Evidence-Grounded AI for Musculoskeletal Care

Musculoskeletal care is inherently longitudinal, requiring clinicians to integrate evolving patient data—such as imaging, laboratory results, and surgical findings—with external medical knowledge over months or years. Current clinical AI systems often fail to support this continuity, focusing instead on isolated diagnostic tasks or single-point medical question answering. The researchers developed OrthoPilot, an agentic AI system designed to manage the full musculoskeletal care pathway, from admission through surgery to rehabilitation.

OrthoPilot is powered by CHEESE (Clinical Holistic Evidence-to-Execution Synthesis Engine), a 32-billion-parameter LLM trained on expert care trajectories. The system utilizes a "Tool Plaza" architecture, which allows it to act as an agent that identifies missing information, retrieves real-time patient data from hospital systems, and cross-references this with authoritative external medical knowledge (e.g., guidelines, literature, and textbooks) to generate traceable, evidence-based recommendations.

To assess the system, the authors created OrthoBench, a comprehensive benchmark derived from 21 years of longitudinal electronic health records, covering 1,000 disease codes. They also introduced ORACLE, an evaluation framework that scores open-ended clinical responses based on physician-defined decision elements rather than simple text matching.

In retrospective evaluations, OrthoPilot outperformed general-purpose and medical-specific LLMs across all tasks. A reader study involving 81 orthopaedic physicians demonstrated that the system’s reasoning surpassed that of experts with 25 years of experience. In a prospective study of 1,870 complex cases, the system increased full-chain management success by 10.6%. Furthermore, an 8-month randomized deployment in a hospital setting showed that OrthoPilot increased cumulative cases per bed by 9.7% and improved patient-reported access to health information, suggesting that the system effectively reduces physician workload while enhancing care quality.