Collaborators at the University of Texas School of Biomedical Informatics, in conjunction with members of the Melax Tech team, recently published work demonstrating the feasibility of using AI-based approaches incorporating deep learning to facilitate automated and accurate information extraction from postmarketting vaccine safety reports. The US Vaccine Adverse Event Reporting System (VAERS) collects data on possible problems with vaccines in conjunction with the CDC and the FDA and is a major resource for those studying the incidence and possible causes of adverse events post vaccination. As of 2020, the system contained over 650,000 reports submitted by healthcare providers, vaccine manufacturers, and the public. While VAERS contains structured data, much of the detailed data is in unstructured text. Given the size, scale, and rate of growth of this data, automated analysis of these reports is important to allow researchers and physicians to understand the progression of rare but severe vaccine adverse events.
In our team’s work we focused on Guillain-Barré syndrome (GBS) occurrence post influenza vaccination. We implemented and evaluated a variety of state-of-the-art deep learning algorithms for named entity recognition (NER) from the GBS-related influenza vaccine safety reports. To date, few other machine learning-based studies have been reported on adverse event extraction from VAERS. Melax Tech’s AI-based NLP system, CLAMP, was used for the model development and NLP work on this project. This work should be of interest to anyone attempting to review and analyse adverse event reports for both new and existing vaccines, including immunology researchers and the pharmaceutical industry.
A full description of our methodology and results is given in “Extracting postmarketing adverse events from safety reports in the vaccine adverse event reporting system (VAERS) using deep learning.” Jingcheng Du, Yang Xiang, et al. Journal of the American Medical Informatics Association, Volume 28, Issue 7, July 2021, Pages 1393–1400,https://doi.org/10.1093/jamia/ocab014