Toward evidence-grounded AI systems for medication safety
Polypharmacy requires drug–drug interaction risk to be assessed in relation to mechanism, severity, patient vulnerability, exposure, monitoring options, and clinical workflow. Feature-based methods, deep learning, knowledge graphs, multimodal fusion, and event-level models provide a computational foundation for predicting DDIs. Their clinical application also requires patient-specific interpretation and evaluation within the intended workflow. This narrative review examines how pairwise prediction can be integrated with these requirements in medication-safety systems. Multimodal biomedical foundation models may connect chemical structures, targets, pathways, knowledge graphs, literature, pharmacovigilance, real-world evidence, and patient trajectories. Large language models can assist with retrieval, summarization, explanation, and communication under source verification and professional review. Bounded AI agents may coordinate planning, tool use, verification, escalation, documentation, monitoring, and audit. We organize these components into a human-supervised architecture and an evaluation pathway linking model performance with workflow and clinical outcomes. Key challenges include data leakage, distribution shift, poorly documented evidence sources, hallucination, workflow burden, governance, and limited prospective validation. Prospective evaluations should assess how these systems affect medication decisions, monitoring, and patient outcomes, alongside the reliability of their evidence and the workload they create.
Authors
- Yiming Chen (ORCID: https://orcid.org/0009-0009-4847-3847)
- Lei Wang (ORCID: https://orcid.org/0000-0002-7014-2149)
- Zhen Zhang (ORCID: https://orcid.org/0000-0001-9629-9614)
- Xianjun Hu
- Tianyu Li
- Jiayi Zhou
Institutions
- Jishou University (CN)
- Changsha University (CN)
Publication Details
- Journal
- Frontiers in Pharmacology
- Published
- 2026-11-20
- DOI
- https://doi.org/10.3389/fphar.2026.1949488
- Primary Topic
- Pharmaceutical Practices and Patient Outcomes
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Natural Science Foundation of China