Multi-Agent Collaborative ADR Prediction With Human-Machine Decision Comparison

Multi-Agent Collaborative Framework for Adverse Drug Reaction Prediction: Evidence-Based Verification and Human-Machine Decision Comparative Study

This study develops a multi-agent collaborative prediction model to forecast adverse drug reactions using real-world clinical medical records. It validates model performance via evidence-based data and compares decision outputs between the AI model and clinical physicians, aiming to improve early identification of drug adverse events. Only de-identified historical medical data will be analyzed; no new clinical interventions will be conducted, with no additional risks to participants.

This observational study first retrospectively collects desensitization cases related to adverse drug reactions to construct a predictive model, and then prospectively enrolls patients to evaluate model efficacy and conduct comparative research with expert blind assessment.

Inclusion Criteria:

• Cases shall involve drug categories including anti-infectives, cardiovascular agents, anti-tumor drugs, central nervous system drugs, digestive system drugs, etc. Each case must contain at least one definite adverse drug reaction (ADR) event, with complete supporting documentation (medical history, medication history, ADR occurrence process, and clinical outcome).

Exclusion Criteria:

• Cases with incomplete supporting documentation lacking medical history, medication history, ADR occurrence process or clinical outcome.
• Cases only with suspected or possible ADRs without definite clinical confirmation.