AI Agent MIRA Matches or Beats Doctors in Simulated EHR Diagnostic Tests

In a Nature paper, Ferber and colleagues presented MIRA (Medical Intelligence for Reasoning and Action), an autonomous agent operating in a sandboxed electronic health record system. Tested on simulated patients derived from the MIMIC-IV database, MIRA could collect histories, order and interpret diagnostics, formulate differential diagnoses, and propose treatment plans. The AI's diagnostic accuracy matched or surpassed that of physician groups, and its therapeutic choices showed higher concordance with established clinical guidelines, particularly in surgical and medication contexts.

MIRA also generated safe prescription recommendations and appropriate emergency department admission or discharge advice. Robustness tests demonstrated consistent performance across diverse patient characteristic perturbations. The findings suggest that such AI agents could be integrated into clinical workflows, though real-world validation remains pending.