Data mining for potential adverse drug-drug interactions

Introduction: Patients, in particular elderly ones, frequently receive more than one drug at a time. With each drug added to a regime, the number of potential drug-drug interactions (DDIs) increases by a power law. Early prediction of relevant interactions by computerized tools greatly aids clinicians and can guide their prescribing choices. Areas covered: In this article, we discuss different types of DDIs, on which levels they can arise and what efforts have been made in the past to detect and predict them. The emphasis is on data mining technology and network analysis, but overlaps with traditional pharmacovigilance are also discussed. Finally, we discuss strategies to focus and simplify mining efforts to get meaningful results with less effort. Expert opinion: The necessary technology for detecting adverse DDIs exists and is quite refined, although it is more often implied in lower risk scenarios (such as syntactic analysis in web searches and online libraries). Data mining for DDIs, on the other hand, still requires a great deal of human intervention, not only to validate the results but also, more importantly, to separate the relevant from the spurious. The fields of network analysis and graph theory show great promise but have not yet shown much beyond descriptive analyses.