This study has some limitations. First, CPRD prescriptions represent those written by general practitioners, and thus treatment adherence is unknown, although this misclassification tends to bias the point estimates toward the null hypothesis. Second, to our knowledge, HES-defined hypoglycemia has not been formally validated, although it has been used as an outcome in a previous unrelated study.However, similar results were obtained after the case definition was limited to those diagnoses in primary position. Third, because of the observational nature of the study, residual confounding needs to be considered. Reassuringly, we observed consistent results using different study design and analytic approaches (such as cohort analysis adjusted for HD-PS quartiles and casecrossover analysis). Moreover, it is important to note that, given the strong observed associations (point estimates ranging between 1.52 and 3.80), any unmeasured or unknown confounder would need to be strongly associated with both the exposure and outcome to completely confound the observed association. It is unclear whether such a confounder exists beyond those considered in the models. Finally, despite the large sample size, the rarity of the outcome led to wide confidence intervals in secondary analyses, and thus these should be interpreted with caution.