Our analysis of hypoglycemic events in the ACCORD trial shows that a large portion of events are preceded by behavioral-related situations that are potentially modifiable or preventable. Over half of all hypoglycemia episodes in both the standard and intensive group were related to either delaying or missing a meal or eating less food than normal. Additional behaviors that preceded hypoglycemic events included increased physical activity and incorrect medication use although incorrect medication use was listed as an antecedent (1-8% depending on medication) much less frequently than diet related behaviors. At the time of reporting severe hypoglycemic events, participants and staff were also asked about changes in medications that were not used to treat diabetes but known to either lower blood glucose or mask the symptoms of low blood glucose (e.g. beta blockers or ACE inhibitors). While there was differential reporting of these as antecedents between the intensive interventional study group (2%) and the standard interventional study group (8%), overall, this was infrequently listed as an antecedent in comparison to behavioral factors.
The consequences of severe hypoglycemia were frequently clinically significant. A large number of the hypoglycemic events reported in this paper resulted in hospitalization. This is in part due to the definition of severe hypoglycemia used in the ACCORD study (low blood glucose and treatment by medical personnel). Importantly, approximately 25% of the severe hypoglycemic incidents reported by the ACCORD participants resulted in loss of consciousness, about a third in confusion/irrational behavior, and 6% resulted in a personal accident or injury. Also of note, and perhaps associated with the more severe complications, the frequency of the neuroglycopenic symptoms of confusion/disorientation was equal to the adrenergic sweaty and shaky symptoms [12
]. This combination of potentially dangerous consequences combined with the high frequency of preventable antecedents illustrates the importance of appropriate education of patients with diabetes on techniques to prevent hypoglycemia.
More events occurred among those who were prescribed insulin. This finding has been reported in other papers [13
]. This may be due to the action of insulin or due to the clinical characteristics of participants for whom insulin is prescribed or both. ACCORD was a clinical trial testing strategies for glycemic control and a variety of medications were available for use to obtain the target A1Cs in each arm of the trial. The recommended study algorithm of treatments suggested the use of oral medications before addition of insulin. Thus, participants who were prescribed insulin may have represented those participants whose disease was more difficult to control. They may also represent participants who had contraindications, such as impaired renal function, to some oral agents.
As expected, we found a low rate of severe hypoglycemic events associated with the biguanide class. This has also been reported in other papers [13
]. In contrast, we unexpectedly found high rates of hypoglycemia among those participants treated with TZD. Few participants were prescribed TZDs alone so we cannot determine the rate among those who used this medication alone. When we examined the use of TZD in conjunction with sulfonylureas and insulin (two of the most commonly used medications in ACCORD), we found the highest rates of hypoglycemia among those who were prescribed TZDs in combination with insulin as compared to the combination of TZD with sulfonylureas. Among those participants who were prescribed TZD but not sulfonylurea or insulin, the incidence rate of severe hypoglycemia was similar to that reported in a prescription monitoring program [16
]. Although the mechanism for this finding is not clear, it is possible that the insulin sensitizing action of the TZD accentuates the risk of hypoglycemia of insulin when these two are used in combination. Alternatively, rosiglitizone has been shown to preserve [17
] and possibly improve pancreatic beta cells function [18
]. This preservation/improvement in the responsiveness of the beta cells in combination with exogenous insulin and improved insulin sensitivity could lead to increased risk of hypoglycemia.
There are several strengths to this analysis. We collected a large amount of data about symptoms, antecedents, and consequences on all severe hypoglycemic events. However, we did not obtain similar information in those participants that did not experience hypoglycemia and cannot determine the risk of hypoglycemia with the various antecedents. ACCORD had a large study population and used a number of common diabetes medications allowing us to examine the rate of hypoglycemia in a variety of medications and combinations. However, the study tested strategies, and participants were not randomly assigned to particular medications. Thus, this analysis can only provide associations. We focused on first events in this paper; when we conducted similar analysis on symptoms, antecedents and consequences of the second reported severe hypoglycemic event, we found no significant difference between study arms.
There are some limitations that should be discussed. There is considerable confounding between participant characteristics and the prescription of certain medications. While we have attempted to account for this in our analysis, it is likely there is still some residual confounding. ACCORD did not capture actual medication use. This analysis instead utilizes prescription of medication. It is possible that some medications prescribed were not actually taken by the study participant. Symptoms, antecedents and consequences of the hypoglycemic events rely to some degree on patient self-report and thus may be subject to under or over-reporting.