6D-10 - The usability of active comparator and propensity scores in pharmacovigilance signal detection studies with disproportionality analysis: An example of sodium-glucose cotransporter-2 inhibitors

Background: Disproportionality analysis is a key pharmacovigilance tool to detect safety signals but there is a risk of false positives due to confounding by indication and reporting biases such as notoriety bias. In pharmacoepidemiology, employing active comparators and propensity scores (PS) are some common approaches to minimise confounding that may also have potential for pharmacovigilance studies.


Objectives: To evaluate the role of active comparators and PS analysis in improving signal detection from spontaneous reports, using sodium-glucose cotransporter-2 inhibitors (SGLT2is) as example.


Methods: Individual case safety reports (ICSRs) containing non-insulin anti-diabetic drugs (NIADs) as suspected drug in adult patients with type 2 diabetes were extracted from WHO VigiBase (2012–2023). The association between SGLT2is and common (≥1% of reports) adverse drug reactions (ADRs) was assessed against three comparators: dipeptidyl peptidase-4 inhibitors (DPP4is), glucagon-like peptide-1 receptor agonists (GLP1-RAs) and other NIADs (excluding SGLT2is). Disproportionality analysis was conducted by fitting a multivariable logistic regression model to estimate reporting odds ratios (RORs), adjusting for available confounders. Moreover, a PS stratification analysis was performed using observed covariates with SGLT2is against GLP1-RAs, to estimate stratified (in quintiles) and pooled RORs of selected outcomes. Also, a calendar time-restricted analysis using PS was conducted to show the impact of a regulatory communication (before vs. after) on RORs.


Results: From 1,670,737 extracted ICSR-ADR combinations, 484,835 unique ICSRs were included. Of these, 87,974 (18%) involved SGLT2is, while 396,861 (82%) involved other NIADs combined (excluding SGLT2is), including 76,848 (16%) DPP4is and 188,406 (39%) GLP1-RAs. While RORs for fungal infection were similar with SGLT2is vs. DPP4is (59.5, 95% CI 45.4–78.0), GLP1-RAs (65.7, 95% CI 57.3–75.4), and NIADs (75.8, 95% CI 67.5–85.3), those for weight loss differed: SGLT2is vs. DPP4is (3.0, 95% CI 2.7–3.3), GLP1-RAs (1.0, 95% CI 0.9–1.1), and NIADs (1.4, 95% CI 1.3–1.5). Comparing SGLT2is to GLP1-RAs, RORs tended to increase after the regulatory communication, with acute kidney injury rising from 1.6 (95% CI 1.1–2.3) to 5.7 (95% CI 5.0–6.5), diabetic ketoacidosis from 6.5 (95% CI 3.9–11.0) to 36.9 (95% CI 32.8–41.5), and toe amputation from 24.4 (95% CI 5.6–107.1) to 104.2 (95% CI 60.2–180.2).


Conclusions: Active comparators may be most valuable in pharmacovigilance when assessing drug-event pairs with well-understood reporting patterns and a well-defined comparator. Also, PS with a calendar time-restricted analysis showed promising usability to minimise reporting bias from a regulatory communication in pharmacovigilance.