Discovering Potential Effects of Dietary Supplements from Twitter Data

Abstract

The U.S. Food and Drug Administration uses the Center for Food Safety and Applied Nutrition (CFSAN) Adverse Event Reporting System (CAERS) as the primary tool for identifying new and emerging dietary supplement adverse events. Despite mandatory and voluntary reporting of dietary supplement adverse events to CAERS, many continue to go unreported. Availability of social media has enabled dietary supplement consumers to freely share their concerns and experiences online. Such consumer generated information can be a useful source to further monitor the safety of dietary supplements. To study the usefulness of social media (Twitter in particular) for safety surveillance of dietary supplements, we developed a computational processing pipeline: 1) machine learning based identification of potential Twitter posts (tweets) of personal experiences related to the use of dietary supplements, 2) detection of potential supplement events from these tweets using the medpie open source tool, and 3) mapping detected events to effects through the taxonomy provided in SNOMED CT. Using our pipeline, we identified, from a group of 1,244,661 tweets collected, a total of 17,346 personal experience tweets pertaining to 4 dietary supplements. A total of 191 effects were mapped to SNOMED CT and we discovered that 48 of the 191 effects are not listed in either of the two online sources we referenced. However, the effects discovered from the social media data will need to be verified and confirmed with other sources and/or clinical evidences.