Machine-learning predicted and actual 2-year structural progression in the IMI-APPROACH cohort

Area of Science:

• Orthopedics
• Radiology
• Medical Imaging

Background:

• Knee osteoarthritis (OA) structural progression is a key factor in disease management.
• Predictive models are crucial for identifying individuals at risk of OA progression.
• The IMI-APPROACH study aimed to evaluate machine learning-based predictions against observed structural changes.

Purpose of the Study:

• To assess the predictive accuracy of machine learning-derived s-scores and Kellgren-Lawrence (KL) grades for knee OA structural progression over two years.
• To compare the performance of different radiographic and MRI-based structural parameters in predicting OA progression.
• To evaluate the rate of structural progression in participants of the IMI-APPROACH study.

Main Methods:

• Trained machine learning models to predict structural progression (s-score) based on >0.3 mm/year joint space width (JSW) decrease.
• Acquired radiographs and MRI scans at baseline and 2-year follow-up for 237 participants.
• Measured radiographic (JSW, bone density, osteophytes) and MRI (cartilage thickness, bone marrow lesions, cartilage damage) parameters, analyzing progression rates and prediction accuracy using logistic regression.

Main Results:

• Approximately 1 in 6 participants exhibited structural progression based on the JSW criterion.
• Highest progression rates were observed in radiographic bone density (39%), MRI cartilage thickness (38%), and radiographic osteophyte size (35%).
• Kellgren-Lawrence grades were significantly better predictors of structural progression across most radiographic and MRI parameters compared to machine learning-based s-scores.

Conclusions:

• Between 1/6 and 1/3 of participants demonstrated structural progression over the 2-year follow-up period.
• Kellgren-Lawrence grading outperformed machine learning s-scores in predicting knee OA structural progression.
• The extensive dataset offers potential for developing more sensitive and comprehensive joint-wide prediction models for OA progression.