Safety of very tight blood glucose control in type 2 diabetes

On 8 February 2008, the glucose lowering arm of a large ongoing randomised controlled trial (ACCORD) of people with type 2 diabetes at high risk of cardiovascular events was stopped 18 months early because of concerns about safety. Intensively lowering blood glucose to a target below current recommendations (glycated haemoglobin (HbA_1c) <6.0%) increased the risk of death compared with a less intensive standard treatment strategy (HbA_1c 7.0-7.9%). This amounted to an excess of deaths of 3/1000 participant years on a background control rate of 10/1000 participant years.1

So what do these findings mean for clinical practice? Several evidence based or consensus guidelines in recent years have recommended target values of HbA_1c <6.5% or <7.0%.2 3 Targets of this kind, however, are rarely tested in clinical trials, which usually compare strategies of different intensity rather than treatment to different targets. Accordingly, the evidence used in target setting is usually secondary—it comes from findings embedded within the results of such treatment trials. This evidence is often supplemented by data from observational studies and within study analyses.

For type 2 diabetes, the core data used in target setting come from the United Kingdom prospective diabetes study. This study achieved HbA_1c values around 6.5% for the first five years in both the main glucose control study and the metformin study, with benefits for vascular outcomes of more intensive treatment in the longer term.4 5 In addition, a within study analysis showed that vascular event rates were lower at HbA_1c values as low as 5.5%. In clinical practice, an HbA_1c of around 6.5% can be achieved in many people for a variable number of years. Once insulin has to be started, however, studies that treat to target have struggled to achieve average values much below 7.0%.6 7

This last finding raised the question of whether very strict blood glucose control, using the full range of modern mealtime and basal insulins and multiple oral agents if necessary—aimed at a target HbA_1c value within the normal range (<6.0%)—would achieve better vascular outcomes than a higher target HbA_1c, such as that used in people with type 1 diabetes (<7.5%). The glucose control arm of the ACCORD study was such a study.8 It recruited people with high cardiovascular risk in order to improve the power of the study through a higher background event rate.

At first sight, the death rate in the control arm of the ACCORD study (10/1000 participant years) is astonishingly low—below the background population rate for people of the same age in the UK.1 Study recruitment, however, tends to exclude all people who are likely to die of other disease within the next five years. It may not be safe therefore to dismiss the effect solely on the basis of an inappropriately low death rate in the control group.

No indication has been given as to the cause of the difference in deaths, but two possibilities have been considered. It was announced that the difference was not caused by hypoglycaemia (which, if anything, might be expected to result in an excess of sudden death as a result of cardiac dysrhythmia) or the use of a particular drug, including rosiglitazone.1 Indeed, with an excess of 54 deaths on a background of 203 deaths it may not be possible to ascertain any useful pointers. This leaves us with the intervention itself as the possible culprit—the highly intensive treatment (often multiple insulin injection regimens combined with multiple oral agents) aimed at reaching the target of HbA_1c <6.0%.

In some ways, the possibility that very intensive treatment caused the increase in deaths in ACCORD is supported by the press release on the ADVANCE study, a study of similar size but of less intensity, which did not have an increased death rate in its more intensive group.9 Despite achieving comparable blood glucose control as ACCORD, only a small proportion (around 30%) of people in the ADVANCE intensive arm were using insulin when the blood pressure arm of the study was reported last year.10 Because the study is complete (but the database has not been locked for final analysis), the lack of excess deaths in the intensive group (average HbA_1c around 6.5%) means that the rate is either similar to that in the control group or better.

The results from 13 years of follow-up of the multi-intervention Steno 2 study were also reported in February.11 Glucose control in the intensive intervention arm was not so good as in ACCORD and ADVANCE, although by the end of the study, target attainment was not much worse than for systolic blood pressure. Nevertheless, the absolute reductions in mortality, cardiovascular events, and microvascular progression are extraordinary—between 15% and 30%.

What should we conclude pending full publication of the ACCORD and ADVANCE studies? It seems that moderately intensive management to targets of HbA_1c <6.5% or lower—if easily attained—need not be abandoned. Meanwhile it would be wise to avoid highly intensive management that combines multiple insulin injection regimens with multiple oral agents.

The problem of increased deaths may in some way be linked to higher doses of injected insulin in combination with stimulation of endogenous insulin secretion (by sulfonylureas) or use of insulin sensitisers (metformin and thiazolidinediones), or both. If insensitivity to insulin is actually a protective mechanism, rather than the pathological outcome of overeating as it is perceived today, then perhaps trying aggressively to overcome it may have adverse cellular effects that we have not yet begun to understand.