Data-Driven Prediction of Beneficial Drug Combinations in Spontaneous Reporting Systems.

Post-market withdrawal of medications because of adverse drug reactions (ADRs) could result in loss of effective compounds which are effective for treating a specific disease but have unfavorable benefit-to- harm ratio. Recent therapeutic successes have renewed interest in drug combinations, which could work synergistically to improve therapeutic efficacy or work antagonistically to alleviate the risk of the ADRs. However, experimental screening approaches are costly and often can identify only a small number of drug combinations. Spontaneous reporting systems (SRSs) routinely collect adverse drug events (ADEs) from patients on complex combinations of medications and provide an empirical and cost-effective source to detect unexpected beneficial drug combinations. In this paper, we proposed a novel data-driven method for the prediction of drug combinations where one drug could reduce the ADRs of the other, based on data from SRSs. The predictive model was shown to be effective using a commonly used evaluation approach in pharmacovigilance by constructing a known drug-drug interaction (DDI) reference standard. The method was applied to perform large-scale screening on SRS data for drug-ADR-drug triples where polypharmacy could potentially reduce the ADR. Analysis of the top ranking candidates showed high level of clinical validity.