Date of Award


Document Type


Degree Name

Doctor of Philosophy (PhD)


Complex Systems and Data Science

First Advisor

Ryan S. McGinnis

Second Advisor

Kimberly R. Bauerly


Multiple sclerosis is an immune-mediated disease of the central nervous system that commonly results in impaired sensory function, balance, coordination, and fatigue. Persons with multiple sclerosis (PwMS) experience high rates of falls, with over half experiencing a fall in any three-month period. Presently, fall risk is assessed at biannual office visits, but symptoms are known to fluctuate and it is not clear that these assessments provide an adequate picture of a patient’s fall risk. Remote monitoring of a patient’s balance and mobility with wearable sensors may provide a way to better characterize fall risk, but technologies for making these measurements are just emerging. The purpose of this work was to advance the state of science in remote monitoring of balance and mobility, and to use the resulting technology to identify potential predictors of fall risk in PwMS.The primary technical contribution of this work was a data analysis platform that allows for remote characterization of balance and mobility impairment. The platform detects walking and standing bouts from free-living wearable accelerometer data and computes metrics that describe how patients are engaging in these balance-challenging activities. This platform was leveraged to examine data from two cohorts of PwMS. First, data from the platform were used to better understand the relationship between walking bout duration and measures that describe how a patient is walking. Walking metrics were significantly different between bouts of differing lengths, and between walking bouts observed in and out of the clinic. Long remote bouts were the closest to in-clinic measurements and were best able to identify PwMS at higher risk for falls using deep learning models. Interestingly, short remote bouts were best when using more traditional machine learning techniques. Data from the platform were then used to investigate how much data is enough for capturing valid measures of balance and mobility impairment remotely. Analysis revealed only two days of data are needed to capture most measures of gait and postural sway in our cohorts of PwMS. In general, minimum wear duration was predicted by the daily variability of a measurement and number of daily observations. Finally, data from the platform were used to further establish remote postural sway, measured by chest-worn accelerometer, as a digital endpoint for balance impairment. Chest-derived measures of sway were validated relative to gold-standard force platforms. A new analysis approach, which builds individualized distributions of each postural sway measure, was introduced that increased accuracy for classifying PwMS’ risk for falls and the strength of associations with patient-reported measures (PRMs) of balance impairment. Remote measures of sway differed from these lab measures but had stronger associations to PRMs. A patient-specific clustering approach for analyzing remote sway further strengthened associations and enabled detection of PwMS at higher risk for falls. Overall, this body of work addresses several key challenges of remote wearable sensor data analysis and introduces remote postural sway as a novel digital endpoint for balance impairment.



Number of Pages

129 p.

Available for download on Saturday, February 17, 2024