The dynamic relationship between current and previous severe hypoglycemic events: a lagged dependent variable...

Past studies have found episodes of severe hypoglycemia (SH) to be serially dependent. Those studies, however, only considered the impact of a single (index) event on future risk; few have analyzed SH risk as it evolves over time in the presence (or absence) of continuing events. The objective of this study was to determine the dynamic risks of SH events conditional on preceding SH events among patients with type 2 diabetes (T2D) who have initiated basal insulin.

We used an electronic health records database from the United States that included encounter and laboratory data and clinical notes on T2D patients who initiated basal insulin therapy between 2008 and 2011 and to identify SH events. We used a repeated-measures lagged dependent variable logistic regression model to estimate the impact of SH in one quarter on the risk of SH in the next quarter.

We identified 7235 patients with T2D who initiated basal insulin. Patients who experienced ≥1 SH event during any quarter were more likely to have ≥1 SH event during the subsequent quarter than those who did not (predicted probabilities of 7.4% and 1.0%, respectively; p < 0.01). This effect was stronger than the impact of history of SH before starting basal insulin (predicted probabilities of 1.0% and 3.2%, respectively; p < 0.01) or of a history of SH during the titration period (predicted probabilities of 1.1% and 2.8%, respectively; p < 0.01).

The risk of experiencing a SH event is highly dependent on a patient's immediate history of SH events and therefore the value of preventing one SH event may be substantial. These results can inform patient care by providing clinicians with dynamic data on a patient's risk of SH, which in turn can facilitate appropriate adjustment of the risk-benefit ratio for individualized patient care. These results should, however, be interpreted in light of the key limitations of our study: not all SH events may have been captured or coded in the database, data on filled prescriptions were not available, we were unable to adjust for basal insulin dose, and the post-titration follow-up period could have divided into time units other than quarters (3 month blocks) resulting in potentially different conclusions. Further real-world studies on how to best to identify patients at risk for SH events based on the presence of recent SH events, rather than on more distant 'prior' events, can help healthcare providers to better manage patients starting basal insulin.