Development of a Machine-Learning Model for Anterior Knee Pain After Total Knee Arthroplasty With Patellar Preservation Using Radiological Variables

Abstract

Background

Anterior knee pain (AKP) following total knee arthroplasty (TKA) with patellar preservation is a common complication that significantly affects patients' quality of life. This study aimed to develop a machine learning model to predict the likelihood of developing AKP after TKA using radiological variables.

Methods

A cohort of 131 anterior stabilized TKA cases (105 patients) without patellar resurfacing was included. Patients underwent a follow-up evaluation with a minimum one-year follow-up. The primary outcome was AKP, and radiological measurements were used as predictor variables. There were two observers who made the radiological measurement, which included lower limb dysmetria, joint space, and coronal, sagittal, and axial alignment. Machine learning models were applied to predict AKP. The best-performing model was selected based on accuracy, precision, sensitivity, specificity, and Kappa statistics. Python 3.11 with Pandas and PyCaret libraries were used for analysis.

Results

A total of 35 TKA had AKP (26.7%). Patient-reported outcomes were significantly better in the patients who did not have AKP. The Gradient Boosting Classifier (GBC) performed best for both observers, achieving an area under the curve (AUC) of 0.9261 and 0.9164, respectively. The mechanical tibial slope was the most important variable for predicting AKP. The Shapley test indicated that high/low mechanical tibial slope, a shorter operated leg, a valgus coronal alignment, and excessive patellar tilt increased AKP risk.

Conclusions

The results suggest that global alignment, including sagittal, coronal, and axial alignment, is relevant in predicting AKP after TKA. These findings provide valuable insights for optimizing TKA outcomes and reducing the incidence of AKP.