A Bayesian belief network model for predicting post-operative rehabilitation outcomes after ACL reconstruction

Abstract

Anterior cruciate ligament reconstruction (ACLR) is routinely performed to restore knee stability after injury; however, postoperative recovery and re-injury risk remain highly variable and difficult to predict using conventional rehabilitation protocols. Existing statistical and machine-learning models provide useful risk stratification but often lack clinical interpretability and the ability to integrate evolving patient information. This research outlines a constructible Bayesian Belief Network that was developed to depict causal factors that are related to the outcomes of post-ACL reconstruction. The hierarchical design connects fourteen demographic, biomechanical, surgical, psychological, and rehabilitative variables with seven intermediate functional variables, and finally results in the measurement of ACL re-injury risk. This framework was based on the available empirical literature and then developed through systematic elicitation of experts. Prior and conditional probabilities were parameterised using a hybrid of epidemiological data and structured expert judgement. Probabilistic inference and causal reasoning were implemented in the GeNIe modelling environment using exact Bayesian algorithms. Causal inference yielded an overall re-injury probability of 53% high risk and 47% low risk. Diagnostic, inter-causal, and predictive reasoning demonstrated that premature return-to-sport, poor neuromuscular control, weak muscle strength and joint stability, and non-anatomic graft technique markedly increased the posterior probability of re-injury, whereas high rehabilitation adherence, favourable psychological readiness, and delayed criterion-based return-to-sport shifted inference toward low-risk outcomes. Scenario-based simulations showed that modifying critical factors such as neuromuscular control and return-to-sport timing produced clinically meaningful reductions in predicted re-injury risk. Unlike black-box machine-learning approaches, the proposed BBN provides transparent causal explanations and supports dynamic updating as new patient data become available. This Bayesian causal framework offers a robust decision-support tool for personalised ACL rehabilitation, enabling clinicians to identify modifiable risk factors, optimize return-to-sport decisions, and reduce the likelihood of re-injury following ACL reconstruction.