Unraveling the Mystery of the ACCORD Study


When Savannah Mitchell was younger, she often heard her parents talking about their “sugars,” or being very thirsty, but she never fully understood what diabetes was, or what the risks were to her as the daughter of two type 2 diabetics.  And when her husband, suffering from untreated diabetes, had his leg amputated, she did not understand the cause, or what he could have done differently.  So when in light of her own symptoms of diabetes, her family doctor asked her about participating in a national trial that would teach her about the risks and proper treatment of type 2 diabetes, she jumped at the chance.

Mitchell was enrolled by her doctors in the nationwide study of best practices in diabetes, Action to Control Cardiovascular Risk in Diabetes (ACCORD). The study, funded by the National Institutes of Health (NIH), looked closely at a number of different possible ways of reducing cardiovascular risk among cohorts of type 2 diabetics.  In 2008, this study created a stir when the NIH prematurely halted part of it due to an unexpected number of deaths among patients receiving intense therapy to lower their blood sugar levels.  Since then, researchers have been trying to understand the cause behind the unexpected deaths, and at the American Diabetes Association Scientific Sessions in June of this year, further results of the five-year  ACCORD study were released.  For participants like Mitchell, these further results are especially relevant details about the expected—and the unexpected—outcomes of the trial.

The participants in this study were split up into three groups, each of which consisted of one control subset and one intensive care subset. In one group, the intensive treatment focused on lowering blood pressure; in the second group, the treatment aimed to lower cholesterol; and in the third, the treatment was designed to help patients reduce their blood glucose as measured by their HbA1c levels. Each treatment arm was evaluated first in terms of the likelihood of cardiovascular death, non-fatal heart attack, or non-fatal stroke, and second in terms of the rate of hospitalization, total mortality, and measures of microvascular health.

Recruitment for the ACCORD trial began in January 2001, and continued through October of 2005; each participant would be followed for five years during which he would receive either intensive treatment according to the selected cohort, or standard tracking and measurement within one of the control cohorts. Participants were selected for having high base cardiovascular risk levels, and collectively had a mean age of 62 and a mean Body Mass Index (BMI) of 32.2.

Mitchell was selected for the HbA1c cohort, and though each different treatment group revealed important data about cardiovascular risk factors, the results of the HbA1c group proved particularly interesting in that they seemed to prevent some counterexamples to the general perception that lower overall blood glucose levels lead to reduced cardiovascular risk and reduced overall risk of death. Mitchell’s experience in the non-intensive cohort was fairly straightforward—she recalls being told to lower her HbA1c to 7.0%, and she was taught about the important lifestyle changes that would help her achieve this goal. Adjusting to a new eating regimen was daunting at first, but with the help of her doctor and family, she began to change her habits and lose excess weight.  Her doctor also taught her how to measure her blood sugar and inject insulin, and she began to give herself shots of Novalog to supplement the new diet and exercise.

Looking from the top down, however, the picture was not so clear. Patients like Mitchell seemed to respond well to the treatment, but others did not, and physicians saw an unexpected increase in the number of hospitalizations and adverse events with some of the patients in the intensive subset of the HbA1c cohort. And it was this intensive treatment in the glycemic arm of the study that was stopped for all patients in March of 2008, an average of 1.4 years early for each patient in the trial, because there was a 22% increase in the number of deaths for participants in the intensive glycemic care cohorts over the participants in the control cohorts.

Clearly, a closer look at the results of this arm of the study was required, and a number of explanations and proposals have since been offered, many of which focus on the possibility of extreme hypoglycemia regardless of HbA1c measurements. However, to Matthew Riddle of Ohio State University and John Lachin of George Washington University, these explanations do not prove satisfying, and so at the ADA Scientific Sessions on June 28th, they presented two alternate explanations of the unexpected results.

The results were not entirely misaligned with expectations. In fact, in the control cohort of the glycemic arm of the study, higher measured HbA1c over the five years of study was directly correlated with a higher risk of mortality or major cardiovascular incident. So for patients like Mitchell, who successfully adapted their habits and saw lower HbA1c levels, health improved and morbidity rates were lower. This finding is exactly in line with other studies that have demonstrated the risks to the heart and cardiovascular system of extended periods of elevated blood glucose levels.

The anomalous findings were within the intensive care cohort. While the control cohort had regular checkups and received the standard treatment for type 2 diabetics aiming for HbA1c levels of less than 8.0%, the intensive care cohort received more frequent checkups and earlier intervention with insulin with the aim of achieving an HbA1c not just below 8.0%, but below 7.0%. It was hoped that the relatively low desired HbA1c level, a full percent beneath the standard care level, would demonstrate a substantive decrease in cardiovascular risk for the selected cohort of patients.

It was an unwelcome surprise, therefore, when the intensive care cohort saw a higher death rate than the control cohort beginning in the second year after the recruitment started, and even more so when the death rate jumped up in the third year. The fourth year saw the two cohorts’ death rates return to parity, but the fifth year’s rate of deaths was comparable to the third year’s. When the rate seemed to hold steady there, the trial was suspended in order to avoid subjecting participants to a potentially dangerous treatment regimen.

Researchers were then faced with a complicated quandary; why would the lowering of HbA1c reduce death risk for the control group and for participants in earlier studies, but increase death risk for the intensive care cohort of the ACCORD trial? What had gone wrong when patients attempted to bring down their HbA1c levels to below 7.0%?

The first possible explanation was that severe hypoglycemic incidents were the cause of the increase in the death rate; perhaps the participants were reaching dangerously low blood glucose levels in the course of lowering their HbA1c. Looking closely at the data, Riddle and his researchers found that indeed, the intensive care participants had about three times as many hypoglycemic incidents requiring medical intervention as the control participants.  Very few of the deaths, however, were actually caused by a hypoglycemic incident; only 1% of the deaths that occurred could be attributed to hypoglycemic events, so though the rate of severe hypoglycemia was higher, there was no evidence that this was a direct cause of the increased rate of mortality.

Further, when each cohort was broken down even more, researchers found that the increased rate of hypoglycemia in the intensive care cohort fell primarily on the shoulders of participants within the cohort who had higher HbA1c levels. In other words, the participants who experienced the majority of the severe hypoglycemic incidents were those struggling to even reach the desired HbA1c of below 7.0%. Likewise, in the control group, those with higher HbA1cs levels proved to have a greater rate of hypoglycemic incidents.

This observation led scientists to consider that perhaps the problem was not reaching or maintaining an HbA1c below 7.0%; perhaps the problem was the rapid decrease of HbA1c for patients starting with higher values and too quickly plummeting downwards, causing severe hypoglycemia and other physical damage. Upon closer investigation, this possibility also proved unlikely. The death rate of the intensive care unit did not begin to rise above that of the control group until after the first year, and the deaths in question did not co-occur with the rapid reduction in HbA1c. The deaths, the researchers concluded, therefore were unlikely to be directly caused by hypoglycemia or any other damage resulting from rapid HbA1c adjustment.

Additionally, as it turned out, not only did those who maintained lower HbA1c levels have fewer severe hypoglycemic incidents, those who reached a lower HbA1c faster had fewer incidents of hypoglycemia than those who took longer to reach or could never reach the target range. In fact, looking again at the death rates within specific subgroups, Riddle found that a patient in the intensive care cohort who maintained an HbA1c of 8.0% was 66% more likely to die in a given period than a patient who maintained an HbA1c of 7.0%.

Contrary to the initial interpretation, then, it seems that the problem is not that the intensive care cohort had HbA1c levels lower than 7.0%, or even that the intensive care cohort went through the process of lowering their HbA1c levels to a relatively tight target range. Rather, the problem seems to be that a subset of the intensive care cohort was unable to successfully lower HbA1c values, even in the face of intensive care.

Unfortunately, this interesting revelation is not a final answer so much as an introduction to new questions. Why might intensive care participants who could not maintain HbA1c levels lower than 7.0% have greatly increased death rates? Are the problems behavioral? A factor of complex drug interactions? A result of the degree to which these participants were able to successfully lose weight? These questions will have to be investigated further, but the process will likely be arduous. Untangling the possible vectors of drug interaction is very difficult given the variability in number, dosage, and unknown external interactive elements.  And tracking the effects of weight gain or loss is complicated by the fact that weight data was not collected in a systematic manner for all participants.

As if this interpretive task weren’t hard enough on those counts, John Lachin had one last monkey wrench to throw in. He agreed that the problem was not severe hypoglycemia, and was not rapid lowering of HbA1c. He even recognized the possibility that the problem was drug interactions or weight or some other unknown factor. But he also proposed an alternate solution: there was no problem to begin with. The intensive care cohort did not experience an increased rate of death at all; rather, there was a certain chance that the measured death rate was not due to any systematic differences between the two groups, but rather just random variation. In other words, as with any statistical prediction, the reported 22% increase in the rate of death in the intensive control cohort could have just been the random chance that in that sample of people, 22% more happened to die, and their deaths were completely unrelated to the treatment they received.

And, to supplement his claim of statistical instability, Lachin pointed out the curious nature of the variation in the death rate over time. Why, he asked, would the death rate in the intensive care cohort match the control cohort in year one, grow in years two and three, snap back together in year four, and then vary again in year five? The pattern of variation before the trial was ended, neither predictable nor regular, implies that perhaps the increase in death rate is not a reliable result of a cause-and-effect relationship, but rather a random statistical wobbling that likely would have evened out over time.

What, then, does the ACCORD trial data imply about the relationship between cardiovascular risk and HbA1c? It says what many other studies say—lower HbA1c levels are correlated with reduced cardiovascular risk—and for participants like Savannah Mitchell, that means quite a bit. Since the end of the trial, Mitchell has not followed the news about the results, but she has continued to follow the regimen that she and her doctors had set up. She has continued to eat well, exercise every day, and lose weight, and she has seen these changes reflected in her HbA1c levels. And, what’s more, she feels better, and has been able to explain fully to her own children about diabetes, its risks and complications, and what they can do to prevent getting type 2 diabetes. The ACCORD trial, she concluded, was immensely educational and useful for her—though that analysis, unfortunately for Riddle, Lachin, and the other researchers, still leaves behind very many statistical questions—questions that hopefully will have clearer answers after researchers take some more time to dissect and re-dissect the data from the trial.


Some further reading on Lachin’s analysis:

Lachin JM, Genuth S, Nathan DM, Zinman B, Rutledge BN: The effect of glycemic exposure on the risk of microvascular complications in the Diabetes Control and Complications Trial—revisited. Diabetes 57:995–1001, 2008


Notify of
Inline Feedbacks
View all comments
Copyright © 2009-2021 Diabetes Media Foundation, All Rights Reserved.
ASweetLife™ is a trademark of the Diabetes Media Foundation, All Rights Reserved.
Would love your thoughts, please comment.x