Autonomous AI Prescribing a Drug to Prevent Severe Acute Graft-versus-Host Disease in HLA-Haploidentical Transplants

Comprehensive Summary

This prospective study represents the first clinical implementation of conditional autonomous artificial intelligence for pharmaceutical intervention. The daGOAT algorithm was deployed to monitor 141 dynamic clinical covariates in patients receiving HLA-haploidentical hematopoietic cell transplants and autonomously prescribed ruxolitinib upon detection of intermediate-to-high risk for severe acute graft-versus-host disease (GVHD). Among 152 eligible patients, 85% were offered enrollment by physicians and 88% of invited patients consented to participate. Initial compliance with AI-generated prescriptions reached 98% (56/57 participants). The cumulative incidence of severe acute GVHD was 5.5% in the AI focus group compared to 16% in covariate-matched controls. The model's successful integration into clinical workflow was facilitated by transparency, minimal workflow disruption, and preservation of physician override authority.

Outcomes and Implications

This study demonstrates the feasibility and acceptance of conditional autonomous AI systems for pharmaceutical decision-making in high-stakes clinical scenarios where conventional predictive methods are inadequate. The high rates of physician adoption and patient compliance suggest that the medical community may be prepared for this level of AI autonomy when specific conditions are met: demonstrated clinical need, algorithmic transparency, workflow compatibility, and maintained physician control. These findings establish a potential framework for deploying similar autonomous AI systems in other clinical contexts requiring early intervention and lacking reliable predictive tools. However, critical issues remain unresolved, including liability determination, implementation of closed-loop feedback mechanisms for continuous model refinement, and validation across diverse patient populations and institutional protocols. Further investigation is warranted to assess efficacy in controlled trials and to establish regulatory and ethical frameworks for autonomous AI pharmaceutical intervention.