PRISMATIC: specialised AI agents work together to prevent drug interactions in patients with multiple conditions

Each agent plays the role of a healthcare professional involved in the patient’s care pathway; together, they draw up and amend the prescription. Tested on MIMIC-IV, the system significantly reduces drug-drug interactions.

Misprescriptions in patients with several diseases often arise from a lack of coordination: each specialist adds his or her treatments without an overview, to the risk of adverse reactions and re-hospitalization. Researchers propose PRISMATIC, a multi-agent architecture where each agent plays a role in the care circuit. Sharing a common workspace, they start by structuring the information from clinical notes, and then develop and refine a safe and justifiable treatment step by step.

Two arbitration mechanisms

Cooperation is based on two mechanisms. The first adjusts over time the weight given to each agent according to their results on past cases, so that the system relies more on the most reliable "experts". The second acts on the patient in progress: the agents refer to targeted returns until they resolve their disagreements and remove the incompatibilities of the prescription.

Based on records from the MIMIC-IV hospital base, PRISMATIC significantly reduces drug conflicts, better covers the pathologies to be treated and produces more coherent plans than solid language models used alone. This is a research work, not a clinically validated tool.