Risk Prediction Model for Knee Arthroplasty

To overcome the dilemma in unicompartmental knee arthroplasty (UKA) revision caused by the degeneration of knee articular osteochondral tissue in the untreated compartment, this paper presents and validates a reliable system to evaluate which osteoarthritis patients may suffer revision after having UKA. We conducted a retrospective cohort study by collecting all revision cases available (n = 1) and randomly selecting 74 UKA cases to keep the revision prevalence of almost 14%. The finite element method was first applied to calculate the strain biomechanical features. We then simulated strains of each tissue node induced by the contact force during gait for five movement points through the stance phase of walking. The biological factors such as C-reactive protein and patient's behaviors such as pre- and post-operative maximum strains change during gait, body mass index (BMI), and age were combined and analyzed with the kernel least mean square (KLMS) method. These data were used to model the relationships among the biomechanical, biological factor, and patient's behaviors to predict the risk of UKA revision. The five-fold cross-validation was conducted to assess the prediction accuracy. As a result, the average prediction accuracy for correctly predicting UKA revision was 90.69% for all cases, providing substantial evidence that this model can serve as a potential tool in decision-making for UKA surgical planning. Although knee cartilage micro-degeneration cannot be measured directly in vivo to evaluate the risk of UKA revision, the process can be estimated by combining biomechanical, biological factor, and patient's behaviors in a statistical learning model.