Predicting total knee replacement at 2 and 5 years in osteoarthritis patients using machine learning

Area of Science:

• Orthopedics and Biomedical Engineering
• Machine Learning in Healthcare
• Osteoarthritis Research

Background:

• Knee osteoarthritis (OA) significantly impairs quality of life and physical function.
• Total knee replacement (TKR) is a common treatment for end-stage knee OA.
• Predictive models for TKR are crucial for timely intervention and resource planning.

Purpose of the Study:

• To develop and externally validate a machine learning model for predicting TKR.
• To forecast the need for TKR at 2-year and 5-year intervals.
• To utilize routinely collected health data for predictive modeling.

Main Methods:

• Prospective study utilizing Osteoarthritis Initiative (OAI) and Multicentre Osteoarthritis Study (MOST) datasets.
• Feature selection curated 45 candidate features from demographics, medical history, imaging, and interventions.
• Gradient boosting machine model trained on OAI data and validated on MOST data.

Main Results:

• The best-performing model achieved an AUC of 0.913 at 2 years and 0.873 at 5 years.
• Key predictors included radiographic features, questionnaire data, and educational attainment.
• The model demonstrated clinically acceptable accuracy (AUC > 0.7) and was externally validated.

Conclusions:

• Routinely collected patient data can effectively drive a predictive model for TKR.
• This externally validated model represents a novel tool for predicting knee OA progression.
• The model's accuracy surpasses previous methods relying on non-routine MRI data.