Abstract
Background: Some patients undergoing total knee arthroplasty successfully manage their condition postoperatively, whereas others encounter challenges in regaining function and controlling pain during recovery at home.
Objective: The aim of this study was to use traditional statistics and machine learning to develop prediction models that identify patients likely to have increased care needs related to managing function and pain following total knee arthroplasty.
Methods: This study included 201 patients. Outcomes were changes between baseline and follow-up in the functional and pain subcomponents of the Oxford Knee Score. Both classification and regression modeling were applied. Twenty-one predictors were included. Tenfold cross-validation was used, and the regression models were evaluated based on root-mean-square error, mean absolute error, and coefficient of determination. Classification models were evaluated based on the area under the receiver operating curve, sensitivity, and specificity.
Results: In classification modeling, random forest and stochastic gradient boosting provided the best overall metrics for model performance. A support vector machine and a stochastic gradient boosting machine in regression modeling provided the best predictive performance. The models performed better in predicting challenges related to function compared to challenges related to pain.
Discussion: There is valuable predictive information in the data routinely collected for patients undergoing total knee arthroplasty. The developed models may predict patients who are likely to have enhanced care needs regarding function and pain management. Improvements are needed before the models can be implemented in routine clinical practice.