sitagliptin (AWARD-5; Fig.
1a). These results are consistent with the primary analyses, which assumed the same effect of baseline HbA
1c for the whole ITT population [
13‐
15].
The current analysis complements previous post hoc analyses [
24,
25], which have categorised patients into distinct subgroups based on baseline HbA
1c cutoff values (e.g. < 8.5% vs ≥ 8.5% [69.4 mmol/mol]). It has shown that in dulaglutide 1.5 mg and 0.75 mg treatment groups, HbA
1c was reduced in patients with baseline HbA
1c levels that ranged from 6.5% (47.5 mmol/mol) to 11.0% (96.7 mmol/mol) (Fig.
1). Once-daily sitagliptin and liraglutide 1.8 mg, and twice-daily exenatide, were also associated with reduced HbA
1c in patients with a range of baseline HbA
1c levels (sitagliptin, HbA
1c 7.0–9.5% [53.0–80.3 mmol/mol]; liraglutide 1.8 mg, HbA
1c 7.0–10.0% [53.0–85.8 mmol/mol]; exenatide, HbA
1c 6.5–11.0% [47.5–96.7 mmol/mol]; Fig.
1). These findings suggest that improvements in glycaemic control can be achieved with incretin-based therapies irrespective of the patient’s HbA
1c value at the time therapy is initiated. Such improvements in glycaemic control will help to decrease the risk of microvascular complications across the broad spectrum of patients with T2DM [
28,
29].
Moreover, in this post hoc analysis, there was a positive linear relationship between HbA
1c reduction at 26 weeks and baseline HbA
1c in patients treated with dulaglutide or liraglutide in AWARD-6, in patients treated with dulaglutide or exenatide twice-daily in AWARD-1, and in patients treated with dulaglutide or sitagliptin in AWARD-5 (Fig.
1). In dulaglutide 1.5 mg-treated patients with better glycaemic control at baseline (HbA
1c = 7% [53.0 mmol/mol]), the reductions were < 1% (10.9 mmol/mol) in each of the AWARD-5, AWARD-6 and AWARD-1 studies at 26 weeks. In contrast, mean reductions ranged from 1.8% to 3.2% (19.8–35.0 mmol/mol]) at 26 weeks in patients with poorer control at baseline (HbA
1c ≥ 9.5% [80.3 mmol/mol]) who received dulaglutide 1.5 mg. These outcomes are consistent with another post hoc analysis, which found that a higher baseline HbA
1c value reflecting poor glycaemic status was a major factor associated with a greater reduction in HbA
1c [
11]. A positive relationship between baseline HbA
1c and the magnitude of HbA
1c reduction has been reported in meta-analyses involving insulin analogues [
6] and ten categories of glucose-lowering therapies [
30], and in the nomogram developed by Esposito et al. [
8] to estimate the HbA
1c to estimate the HbA
1c response to different DPP-4 inhibitors.
The positive linear relationship that exists between incretin-associated HbA
1c reduction and baseline HbA
1c was evident with dulaglutide and liraglutide treatments in AWARD-6, with dulaglutide and exenatide twice-daily treatments in AWARD-1 and with dulaglutide and sitagliptin treatments in AWARD-5. There was no evidence of a significant treatment-by-baseline HbA
1c interaction in AWARD-6 or AWARD-1 (interaction
P values, 0.605 and 0.443, respectively); however, a significant interaction was demonstrated in patients treated with dulaglutide versus sitagliptin in AWARD-5 (
P value = 0.001), with a progressively greater HbA
1c reduction associated with dulaglutide 1.5 mg compared with sitagliptin as baseline HbA
1c increased. This significant interaction was confirmed at 52 weeks [
27]. The greater HbA
1c reduction in patients with higher baseline HbA
1c seen in those treated with dulaglutide 1.5 mg, compared with sitagliptin, could be related to the greater effect that GLP-1 RAs have on both fasting and postprandial glucose levels [
13,
30].
This post hoc analysis indicates that consideration of a patient’s baseline HbA
1c level is important because the glycaemic response varies among patients treated with different antihyperglycaemic agents, and these variances may be magnified at higher baseline HbA
1c levels (Fig.
1). In light of the relationship between the patient’s baseline HbA
1c and their changes in HbA
1c, this analysis highlights an opportunity for personalising therapy. Such personalisation of therapy is increasingly being seen as important in the optimal clinical management of patients with T2DM [
31], with the patient’s age, their attitude to specific adverse effects of therapy, the presence of comorbid conditions and the duration of T2DM all being considered when a therapeutic strategy is being considered [
32].
This analysis has a number of limitations. These include the open-label design of AWARD-6 and AWARD-1 (which could have affected physicians’ and patients’ behaviour), the exploratory post hoc nature of the analysis and the absence of placebo-based correction for the absolute reductions in HbA
1c (only two of the three original AWARD trials included a placebo arm [
13‐
15] and the data must be interpreted in the context of the absence of placebo data). In terms of the ANCOVA models used in the analysis, these did not include factors potentially associated with HbA
1c such as age and duration of diabetes [
33,
34]. Although this may be regarded as a limitation, use of the simpler models allowed us to remain consistent with the original, pre-planned statistical analysis. Moreover, baseline HbA
1c is the major factor associated with change in HbA
1c, irrespective of treatment [
10]. A post hoc analysis of studies involving dulaglutide showed that baseline HbA
1c accounted for 49% of relative influence, whereas the relative influence of other baseline characteristics such as age, fasting serum glucose or duration of diabetes was very small or of limited clinical significance [
11]. Consistency with the original analysis was also maintained by focussing solely on HbA
1c change from baseline; this outcome was the primary endpoint in the AWARD trials and therefore the most suitable for exploration in this analysis. It should be noted that, because this analysis presents data only for the baseline HbA
1c ranges specified by the inclusion criteria in each study, no conclusions can be drawn regarding the effect of incretin-based therapy in patients with HbA
1c levels outside these ranges.