An interpretable knee replacement risk assessment system for osteoarthritis patients

Area of Science:

• Orthopedics
• Biostatistics
• Data Science

Background:

• Knee osteoarthritis (KOA) is a complex condition with varied presentations.
• Effective prevention and early treatment are possible, but reliable prognostic tools are lacking.
• A quantitative, self-administrable risk stratification system for knee replacement (KR) is needed for KOA patients.

Purpose of the Study:

• To develop a quantitative and self-administrable risk stratification system for knee replacement (KR) in knee osteoarthritis (KOA) patients.
• To identify key clinical features influencing KR risk.
• To create an interpretable model for patient self-assessment and personalized healthcare.

Main Methods:

• Utilized 14 baseline clinical features from 9592 cases in the Osteoarthritis Initiative (OAI) cohort.
• Constructed a survival model using the Random Survival Forests algorithm.
• Evaluated model performance using concordance index (C-index) and average AUC; employed SHAP for interpretability.

Main Results:

• The model accurately predicted KR incidence with a C-index of 0.770 and average AUC of 0.807.
• A ten-point system stratified patients into low, medium, and high-risk groups with distinct four-year KR rates (0.79%, 5.78%, 16.2%).
• Key predictors for KR included pain medication use, age, surgery history, diabetes, and high BMI.

Conclusions:

• Developed a self-administrable and interpretable KR survival model and risk scoring system for KOA patients.
• The system facilitates patient self-assessment and personalized healthcare for KOA prevention.
• Aims to improve primary and secondary prevention strategies for knee osteoarthritis.