Response to Comment on Novodvorsky et al. Diurnal Differences in Risk of Cardiac Arrhythmias During Spontaneous Hypoglycemia in Young People With Type 1 Diabetes. Diabetes Care 2017;40:655–662

We welcome the opportunity to respond to the comments from Home and Lachin (1) on our article (2). One of us (M.C.) has independently reexamined the nighttime bradycardia data using the negative binomial generalized estimated equations approach (SPSS version 22; IBM, Chicago, IL) described in our article. With autocorrelated error terms, he could approximately replicate our results (incident rate ratio [IRR] 5.7 [95% CI 5.0, 6.5], P < 0.001 vs. IRR 6.4 [95% CI 6.3, 6.6], P < 0.001) as previously reported. The minor differences are probably because in the article, autocorrelation within subjects between days was accounted for, as was autocorrelation within subjects between hours. He agreed that the model that we used is a reasonable approach as it models over dispersed count data and allows both for the correlation within subjects and that the data are ordered as a time series within each person. He did however identify an unusual distribution of the data. We have therefore conducted additional analyses.

Table 1 shows the number and percentage of hours where subjects have no bradycardia counts and at least one bradycardia count for hypoglycemia and euglycemia. A total of 12.4% (47/378) of hours of euglycemia had bradycardia, whereas 6.9% (4/58) hours of hypoglycemia had bradycardia. This simple analysis does not account for within-subject correlations—so it would be incorrect to apply an inference—but it suggests that an hour is approximately twice as likely to have bradycardia for euglycemia compared with hypoglycemia.

Table 2 shows the distribution of bradycardia beats for hours of euglycemia and hypoglycemia. The number of beats per hour has been categorized to allow comparison. From this data we can see that for the hypoglycemia group there are 5.2% (3/58) of hours with over 1,000 bradycardia beats compared with 3.2% (12/378) of hours during euglycemia. This would suggest that hours with over 1,000 bradycardia beats are more common during hypoglycemia than during euglycemia.

We then looked for unusual or outlying patterns of bradycardia and identified one subject who accounted for all hours with over 1,000 bradycardia beats. Table 3 shows the pattern of bradycardia beats per hour for subject 140 (a male). We can see that for hypoglycemia there are 100% (3/3) of hours with more than 1,000 bradycardia beats compared with 80% (12/15) hours during euglycemia. This is the only subject in the study experiencing more than 10 bradycardia beats for all hours of their recorded euglycemia and hypoglycemia (range 29–2,306).

Finally, we performed a sensitivity analysis of our original model, removing the subject from Table 3. When we remove this subject from the analysis, the IRR is 0.2 (95% CI 0.12, 0.21, P < 0.001), such that there is no increased risk of bradycardia during hypoglycemia. This indicates that this individual alone is responsible for the increase in IRR and behaves very differently in his heart rate response during hypoglycemia than others in the study.

On reflection, we conclude that although the calculation of the IRR is correct, the inference that this increased risk applies to all patients conveys the wrong impression. Importantly, our additional analysis suggests that an increased risk of bradycardia may apply to only a few individuals who are highly susceptible to bradycardia during hypoglycemia. It remains to be established whether bradycardia or other arrhythmias contribute to sudden nocturnal death during hypoglycemia, although a recent case report suggests an arrhythmic mode of death (3). Nevertheless, these analyses would fit the epidemiological data, which while suggesting a 10-fold increased risk in sudden death among young people with hypoglycemia compared with individuals without diabetes, with a male predominance (4), show clearly that fatal episodes are rare, although nocturnal hypoglycemia is common.

Our results also support the findings of a previous study that reported bradycardia during nocturnal hypoglycemia in 3 of 25 individuals with type 1 diabetes (5). This study used summary statistics only and did not control for an increased risk of bradycardia at night by comparing risks of bradycardia in the same individual during nocturnal euglycemia.

If further studies, which we are currently planning, confirm individual susceptibility, then in an era where continuous glucose monitoring has become a clinical tool, it offers the prospect of routine screening in young people with type 1 diabetes to identify those at increased risk. We will also explore other approaches that might better summarize this clinical picture.

We are grateful to Home and Lachin (1) for emphasizing the challenges of finding the ideal statistical approach when investigating this syndrome. In encouraging us to reanalyze our data, this has suggested to us a potential clinical approach to a rare but tragic condition.