Hip Morphology-Based Osteoarthritis Risk Prediction Models: Development and External Validation Using Individual...

Area of Science:

• Orthopedics
• Radiology
• Biostatistics

Background:

• Radiographic hip osteoarthritis (RHOA) is a degenerative joint disease.
• Early risk prediction is crucial for timely intervention.
• Hip morphology is a potential risk factor for RHOA development.

Purpose of the Study:

• To develop hip morphology-based RHOA risk prediction models.
• To assess the added predictive value of hip morphology measurements.
• To evaluate the generalizability of these models across diverse populations.

Main Methods:

• Combined data from nine prospective cohort studies (World COACH consortium).
• Harmonized RHOA grades and defined incident RHOA.
• Quantified hip morphology using lateral center edge and alpha angles on radiographs.
• Utilized generalized linear mixed models (GLMM) and Random Forest (RF) for prediction, with stratified and leave-one-cohort-out cross-validation.

Main Results:

• 4.7% of hips without definite RHOA developed incident RHOA within 4-8 years.
• GLMM with hip morphology achieved a mean AUC of 0.80 in stratified cross-validation.
• Leave-one-cohort-out cross-validation showed heterogeneous performance (AUC 0.56-0.88) for morphology-based models.

Conclusions:

• RHOA risk models demonstrated good predictive performance in similar populations.
• The incremental predictive value of hip morphology measurements was limited.
• Model generalizability across different populations was variable.