Related Experiment Video
Updated: Aug 16, 2026
05:47
Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems
Published on: June 13, 2025
Advancing Evidence-Based Medicine for Population, Intervention, Comparison, and Outcome Element Recognition and
Zeyuan Hao1, Yifan Duan2, Yu Wang3
1School of Software Engineering, Beijing Jiaotong University, 3 Shangyuan Village, Haidian District, Beijing, 100044, China, 86 18404966218.
Journal of Medical Internet Research
|August 14, 2026
Summary
Automating PICO extraction from biomedical literature is crucial. This study shows that both prompt engineering and parameter-efficient fine-tuning (PEFT) of large language models (LLMs) effectively extract population, intervention, comparison, and outcome (PICO) elements.
Area of Science:
- Biomedical Informatics
- Natural Language Processing
- Artificial Intelligence in Medicine
Background:
- The rapid growth of biomedical literature necessitates efficient methods for extracting Population, Intervention, Comparison, and Outcome (PICO) elements.
- PICO elements are fundamental for evidence-based medicine.
Purpose of the Study:
- To systematically evaluate prompt engineering optimization and parameter-efficient fine-tuning (PEFT) for automating PICO recognition and extraction.
- To compare the effectiveness of different prompt engineering strategies and PEFT techniques.
Main Methods:
- Developed a dual-phase framework: systematic prompt optimization (in-context learning, chain of thought, multipath reasoning) and PEFT (LoRA, quantized LoRA, freeze).
- Utilized PubMed-PICO and NICTA-PIBOSO datasets for recognition, and EBM-NLP for extraction.
- Evaluated performance using precision, recall, and F1-score, with F1-score as the primary metric.
Main Results:
- Chain of Thought (COT) prompt engineering yielded the best overall performance for recognition and extraction tasks.
- LoRA achieved top recognition performance (91.7% F1 for population), while quantized LoRA excelled in extraction (79.3% F1 for intervention).
- PEFT methods enhanced performance beyond prompt engineering, demonstrating significant gains on specific datasets.
Conclusions:
- Large Language Models (LLMs) can effectively automate PICO recognition and extraction using prompt engineering and PEFT.
- PEFT further optimizes LLM performance, offering critical insights for clinical applications involving PICO extraction.
- This research advances methodological approaches for processing biomedical literature.
