JMIR Publications

ABSTRACT

Background:

Early identification of modifiable risk factors for clinical knee osteoarthritis (KOA) is crucial for implementing evidence-based preventive rehabilitation interventions in older adults.

Objective:

To develop and validate machine learning models predicting 24-month incident clinical KOA using baseline biomechanical characteristics in asymptomatic older adults and identify key modifiable targets for rehabilitation.

Methods:

A total of 125 community-dwelling adults aged 60-80 years without baseline knee symptoms were recruited in Hong Kong (November 2018-October 2019). Baseline assessments included quadriceps and patellar tendon stiffness via shear-wave ultrasound elastography, isokinetic knee extensor/flexor strength, and handheld dynamometry for hip abductors. Clinical KOA was defined using American College of Rheumatology criteria at 24-month follow-up. Five machine learning algorithms were developed using Harris Hawks Optimization for feature selection, with bootstrap validation and independent test set evaluation.

Results:

Of 125 participants completing follow-up, 17 (13.6%) developed clinical KOA. CatBoost demonstrated superior predictive performance with area under the curve 0.766 (95% CI: 0.592-0.902) on external validation. Seven predictive features were identified: weight, body mass index, knee extensor strength, rectus femoris stiffness, vastus medialis stiffness, patellar tendon stiffness, and cardiopulmonary disease. Body mass index emerged as the most influential predictor with distinctive bimodal risk distribution at 21.8-24 kg/m² and ≥28.2 kg/m² ranges. Vastus medialis stiffness demonstrated a critical threshold effect at 4.5 kPa, below which KOA risk increased substantially.

Conclusions:

Machine learning effectively predicts incident clinical KOA using modifiable biomechanical parameters, providing evidence-based targets for rehabilitation interventions. The 24-month prediction window enables implementation of personalized prevention strategies focusing on weight management and muscle quality enhancement in at-risk individuals before irreversible joint damage occurs.