Towards automating systematic reviews on immunization using an advanced natural language processing-based extraction...

Area of Science:

• Public Health
• Biomedical Informatics
• Artificial Intelligence

Background:

• Systematic reviews are crucial for evidence-informed decision-making but are resource-intensive and slow.
• The growing volume of unstructured evidence challenges traditional systematic review methods.
• Automation of evidence synthesis is needed to improve efficiency and scalability.

Purpose of the Study:

• To develop and evaluate a novel machine learning-based system for automating stages of evidence synthesis.
• To leverage natural language processing (NLP) advancements for efficient extraction of PICO data from scientific publications.
• To address the scalability issues in conducting systematic reviews for public health research.

Main Methods:

• Developed a machine learning system utilizing advanced NLP models like BioBERT.
• Optimized the NLP model using a specialized immunization document database.
• The system identifies and extracts PICO (population, intervention, control, outcomes) fields from text.

Main Results:

• The optimized NLP model achieved an average accuracy of 88% across five text classes for PICO field extraction.
• The system demonstrates a significant improvement in the efficiency of evidence synthesis.
• Functionality is accessible via a user-friendly web interface.

Conclusions:

• The developed machine learning system offers a promising solution for automating systematic reviews in public health.
• NLP advancements can effectively extract critical PICO data, enhancing the speed and scalability of evidence synthesis.
• This approach can help manage the increasing volume of scientific literature for better public health decision-making.