Evaluation of the Robustness of Disproportionality Analyses from National and International Spontaneous Reporting Databases

Introduction: Disproportionality analyses are widely used to detect new safety signals but there is a wide heterogeneity of implementation strategies that could lead to misinterpretations by readers, searchers, or stakeholders. Objective: The objective of this study was to assess the robustness of disproportionality analyses, both from national and international spontaneous reporting databases. Methods: A systematic literature review was performed through Pubmed and ISI Web of Science, to select disproportionality analyses published between 2017 and 2018, concerning any drug available in a post-marketing setting. The outcome of articles was any detected safety signal, and selected articles concerned only standalone disproportionality analyses. Articles' characteristics were extracted, and the robustness assessment table was assessed through seven criteria selected and reviewed by the READ-US executive committee (www. readus-statement.org). These criteria were shared into three different sub-groups, relating to their domain investigated. The first sub-group concerned methods: it contains four criteria related to study rationale, drug identification, event identification, and signal method and threshold. The second sub-group concerned the interpretation, repre sented by two criteria, the report of strategy for accounting for most common biases in spontaneous reporting databases and the use of positive and/or negative controls. An additional criterion was defined as the justified selection of comparator if it was different from the whole database. Each domain could be reported or not. The assess ment was conducted by two authors independently; robustness evaluation and author's justification were extracted using Microsoft Access® dedicated extraction form. Results: Following the literature review, 860 articles were screened, and 86 articles were finally included. FAERS was the most used data source (34 studies), nervous system drugs were the most investigated (17 studies), while neurological and cardiac events (9 studies each), were the most frequently studied events. ROR or PRR were the most frequently used disproportionality measures (75 studies). Concerning robustness, only 50.0% of articles mentioned the role of the drug (suspect, concomitant, or interacting drug), 27.9% of articles used a strategy for accounting for the most common biases in spontaneous reporting databases, and only 8.1% of articles mentioned the use of positive and/or negative controls