Introduction

Type 2 diabetes is generally regarded as a slowly progressive disease characterised by a reduction in ability to maintain glucose homeostasis over time [1] and a concomitant increase in the need for pharmacological therapy to do so. Recent observations that bariatric surgery seems to slow or even reverse this process [2, 3] suggest that diabetes is not necessarily progressive and that beta cell function and the capacity to make insulin can be improved [4, 5] in association with reduced body mass and food intake. Additional evidence suggests that short-term intensive insulin therapy early in the course of diabetes [6] plus lifestyle modification [7] may have persistent effects on glycaemic control by reducing the demand on the beta cells to secrete insulin, thereby preserving their function [8, 9], or by improving adherence to health-promoting behaviours. Other agents, including thiazolidinediones and incretin mimetics, also improve beta cell function [8, 9]. It is not known whether a strategy of prolonged multimodal intensive glucose lowering has durable effects.

The Action to Control Cardiovascular Risk in Diabetes (ACCORD) Trial (NCT00000620) was designed to study the cardiovascular effects of intensive glycaemic control, intensive blood pressure control, and fibrate use among people with type 2 diabetes. As previously reported [10], the intensive glycaemic intervention was discontinued after the independent data and safety monitoring board noted excess mortality in that group. As a result, participants who had been randomised to intensive glycaemic management (i.e. targeting an HbA1c of <6.0% [42 mmol/mol]) and who, as a group, had reduced their entry median HbA1c from 8.1% (65 mmol/mol) to 6.4% (46 mmol/mol) were all transitioned to the standard glycaemic management strategy (i.e. targeting an HbA1c of 7.0–7.9% [53–63 mmol/mol]), and continued to be followed until the blood pressure and lipid trials that were part of the initial ACCORD design were completed about 1 year later. By the end of the follow-up period, the HbA1c in this group was 7.4 ± 1.2% compared with 7.8 ± 1.2% in the group randomised to standard care [11]. This situation offered a unique opportunity to observe the glycaemic effects of intentional relaxation of glycaemic therapy in a group of patients who started out with poorly controlled (HbA1c >7.5% [58 mmol/mol]), longstanding (median 10 years) type 2 diabetes, and who achieved near-normal HbA1c levels during more than 3 years of intensive glucose lowering. This report therefore describes the metabolic course of participants randomised to the intensive glycaemia group who had at least one visit before transition and one visit after transition to a standard glycaemic management approach, and determines whether a period of intensive glucose lowering that achieved normal or near-normal glucose levels is associated with persistence of glycaemic control after relaxation of therapy.

Methods

ACCORD Trial

The design of the ACCORD Trial has been described previously [12, 13]. Briefly, 10,251 participants aged 40–79 years with type 2 diabetes, an HbA1c ≥7.5% and additional cardiovascular risk factors were recruited at 77 sites in North America. Participants were randomised to either intensive glycaemic management targeting an HbA1c of <6.0% or standard glycaemic management targeting an HbA1c of 7.0–7.9%. In a double 2 × 2 factorial design, 4,733 participants were also randomised to either intensive blood pressure lowering (<120/80 mmHg) or standard blood pressure lowering (130–139/80–90 mmHg), and the remaining 5,518 participants were randomised to the addition of fenofibrate or placebo to statin therapy. The trial was approved by ethics committees at all participating sites, and all participants gave informed consent.

The intensive glycaemic management strategy targeting an HbA1c of <6.0% involved monthly visits for the first 4 months followed by visits every 1–2 months, which included review of logs of self-monitoring of blood glucose (SMBG; performed two to eight times per day) and point-of-care HbA1c (every 2 months), review and prescription or titration of glucose-lowering medication from all approved classes in any combination to further lower HbA1c if it could be achieved safely while avoiding hypoglycaemia, and education regarding diabetes self-management including diet, physical activity, insulin self-titration, hypoglycaemia avoidance and management, with additional supportive phone calls between visits.

The standard management strategy targeting an HbA1c of 7.0–7.9% included visits every 4 months (except for participants also randomised to intensive blood pressure control, who had visits every 2 months) to review logs of SMBG (performed a few times per week to three times per day), education regarding diabetes self-management as needed, and review and prescription or titration of glucose-lowering medication including down-titration for any of the following reasons: severe or frequent symptomatic hypoglycaemia; 50% or more SMBG values below 5 mmol/l (90 mg/dl); or in the setting of either one HbA1c <6.5% or consecutive HbA1c values <7.0%, any of use of insulin or a secretagogue, any symptomatic hypoglycaemia or any SMBG below 5 mmol/l (90 mg/dl).

An extra visit was scheduled for all intensive group participants at the time of the transition to explain the new HbA1c goal of 7.0–7.9%, and their conversion to the standard management strategy of the trial as described above. At this transition visit, research staff reviewed SMBG values, point-of-care HbA1c and medications, and down-titrated medications for individuals with an HbA1c <7.0%.

Glucose-lowering medications, BMI and a centrally measured HbA1c were documented prior to randomisation and at least every 4 months for all participants throughout the study. Glycaemic status of all participants was reviewed centrally and reported to investigators regularly to promote adherence to the standard glycaemic management strategy after transition. In particular, site investigators were sent regular reminders of the down-titration algorithm, reports indicating the proportion of formerly intensive group participants at their site whose medications were down-titrated according to the standard management algorithm, and listings of intensive group participants whose HbA1c remained <7.0%. Extra visits or phone calls were completed to further down-titrate glucose-lowering medications as needed to achieve an HbA1c of 7.0–7.9%.

The last study visit on or before the date of the transition at which HbA1c was measured centrally was considered the pre-transition visit, and pre-transition values were obtained from that visit. Thereafter, the last study visit at which HbA1c was measured centrally was considered the final post-transition visit, and post-transition values were obtained from that visit.

Participants

All participants allocated to the intensive group who were alive at the time of transition and who had at least one HbA1c measurement at or before the date of transition and at least one measurement after the transition date were included in this analysis.

Statistical methods

All analyses were conducted using SAS software version 9.1 (SAS Institute, Cary, NC, USA). Two-sided p values <0.05 were considered nominally significant. Before any analyses were conducted, a decision was made to assess the effect of achieving a pre-transition HbA1c <6.5% on the final HbA1c level. This threshold was chosen because it was just above the median HbA1c level achieved in the intensive group during the ACCORD Trial (i.e. 6.4%) and thus represented approximately half of the intensive group participants. It is also the threshold used to diagnose diabetes [14].

Characteristics of intensive group participants whose last pre-transition HbA1c was <6.5% vs ≥6.5% were compared by t tests or χ 2 tests. Mean insulin doses were calculated based on all participants; those not on insulin were assigned a dose of 0 units for the analyses. Participants whose final post-transition HbA1c was <6.5% vs ≥6.5% were compared for mean final HbA1c, post-transition change in BMI (three categories), change in insulin dose (three categories), and change in use of other glucose-lowering medications (added, continued, never prescribed or discontinued for each medication class) by χ 2 tests.

RRs with 95% CI of achieving a final post-transition HbA1c <6.5% were calculated for a pre-transition HbA1c <6.5% vs ≥6.5% using Poisson regression, both before and after adjustment for baseline pre-randomisation HbA1c and demographic, anthropometric, co-interventional and pharmacological covariates listed in Tables 1 and 2. To assess the relationship between the pre-transition and final HbA1c levels, RRs were calculated for 0.1% increments of the pre-transition HbA1c level. Comparisons were relative to those who had a pre-transition HbA1c equal to 6.5% using unadjusted and adjusted Poisson regression. Lines were fitted assuming a single linear term for pre-transition HbA1c in a log-linear model. Finally, to determine if HbA1c levels ‘drifted’ up with a greater duration of exposure to the standard glycaemic intervention after transition, a second degree penalised B-spline was fitted for change in HbA1c vs time between pre-transition and final HbA1c measurements [15].

Table 1 Baseline and pre-transition characteristics of intensive group participants who were transitioned to standard care and followed-up after transition
Full size table
Table 2 Changes in BMI and medication use from the last pre-transition visit to study completion
Full size table

Results

A total of 4,119 participants who were allocated to intensive glycaemic management and who had at least one HbA1c level measured before and after transition to the approach used in the standard glycaemic group were analysed. Excluded intensive group participant characteristics are summarised in the electronic supplementary material (ESM) Table 1 for comparison. The mean ± SD duration of intensive glycaemic management in the pre-transition period was 4.0 ± 1.2 years (range 2.3–7.0). As noted in Table 1, compared with the 1,786 intensive group participants whose HbA1c before transition was ≥6.5%, the 2,333 who achieved an HbA1c before transition of <6.5% were more likely to be male, older and have shorter-duration diabetes, and have access to a certified diabetes educator (CDE) at their investigative site at baseline. Prior to initiating intensive therapy at the time of randomisation, this group also had lower HbA1c levels and less use of insulin, but greater use of sulfonylureas and a higher BMI. At the pre-transition visit, they continued to require less insulin, but were more likely to be taking other glucose-lowering medications, including metformin, thiazolidinediones and secretagogues, than those who did not achieve an HbA1c <6.5% during intensive management.

These 4,119 participants were followed for a mean ± SD of 1.1 ± 0.2 years (range 0.4–1.4) after their glycaemic management approach was relaxed to the standard glycaemic approach. At the time of the final visit, 711 participants continued to have an HbA1c <6.5%, 1,622 had a rise from <6.5% to ≥6.5%, 112 participants had a fall from ≥6.5% to <6.5%, and 1,674 maintained an HbA1c ≥6.5%. Among the 823 participants with a final HbA1c <6.5%, mean HbA1c was 6.0 ± 0.3% (42 ± 3 mmol/mol), and among those with a final HbA1c ≥ 6.5%, the mean was 7.7 ± 1.1% (61 ± 12 mmol/mol). More participants whose final HbA1c was <6.5% had lost weight and reduced their dose of insulin and use of secretagogues and acarbose from the pre-transition visit to the final post-transition visit compared with those who did not achieve a final HbA1c of <6.5% (Table 2).

Achieving a pre-transition HbA1c <6.5% was a strong predictor of maintaining a final HbA1c <6.5% (crude RR 4.9 [95% CI 4.0, 5.9], p < 0.0001) (Fig. 1). This strong association was maintained even after adjustment for age, sex, diabetes duration, availability of a CDE at the study site, allocation to intensive blood pressure control, allocation to receive fibrate, pre-randomisation HbA1c, BMI and glucose-lowering medication use, and pre-transition BMI and glucose-lowering medication use (RR 4.0 [95% CI 3.3, 5.0], p < 0.0001). Adjustment for change in both BMI and glucose-lowering medication use during the post-transition follow-up also did not alter this association (RR 3.9 [95% CI 3.2, 4.8], p < 0.0001). Duration of intensive management was not a significant determinant of final HbA1c (RR 1.05 per year [95% CI 0.99, 1.12], p = 0.08).

Fig. 1
figure 1

Effect of having an HbA1c <6.5% (48 mmol/mol) after 4.0 years of intensive diabetes management on the likelihood of having an HbA1c <6.5% 1.1 years after relaxation of glycaemic management. Demographics = sex, age and diabetes duration; Co-interventions = CDE availability at study site, randomisation to fibrate and randomisation to intensive blood pressure control; Pre-Rand = at ACCORD baseline pre-randomisation; Pre-T = last pre-transition; Post-T = post-transition; Pre-T medications = insulin dose (U kg−1 day−1) and use of thiazolidinedione, metformin, secretagogue, acarbose and incretin mimetic before transition; Post-T change in BMI = category of BMI change (as in Table 2); Post-T change in medications = category of change in insulin dose (as in Table 2) and in use of thiazolidinedione, metformin, secretagogue, acarbose and incretin mimetic from before transition to study completion (added, continued, never prescribed, discontinued)

Full size image

Figure 2 shows that there was no clear threshold for the effect of progressively lower pre-transition HbA1c on the likelihood of achieving an HbA1c <6.5% after relaxation of glycaemic management. Indeed, there was a graded relationship such that those with lower pre-transition HbA1c were more likely to maintain an HbA1c <6.5% over time. Compared with those with a pre-transition HbA1c equal to 6.5%, those with a pre-transition HbA1c of 6.0% or ≤5.5% (37 mmol/mol) were, respectively, 2.2 times [95% CI 1.4, 3.5] or 4.2 times [95% CI 2.8, 6.5] more likely to have an HbA1c <6.5% at the end of follow-up. Adjustment for age, sex, known diabetes duration, availability of a CDE at the study site, allocation to intensive blood pressure control, allocation to receive fibrate, pre-randomisation HbA1c, BMI and glucose-lowering medication use, and change in both BMI and glucose-lowering medication use during the post-transition follow-up minimally affected these estimates (2.0 [95% CI 1.3, 3.2] and 3.7 [95% CI 2.4, 5.7], respectively).

Fig. 2
figure 2

Prolonged effect of the achieved HbA1c with intensive glycaemic therapy on the likelihood of achieving a final HbA1c <6.5%. Models are unadjusted (blue circles) and adjusted (for demographics, co-interventions, baseline pre-randomisation HbA1c, BMI and medications, and post-transition change in BMI and medications) (red circles). The y-axis displays risks relative to a pre-transition HbA1c of 6.5%. The lines represent a log-linear model assuming a single linear term for pre-transition HbA1c. To convert values for HbA1c in % into mmol/mol, subtract 2.15 and multiply by 10.929

Full size image

As noted in Fig. 3, there was no relationship between change in HbA1c from pre-transition to the final measurement and the time interval between those two measurements.

Fig. 3
figure 3

Effect of time since transition to a standard glycaemic approach on the change in HbA1c since transition. To convert increments for HbA1c in % into mmol/mol, multiply by 10.929

Full size image

Discussion

Achievement of tight glycaemic control for a mean of 4 years was associated with persistent glycaemic control thereafter. This group of people with a median self-reported duration of diabetes of 10 years had an HbA1c >7.5% at baseline, and were intensively managed to a mean of 6.5% at the time of transition. One-fifth of them had an HbA1c <6.5% after a mean of 1.1 years of relaxation from the intensive glycaemic therapy to a standard glycaemic management approach. Those who achieved an HbA1c <6.5% during intensive therapy were four times more likely to have a final post-transition HbA1c below 6.5% compared with those who were not able to achieve such tight control on intensive therapy. This analysis indicates that, in some patients who have established and initially suboptimally controlled type 2 diabetes, achieving tight glycaemic control with intensive diabetes management is associated with improved maintenance of subsequent glycaemic control despite transition to standard therapy.

There are several potential explanations for the findings, and more than one may have contributed to the observed outcome. First, those who achieved and maintained better glycaemic control may have had physiologically ‘milder’ diabetes with more beta cell reserve or less insulin resistance at entry into the trial. The group with pre-transition HbA1c <6.5% did have shorter known duration of diabetes, and, before initiating intensive management, they had lower HbA1c levels with less insulin use, but more sulfonylurea use and their BMI was higher. Nevertheless, even after adjustment for pre-randomisation and pre-transition characteristics, there was a strong effect of pre-transition HbA1c on post-transition HbA1c. Second, despite the fact that the standard glycaemia approach was used to manage intensive group participants after transition, participants may have continued some unmeasured behaviours and therapies that they were using before transition. It is notable that participants who achieved a final HbA1c <6.5% did so despite greater down-titration of glucose-lowering medications than those who did not achieve a final HbA1c <6.5%. These lifestyle changes and decrease in medication use may have led to more weight loss after transition which reduced insulin resistance, allowing their endogenous insulin to be more effective.

Another possibility is that the period of intensive glycaemic control with multiple agents which resulted in near-normal glycaemia may have ‘rested’ the beta cells and allowed a sustained recovery of beta cell function and/or mass leading to improved glucose homeostasis. Indeed, very early in the course of type 2 diabetes, there is supportive evidence for restoration of glucose homeostasis to the point of remission of diabetes for up to 2 years after 2–6 weeks of intensive management using insulin [6] or oral agents (gliclazide and metformin) [16]. In more longstanding diabetes, short-term continuous subcutaneous insulin infusion [17] and intensive lifestyle intervention [7] have had significant but modest effects. Furthermore, trials in people with impaired fasting glucose and/or impaired glucose tolerance have shown that diabetes can be prevented and/or normal glucose tolerance restored after several years of treatment with therapies that reduce the need to secrete insulin including lifestyle intervention, metformin [18, 19], thiazolidinediones [20, 21], acarbose [22] and insulin [23]. In the present study, this paradigm of therapeutic near-normalisation of glucose is extended to a population with longstanding diabetes using an intensive combination therapy approach. Moreover, this study shows that the nearer to normal that glycated haemoglobin can be brought initially, the more likely sustained maintenance of glucose homeostasis is to be, and this is not simply because it takes longer for HbA1c to drift up from lower levels.

The role of each of the medications used cannot be established in this study. Greater use of metformin, thiazolidinediones and secretagogues was associated with better pre-transition glycaemic control. However, in multivariable regression models, none of the medications or changes in medication was favourably associated with lower post-transition HbA1c (data not shown). Furthermore, adjustment for medications or changes therein did not affect the relationship between pre-transition and post-transition HbA1c, suggesting the types of medications used may be less important than the glycaemic target.

This post hoc analysis was not part of the ACCORD protocol, and must be interpreted with caution. However, the observation is consistent with the existing literature [2–7]. Additional limitations include the absence of a comparison group that did not relax glycaemic management. Nonetheless, both the intensive phase and the relaxation to a standard approach were conducted according to a structured protocol, and information about important covariates was systematically collected for all the 4,119 participants over the full intensive intervention period and the relatively long post-transition follow-up to allow adjustment for potential confounders. Finally, the postulated protection of beta cell function cannot be directly confirmed because of lack of physiological measurements in the ACCORD population.

Questions remain about the optimal duration and type of therapy, and the best candidate patients for this approach, especially in light of the ACCORD Trial finding of increased mortality in the intensive management group [10]. Of note, however, the subgroup of intensively treated participants that benefited here (those who achieve an HbA1c <6.5%) did not have an increased risk of mortality in the ACCORD Trial [24]. The potential benefits and harms of maintaining an HbA1c <6.5% despite relaxing therapy to target a higher HbA1c are unknown.

In summary, the results show that attainment of HbA1c levels below the diabetes threshold level of 6.5% during a mean 4 year period of intensive glycaemic control using a combination of agents is associated with subsequent durable glycaemic control. The relative contributions from physiological vs behavioural changes remain unknown, but the main observation is consistent with other lines of evidence suggesting that the progressive deterioration in glucose homeostasis is not necessarily irreversible in people with established type 2 diabetes.