Adverse event detection by integrating twitter data and VAERS.

Area of Science:

• Public Health Surveillance
• Computational Epidemiology
• Pharmacovigilance

Background:

• Vaccines are vital public health tools but can cause adverse events (AEs).
• Traditional AE reporting systems lack timeliness.
• Social media offers timely data but faces labeling and class imbalance challenges.

Purpose of the Study:

• To develop a framework for detecting vaccine AEs by integrating traditional reports and social media data.
• To address timeliness, labeling, and class imbalance issues in AE detection.
• To improve the accuracy and efficiency of post-vaccination AE surveillance.

Main Methods:

• Developed a combinatorial classification approach integrating Twitter data and Vaccine Adverse Event Reporting System (VAERS) information.
• Employed a max-rule based multi-instance learning method to address class imbalance and bias positive users.
• Combined formal reports with social media data to reduce manual labeling costs.

Main Results:

• The proposed multi-instance learning methods outperformed baselines using only Twitter data.
• Integrating formal reports improved performance metrics for multi-instance learning, especially with smaller training datasets.
• Case studies confirmed the model's accuracy in labeling users and tweets related to vaccine AEs.

Conclusions:

• A novel framework effectively detects vaccine AEs by combining formal reports and social media data.
• Formal reports significantly enhance AE detection performance, particularly when social media data is limited.
• The developed model demonstrates effectiveness and accuracy in real-world AE surveillance scenarios.