Mechanism-based simulation of comparator selection in pharmacovigilance signal detection

Background: Reference set choice can influence automated disproportionality analysis of spontaneous reporting data. Active comparators may reduce confounding by indication but can increase susceptibility to masking and reporting differences. This study examined comparator effects under different clinical and reporting conditions. Methods: A controlled simulation separately modelled event generation, reporting, and analysis. Full-data, active-comparator, and class-exclusion reference sets were compared across nine prespecified scenarios involving confounding by indication, masking, differential reporting, and their combinations. Each scenario included 5,000 Monte Carlo replicates. The primary outcome was the pairwise difference in mean log-ROR. Sensitivity analyses examined comparator similarity, while exploratory analyses assessed masking intensity and consistency using PRR and BCPNN. A published FAERS analysis provided an illustrative external comparison. Results: Under masking, mean RORs were 1.481, 0.759, and 1.933 for full-data, active, and class-exclusion comparators, respectively. The active versus full difference in mean log ROR was −0.670. Under combined confounding and masking, corresponding RORs were 1.543, 0.771, and 2.331, with a difference of −0.696. With confounding alone (true OR=1.0), the active comparator reduced mean ROR from 1.126 to 1.008. Findings were consistent across comparator similarity levels and with PRR and BCPNN. Conclusions: Comparator effects depended on the underlying conditions. Active comparators reduced distortion from confounding by indication but could lower disproportionality estimates when masking occurred. Comparator selection should therefore consider the clinical and reporting factors that shape spontaneous reporting data rather than assume active comparators are always preferable.