The Approaches to Assess Robustness of Disproportionality Findings in Pharmacovigilance. A Meta-Epidemiological Research

Introduction: Disproportionality analyses aim to identify potential adverse drug reactions from spontaneous reporting systems in a timely and cost-effective manner. For their apparent simplicity, their use is rapidly expanding as a source of safety evidence complementary to clinical trials. In fact, most disproportions in reporting are not robust enough to identify a safety signal, and strategies used to assess robustness are heterogeneous and inconsistently used (1-4). Documenting these strategies would move forward harmonization in pharmacovigilance studies and simplify signals prioritization for regulatory interventions. Objective: To describe the approaches used in the last two decades to assess the robustness of disproportions. Methods: One hundred studies were randomly selected through a systematic literature search performed on Medline to identify all published disproportionality analyses since inception up to January 1, 2020 (search terms: "case-non case", "disproportionality analysis", "pharmacovigilance analysis", "pharmacovigilance study"). We designed an extraction table to gather, for each article, general information, disproportionality techniques, other new clinical data, integration with other sources, and literature support (see Picture 1). A pilot analysis was performed to train 3 operators (MF, MI, ER). Disagreements were solved through discussion and consensus. Results: This pilot study was based on 35 articles, of which 20% used more than 1 database, 26% employed a Bayesian method, and 54% used standardized or ad hoc queries for a more sensitive case retrieval. Techniques used to increase the robustness of the disproportionality were adjustment (29%), consistency on subpopulations (17%), positive and negative controls (26 and 20%), disproportionality time-trend (14%). Clinical elements of the reporting systems were investigated in a minority of studies: time to onset (34%), dose (14%), dechallenge/rechallenge (11%). Only 14% performed a causality assessment and 31% of the studies accounted for at least one bias (11% indication, 9% comedication, and 9% notoriety bias). Drug utilization data and systematic reviews were used to support robustness only in 1 study each. Most studies stemmed from regulatory reasons (51%) or case series (43%) as rationale. The main literature support presented in the discussion came from clinical trials (34%). Biological plausibility was discussed by 60% of the studies, but only 11% employed a dedicated pharm