Relation of gait measures with mild unilateral knee pain during walking using machine learning

Kathryn L. Bacon, David T. Felson, S. Reza Jafarzadeh, Vijaya B. Kolachalama, Jeffrey M. Hausdorff, Eran Gazit, Neil A. Segal, Cora E. Lewis, Michael C. Nevitt, Deepak KumarScientific 2022 · DOI

TL;DR

This study leverages machine learning on wearable sensor data to identify subtle gait alterations associated with mild unilateral knee pain. The findings suggest that reduced gait regularity and complexity are early indicators of knee pain, offering potential for non-invasive monitoring and intervention strategies.

What Problem Does This Paper Address?

Knee osteoarthritis (OA) is a prevalent condition often accompanied by pain and altered gait. While significant gait deviations are known in advanced OA, understanding the subtle changes associated with mild unilateral knee pain is crucial. This early pain might be a precursor to more severe OA and bilateral involvement. Current methods for gait analysis often rely on expensive and time-consuming laboratory equipment, limiting their widespread application. This paper aims to bridge this gap by using wearable sensors and machine learning to identify gait markers of mild knee pain in a large, real-world cohort.

Methods

The researchers employed a two-step machine learning approach. First, an ensemble machine learning technique called "super learning" was used to identify the most influential gait parameters associated with the presence or absence of mild unilateral knee pain. This method combines multiple algorithms to achieve optimal predictive accuracy and was run 100 times to ensure robustness. The input variables included a comprehensive set of spatiotemporal gait features derived from trunk and ankle sensors, along with demographic and clinical covariates.

Following the variable selection, logistic regression was used to quantify the association between the identified gait parameters and mild unilateral knee pain. This allowed for adjusted analyses, controlling for factors like age, BMI, and radiographic OA severity. The study also conducted sensitivity analyses to explore associations with moderate/severe pain and to assess the impact of gait speed.

Datasets

The study utilized data from the Multicenter Osteoarthritis (MOST) study, a large, NIH-funded cohort of individuals at risk of or with knee OA. The analysis included participants from the 144-month clinic visit, who underwent a 20-m walk test while wearing inertial sensors. The dataset comprised 2066 participants with gait data and pain outcome information. No new benchmarks were introduced; the study leverages existing, well-characterized observational data.

Key Insights and Ideas

  • Early Gait Signatures of Pain: The study highlights that even mild unilateral knee pain can manifest as detectable changes in gait, specifically in variability and complexity, rather than just speed. This suggests that the nervous system is adapting to pain even at its earliest stages.
  • Wearable Sensors as a Powerful Tool: The successful application of machine learning to wearable sensor data demonstrates the potential for efficient, large-scale gait analysis outside of traditional lab settings. This opens doors for more accessible and frequent monitoring.
  • Complexity and Regularity as Key Indicators: Reduced step regularity (increased variability) and lower sample entropy (decreased complexity) were consistently linked to a higher likelihood of mild knee pain. This implies that the body's control over movement becomes less consistent and adaptable when experiencing early knee discomfort.
  • Distinguishing Mild vs. Severe Pain: While variability and complexity are key for mild pain, slower gait speed emerged as a significant factor for moderate to severe pain. This suggests a progression in gait adaptations as pain intensifies.

Key Results

In adjusted cross-sectional analyses, the study found that:

  • Lower step regularity (greater gait variability) was associated with an increased likelihood of mild unilateral knee pain while walking (aOR 0.80, 95% CI 0.64-1.00).
  • Lower sample entropy (lower gait complexity) was also associated with an increased likelihood of mild unilateral knee pain while walking (aOR 0.79, 95% CI 0.66-0.95).

These associations remained significant after controlling for age, sex, BMI, and radiographic OA status. The machine learning model achieved a median AUC of 0.75, indicating a reasonable ability to distinguish between individuals with and without mild knee pain.

Limitations and Future Work

The authors acknowledge the cross-sectional nature of the study, which prevents the determination of causality. It's possible that individuals adapt their gait in response to pain, rather than gait alterations directly causing pain. Residual confounding, despite adjustments, is also a possibility in observational studies.

Future work should focus on longitudinal studies to establish the temporal relationship between gait changes and the onset or progression of knee pain. Investigating interventions aimed at improving gait regularity and complexity, such as mindfulness meditation or gait adaptability training, is also a promising avenue. The authors also suggest that advances in wearable sensors could facilitate rapid gait assessment in clinical and real-world settings, potentially aiding in early detection and management of knee OA.

Summary updated

Abstract

Gait alterations in those with mild unilateral knee pain during walking may provide clues to modifiable alterations that affect progression of knee pain and osteoarthritis (OA). To examine this, we applied machine learning (ML) approaches to gait data from wearable sensors in a large observational knee OA cohort, the Multicenter Osteoarthritis (MOST) study. Participants completed a 20-m walk test wearing sensors on their trunk and ankles. Parameters describing spatiotemporal features of gait and symmetry, variability and complexity were extracted. We used an ensemble ML technique ("super learning") to identify gait variables in our cross-sectional data associated with the presence/absence of unilateral knee pain. We then used logistic regression to determine the association of selected gait variables with odds of mild knee pain. Of 2066 participants (mean age 63.6 [SD: 10.4] years, 56% female), 21.3% had mild unilateral pain while walking. Gait parameters selected in the ML process as influential included step regularity, sample entropy, gait speed, and amplitude dominant frequency, among others. In adjusted cross-sectional analyses, lower levels of step regularity (i.e., greater gait variability) and lower sample entropy(i.e., lower gait complexity) were associated with increased likelihood of unilateral mild pain while walking [aOR 0.80 (0.64-1.00) and aOR 0.79 (0.66-0.95), respectively].
BibTeX
@inproceedings{bacon2022relation,
  author = {Kathryn L. Bacon and David T. Felson and S. Reza Jafarzadeh and Vijaya B. Kolachalama and Jeffrey M. Hausdorff and Eran Gazit and Neil A. Segal and Cora E. Lewis and Michael C. Nevitt and Deepak Kumar and David T. Felson and Neil A. Segal and Cora E. Lewis and Michael C. Nevitt},
  title = {{Relation of gait measures with mild unilateral knee pain during walking using machine learning}},
  booktitle = {Scientific},
  year = {2022},
  doi = {10.1038/s41598-022-21142-2},
}
Figure