Introduction

Diabetes is a serious chronic disease that is growing rapidly in prevalence and that now affects >10% of adults in developed countries [1, 2]. People with type 2 diabetes are two to four times more likely to develop a serious cardiovascular outcome compared with those without diabetes [3, 4]. Despite risk-reduction strategies that include lowering of cholesterol and BP, and smoking cessation, the majority of those with diabetes continue to die from cardiovascular causes [5]. The degree to which improved glucose control could help address this residual cardiovascular risk remains uncertain.

The UK Prospective Diabetes Study (UKPDS) showed that hyperglycaemia, as assessed by HbA1c levels, was a statistically independent and potentially modifiable risk factor for cardiovascular disease, in addition to LDL-cholesterol, HDL-cholesterol, BP and smoking [6]. Findings from other large observational studies have confirmed the continuous and positive association between various measures of glycaemia (including fasting and post-load glucose levels and HbA1c) and the risk of cardiovascular disease [7, 8]. However, despite achieving a median 0.9% HbA1c difference for a median of 10 years, the UKPDS did not demonstrate a statistically significant risk reduction for myocardial infarction.

The results of three other clinical trials [9–11] designed primarily to determine whether targeting lower vs higher glucose levels can reduce the risk of cardiovascular events in patients with type 2 diabetes were published in 2008. Of these, the ACCORD trial’s intensive glycaemic intervention [9] was terminated early after a median of 3.5 years because of higher mortality among participants assigned to an HbA1c target of <6.0%.

In order to provide more precise estimates of the effects of glucose-lowering on major cardiovascular events, the Action to Control Cardiovascular Risk in Diabetes (ACCORD) [9], Action in Diabetes and Vascular Disease: Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE) [10], UKPDS [12] and Veterans Affairs Diabetes Trial (VADT) [11] investigators have established a collaboration to facilitate a formal meta-analysis of the results from each trial and to explore any differences among trials.

Methods

Trial inclusion and exclusion criteria

Trials were eligible for inclusion in these meta-analyses if they were designed to assess directly the impact of achieving lower vs higher levels of glycaemia on cardiovascular outcomes in adult patients with type 2 diabetes and had the following features: large size, defined as at least 1,000 person-years of follow-up in each treatment arm and a minimum of 2 years median post-randomisation follow-up; randomised and controlled; double-blind or blind assessment of endpoint design; pre-specified cardiovascular outcomes; analysed using an intention-to-treat approach; and follow-up of ≥90% of randomised participants for vital status.

Trials randomising individuals to comprehensive cardiovascular risk-reduction strategies were excluded unless there was a separate randomisation to different levels of glycaemic control. Trials conducted in patients with type 1 diabetes, with gestational diabetes or in children aged ≤16 years and trials conducted in acute or critical-care settings were all excluded.

Search strategy

A comprehensive literature search was performed to determine if any trials in addition to ACCORD, ADVANCE, UKPDS and VADT met the inclusion and exclusion criteria for this review. Potentially eligible trials were identified using literature searches, scrutiny of clinical trial registers and abstracts of proceedings, and enquiry among colleagues and industry representatives, in accordance with internationally accepted norms. MEDLINE, controlled clinical trial registers and the Cochrane Database were searched for articles published in English up until January 2009. Studies were identified through PubMed searches of the MEDLINE database with the MeSH headings ‘blood glucose’, ‘diabetes mellitus’, ‘clinical trial’, and the non-MeSH terms ‘glycaemic/glycemic/blood glucose control’ and ‘aggressive/tight/intensive’. Reference lists of the retrieved articles were also searched to identify other eligible studies, and information from colleagues was used to identify more-recently published articles.

Outcomes

All outcomes were pre-specified in the study protocol for this meta-analysis. The primary outcome was a composite of major cardiovascular events, defined as death from cardiovascular causes (including sudden death), non-fatal myocardial infarction and non-fatal stroke.

Secondary outcomes were stroke (non-fatal and fatal), myocardial infarction (non-fatal and fatal), heart failure resulting in hospitalisation or death, cardiovascular death, non-cardiovascular death, all-cause mortality and major hypoglycaemia. Major hypoglycaemia was defined as an episode with typical symptoms and signs of hypoglycaemia, without other apparent cause, where the individual was unable to treat him/herself.

Data collection

Each trial group collated data from its own trial according to pre-specified variable definitions and shared group data with each of the other three groups. This enabled collaborators to cross-check the data and to perform the analyses independently. UKPDS data were censored at 5 years after randomisation, so as to provide a follow-up duration that was similar to that of the other three trials. Individual patient data meta-analysis was not done, as ongoing within-trial analyses are in progress and one trial (ACCORD) [9] is continuing to follow participants within the active-treatment phase of the study.

Statistical analysis

The overall mean age of participants, duration of follow-up and proportion of women was calculated using the mean values for each trial weighted by the trial’s number of participants. The glycaemic separation achieved between randomised groups for mean HbA1c and for mean fasting plasma glucose in each trial was calculated as the differences in the reductions between baseline and the last recorded visit. The overall glycaemic separation between randomised groups was calculated as the inverse-variance weighted mean of the individual trial differences using a random-effects model.

HRs for the impact on outcomes of intensive vs less-intensive glycaemic control were estimated separately for each trial using Cox proportional hazards models and according to intention-to-treat. Overall effect estimates, and 95% CIs, were calculated using the random-effects model. (i.e. weighting by the statistical precision of the estimate in each trial). Sensitivity analyses were performed using fixed-effects models. The cut-off for significance with respect to the primary outcome of macrovascular disease was set at p < 0.05 (95% CIs exclude 1.0); the same cut-off was identified for the secondary outcomes, in the event of a significant primary outcome. All tests were two-sided. There were no adjustments made for multiple comparisons. Heterogeneity across studies was estimated using the I 2 statistic, which measures the percentage of variability across studies that is attributable to heterogeneity rather than chance, and was tested using the Q test with a cut-off for significance of p = 0.1 chosen before the analyses began [13]. For each trial, HRs and 95% CIs were also calculated cumulatively for each year of follow-up, to examine the effect of length of follow-up on the primary cardiovascular outcome.

Pre-defined subgroup analyses were performed to explore the effect of therapy on major cardiovascular events. Common definitions for the subgroups were used across all the trials and the subgroups assessed included sex, age, duration of known diabetes, pre-existing macrovascular disease, pre-existing microvascular disease and baseline HbA1c. Consistency of treatment effects across the subgroups was tested using χ 2 tests of homogeneity. Because no adjustments for multiple comparisons were made for secondary outcomes or for subgroups, these analyses should be regarded as exploratory.

Analyses were carried out using STATA (Release 9.2; Stata Corporation, College Station, TX, USA).

Results

Of 163 studies that were identified from the literature search, only four (ACCORD [9], ADVANCE [10], UKPDS [12] and VADT [11]) satisfied the study eligibility criteria for the meta-analysis (Electronic supplementary material [ESM] Fig. 1). Key characteristics of these four trials are shown in Table 1. Between them they randomised a total of 27,049 participants, the majority of whom were selected on the basis of having type 2 diabetes in conjunction with at least one other risk factor for cardiovascular disease. The ACCORD, ADVANCE and VADT trials allocated participants equally to intensive vs less-intensive glycaemic treatment groups; the UKPDS allocated 70% of its participants to an intensive group and 30% to the less-intensive group. Median participant follow-up ranged from 3.4 years for ACCORD to 5.6 years for VADT. The average duration of follow-up (weighted by study size) was 4.4 years.

Table 1 Key characteristics of trials and length of follow-up
Full size table

Baseline participant characteristics are shown in Table 2. The overall mean age was 62 years and 38% were women. Excluding UKPDS participants, who all had newly diagnosed type 2 diabetes, the median duration of known diabetes was 9 years with over one-third of participants having a history of macrovascular disease. Baseline glycaemia, BP and lipid profiles were similar among the studies, although UKPDS participants were more often smokers and ADVANCE and UKPDS participants had a lower mean BMI than did participants in the other studies. At the last follow-up visit, participants allocated to more-intensive glycaemic control were taking more glucose-lowering therapies but the proportions taking other risk-factor treatments did not differ between randomised groups, although there were differences among the trials (Table 3).

Table 2 Participant characteristics at baseline
Full size table
Table 3 Glucose-lowering and cardioprotective therapies at follow-upa
Full size table

Table 4 shows the mean differences achieved in glycaemic control and the mean differences observed in other major risk factors between randomised groups at the last clinic visit. The overall weighted mean HbA1c and fasting plasma glucose differences between those allocated to more- compared with less-intensive glycaemic control were 0.88 percentage points (Table 5) and 1.53 mmol/l (data not shown), respectively.

Table 4 Mean differencesa in glycaemic levels and other risk factors at follow-upb between groups randomised to more- or less-intensive glucose control
Full size table
Table 5 Effects of more- vs less-intensive glycaemic control on severe hypoglycaemia
Full size table

Clinical outcomes

A total of 2,370 major cardiovascular events (1,194/14,320 in the more-intensive group, 1,176/12,729 in the less-intensive group) contributed to the meta-analyses. The risk of a major cardiovascular event was reduced by 9% (Fig. 1) in those allocated to more-intensive compared with less-intensive glycaemic control (HR 0.91, 95% CI 0.84–0.99) with no evidence of heterogeneity among trials (p = 0.72). The cumulative HRs by year of follow-up (ESM Table 1) did not differ appreciably and showed no systematic trend.

Fig. 1
figure 1

Effects of more- vs less-intensive glycaemic control on major cardiovascular events (cardiovascular death or non-fatal stroke or non-fatal myocardial infarction), stroke (fatal or non-fatal), myocardial infarction (fatal or non-fatal) and heart failure resulting in hospitalisation or death. The diamond incorporates the point estimate, represented by the vertical dashed line, and the 95% CI of the overall effect for each outcome. The HRs are given for more-intensive compared with less-intensive glucose control. ΔHbA1c = mean HbA1c of more-intensive group minus mean HbA1c of less-intensive group. UKPDS follow-up truncated at 5 years from the time of randomisation

Full size image

The risk of non-fatal/fatal myocardial infarction was reduced by 15% (Fig. 1) in those allocated to more-intensive compared with less-intensive glycaemic control (HR 0.85, 95% CI 0.76–0.94), with a non-significant reduction in the risk of non-fatal/fatal stroke (HR 0.96, 95% CI 0.83–1.10) and no difference for hospitalised fatal heart failure (HR 1.00, 95% CI 0.86–1.16).

A total of 1,864 participants died. The HR for all-cause mortality among randomised groups was 1.04 (95% CI 0.90–1.20) (Fig. 2). Although the Q test for heterogeneity did not reach statistical significance (p = 0.13), almost 50% of variability across studies was estimated to be attributable to heterogeneity rather than chance (I 2 = 47.5%). Cause of death could be categorised as cardiovascular or non-cardiovascular in all but 18 (0.1%) participants. The estimated HR for cardiovascular death varied among trials (p for heterogeneity = 0.04) but the overall estimate did not differ significantly from unity (HR 1.10, 95% CI 0.84–1.42). Re-analysis using the fixed-effect model did not alter the results (ESM Table 2).

Fig. 2
figure 2

Effects of more- vs less-intensive glycaemic control on all-cause mortality, cardiovascular death and non-cardiovascular death. Conventions as for Fig. 1

Full size image

Effects on severe hypoglycaemia

Overall, there were 1,443 events (1,071 in the more-intensive group, 372 in the less-intensive group). Allocation to more-intensive glycaemic control was associated with a more than doubling in the risk of severe hypoglycaemia (HR 2.48, 95% CI 1.91–3.21) and there was significant heterogeneity among the trials (Q = 10.74, p = 0.01, I 2 = 72.1%) (Table 5).

Subgroup analyses

The effect of more-intensive glycaemic control on major cardiovascular events was consistent across pre-specified participant subgroups (Fig. 3) with one exception: participants who did not have a history of macrovascular disease prior to randomisation appeared to benefit from more-intensive glycaemic control, whereas those with a history of a macrovascular disease did not appear to benefit (test for homogeneity p = 0.04). The UKPDS only enrolled participants with newly diagnosed diabetes and therefore this trial contributed data to one subgroup only for the analyses of ‘duration of diabetes’ and very few to the subgroups for ‘history of macrovascular disease’ and ‘history of microvascular disease’. Sensitivity analyses excluding UKPDS data showed no material difference in the subgroup findings.

Fig. 3
figure 3

Effects of more- vs less-intensive glycaemic control on major cardiovascular events for pre-specified participant subgroups. The diamond incorporates the point estimate and the 95% CI of the overall effect for each outcome in each participant subgroup. The hazard ratios are given for more-intensive compared with less-intensive glucose control. ACCORD data were unavailable for the ‘duration of diabetes’ subgroup analyses. UKPDS follow-up truncated at 5 years from the time of randomisation

Full size image

Discussion

This meta-analysis of four large, randomised controlled trials of more- vs less-intensive glycaemic control in people with type 2 diabetes demonstrates a modest reduction in major macrovascular events with greater glucose lowering. Overall, intensive glycaemic control reduced the final visit HbA1c by a mean of 0.88 percentage points more than less-intensive glycaemic control, with an associated 9% (95% CI 1–16%) RR reduction for the composite major cardiovascular outcome of cardiovascular death or non-fatal stroke or non-fatal myocardial infarction during an average follow-up of 4.4 years. For fatal/non-fatal myocardial infarction alone, the RR reduction was 15% (95% CI 6–24%). The magnitude of these macrovascular risk reductions are consistent with the epidemiological relationship between HbA1c and cardiovascular events reported from observational studies in persons with diabetes [14–17].

Meta-analysis of the other secondary endpoints showed no significant overall effect on the risk of fatal/non-fatal stroke, hospitalised or fatal congestive heart failure or all-cause mortality. No significant effect was seen on cardiovascular death, although there was a 10% trend for an RR increase (HR 1.10, 95% CI 0.84–1.42), including point estimates of 1.35 and 1.32 for the ACCORD and VADT trials, respectively. These were the two trials that achieved and maintained the greatest differences in HbA1c. The significant heterogeneity among the four studies suggests that the possibility of harm with more-intensive glycaemic treatment cannot be ruled out.

Exploratory subgroup analyses of the impact of more-intensive glycaemic control on the composite major cardiovascular outcome showed no significant differences with respect to sex, age, initial HbA1c, duration of known diabetes or history of microvascular disease. Although of borderline significance, there was a suggestion that participants with no history of macrovascular disease achieved benefit, whereas those with prior macrovascular disease did not.

This meta-analysis also shows that allocation to a more-intensive glycaemic control regimen is associated with a more than twofold risk of major hypoglycaemia and that many glucose-lowering medications are generally required to achieve lower glycaemic targets. Whether these factors play a role in the effect of glycaemic control on cardiovascular outcomes clearly requires further investigation. Nevertheless, in conjunction with other reported benefits of glycaemic control in patients with type 2 diabetes [12, 18] the results presented here suggest some cardiovascular benefit for people with diabetes. This does not preclude the possibility that the balance of risks and benefits may vary for different patient groups. Indeed, evidence of statistical heterogeneity with respect to cardiovascular death among the trials (with the highest and lowest point estimates for the HR occurring in the ACCORD and ADVANCE studies, respectively), and the benefit on the composite major cardiovascular outcome in participants without, but not in participants with, prior macrovascular disease, suggest that either patient characteristics, the approach to glucose lowering or other measured (or unmeasured) variables may affect cardiovascular risk. Avoidance of severe hypoglycaemia in the setting of an intensive glycaemic control regimen, for example, clearly requires a particular set of patient capabilities.

The chief strengths of this analysis include its focus on the key primary and secondary outcomes in the trials, the large size of the trials, the consistency of results when analysed using different approaches, and the collaboration of the original trial investigators to produce data of the highest quality. In particular, it has been possible to ensure that definitions of outcomes and exposures are directly comparable, that analytic techniques are identical across the trials, and that subgroups are defined consistently. Furthermore, sharing of the data among study groups allowed for the independent analysis and confirmation of the results.

The ability to understand the heterogeneity among trials for some outcomes, however, remains limited by the number of trials and the limited power of the subgroup analyses. However, the fact that these trials differed in several ways is apparent in Tables 1, 2, 3, 4 and 5 and some of these differences may account for the observed heterogeneity. For example, the UKPDS was completed 10 years earlier than the other trials and studied patients with newly diagnosed vs established diabetes who were younger, lighter, more likely to be smokers, and on fewer cardioprotective drugs but at lower cardiovascular risk, and achieved a lesser contrast in the final visit HbA1c than the other three trials. Moreover, the four trials differed with respect to their mean on-trial difference in HbA1c (data not shown), the speed of HbA1c lowering, the mean difference in HbA1c reduction from baseline (Table 5), the methods by which this difference was achieved (Table 3), and the incidence rates of severe hypoglycaemia, which were also statistically heterogeneous (Table 5). They also differed with respect to the duration of exposure to the intervention, which may be an important determinant of its risks and benefits. Meta-analyses of trials using individual patient data can further explore these and other differences as possible explanations for the differential effect of intensive glucose control on myocardial infarctions vs cardiovascular death. However, these analyses were not available, as the ACCORD trial is still ongoing. Nevertheless, the exploration of the pooled data made possible by this collaboration of the four studies has generated the best estimate that is available currently of the cardiovascular benefits of more-intensive glycaemic control.

In conclusion, the recent publication of the findings from three major new trials of glucose-lowering and the UKPDS 10 year post-trial follow-up has provided important new insights into the balance of risks and benefits associated with the use of more-intensive glycaemic control in patient with type 2 diabetes. The meta-analysis presented here shows that more-intensive glycaemic control affords a modest but significant cardiovascular benefit in the short-to-medium term, although all-cause and cardiovascular mortality are not benefited. The effect on cardiovascular events is driven primarily by a 15% reduction in the risk of myocardial infarction.