Digital pharmacovigilance: The MedWatcher system for monitoring adverse events through automated processing of Internet social media and crowdsourcing.

摘要

Half of Americans take a prescription drug, medical devices are in broad use, and population coverage for many vaccines is over 90%. Nearly all medical products carry risk of adverse events (AEs), sometimes severe. However, pre-approval trials use small populations and exclude participants by specific criteria, making them insufficient to determine the risks of a product as used in the population. Existing post-marketing reporting systems are critical, but suffer from underreporting. Meanwhile, recent years have seen an explosion in adoption of Internet services and smartphones. MedWatcher is a new system that harnesses emerging technologies for pharmacovigilance in the general population. MedWatcher consists of two components, a text-processing module, MedWatcher Social, and a crowdsourcing module, MedWatcher Personal. With the natural language processing component, we acquire public data from the Internet, apply classification algorithms, and extract AE signals. With the crowdsourcing application, we provide software allowing consumers to submit AE reports directly.;Our MedWatcher Social algorithm for identifying symptoms performs with 77% precision and 88% recall on a sample of Twitter posts. Our machine learning algorithm for identifying AE-related posts performs with 68% precision and 89% recall on a labeled Twitter corpus. For zolpidem tartrate, certolizumab pegol, and dimethyl fumarate, we compared AE profiles from Twitter with reports from the FDA spontaneous reporting system, We find some concordance (Spearman's rho = 0.85, 0.77, 0.82, respectively, for symptoms at MedDRA System Organ Class level). Where the sources differ, milder effects are overrepresented in Twitter. We also compared post-marketing profiles with trial results and found little concordance.;MedWatcher Personal saw substantial user adoption, receiving 550 AE reports in a one-year period, including over 400 for one device, Essure. We categorized 400 Essure reports by symptom, compared them to 29 reports from the FDA spontaneous reporting system, and found high concordance (rho = 0.65) using MedDRA Preferred Term granularity. We also compared Essure Twitter posts with MedWatcher and FDA reports, and found rho = 0.25 and 0.31 respectively.;MedWatcher represents a novel pharmacoepidemiology surveillance informatics system; our analysis is the first to compare AEs across social media, direct reporting, FDA spontaneous reports, and pre-approval trials.