Predicting Anterior Cruciate Ligament Reconstruction Revision Risk

Background: Predicting anterior cruciate ligament reconstruction (ACLR) revision risk using machine learning (ML) regression analyses of large-scale registry data offers an evidence-based approach for clinical decision-making and management at a patient-specific level. We examined the performance of an enhanced ML-Cox regression analysis of the Danish Knee Ligament Reconstruction Registry (DKRR) for predicting ACLR revision risk. Methods: We analyzed surgical and patient-reported outcome measure data from 18,753 patients in the DKRR who underwent primary ACLR between 2005 and 2023. Enhanced ML-Cox regression analyses, using the least absolute shrinkage and selection operator (LASSO) and stable iterative variable selection (SIVS) approaches, were applied to predict the risk of ACLR revision (i.e., the risk of repeat surgery to reconstruct the ACL). The SIVS procedure identified key variables, including age at the time of primary ACLR and several Knee injury and Osteoarthritis Outcome Score (KOOS) items from 12-month follow-up surveys, as inputs for the best-performing regression models for predicting ACLR revision risk. The resultant Cox regression models for the prediction of ACLR revision risk, therefore, did not involve an analysis of patients with incomplete 12-month follow-up survey data, including patients with graft ruptures within 12 months after the primary surgery. Results: The best-performing Cox regression model for predicting ACLR revision risk incorporated age at the time of primary ACLR and 3 KOOS items (Pain P1 and Quality of Life Q2 and Q3) from the 12-month postoperative follow-up assessment. This model demonstrated good prediction accuracy 1, 2, and 5 years after the 12-month follow-up assessment (C-index [and standard error], 0.73 [0.03], 0.73 [0.02], and 0.74 [0.02], respectively). This 4-variable Cox regression model was well-calibrated across these time points. An online clinical point-of-care tool, the Danish KOOS 3 Risk Monitoring Tool (DK 3 ), was developed for predicting ACLR revision risk. Conclusions: Enhanced ML-Cox regression, incorporating patient age and 3 KOOS items obtained 12 months postoperatively, provided good prediction accuracy for ACLR revision risk from 1 to 5 years after the 12-month follow-up assessment, a period that has been associated with the vast majority of ACLR revisions. The newly developed DK 3 point-of-care tool offers a direct-input method to predict and monitor the risk of ACLR revision. Level of Evidence: Prognostic Level III . See Instructions for Authors for a complete description of levels of evidence.