Analysis of COVID-19 vaccine adverse event using language model and unsupervised machine learning

Area of Science:

• Vaccinology
• Data Science
• Public Health

Background:

• The COVID-19 pandemic necessitated widespread vaccination efforts.
• While vaccines are crucial for disease control, some individuals experience adverse events.
• Understanding these side effects is vital for public confidence and safety.

Purpose of the Study:

• To analyze COVID-19 vaccine adverse events using real-world data.
• To identify demographic and vaccine-specific factors associated with adverse events.
• To characterize symptom clusters and discover associations between adverse events.

Main Methods:

• Utilized the Vaccine Adverse Event Reporting System (VAERS) dataset.
• Employed language models for symptom vectorization and dimensionality reduction.
• Applied unsupervised machine learning for symptom clustering and data mining for association rule discovery.

Main Results:

• Adverse events were more frequent in women, with Moderna vaccines, and after the first dose.
• Symptom clusters varied by gender, vaccine manufacturer, age, and underlying conditions.
• Fatal adverse events were significantly associated with a cluster involving hypoxia.
• Strongest association rules found were {chills ↔ pyrexia} and {vaccination site pruritus ↔ vaccination site erythema}.

Conclusions:

• Provides data-driven insights into COVID-19 vaccine adverse events.
• Aims to alleviate public anxiety by offering accurate information.
• Highlights the importance of detailed analysis for vaccine safety monitoring.