Taghi Ramazanian - ۱ Department of Health Sciences Research, Mayo Clinic, ۲۰۰ First St SW Rochester, Rochester, Minnesota, USA ۲ Department of Orthopedics, Mayo Clinic, ۲۰۰ First St SW Rochester, Rochester, Minnesota, USA
Sunyang Fu - Department of Health Sciences Research, Mayo Clinic, ۲۰۰ First St SW Rochester, Rochester, Minnesota, USA
Sunghwan Sohn - Department of Health Sciences Research, Mayo Clinic, ۲۰۰ First St SW Rochester, Rochester, Minnesota, USA
Michael Taunton - Department of Orthopedics, Mayo Clinic, ۲۰۰ First St SW Rochester, Rochester, Minnesota, USA
Hilal Maradit Kremers - ۱ Department of Health Sciences Research, Mayo Clinic, ۲۰۰ First St SW Rochester, Rochester, Minnesota, USA ۲ Department of Orthopedics, Mayo Clinic, ۲۰۰ First St SW Rochester, Rochester, Minnesota, USA

Background: Knee osteoarthritis (OA) is a prevalent joint disease. Clinical prediction models consider a wide range of risk factors for knee OA. This review aimed to evaluate published prediction models for knee OA and identify opportunities for future model development.Methods: We searched Scopus, PubMed, and Google Scholar using the terms knee osteoarthritis, prediction model, deep learning, and machine learning. All the identified articles were reviewed by one of the researchers and we recorded information on methodological characteristics and findings. We only included articles that were published after ۲۰۰۰ and reported a knee OA incidence or progression prediction model.Results: We identified ۲۶ models of which ۱۶ employed traditional regression-based models and ۱۰ machine learning (ML) models. Four traditional and five ML models relied on data from the Osteoarthritis Initiative. There was significant variation in the number and type of risk factors. The median sample size for traditional and ML models was ۷۸۰ and ۲۹۵, respectively. The reported Area Under the Curve (AUC) ranged between ۰.۶ and ۱.۰. Regarding external validation, ۶ of the ۱۶ traditional models and only ۱ of the ۱۰ ML models validated their results in an external data set. Conclusion: Diverse use of knee OA risk factors, small, non-representative cohorts, and use of magnetic resonance imaging which is not a routine evaluation tool of knee OA in daily clinical practice are some of the main limitations of current knee OA prediction models. Level of evidence: III

Artificial intelligence, Knee Osteoarthritis, Machine Learning, Prediction models