Machine Learning Prediction of Poor Flexion-Extension Outcome After Open Elbow Arthrolysis: Identifying Individual...

Area of Science:

• Orthopedic Surgery
• Biomedical Engineering
• Machine Learning in Medicine

Background:

• Elbow stiffness significantly impairs mobility and quality of life.
• Patients undergoing open elbow arthrolysis face risks of poor postoperative range of motion.
• A unique 'elbow stiffness predisposition' is suspected in these patients.

Purpose of the Study:

• To develop a predictive model for poor elbow flexion-extension outcomes post-arthrolysis.
• To interpret the 'elbow stiffness predisposition' using clinical and laboratory data.
• To enhance patient outcomes through early risk identification.

Main Methods:

• A cohort study involving 254 patients for training/validation and 35 for testing.
• Analysis of 19 clinical features and 58 laboratory parameters.
• Comparative evaluation of machine learning models, including XGBoost, with SHAP for factor prioritization.

Main Results:

• An XGBoost model achieved an area under the curve of 0.909 on the test dataset.
• Fourteen key variables were selected using LASSO regression.
• Lipoprotein(a), alkaline phosphatase, and visual analog scale score were among the top predictors identified by SHAP.

Conclusions:

• A machine learning model effectively predicts poor outcomes in post-arthrolysis elbow stiffness.
• The model, utilizing 14 indicators, offers preliminary insights into the 'elbow stiffness predisposition'.
• This predictive tool can aid in managing patients at risk for poor elbow function.