The Archives of Bone and Joint Surgery

The Archives of Bone and Joint Surgery

From Model Performance to Patient Benefit: What Must Orthopedic Artificial Intelligence Prove?

Document Type : EDITORIAL

Authors
1 Department of Biomedical Engineering, Faculty of Engineering, University of Isfahan, Isfahan, Iran
2 Orthopedic Research Center, Department of Orthopedic Surgery, Ghaem Hospital, Mashhad University of Medical Sciences, Mashhad, Iran
10.22038/abjs.2026.99463.4436
Abstract
Artificial intelligence (AI) has demonstrated promising accuracy in orthopedic fracture detection, alignment measurement, and outcome prediction. However, high technical performance does not necessarily translate into improved clinical decisions or patient outcomes. Evidence reveals a substantial translation gap: a systematic review found that only 10 of 59 machine-learning models for orthopedic surgical outcomes had undergone external validation, while a 2026 review of 146 orthopedic-trauma AI studies reported that approximately 99% were retrospective and only two involved prospective clinical evaluation. This editorial argues that the appropriate unit of evaluation is the AI-enabled clinical pathway rather than the algorithm in isolation. We propose five interconnected requirements for establishing clinical readiness: (1) clearly defined intended use and demonstrated technical and clinical validity; (2) external validation and transportability across healthcare settings and patient populations; (3) calibrated predictions, clinically meaningful decision thresholds, and demonstrated decision utility; (4) evaluation of clinician–AI interaction, including workflow effects and automation bias; and (5) evidence supporting the claimed clinical and patient-level benefits. Importantly, when AI recommendations involve biomechanical mechanisms, predictions related to loading, strain, fixation, alignment, or movement require independent mechanical validation and appropriate linkage to clinical outcomes. Continuous post-deployment monitoring, human oversight, and risk-proportionate evaluation are also essential. Ultimately, orthopedic AI should be judged not merely by discrimination accuracy or computational sophistication, but by its demonstrated ability to deliver safer, more effective, and meaningfully more efficient patient care.
Keywords
Subjects


Articles in Press, Accepted Manuscript
Available Online from 16 November 2026

  • Receive Date 11 October 2026
  • Accept Date 01 November 2026