Therapeutic drugs often have side effects that are not picked up during pre-market clinical trials. For this reason, even after their market release, they need to be continuously monitored for new or incompletely documented adverse drug reactions (ADRs), known as “signals”. Post-market drug safety surveillance has traditionally relied on individual voluntarily reporting of adverse events. For a more complete picture of drug safety profiling, however, it has become clear lately that additional data sources shall be considered for surveillance, like electronic health records and social media platforms. The EU-funded SAFER (Semantic integration and reasoning framework for pharmacovigilance signals research) project developed a method to leverage and enrich post-market signal detection. To detect signals, researchers typically use computational methods. However, current methods are not highly accurate, as they often exhibit many false positive indications. SAFER identified such gaps in current signal detection approaches and developed ways to overcome them through the complementary analysis of data obtained from different sources, with appropriate detection methods applied in parallel. In SAFER, researchers used publicly available data, obtained from an adverse event reporting system, a database of published medical papers and a social media platform. To identify signals by analysing data from these sources, researchers developed a computational workflow that encompasses: (a) an appropriate search mechanism for the drug(s)-effect(s) of interest applied on the targeted sources, (b) data acquisition, (c) execution of respective signal detection methods on the obtained data, (d) results aggregation, (e) filtering results based on evidence from electronic resources documenting package inserts and drug interactions, and (f) ranking the remaining indications based on their significance. The outcome, in terms of prioritised potential ADR signals, is then provided to drug safety experts for causality assessment. Parts of this workflow were developed via semantic technologies. Finally, researchers evaluated their method using a representative set of test cases concerning important drug safety issues. An example of such a case is the assessment of risk for brain bleeding potentially induced by new oral anticoagulants. SAFER's drug safety surveillance tool will likely increase signal detection accuracy, allowing for timely interventions and mitigation strategies.
Drug safety, adverse drug reactions, pharmacovigilance, signal detection, semantic technologies