BIOMECHANICAL AND BIOLOGICAL PHENOTYPES EMERGING WITHIN ONE YEAR POST-ANTERIOR CRUCIATE LIGAMENT RECONSTRUCTION: A MACHINE LEARNING APPROACH TO UNDERSTANDING EARLY KNEE OSTEOARTHRITIS

Context: Anterior cruciate ligament (ACL) injury and reconstruction (ACLR) are associated with an increased risk of knee osteoarthritis (KOA), yet the early mechanobiological factors contributing to KOA risk post-ACLR are not well understood. Objective: This dissertation sought to characterize variability in biomechanical and biological profiles following ACLR and determine the relationships between biomechanical and biological factors with early KOA-related outcomes using robust machine learning methods. Aim 1: Bayesian joint trajectory modeling was used to identify integrated biomechanical recovery trajectories of vertical ground reaction force (vGRF) and knee flexion excursion (KFE) 2-6 months post-ACLR. Two integrated biomechanical trajectories emerged: i) less dynamic vGRF loading with reduced KFE, and ii) more dynamic vGRF loading with greater KFE. Individuals comprising the less dynamic trajectory demonstrated greater declines in cartilage composition (T1ρ relaxation times) and higher concentrations of 12-month u-CTX-II, despite no differences in patient-reported outcomes. Aim 2: Gaussian mixture modeling was used to evaluate longitudinal biomarker trajectories (s-MCP-1, s-MMP-3, s-COMP) from preoperative to 2 and 6 months post-ACLR. Distinct high and low biomarker trajectory groups were identified for inflammatory (s-MCP-1) and cartilage metabolism (s-MMP-3, s-COMP) biomarkers. Individuals in higher inflammatory (s-MCP-1) and matrix metabolism (s-MMP-3) trajectories demonstrated greater deleterious cartilage changes between preoperative and 12 months post-ACLR (%ΔT1ρ). Aim 3: Functional data analysis approaches identified variation patterns in 6-month vGRF waveforms. Machine learning methods (Data Integration Via Analysis of Subspaces) were used to identify shared biomechanical-biological phenotypes at 6 months post-ACLR. A biomechanical-biological phenotype revealed that greater early stance loading was associated with lower KOA-related biomarker concentrations (s-MCP-1, s-MMP-3, s-COMP) 6-months post-ACLR. This shared biomechanical-biological phenotype associated with better cartilage composition (lower T1ρ relaxation times) in the lateral tibiofemoral compartment at 12 months, suggesting a link between early loading patterns and biological responses with future joint tissue health. Conclusion: These findings demonstrate that distinct biomechanical and biological recovery phenotypes emerge within one year post-ACLR and are associated with early KOA-related cartilage compositional changes. These results highlight the importance of integrating longitudinal biomechanical and biological assessments to better understand early KOA-related changes and identify individuals at highest-risk for KOA-related outcomes following ACLR.