Physical Frailty Prediction Using Cane Usage Characteristics during Walking.

Area of Science:

• Gerontology
• Biomedical Engineering
• Rehabilitation Science

Background:

• Physical frailty is a significant concern in aging populations, impacting mobility and independence.
• Objective assessment of physical frailty in community-dwelling older adults is crucial for timely intervention.
• Gait analysis using wearable sensors offers a promising avenue for non-invasive frailty assessment.

Purpose of the Study:

• To characterize gait parameters (accelerations, angular velocities) measured by an inertial measurement unit (IMU) on a cane in older adults with and without physical frailty.
• To evaluate the effectiveness of machine learning models in distinguishing between frail and non-frail older adults based on IMU data.
• To identify specific gait characteristics associated with physical frailty.

Main Methods:

• Community-dwelling older adults walked with a cane equipped with an IMU.
• Physical frailty was assessed using exercise-related items from the Kihon Check List.
• Five machine learning models were trained and tested to classify physical frailty using IMU-derived gait metrics.

Main Results:

• Older adults with physical frailty exhibited smaller root mean square values in vertical and anteroposterior directions and reduced anteroposterior angular velocity compared to non-frail individuals.
• Frail participants showed a larger mean power frequency in the vertical direction.
• The decision tree model achieved the highest classification performance, with 78.6% accuracy, 91.8% F1 score, and 0.81 AUC.

Conclusions:

• Gait characteristics captured by an IMU-attached cane differ between physically frail and non-frail older adults.
• IMU-based gait analysis holds potential for effective and objective evaluation of physical frailty in real-world settings.
• Machine learning, particularly decision trees, can accurately identify physical frailty using cane-based IMU data.