Degenerative cervical myelopathy (DCM) is the leading cause of nontraumatic spinal cord dysfunction and remains clinically heterogeneous in presentation and course. This review synthesizes current evidence on predictors of neurological outcomes across conventional prognostic factors (clinical and macrostructural metrics) and quantitative neuroimaging (microstructural metrics), as well as how machine learning (ML) models integrate these predictors into a precision medicine framework to aid in DCM management. We explore evidence on conventional clinical and radiographic factors. Although several signs and scales are associated with clinical outcomes, cross-study inconsistency and the limits of linear models blunt their standalone utility, underscoring the need for multifactorial modelling. We then assess quantitative MRI biomarkers, including diffusion tensor imaging, magnetization transfer, and myelin water imaging, which index axonal integrity and myelination, thereby enriching risk stratification and prediction. Building on these measurements, we examine ML models combining clinical, imaging, and demographic features to predict postoperative outcomes and, increasingly, the natural history of mild DCM. Finally, current gaps and necessary future directions are outlined, including protocol harmonization, prospective multicentre validation, and clinician–patient education to support equitable uptake. Collectively, this review advances in DCM diagnosis and prognosis, highlighting the role of precision medicine tools for personalized patient care.

BACKGROUND Large language models (LLMs) have entered medical education quickly, with students utilizing them for clinical reasoning and self-assessment. Whether they can be trusted on written examinations that anchor residency assessment remains unsettled, particularly in image-rich specialties such as orthopedic surgery. OBJECTIVE Synthesize published evidence on LLM performance on orthopedic written examinations. METHODS We conducted a rapid review following PRISMA 2020 guidance. We searched PubMed, Embase, and Google Scholar from 1 January 2023 through 1 August 2025 for primary studies evaluating any LLM on orthopedic board, in-training, licensure, or commercial-bank items reporting quantitative accuracy. We appraised bias risk across six pre-specified domains and synthesized results. As a review of previously published literature, this study did not require institutional review board approval. RESULTS Thirty-nine studies (2023-2025) evaluating approximately 14,700 questions across 74 model-level arms met inclusion criteria. GPT-3.5-series models achieved 29% to 61% accuracy, below pass thresholds. GPT-4 reached 60% to 82% on text-only items, with multiple studies meeting the American Board of Orthopedic Surgery (ABOS)-Orthopedic In-Training Examination (OITE) linking benchmark of 68.6%; GPT-4o ranged from 58% to 88%. Image-based accuracy was lower (38% to 74% for GPT-4). Eighteen of 39 studies (46%) carried elevated risk of bias for index-test conduct. CONCLUSIONS Current LLMs approach passing-level accuracy on text-only orthopedic examination items but remain unreliable on image-dependent questions. Included studies did not assess the fidelity of LLM-generated explanations. We recommend program-level guidelines for supervised LLM use, examination-body attention to item-security risks, and learner-outcome research before broader integration into orthopedic residency curricula.