JMIR Publications

ABSTRACT

Background:

Knee pain affects 22.9% of individuals aged over 40 globally, with higher prevalence in Asian older adults (33.3%). Traditional analytical approaches examining isolated risk factors inadequately capture knee pain's complex etiology, where pain functions as part of interconnected health networks rather than isolated symptoms. Current assessment methods lack integration of machine learning with network analysis for comprehensive understanding of multidimensional pain determinants in aging populations.

Objective:

To integrate machine learning classification with network analysis for comprehensive understanding of knee pain determinants in community-dwelling older adults, utilizing smartphone-based physical performance assessment within the International Classification of Functioning framework.

Methods:

This cross-sectional study analyzed 852 community-dwelling adults aged ≥60 years using integrated machine learning and network analysis methodologies. Comprehensive assessment included 37 predictor variables across sociodemographic, physical performance, quality of life, mental health, and lifestyle domains. Physical performance was measured using smartphone-based computer vision pose estimation. Harris Hawks Optimization identified optimal 10-feature subsets. Five machine learning algorithms classified knee pain presence with 5-fold cross-validation. Network analysis examined interconnectedness among machine learning-identified features using graphical LASSO methods with bootstrap validation, quantifying variable relationships through density measures and betweenness centrality.

Results:

Knee pain prevalence was 36.6% (n=256/852). Random Forest achieved optimal classification performance (AUC=0.705, accuracy=68.6%) with predictors including housing estate type, EQ-5D utility scores, sitting time, and walking speed. Critical actionable thresholds were identified: sitting time >150 minutes/day, walking speed <1.3 m/s, sit-to-stand performance <12 repetitions. Network analysis revealed knee pain as a central mediator (2nd highest betweenness centrality=0.089) connecting six health domains, with 41.8% network density demonstrating substantial interconnectedness among aging-related health determinants.

Conclusions:

Knee pain functions as a central mediator in interconnected health networks among aging populations. This integrated machine learning and network analysis approach demonstrates that pain represents a multidimensional network phenomenon requiring comprehensive assessment rather than isolated symptom management, supporting implementation of multi-modal intervention approaches in digital health platforms.