PREOPERATIVE BIOMECHANICAL PATIENT STRATIFICATION BY MACHINE LEARNING-BASED CLUSTER ANALYSIS AND FUNCTIONAL OUTCOMES AFTER TOTAL KNEE ARTHROPLASTY

Introduction

Identifying knee osteoarthritis patient phenotypes is relevant to assessing treatment efficacy. Biomechanical variability has not been applied to phenotyping, yet features may be related to outcomes of total knee arthroplasty (TKA), an inherently mechanical surgery. This study aimed to i) identify biomechanical phenotypes among TKA candidates based on demographic and gait mechanic similarities, and ii) compare objective gait improvements between phenotypes post-TKA.

Methods

TKA patients underwent 3D gait analysis one-week pre (n=134) and one-year post-TKA (n=105). Principal component analysis was applied to frontal and sagittal knee angle and moment gait waveforms, extracting major patterns of variability. Demographics (age, sex, BMI), gait speed, and frontal and sagittal pre-TKA angle and moment principal component (PC) scores previously found to differentiate sex, osteoarthritis (OA) severity, and symptoms of TKA recipients were standardized (mean=0, SD=1, [134×15]) to perform multidimensional scaling and machine learning based hierarchical clustering. Final clusters were validated by examining inter-cluster differences at baseline and gait changes (Post_PCscore–Pre_PCscore) by k-way Chi-Squared, and ANOVA tests.

Results

Four (k=4) TKA candidate groups yielded optimum clustering metrics, interpreted as 1) high-functioning males, 2) older stiff-kneed males, 3) slower stiff-kneed females, and 4) high-functioning females. Pre-TKA, higher-functioning clusters (1 & 4) had more dynamic loading/un-loading kinetic patterns during stance (flexion moment PC2, 3<2<4<1, P<0.001; adduction moent PC2; 3,2<4<1; P<0.001). Post-TKA, higher-functioning clusters demonstrated less gait improvement (flexion angle ΔPC2, 1,2,4<3, P<0.001; flexion moment ΔPC2, 4<2,3, P<0.001; adduction moment ΔPC2, 1<3, P=0.01).

Conclusions

TKA candidates can be characterized by four clusters, interpreted as 1) high-functioning males, 2) older stiff-kneed males, 3) slower stiff-kneed females, and 4) high-functioning females, differing by demographics and biomechanical severity features. Functional gains after TKA were cluster-specific; stiff-gait clusters experienced more improvement, while higher-functioning clusters demonstrated some functional decline. Results suggest the presence of cohorts who may not benefit functionally from TKA. Cluster profiling may aid in triaging and developing osteoarthritis management and surgical strategies that meet individual or group-level function needs.

For figures, tables, or references, please contact authors directly.