Among ambulatory type 2 diabetes patients receiving care in a well-integrated, managed-care system, we found that hypoglycemia is a significant problem. To our knowledge, this is the largest observational study in the U.S. to assess significant hypoglycemia. As prior studies would suggest40
, the frequency of significant hypoglycemia reported in this population was significantly higher than reported in landmark trials. Annual assistance-requiring hypoglycemia rates range from 0.7 per 100 overall for ADVANCE1
and 5.1% in the standard arm for ACCORD41
. In UKPDS, the highest rate of assistance-requiring hypoglycemia was among insulin users in the intensive arm, and was only 1.8%/year1
. The prevalence of self-reported hypoglycemia relative to emergency department visits or hospitalizations coded for hypoglycemia suggests that (a) the vast majority of events are cared for outside of the medical setting, (b) clinicians and health systems may not fully appreciate the scope of this problem and that (c) most patients are choosing to self-manage any immediate consequences of hypoglycemia without medical assistance. Whether patients are developing and implementing strategies to avoid future hypoglycemia events on their own, or in consultation with their treating clinicians, is not known.
Hypoglycemia, while an important adverse event in and of itself, contributes to poorer diabetes outcomes in other domains. Recent work has shown that acute care visits for hypoglycemia predict future risk for dementia among type 2 diabetes patients42
. Many have suggested that those who experience hypoglycemia may have lower rates of hypoglycemic medication adherence for fear of recurrence43,44
Significant hypoglycemia was especially common among patients reporting limited HL, and this association persisted even after adjusting for known and suspected correlates of HL and hypoglycemia. There are several possible mechanisms by which limited HL could increase hypoglycemia risk. First, limited HL could lead to patient errors in timing or dosage of diabetes medications16,45
. Second, limited HL could interfere with proper adjustment of diabetes medication in response to changing caloric intake or energy expenditure46
. Third, those with limited HL may not correctly interpret and respond to their blood glucose values and trends obtained through self-monitoring, potentially missing opportunities to adapt self-management behaviors so as to prevent significant hypoglycemia47
. Importantly, mastery of such self-management skills depends, in part, on adequate patient-clinician communication. Limited HL has been shown to impede both patients’ understanding of clinician information and instructions, and patients’ reporting of symptoms and disease state to clinicians18
Study strengths include a large, diverse population with uniform access to care, and detailed assessments of HL and hypoglycemic events within a real world setting. Nevertheless, this study has several limitations. First, the KPNC system provides coordinated and uniform access to care, including diabetes education and comprehensive care management. As such, our results may under-estimate hypoglycemic event rates in the under- and uninsured,48
or among populations cared for in non-integrated health systems. Second, because this is a cross-sectional analysis, we cannot infer that limited HL was the cause of increased hypoglycemia risk; although reverse causation is improbable, there may be underlying factors affecting both. Third, we did not directly measure HL, although the self-report HL items we used have been validated against direct measurement instruments29,30
. Moreover, the degree to which these items exactly correlate with HL as assessed by standard, ‘direct’ measurement may be less relevant in practice, insofar as these questions have predictive validity as they appear to identify subgroups at risk for a clinically important adverse event. Fourth, we did not measure numeracy in this study, so we could not address the relative contribution of numeracy versus literacy. Numeracy and literacy have been shown to be highly correlated. Fifth, we could not capture attendance at diabetes education. Sixth, not all participants answered every question in this lengthy survey (184 questions, 52 pages). Nevertheless, this remains the largest observational study of hypoglycemia, and the subjects included in the analysis did not differ from the overall sample in terms of age, race/ ethnicity, educational attainment, which are the best predictors of health literacy, or HbA1c, renal function, or medication type, the clearest correlates of hypoglycemia. In addition, we used multiple imputation to minimize the bias associated with missing items36,38,39,49
. Moreover, results using multiply imputed data did not differ from the complete data-only approach. Seventh, we could not explore individual medications within each class. Finally, consistent with prior studies,1,10-13,50
we obtained measures of hypoglycemia via patient self-report, albeit an imperfect measure. Since self-report of mild hypoglycemia may be less accurate, we queried only clinically significant episodes of losing consciousness or requiring treatment assistance. The fact that we observed a stepwise increase in hypoglycemia consistent with the established pharmacologic risks of hypoglycemia from different medication regimens (i.e., highest in insulin-treated patients) suggests the validity of patients’ reports. We also found that self-report of hypoglycemia was strongly associated with likelihood of making an emergency room visit or requiring hospitalization for hypoglycemia during the year prior to the survey, providing additional validation of this self-reported measure. Furthermore, while it is possible that those with limited HL were more likely to misattribute non-hypoglycemic events to hypoglycemia, the results of our sensitivity analysis that involved only those participants who demonstrated accurate knowledge of hypoglycemia symptoms and treatment suggest that outcome misclassification does not explain our findings. It is theoretically possible that those with limited HL reported more significant hypoglycemia because they may require help for hypoglycemic episodes while those with adequate HL can manage themselves. Such a heightened need for assistance would, in and of itself, place them at greater risk.
Our findings have several clinical and health policy implications. First, while it is unclear whether treating clinicians’ actions (such as intensifying medications or inadequate communication of safety precautions) contributed to these events or whether treating clinicians were aware of patients’ hypoglycemic episodes, our results suggest that hypoglycemic symptoms should be routinely assessed in patients with type 2 diabetes receiving hyperglycemia-lowering agents. Second, because insulin use confers a particularly high risk for hypoglycemia, we recommend formal hypoglycemia prevention counseling that is appropriate for those with limited health literacy, as well as more vigilant self-management support that includes training family members and caregivers, at the time of insulin initiation. More research is needed to (a) identify whether errors in patient self-management are contributing to hypoglycemia, and (b) determine what actions by patients, families, caregivers and clinicians can reduce hypoglycemia. Third, hypoglycemia risk profile should be an important criteria when choosing medication regimen, favoring medications which are least likely to cause hypoglycemia.
Hypoglycemia has been described as a sentinel adverse event7
in type 2 diabetes that should trigger systems-based approach to reduce recurrences. System-level surveillance for hypoglycemia, perhaps through use of patient-directed health information technology, may promote patient safety. Between-visit surveillance mechanisms, utilizing technologies such as interactive voice response technology,30,5,52
personal health internet portals,53-55
or continuous blood glucose monitors56,57
may be more sensitive in capturing hypoglycemia than are current practices in diabetes management. For example, patients receiving periodic automated telephone self-management support for diabetes could input a low blood glucose value into their telephone touchpad, which would lead to a live telephone call with a care manager as well as notification to the primary care provider. Such technologies have the potential to make use of visual and voice modalities so as to be appropriate for those with limited HL, but this requires further investigation. Finally, from a health policy perspective, our results suggest higher rates of hypoglycemia than previously documented, raising the question of whether widely-recommended clinical targets and performance measures derived from efficacy studies, when applied to real world contexts, may jeopardize the safety of many patients with type 2 diabetes3,4