Gait biomechanics phenotypes among total knee arthroplasty candidates by machine learning cluster analysis

Knee osteoarthritis patient phenotyping is relevant to developing targeted treatments and assessing the treatment efficacy of total knee arthroplasty (TKA). This study aimed to identify clusters among TKA candidates based on demographic and knee mechanic features during gait, and compare gait changes between clusters postoperatively. TKA patients underwent 3D gait analysis 1-week pre (n = 134) and 1-year post-TKA (n = 105). Principal component analysis was applied to frontal and sagittal knee angle and moment waveforms, extracting major patterns of variability. Age, sex, body mass index, gait speed, and frontal and sagittal pre-TKA angle and moment PC scores previously identified as relevant to TKA outcomes were standardized (mean = 0, SD = 1, [134 × 15]). Multidimensional scaling and machine learning-based hierarchical clustering were applied. Final clusters were validated by examining intercluster differences pre-TKA and gait feature changes (Post_PCscore  - Pre_PCscore ) by k-way Χ² and ANOVA tests. Four TKA candidate phenotypes yielded optimum clustering metrics, interpreted as higher and lower functioning clusters that were predominantly male and female. Higher functioning clusters pre-TKA (clusters 1 and 4) had more dynamic sagittal flexion moment (p < 0.001) and frontal plane adduction moment (p < 0.001) loading/un-loading patterns during stance. Post-TKA, higher functioning clusters demonstrated less knee mechanic improvements during gait (flexion angle p < 0.001; flexion moment p < 0.001). TKA candidates can be characterized by four clusters, predominately separated by sex and knee joint biomechanics. Post-TKA knee kinematics and kinetics improvements were cluster-specific; lower functioning clusters experienced more improvement. Cluster-based patient profiling may aid in triaging and developing OA management and surgical strategies meeting group-level function needs.