OAI Data.

Data used to understand knee osteoarthritis (KOA) often involves knee-level, rather than person-level information. Failure to account for the correlation between joints within a person may lead to inaccurate inferences. The aim of this study was to develop a flexible, data-driven framework for predicting knee pain outcomes, incorporating the advantages of both random forest (RF) and mixed effects models for correlated data. Specifically, we utilized data from the baseline visit of the Osteoarthritis Initiative (OAI) and applied the Binary Mixed Models (BiMM) algorithm to predict two binary dependent variables. 1) presence of knee pain, stiffness or aching in the past 12 months and 2) presence of knee pain indicated by a KOOS pain score > 85. This novel approach was compared to standard random forests (RF), which do not account for correlations among knees. This study demonstrates the potential of BiMM as a predictive tool for KOA pain, achieving a comparable or slightly improved performance over traditional RF models while simultaneously accounting for within-person correlation among knees. This is a significant advancement, as most machine learning models to date have only considered each knee individually. These findings support the integration of BiMM in KOA outcome prediction, providing a nuanced alternative to existing models and advancing our understanding of important KOA outcomes on the person level. Although demonstrated here for KOA, this method is relevant to any situation where within-person correlations are relevant, including other joints and other musculoskeletal conditions.