LEVERAGING UNSTRUCTURED DATA IN ELECTRONIC HEALTH RECORDS TO DETECT ADVERSE EVENTS FROM PEDIATRIC DRUG USE - A...

Area of Science:

• Pharmacovigilance
• Pediatric Drug Safety
• Health Informatics

Background:

• Adverse drug events (ADEs) in children present significant public health concerns.
• Research on detecting and monitoring pediatric ADEs is notably limited.
• Electronic health records (EHRs) contain valuable unstructured data for pharmacovigilance.

Purpose of the Study:

• To conduct a scoping review on the utilization of unstructured EHR data for identifying pediatric ADEs.
• To assess the current state of research in this domain.
• To identify challenges and future directions for pediatric pharmacovigilance.

Main Methods:

• Systematic search of six major databases (MEDLINE, Embase, IEEE Xplore, etc.) in September 2024.
• Inclusion criteria applied to identify relevant studies on pediatric ADE detection using EHRs.
• Analysis of identified studies focusing on data sources, methodologies, and findings.

Main Results:

• A significant research gap exists, with only nine studies meeting inclusion criteria from 984 records.
• Unstructured EHR data shows potential for enhancing pediatric pharmacovigilance.
• Traditional Natural Language Processing (NLP) methods have limitations in generalizability and context interpretation for ADE extraction.

Conclusions:

• There is a critical need for more research into identifying pediatric ADEs using EHR data.
• Advances in transformer-based models and large language models (LLMs) can overcome current NLP challenges.
• Leveraging LLMs offers a promising avenue for large-scale utilization of EHR data in pediatric pharmacovigilance.