Natural language processing for detecting adverse drug events: A systematic review protocol.

Area of Science:

• Pharmacovigilance and Drug Safety
• Medical Informatics
• Computational Linguistics

Background:

• Detecting Adverse Drug Events (ADEs) is a critical area in healthcare research.
• Automatic extraction of ADEs using Natural Language Processing (NLP) offers significant potential for knowledge distillation and risk prediction.
• Improved management of predisposing factors for ADEs can lead to better patient outcomes.

Purpose of the Study:

• To systematically review and consolidate evidence on Natural Language Processing (NLP) approaches for automated ADE detection from diverse free-text sources.
• To enhance understanding of the current landscape and effectiveness of NLP tools in extracting ADE information.
• To inform the development of advanced NLP methods and support clinical researchers in healthcare.

Main Methods:

• A systematic review was conducted, searching six major academic databases (Embase, Medline, Web of Science, ACM, IEEE, Scopus).
• Studies underwent rigorous title, abstract, and full-text screening.
• Data extraction, tabulation, and summarization of study characteristics and findings were performed, with risk of bias assessed using the PROBAST tool.

Main Results:

• A total of 178 relevant studies were identified and included in the review as of December 2024.
• Data extraction from all included papers has been completed.
• The systematic review paper is currently in the synthesis and writing phase, with ongoing refinement of eligibility criteria and data extraction since August 2022.

Conclusions:

• The majority of reviewed works focus on the detection and extraction of ADEs from text.
• A significant portion of studies also addresses binary text classification for ADE identification.
• Identified challenges include imbalanced datasets, handling abbreviations, and achieving high accuracy for rare ADEs.