A Reciprocal Machine Learning Approach to Defining Physiologic Compensation of the Synovial Joint in Knee Osteoarthritis

Objectives Osteoarthritis (OA) is the most common joint disease worldwide, and one of the most common synovial joints affected is the knee. Synovial joints are organs that permit movement. Like other organs, synovial joints must respond physiologically to biomechanical and molecular stresses to maintain homeostasis. To study the physiologic response of the organ, a definition of synovial organ compensation is required. We propose that physiologic compensation of the synovial organ is defined by an ability to maintain homeostasis in response to load or injury, thereby preserving function and preventing tissue damage. By contrast, synovial organ decompensation results in OA, leading to joint dysfunction, tissue damage, and pain experience. Our objective was to produce a candidate clinical definition of physiologic joint compensation using a reciprocal machine learning (ML) approach. Methods A reciprocal ML approach employing data-driven and ground-truth classifiers was applied. Data was derived from patients in the Western Ontario Registry for Early Osteoarthritis (WOREO; n=774 knees). Included variables were derived from a battery of clinical assessments (n=34). The data-driven classifier first used a principal component analysis (PCA) to reduce dimensionality. A k-means clustering algorithm was applied to the multidimensional space defined by the first 4 principal components and revealed 3 identifiable clusters. A reciprocal ground-truth classifier utilized patient-reported pre/post pain on the 6-minute walk test to define 3 disease clusters (compensatory status). A random forest algorithm identified a subset of measures capable of predicting compensatory status. Results Findings from the PCA revealed that the first 4 principal components (PCs) explained 58.0% of the variance. A permutation test confirmed that PC1-4 robustly explained dataset variance (p < 0.001). Post-hoc clustering showed that the features with the most variance between clusters were knee injury and osteoarthritis outcome score (KOOS; 66.2 ± 15.6), mechanical axis angle (MAA; −3.5 ± 13.2), BMI (32.1 ± 2.12), age (62.5 ± 4.3), and uric acid (321.3 ± 57.4). The random forest algorithm revealed that KOOS pain, MAA, BMI, age, and peak knee flexion were most predictive of compensatory status (vector similarity = 0.80; Ɵ = 36.87). Conclusion A clinical definition of synovial organ compensation is required to study the underlying physiology. Our findings support the proposed model of physiologic compensation and offer candidate clinical measures capable of determining the compensatory status of the synovial organ in the knee. Next steps involve validating the model via analysis of the physiologic profiles associated with compensatory status (i.e., synovial fluid proteomics).

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.