Reinjury risk after ACL reconstruction: a scoping review of artificial intelligence-based prediction models

Abstract

BACKGROUND: ACL reinjury after reconstruction remains a major challenge, affecting long-term function, return to sport and healthcare costs. Although risk factors are known, accurate patient-specific prediction is limited. This scoping review synthesises evidence on supervised AI/ML models for predicting ACL reinjury or related adverse outcomes after primary ACL reconstruction, focusing on outcome definitions, validation and clinical applicability. METHODS: PubMed and Embase were searched using predefined eligibility criteria informed by PRISMA-ScR. Ten studies were included and synthesised qualitatively. RESULTS: Studies were heterogeneous in outcomes, data sources, modelling approaches and validation. Most used supervised ensemble methods and reported moderate-high discrimination within development datasets, but relied mainly on internal validation; external validation was uncommon. Calibration and clinical-utility reporting were inconsistent. Predictors spanned anatomical, patient-related and surgical factors, with variable availability across studies. CONCLUSIONS: AI/ML models show potential, but the evidence base is methodologically heterogeneous and largely developmental. Limited external validation, inconsistent calibration assessment and variable outcome definitions constrain clinical applicability. Current models should therefore be considered exploratory.