Date of Award

2021

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Biomedical Engineering

First Advisor

Ryan McGinnis

Abstract

Multiple sclerosis is a chronic, neurodegenerative disease characterized by inflammation and demyelination of the central nervous system that leads to impaired coordination, muscle strength, and sensation. Balance and postural control deficits are common in people with multiple sclerosis (PwMS) as 50% will fall in any three-month period. Clinical exams have evolved to include functional assessments such as the 30-second chair stand test (30CST) to evaluate balance, fall risk and provide insight into functional mobility. However, psychometric properties are limited and there remain gaps for statistically characterizing change over time and generalizing performance to daily life. While advances in wearable sensor technology allow for longer term monitoring and biomechanical analysis outside the clinic, our methods for integrating contextual information from daily life to identify people at risk for falls and preemptively intervening before a fall occurs remain suboptimal. The work herein is focused on developing quantitative biomarkers of impairment and fall risk using wearable sensor data captured during sit-to-stand (STS) transitions and represents a comprehensive evaluation of the STS during structured (30CST) and unstructured tasks in supervised and unsupervised (daily life) settings.

Triaxial accelerometer data were collected for 40 PwMS from two wearable sensors during supervised and unsupervised monitoring periods. PwMS completed 30CSTs during supervised visits then participated in two days of unsupervised monitoring in which they performed bi-hourly 30CSTs. An automated algorithm was developed to process raw accelerometer data and delineate the STS into regions of interest. Accelerometer-derived performance metrics were developed and evaluated for their ability to classify fall risk, and impairment through multiple techniques. First, metrics derived from supervised 30CSTs were used to inform machine learning models to classify fall risk and model performance was compared to models trained on current standards of care. Then unsupervised 30CST performance was compared to supervised performance and performance variability was characterized. Finally, STS transitions were identified during daily life and associated metrics were compared to 30CST metrics to classify fall risk. The feasibility of using metrics to discriminate pyramidal and sensory impairment was also explored.

The metric that best discriminated high and low fall risk was the average sit-stand time from daily life STS transitions (AUC=0.85 vs AUC=0.69 for the current standard of care). A threshold of 13 repetitions was proposed to identify PwMS at high risk for falls in the clinic using the 30CST. While unsupervised STS metrics optimized fall risk classification, supervised 30CST metrics best classified high/low pyramidal impairment (AUC=0.85) and only supervised 30CST features classified high/low sensory impairment. This work underscores the benefits of instrumented analysis for both structured and daily life tasks as well as short bouts of unsupervised monitoring to inform clinical decision making. Future work should build on these findings to improve the clinical adoption of wearable sensors for patient monitoring and explore physiological significance of performance metrics to inform personalized intervention.

Language

en

Number of Pages

96 p.

Download

Share

COinS