Identifying diabetogenic drugs using real world health care databases: A Danish and Australian symmetry analysis

Abstract

AbstractAimsDrug‐induced diabetes is underreported in conventional drug safety monitoring and may contribute to the increasing incidence of type 2 diabetes. Therefore, we used routinely collected prescription data to screen all commonly used drugs for diabetogenic effects.MethodsLeveraging the Danish nationwide health registries, we used a case‐only symmetry analysis design to evaluate all possible associations between drug initiation and subsequent diabetes. The study was conducted among individuals aged ≥40 years with a first‐ever prescription for any antidiabetic drug 1996‐2018 (n = 348 996). Sequence ratios (SRs) and 95% confidence intervals (CIs) were obtained for all possible drug class‐diabetes combinations. A lower bound of the 95% CI >1.00 was considered a signal. Signals generated in Denmark were replicated using the Services Australia, Pharmaceutical Benefits Scheme 10% data extract.ResultsOverall, 386 drug classes were investigated, of which 70 generated a signal. In total, 43 were classified as previously known based on the SIDER database or a literature review, for example, glucocorticoids (SR 1.67, 95% CI 1.62‐1.72) and β‐blockers (SR 1.20, 95% CI 1.16‐1.23). Of 27 new signals, three drug classes yielded a signal in both the Danish and Australian data source: digitalis glycosides (SR 2.15, 95% CI 2.04‐2.27, and SR 1.76, 95% CI 1.50‐2.08), macrolides (SR 1.20, 95% CI 1.16‐1.24, and SR 1.11, 95% CI 1.06‐1.16) and inhaled β2‐agonists combined with glucocorticoids (SR 1.35, 95% CI 1.28‐1.42, and SR 1.14, 95% CI 1.06‐1.22).ConclusionWe identified 70 drug‐diabetes associations, of which 27 were classified as hitherto unknown. Further studies evaluating the hypotheses generated by this work are needed, particularly for the signal for digitalis glycosides.