Effect of intensive control of glucose on cardiovascular outcomes and death in patients with diabetes mellitus: a meta-analysis of randomised controlled trials
The Lancet, Volume 373, Issue 9677, Pages 1765 - 1772, 23 May 2009
Editors' note: Despite great efforts, there is still no consensus over the role of glycaemic control on cardiovascular outcomes and death in diabetic patients. A meta-analysis of 33 000 patients from five prospective randomised controlled trials shows that intensive control of glucose in diabetes results in fewer cardiovascular events (such as myocardial infarction and coronary heart disease) compared to standard levels of glucsoe control. Importantly, the intensive therapy did not increase the risk of death. This independent analysis suggests that lower HBA1c levels (as a result of intensive therapy) may be beneficial for patients with diabetes at risk of cardiovascular complications.
"As expected, a higher proportion of participants on intensive treatment than standard treatment had a hypoglycaemic episode (weighted averages 38·1% vs 28·6%). Overall, severe hypoglycaemia was much less common than was hypoglycaemia, but almost twice as many participants on intensive treatment compared with those on standard treatment had a severe hypoglycaemic event (weighted averages 2·3% vs 1·2%). Participants receiving intensive treatment were a mean of 2·5 kg (SD 1·2) heavier than those on standard treatment by the end of the study."
from Jules: an interesting question would be to look at outcomes patients with low HbA1c of say <6.0 and low sugars with 2 week average of <118, to see their CVD outcomes and rates of severe and less severe hypoglycemia and mortality. Other recent studies hypothesize that increased mortality associated with low HbA1c could be because reducing glucose so much and perhaps so fast might have a negative affect on the heart.
"Intensive glucose control was associated with adverse effects of 2·5 kg difference in weight gain and nearly double severe hypoglycaemic episodes compared with standard treatment. Two studies-ACCORD8 and VADT21, 22-with increased mortality in the intensive treatment group also had patients with the longest duration since diabetes diagnosis at baseline (≥10 years); the highest HbA1c concentration at baseline; and a greater risk of hypoglycaemia. Additionally, the ACCORD study8 had a significantly increased risk of cardiovascular death and non-coronary cardiovascular death."
"In ACCORD,8 HbA1c fell by around 1·5% within 6 months and the average HbA1c was less than 6·0% by 1 year in intensively treated individuals through early and aggressive use of insulin with the use of bolus doses when necessary. Additionally, a greater proportion of intensively treated participants received rosiglitazone at the end of follow-up (92% [n=4677]) compared with those receiving standard treatment (58% [n=2946]).30 By contrast, in ADVANCE5 HbA1c fell by only 0·5% within 6 months and the target HbA1c concentration of 6·5% or less was achieved much more slowly (about 36 months), with much lower use of insulin and with preparations that were long acting. Participants were also encouraged to adopt a favourable lifestyle and were closely monitored for outcomes and adverse events. Although the data presented in our meta-analysis cannot substantiate or refute such mechanistic associations, a practical clinical approach might be to reduce HbA1c concentration steadily with care taken to avoid severe hypoglycaemia. Furthermore, less stringent targets might be appropriate for patients with more advanced disease of longer duration and higher baseline HbA1c concentration.31"
Effect of intensive control of glucose on cardiovascular outcomes and death in patients with diabetes mellitus: a meta-analysis of randomised controlled trials
Dr Kausik K Ray MD a b Corresponding AuthorEmail Address, Sreenivasa Rao Kondapally Seshasai MD a , Shanelle Wijesuriya BA a , Rupa Sivakumaran BA a , Sarah Nethercott BA a , David Preiss MRCP c, Sebhat Erqou MD a, Prof Naveed Sattar FRCPath c
Whether intensive control of glucose reduces macrovascular events and all-cause mortality in individuals with type 2 diabetes mellitus is unclear. We undertook a meta-analysis of randomised controlled trials to determine whether intensive treatment is beneficial.
We selected five prospective randomised controlled trials of 33 040 participants to assess the effect of an intensive glucose-lowering regimen on death and cardiovascular outcomes compared with a standard regimen. We gathered information about events of non-fatal myocardial infarction, coronary heart disease (fatal and non-fatal myocardial infarction), stroke, and all-cause mortality, and did a random-effects meta-analysis to obtain summary effect estimates for the clinical outcomes with use of odds ratios calculated from the raw data of every trial. Statistical heterogeneity across trials was assessed with the χ2 and I2 statistics.
The five trials provided information on 1497 events of non-fatal myocardial infarction, 2318 of coronary heart disease, 1127 of stroke, and 2892 of all-cause mortality during about 163 000 person-years of follow-up. The mean haemoglobin A1c concentration (HbA1c) was 0·9% lower for participants given intensive treatment than for those given standard treatment. Intensive glycaemic control resulted in a 17% reduction in events of non-fatal myocardial infarction (odds ratio 0·83, 95% CI 0·75-0·93), and a 15% reduction in events of coronary heart disease (0·85, 0·77-0·93). Intensive glycaemic control had no significant effect on events of stroke (0·93, 0·81-1·06) or all-cause mortality (1·02, 0·87-1·19).
Overall, intensive compared with standard glycaemic control significantly reduces coronary events without an increased risk of death. However, the optimum mechanism, speed, and extent of HbA1c reduction might be different in differing populations.
Type 2 diabetes mellitus is a well established risk factor for cardiovascular disease. Several observational studies have shown a positive correlation between measures of glycaemic control and both cardiovascular outcomes and microvascular disease, independent of risk factors known to cluster with diabetes.1-3 Consequently, randomised controlled trials have aimed to assess whether more intensive control of glucose reduces long-term clinical events and lengthens lifetime compared with standard treatment. By contrast with the substantial benefits to microvascular outcomes,4, 5 individually these trials have failed to show consistent beneficial effects on cardiovascular events.5-8
Such inconsistent evidence has resulted in the American Heart Association, the American College of Cardiology, and the American Diabetes Association providing a conservative class IIb recommendation with level of evidence A9 for the benefit of glycaemic control on cardiovascular disease. However, individually these trials might have been underpowered to show clinical benefit-especially if event rates were lower than were expected because of improved control of risk factors; duration of treatment was shorter than was needed to show a clinical benefit;10 or differences in glycaemic control between patient groups were too small to show any benefit. To address such uncertainties, we quantitatively assessed whether intensive glucose-lowering treatment in individuals with type 2 diabetes mellitus resulted in a reduction of cardiovascular events and all-cause mortality. We present data from a meta-analysis of randomised controlled clinical trials, which aimed to assess the effect of differential glycaemic control on cardiovascular outcomes.
We searched Medline, Cochrane Central, and EmBase for articles published in English from January, 1970, to January, 2009, with terms related to diabetes and vascular outcomes (eg, "cardiovascular diseases", "diabetes mellitus", "glucose", and "HbA1c"). We restricted the search to randomised controlled trials. This search provided 2439 articles, which were further screened for inclusion from titles, abstracts, or full texts, or a combination of these. We supplemented the electronic search from reference lists of relevant articles including meta-analyses and reviews, and by discussion with experts.
Our predefined inclusion criteria required clinical trials to: (1) randomly assign individuals with type 2 diabetes mellitus either to an intensive lowering of glucose versus a standard regimen (placebo, standard care, or glycaemic control of reduced intensity), with significantly different glycaemic control (measured by haemoglabin A1c [HbA1c]) between patient groups during follow-up; (2) measure outcome with a primary endpoint based on cardiovascular events, and report complete information about effect estimates or provision of information to allow calculation of effect estimates for non-fatal myocardial infarction, coronary heart disease (fatal or non-fatal myocardial infarction), stroke, and all-cause mortality; and (3) be done in stable individuals only, which excluded studies in an acute hospital setting. 16 articles from 11 trials that met the above inclusion criteria were identified with information about cardiovascular outcomes and glycaemic control sourced from the title or abstract, or both, of primary and secondary published articles, and study websites.
Six trials that were initially screened were excluded: ADOPT11 and RECORD12 did not assess cardiovascular outcomes in the primary endpoint,11, 12 and RECORD had only interim data for some of the outcomes of interest without provision of the change in HbA1c concentration during follow-up; DREAM13 was done in individuals with impaired glucose tolerance; UGDP14, 15 included patients with diabetes and impaired glucose tolerance, and did not provide either separate information about those with diabetes or effect estimates for the outcomes of interest in each treatment group; STENO 216 tested several interventions and therefore did not assess intensive glucose control compared with standard treatment; and Kumamato17 reported a composite endpoint of cardiovascular events, including peripheral vascular disease and angina, rather than the individual endpoints of interest (additionally, events included in the primary endpoint of this study were neither definitive nor clearly adjudicated).
Five randomised controlled trials fulfilled our selection criteria and were included in the meta-analysis (table 1). We combined data from the two United Kingdom Prospective Diabetes Study (UKPDS) reports-UKPDS 334 (intensive glucose control with sulphonylureas or insulin compared with usual care) and UKPDS 347 (intensive glucose control with metformin compared with diet therapy in overweight patients)-into one study (UKPDS). The other four studies were the PROspective pioglitAzone Clinical Trial In macroVascular Events (PROactive),18-20 the Action in Diabetes and Vascular Disease:Preterax and Diamicron Modified Release Controlled Evaluation (ADVANCE),5 the Veterans Affairs Diabetes Trial (VADT),21, 22 and the Action to Control Cardiovascular Risk in Diabetes trial (ACCORD).8
We (SW, RS, SN, and DP) gathered information in duplicate using a standardised format from all relevant studies and, where necessary, another investigator (KKR) adjudicated any discrepancies. Information was obtained for several baseline characteristics of the participants (eg, age, HbA1c concentration, blood pressure, body-mass index), the absolute number of events (non-fatal myocardial infarction, coronary heart disease, stroke, and all-cause mortality), and the event rates for both treatment groups. Additional information was obtained about heart failure, type of death, HbA1c during follow-up, and adverse events including hypoglycaemia and weight gain. Event rates were calculated from published information about average duration of follow-up and the number of participants in each randomisation group. Follow-up duration was reported as mean (SD) in PROactive18-20 and ACCORD,8 and median (IQR) in UKPDS,4, 7 ADVANCE,5 and VADT.21, 22 To estimate the total number of person-years of follow-up in UKPDS,4, 7 ADVANCE,5 and VADT,21, 22 the median was assumed to approximate to the arithmetic mean.
Three studies (UKPDS, PROactive, and ADVANCE) provided hazard ratios and 95% CIs for the four main outcomes of interest, whereas two studies (VADT and ACCORD) provided absolute numbers of events. To standardise reporting of our results, odds ratios (ORs) and 95% CIs were calculated from raw data of every trial. We assessed the effect of intensive glucose-lowering versus standard treatment on the outcomes of interest with a random-effects-model meta-analysis, which assumes that the true underlying effect varies between studies. Statistical heterogeneity across trials was assessed with χ2 (p<0·1) and I2 statistics. The I2 statistic is derived from Cochran's Q-ie, χ2 statistic [(Q-df/Q)x100]-and measures the proportion of overall variation that is attributable to between-study heterogeneity. Additionally we assessed the probability of publication bias with funnel plots and the Egger test.
To calculate the absolute rates of every endpoint of interest in the intensive versus standard treatment groups, we divided the absolute number of events by the number of person-years of follow-up. We obtained summary data for every endpoint by combination of rates across studies. Other summary characteristics are presented as mean values weighted by study size. UKPDS 334 and 347 are combined and reported as UKPDS, with use of the random-effects model or calculation of weighted means as appropriate for each analysis. As a sensitivity analysis, odds ratios from the main analysis were compared with corresponding rate ratios in a random-effects-model meta-analysis. All p values are two-sided (p<0·05). Analyses were done with Stata (version 10.1).
Role of the funding source
There was no funding source for this study. The corresponding author had full access to all the data and final responsibility for the decision to submit for publication.
Table 1 shows the study design, baseline demographic characteristics of participants, duration of follow-up, and mean HbA1c concentration in the five randomised controlled trials. The criteria for diagnosis of type 2 diabetes and eligibility for the studies are shown on webappendix p 2. 33 040 participants were enrolled from predominantly western populations (table 1). UKPDS4, 7 enrolled individuals within 1 year after diagnosis, whereas the remaining four studies enrolled participants with longstanding diabetes, diagnosed at least 8 years earlier (table 1). Four studies reported a history of macrovascular disease in 32-100% of participants; in PROactive,18-20 macrovascular disease was a criterion for eligibility. Participants were aged 53-66 years, and more than half were men (table 1). Baseline LDL concentration, systolic blood pressure, and HbA1c concentration are also shown. At follow-up, participants given intensive treatment had a mean HbA1c concentration of 0·9% (95% CI 0·88-0·92) lower than had those given usual treatment (table 1).
We looked at four cardiovascular endpoints from the five trials (definitions webappendix p 2). During about 163 000 person-years of follow-up, we recorded 1497 events of non-fatal myocardial infarction, 2318 of coronary heart disease, 1127 of fatal and non-fatal stroke, and 2892 of deaths from any cause. Table 2 reports the event rates per 1000 person-years of follow-up in the more versus less intensively treated populations in every trial. 2·3 fewer events of myocardial infarction or 2·9 fewer events of coronary heart disease took place for every 200 patients on intensive treatment for 5 years. However, the event rates for stroke and all-cause mortality were not statistically different between the treatment groups.
We assessed the effect of intensive control of glucose versus standard treatment on non-fatal myocardial infarction (figure 1), coronary heart disease (figure 2), stroke (figure 3), and all-cause mortality (figure 4). Intensive glucose-lowering treatment significantly reduced events of non-fatal myocardial infarction by 17% (OR 0·83, 95% CI 0·75-0·93; figure 1) and events of coronary heart disease by 15% (OR 0·85, 0·77-0·93; figure 2); the effect estimate was not heterogeneous between studies for either of these outcomes (non-fatal myocardial infarction I2=0·0%, 95% CI 0·0-69·3, p=0·61; coronary heart disease I2=0·0%, 0·0-52·7, p=0·78). However, intensive treatment did not significantly affect stroke (figure 3) or all-cause mortality (figure 4); the effect estimate was not heterogeneous for stroke (I2=0·0%, 0·0-62·0, p=0·70), but heterogeneity was high for all-cause mortality (I2=58·0%, 0·0-84·4, p=0·049). Rate ratios also showed that compared with standard treatment, intensive treatment significantly reduced non-fatal myocardial infarction and events of coronary heart disease, but not stroke or all-cause mortality (webappendix p 4). Funnel plots did not show a publication bias (webappendix p 5).
Intensive glucose-lowering treatment did not significantly affect heart failure (OR 1·08, 95% CI 0·90-1·31). However pronounced heterogeneity was recorded between studies (I2=62·9%, 95% CI 1·74-85·96, p=0·029), and between subgroups of studies separated by differential glitazone use (PROactive18-20 and ACCORD8vs UKPDS,4, 7 ADVANCE,5 and VADT21, 22), suggesting that glitazone use was associated with an excess risk of heart failure (I2=89·9%, 95% CI 63·05-97·26, p=0·002) (webappendix p 6).
Data for cardiovascular mortality and non-cardiovascular mortality were restricted to four studies of 28 420 participants because UKPDS4, 7 did not record data for these endpoints. Intensive glucose-lowering treatment did not significantly affect the type of death (webappendix p 7). The effect of intensive glucose-lowering treatment on myocardial infarction, coronary heart disease, stroke, and heart failure in this restricted cohort was consistent with the main results (webappendix p 7).
Additionally, we recorded the effect of intensive glucose-lowering on hypoglycaemia and weight gain (webappendix p 3). As expected, a higher proportion of participants on intensive treatment than standard treatment had a hypoglycaemic episode (weighted averages 38·1% vs 28·6%). Overall, severe hypoglycaemia was much less common than was hypoglycaemia, but almost twice as many participants on intensive treatment compared with those on standard treatment had a severe hypoglycaemic event (weighted averages 2·3% vs 1·2%). Participants receiving intensive treatment were a mean of 2·5 kg (SD 1·2) heavier than those on standard treatment by the end of the study.
This meta-analysis of five relevant randomised controlled trials has shown consistently that intensive glucose-lowering treatment has cardiovascular benefit compared with standard treatment for individuals with type 2 diabetes. During about 5 years of treatment, reduction of HbA1c concentration by 0·9% resulted in a significant 17% reduction in events of non-fatal myocardial infarction, a significant 15% reduction in events of coronary heart disease, and a non-significant 7% reduction in events of stroke, with no significant heterogeneity recorded across studies that varied with respect to participant characteristics, baseline HbA1c concentration, or, more importantly, the hypoglycaemic regimens used. However, intensive treatment did not significantly affect all-cause mortality, and we recorded obvious heterogeneity across studies.
In the early 1970s, the UGDP study23 of intensive glycaemic control with sulphonylureas versus usual care suggested an excess mortality with intensive treatment, but with potential benefits of insulin-based regimens. This study was small-about 200 participants per group-and compared patients in several groups receiving different intensive treatments. By contrast, the much larger UKPDS study4 compared intensive and standard glycaemic control, but failed to show benefit to cardiovascular outcomes. However, a small subgroup of 753 overweight individuals randomly assigned to metformin versus usual care showed a clinical benefit of intensive glucose control.7 Post-hoc observational data from UKPDS3 suggested that for every 1% reduction in HbA1c concentration, risk of myocardial infarction was reduced by 14%. More recently, extension of the initial randomised groups in the UKPDS study10 has shown a reduction in myocardial infarction and all-cause mortality with both metformin and sulphonylurea-insulin regimens. This result was achieved despite the fact that HbA1c concentrations were similar during the extension phase,10 suggesting that these initial studies were underpowered to assess the effect of intensive treatment on cardiovascular outcomes.
Two large studies have suggested that significant differences in HbA1c concentration might not confer substantial benefits to macrovascular events.5, 8 Furthermore, the ACCORD trial8 suggested that lowered HbA1c concentration might cause an excess risk of all-cause mortality. By contrast, an earlier meta-analysis24 of data from UKPDS4, 7 and two small studies17, 25 of 60 cardiovascular events suggested that lowered HbA1c concentration caused a 19% reduction in the combined endpoint of acute and non-acute cardiovascular events that included revascularisation. The absence of convincing data and concerns about possible harm has led consensus groups to provide a conservative endorsement (class IIb recommendation, level of evidence A) for the cardiovascular benefits of intensive glycaemic control: "usefulness and efficacy are less well established by evidence or opinion, with data derived from multiple randomized clinical trials or meta-analyses".9
Our quantitative analysis of randomised controlled trials provides reliable large-scale evidence of a consistent beneficial effect of intensive treatment on non-fatal myocardial infarction and coronary heart disease, without increased risk of all-cause mortality. The reduction of myocardial infarction from a decrease in HbA1c concentration of 0·9% is broadly consistent with observational data from the UKPDS study.3 We recorded a non-significant benefit for stroke, but 370 fewer events of stroke than myocardial infarction were reported, which conferred less power to ascertain whether a significant benefit exists.
The implications and context of these findings with respect to public health policy merit careful consideration in view of the established benefits of intensive glucose control for microvascular disease. Evidence is well established that in individuals with diabetes, statin treatment and intensive blood pressure control reduce both macrovascular events and, by contrast with our findings, all-cause mortality (9% and 27%, respectively).26-29 Despite the benefits of statin treatment and blood pressure control, individuals with diabetes have a heightened risk of cardiovascular events, and the rate of events is even higher for those with diabetes and existing cardiovascular disease.
Our analysis shows that the mean (weighted) mortality rate of participants on standard treatment is 18·6 per 1000 person-years of follow-up, and those who achieve a 0·9% reduction in mean HbA1c concentration over 5 years (from a mean HbA1c concentration of 7·8% at baseline) have about two events of non-fatal myocardial infarction or three of coronary heart disease fewer for about every 200 individuals treated for 5 years. These estimates correspond to a number needed to treat over 5 years of 87 and 69, respectively. This benefit is much more modest than is that from a per mmol/L reduction in LDL cholesterol or from a 4 mm Hg lower blood pressure (8·2 and 12·5 events of cardiovascular disease prevented, respectively).26, 29 In view of the burden of cardiovascular risk in individuals with type 2 diabetes, a general approach to cardiovascular risk that uses several interventions, including stricter glycaemic control, is warranted.16
Intensive glucose control was associated with adverse effects of 2·5 kg difference in weight gain and nearly double severe hypoglycaemic episodes compared with standard treatment. Two studies-ACCORD8 and VADT21, 22-with increased mortality in the intensive treatment group also had patients with the longest duration since diabetes diagnosis at baseline (≥10 years); the highest HbA1c concentration at baseline; and a greater risk of hypoglycaemia. Additionally, the ACCORD study8 had a significantly increased risk of cardiovascular death and non-coronary cardiovascular death.
In ACCORD,8 HbA1c fell by around 1·5% within 6 months and the average HbA1c was less than 6·0% by 1 year in intensively treated individuals through early and aggressive use of insulin with the use of bolus doses when necessary. Additionally, a greater proportion of intensively treated participants received rosiglitazone at the end of follow-up (92% [n=4677]) compared with those receiving standard treatment (58% [n=2946]).30 By contrast, in ADVANCE5 HbA1c fell by only 0·5% within 6 months and the target HbA1c concentration of 6·5% or less was achieved much more slowly (about 36 months), with much lower use of insulin and with preparations that were long acting. Participants were also encouraged to adopt a favourable lifestyle and were closely monitored for outcomes and adverse events. Although the data presented in our meta-analysis cannot substantiate or refute such mechanistic associations, a practical clinical approach might be to reduce HbA1c concentration steadily with care taken to avoid severe hypoglycaemia. Furthermore, less stringent targets might be appropriate for patients with more advanced disease of longer duration and higher baseline HbA1c concentration.31
Our study has several potential limitations. First, meta-analysis is retrospective research that is affected by the methodological rigour of the studies included, comprehensiveness of search strategies, and possibility of publication bias. We tried to keep the probability of bias to a minimum by developing a detailed protocol a priori, doing a thorough search for published and unpublished data, and using explicit criteria for study selection, data collection, and data analysis. Therefore, some notable studies were not eligible for our meta-analysis for legitimate reasons. We believe that we have been robust in our approach and that the results and conclusions can therefore provide reliable recommendations for clinical practice.
Second, as in other meta-analyses, these results should be interpreted with caution because individual studies varied greatly with respect to the demographic characteristics of participants, duration of follow-up, and drugs used for intensive glucose control. Therefore, our report can provide information only about whether intensive glucose-lowering treatment is safe and effective for reduction of macrovascular events compared with standard treatment. The study cannot provide evidence of superiority or harm of a specific glucose-lowering regimen, but we did not record any significant heterogeneity across studies with respect to the effects of different glucose-lowering regimens on non-fatal myocardial infarction, coronary heart disease, or stroke. Combination of such data with the vastly different ancillary metabolic effects of the range of glucose-lowering regimens (eg, metformin, sulphonylureas, insulin, and glitazones) used in the five trials, suggests that the common action to lower glucose is to at least partly bring about the reported benefits in the reduction of the risk of cardiovascular events. Although we did not see an effect on all-cause mortality, significant heterogeneity was recorded across studies, which could not be further clarified without access to individual participant data.
Third, sufficient data were not available to analyse the effects of intensive glycaemic control within various patient subgroups (eg, by age, men vs women, duration of diabetes, baseline HbA1c, prevalence of cardiovasular disease at baseline, comorbitity). Such analyses are most informative when done with individual participant data, to which we did not have access, and similar approaches (adjusted for the same confounders in each study) to establish whether the magnitude of reduction of HbA1c concentration is correlated with cardiovascular events and all-cause mortality. Therefore, our findings will help to encourage pooling of individual participant data into a database, analogous to that of blood pressure and cholesterol, which have proved highly informative.
Fourth, we used odds ratios rather than hazard ratios (which were available in only a proportion of studies), to enable data for all endpoints from all five trials to be incorporated, thus maximising the available data. In sensitivity analyses we did a random-effects-model meta-analysis with rate ratios to calculate effect estimates, which were of similar magnitude to the odds ratio. However, in three studies, we had to assume that the median number of person-years of follow-up was approximate to the arithmetic mean. In variables with a skewed continuous distribution such as follow-up duration, the median is usually not a good approximation of the mean.
Our findings provide reassurance about the effectiveness of glycaemic control for cardiovascular risk reduction, but we have not proven a clear benefit to all-cause mortality. By contrast, strong evidence suggests that lipid-lowering treatment and blood pressure reduction does benefit all-cause mortality reduction, which reinforces the crucial importance of these treatments to reduce cardiovascular events and all-cause mortality in individuals with type 2 diabetes. The optimum methods to achieve glycaemic control need to be established, and guidelines drawn up with specific recommendations for reduction of HbA1c concentration in a range of patient populations.