Identifying adverse reactions following COVID-19 vaccination in Korea using data from active surveillance: a text mining approach.

Abstract

Objectives: Unstructured text data collected through vaccine safety surveillance systems can identify previously unreported adverse reactions and provide critical information to enhance these systems. This study explored adverse reactions using text data collected through an active surveillance system following coronavirus disease 2019 (COVID-19) vaccination.

Methods: We performed text mining on 2,608 and 2,054 records from 2 survey seasons (2023-2024 and 2024-2025), in which participants reported health conditions experienced within 7 days of vaccination using free-text responses. Frequency analysis was conducted to identify key terms, followed by subgroup analyses by sex, age, and concomitant influenza vaccination. In addition, semantic network analysis was used to examine terms reported together.

Results: The analysis identified several common (≥1%) adverse events, such as respiratory symptoms, sleep disturbances, lumbago, and indigestion, which had not been frequently noted in prior literature. Moreover, less frequent (≥0.1 to <1.0%) adverse reactions affecting the eyes, ears, and oral cavity were also detected. These adverse reactions did not differ significantly in frequency based on the presence or absence of simultaneous influenza vaccination. Co-occurrence analysis and estimation of correlation coefficients further revealed associations between frequently co-reported symptoms.

Conclusions: This study utilized text mining to uncover previously unrecognized adverse reactions associated with COVID-19 vaccination, thereby broadening our understanding of the vaccine's safety profile. The insights obtained may inform future investigations into vaccine-related adverse reactions and improve the processing of text data in surveillance systems.