The article by Novodvorsky et al. (1) claims to show an incident rate ratio for bradycardia during periods of nocturnal hypoglycemia versus euglycemia of 6.44 (95% CI 6.26, 6.63). However, with only 37 people studied, and only a proportion of those having hypoglycemia and/or dysrhythmias, it would not be possible for such tight confidence limits to be anywhere near valid. A much larger study of longer duration would be required for this.
Further, it was not clear to us what proportions of the participants were included in the analysis as having hypoglycemia data with paired Holter electrocardiogram data and perhaps matching “euglycemia” data. Although the number of people and hours with dysrhythmias were stated, the total numbers of such people/hours and the numbers without dysrhythmias were not shown. We subsequently communicated with the authors, who responded helpfully and provided us with a complete version of the article’s Supplementary Table 1, from which the missing numbers of people/hours could be obtained.
In summary, their nocturnal bradycardia data (the authors use beats of bradycardia rather than time in bradycardia) included:
4 h of hypoglycemia in 2 participants with 4,742 beats total
54 h of hypoglycemia in 35 participants, all with no beats
47 h of euglycemia in 9 participants with 28,579 beats total
331 h of euglycemia in 28 participants with no beats
These yield crude rates of bradycardia of 81.8 (4,742/58) beats per hour of hypoglycemia versus 75.6 (28,579/378) beats per hour of euglycemia, with a crude rate ratio of 1.08, are substantially different from that published.
One of us (J.L.) then constructed a data set with the above numbers of hours with hypoglycemia or euglycemia and averaging the total number of beats/hours observed (e.g., 4,742/4 in the first category). Analyses were then conducted using the methodology described by the authors in the published article and the supplemental response (negative binomial regression with a repeated hourly effect within subjects), without and with an attempt to account for matching of hypoglycemia and euglycemia hours within subjects. The calculated risk ratios were close to or even less than 1, incompatible at face value with the authors’ findings (6.44) and highly unlikely to be a statistically significant result. We shared these results with the authors. Because these are not precise estimates, we prefer to not give exact figures here to avoid them being quoted by others.
We have direct dualities of interest in the topic: one of us (P.H.) holds skeptical views of the role of hypoglycemia in causation of cardiovascular events in type 2 diabetes, and the other (J.L.) is a co-investigator of a study funded by the National Institute of Diabetes and Digestive and Kidney Diseases on the association of hypoglycemia on continuous glucose monitoring with cardiac dysrhythmic abnormalities.
See accompanying article, p. 664.
Article Information
Duality of Interest. P.H. or institutions with which he is associated with receive funding from manufacturers of insulin therapy for his lecturing, research, and advisory activity, including Biocon, Merck (Merck Sharp & Dohme), Novo Nordisk, and Sanofi. J.L. has research funding as stated above but no commercial funding to report in regard to insulin therapy. No other potential conflicts of interest relevant to this article were reported.
