August 12th, 2014

What Can We Really Deduce from ACCORD?

Post hoc analyses of intensive glucose lowering in the ACCORD trial require greater scrutiny.


About 68% of people with diabetes die of heart disease or stroke. Elevated glucose defines diabetes, and markers of chronic hyperglycemia, such as HbA1c, are well-recognized risk factors for heart disease. Therefore, strategies that lower glucose should theoretically reduce the risk for cardiovascular events in people with diabetes. However, three recent randomized controlled trials — ACCORD, ADVANCE, and VADT ­— showed no significant effect of intensive glucose control on the incidence of major cardiovascular events. In addition, ACCORD’s intensive treatment arm was terminated early because it was associated with increased mortality. The relationship between glucose-lowering and cardiovascular outcomes is clearly complex. Specifically, we cannot easily predict how the effects of interventions on surrogate measures such as HbA1c ultimately affect patient outcomes.

The New Analysis of ACCORD

In a new post hoc analysis, published in The Lancet, investigators examined the effects of the ACCORD glucose-lowering interventions on several ischemic heart disease (IHD) outcomes, including fatal and nonfatal myocardial infarction (which, along with stroke and cardiovascular death, constituted ACCORD’s primary endpoint); unstable and new-onset angina; and coronary revascularization. The researchers analyzed various overlapping combinations of these outcomes during treatment transition and during follow-up and found, for example, that the incidence of coronary revascularization, unstable angina, or any MI was significantly lower in the intensive glucose control group than in the standard-therapy group after full follow-up (hazard ratio, 0.87; 95% CI, 0.79­–0.96). Moreover, when the model included HbA1c concentration during active treatment (i.e., before the intensive glucose control arm was terminated) as a time-dependent variable, the hazard ratios for the various IHD outcomes were no longer significantly lower in the intensive-control group.

What ACCORD Ultimately Tells Us

1. Glucose is still a surrogate measure, not an outcome. The finding that adjustment for HbA1c rendered treatment effects nonsignificant is based on observational HbA1c data from participants in the trial. Therefore, participants with persistently elevated HbA1c levels might have had other characteristics that increased their risk for ischemic heart disease.

2. Despite modest reductions in IHD outcomes, the main trial results suggest that the intensive glucose control strategy used in ACCORD is associated with net harm. Chance findings are more likely in post hoc analyses than for prespecified outcomes of the trial.

3. For some patients, the benefits of ACCORD’s intensive strategy might outweigh any harm. However, given inherent uncertainty about who those patients are, I still consider it prudent to avoid ACCORD-like interventions.


Do you agree with Dr. Lipska’s interpretation of the new analyses from ACCORD?

2 Responses to “What Can We Really Deduce from ACCORD?”

  1. Gervasio Antonio Lamas, MD says:

    As a cardiologist who treats patients with diabetes and coronary disease, I have been frustrated that interventions to alter glucose control have not proven more effective in reducing CV outcomes in patients with diabetes.

  2. I strongly suspect that the lack of benefit is in some way related to increased hypoglycemia in those with more intensive control. I would love to see a secondary analysis among those treated exclusively with metformin, GLP-1 agonists, thiazolidinediones, DPP-4 inhibitors, and/or SGLT-2 inhibitors where the probability of hypoglycemia is minimal. Specifically to avoid subjects on insulin and or sulfonylureas where hypoglycemia is more likely a problem with tight glycemic control.

    In my practice, I avoid sulfonylureas and use insulin as a last resort for those with inadequate glycemic control. I have many diabetic patients and have not seen an MI in a diabetic for almost a decade, maybe more.