Transforming user narratives into vaccine safety signals: Advanced text analysis of syndromic app data in the UK and Sweden
Effective post-authorization pharmacovigilance is essential for monitoring the safety of approved vaccines in the general population. To overcome current methodological challenges, we developed a workflow to extract signals from user-generated free text narratives about COVID-19 vaccination using a syndromic app.
Details
- Funder: Hjärt-Lungfonden, Swedish Research Council,ZOE Limited provided in-kind support for all aspects of building, running and supporting the app and service to all users worldwide.
Project description
Vaccination is a cornerstone of public health, preventing millions of deaths each year by controlling infectious diseases. The COVID-19 pandemic highlighted the importance of rapid vaccine development and mass administration. However, concerns over adverse events following immunization (AEFIs) underscore the need for rigorous post-authorization pharmacovigilance to maintain public trust and safety.
This study develops a workflow to extract vaccine safety signals from user-generated narratives about COVID-19 vaccination using the ZOE Covid Symptom Study app in the UK and Sweden. Signals are early indicators of a potential link between a vaccine and an adverse event. Signals can be used alongside findings from other sources to guide further investigation by medical professionals and pharmacovigilance experts, helping them determine if regulatory action is necessary.
Our study includes data from 24 626 UK participants (October 16, 2020 - November 3, 2021) and 812 Swedish participants (December 14, 2020 - February 12, 2022). We analyzed free-text submissions focusing on narratives two months pre- and post-vaccination. We used Large Language Models and Natural Language Processing pipelines developed by spaCy to perform text processing, including tokenization and lemmatization. We perform topic modeling with BERTopic to create dense word-clusters and coherent topics for context-based semantic pattern analysis. Our approach aims to enhance real-time vaccine safety monitoring.