Using Machine Learning to Understand Text for Pharmacovigilance: A Systematic Review

Background: Pharmacovigilance is a science that involves the ongoing monitoring of adverse drug reactions of existing medicines. Its primary purpose is to sustain and improve public health. The existing systems that apply the science of pharmacovigilance to practice are, however, not only expensive and time-consuming, but they also fail to include experiences from many users. The application of computational linguistics to user-generated text is hypothesized as a pro-active and an effective supplemental source of evidence. Objective: To review the existing evidence on the effectiveness of computational linguistics to understand user-generated text for the purpose of pharmacovigilance. Methodology: A broad and multi-disciplinary systematic literature search was conducted that involved four databases. Studies were considered relevant if they reported on the application of computational linguistics to understand text for pharmacovigilance. Both peer- reviewed journal articles and conference materials were included. The PRISMA guidelines were used to evaluate the quality of this systematic review. Results: A total of 16 relevant publications were included in this systematic review. All studies were evaluated to have a medium reliability and validity. Despite the quality, for all types of drugs, a vast majority of publications reported positive findings with respect to the identification of adverse drug reactions. The remaining two studies reported rather neutral results but acknowledged the potential of computational linguistics for pharmacovigilance. Conclusions: There exists consistent evidence that computational linguistics can be used effectively and accurately on user-generated textual content that was published to the Internet, to identify adverse drug reactions for the purpose of pharmacovigilance. The evidence suggests that the analysis of textual data has the potential to complement the traditional system of pharmacovigilance. Recommendations for researchers and practitioners of computational linguistics, policy makers, and users of drugs are suggested