Identification of Subgroups of Individuals Experiencing Patellofemoral Pain with Kinematic and Kinetic Features During Overground Running

Patellofemoral pain (PFP) is a common running related injury associated with several biomechanical factors such as altered kinematics and kinetics across lower extremity joints. Prior research suggests mechanisms for PFP may differ within those affected, indicating that subgroups characterized by unique factors might exist. Subgroups have previously been found using unsupervised machine learning (ML) models based on clinical tests, imaging, and/or biomechanical stair negotiation variables, however clusters have not yet been identified with 3-dimensional running mechanics. Therefore, the purpose of this study was to leverage an unsupervised ML algorithm to identify clusters of individuals experiencing unique PFP related biomechanical factors and furthermore characterize features of each subgroup. 40 participants experiencing PFP and 20 healthy controls completed overground running trials in a motion capture environment. Kinematic and kinetic variables were reduced with principal component (PC) analysis. PC-based features were fed into a k-means clustering model optimized on a measure of cluster purity. Four unique subgroups were identified, three of which contained mostly individuals experiencing PFP. Each subgroup displayed distinct kinematic and/or kinetic features of running gait related to development of PFP. The first PFP subgroup was characterized by excessive knee abduction and rotation moments, and reduced rearfoot inversion moments. The second PFP subgroup displayed greater knee joint frontal plane loading rates and rearfoot eversion angles. The third PFP subgroup was characterized by greater hip adduction, hip rotation, and knee abduction angles. Identifying and characterizing subgroups is a key step for developing targeted treatments that may be more effective for mitigating PFP. HighlightsO_LIThree distinct subgroups were identified with a clustering model and principal component-based features, each with unique kinematic and kinetic patterns during overground running. C_LIO_LIKey features for differentiating subgroups included the knee abduction moment, rearfoot inversion moment, knee frontal plane loading rate, rearfoot eversion angle, hip adduction angle, hip rotation angle, and knee abduction angle. C_LIO_LIEach subgroup may benefit from targeted intervention strategies as opposed to generalized treatment procedures. C_LI