DeepER-Med: Advancing Deep Evidence-Based Research in Medicine Through Agentic AI

Area of Science:

• Artificial Intelligence in Medicine
• Biomedical Research
• Clinical Decision Support

Background:

• AI adoption in healthcare requires trustworthiness and transparency.
• Current AI systems for scientific discovery lack inspectable evidence appraisal.
• Existing benchmarks fail to assess AI performance on complex medical questions.

Purpose of the Study:

• Introduce DeepER-Med, an AI framework for evidence-based medical research.
• Develop DeepER-MedQA, a dataset for evaluating AI on real-world medical questions.
• Assess DeepER-Med's performance against existing platforms and in clinical settings.

Main Methods:

• DeepER-Med framework with modules for research planning, agentic collaboration, and evidence synthesis.
• DeepER-MedQA dataset with 100 expert-level medical research questions.
• Expert manual evaluation and human clinician assessment.

Main Results:

• DeepER-Med outperforms production-grade AI platforms in evidence-based research.
• The system demonstrates effectiveness in generating novel scientific insights.
• Clinician assessment showed DeepER-Med's conclusions align with recommendations in 7 of 8 cases.

Conclusions:

• DeepER-Med offers a transparent and inspectable AI workflow for medical research.
• The framework shows significant potential for advancing biomedical discovery and clinical decision support.
• DeepER-Med addresses limitations in current AI systems for healthcare applications.