Free-Living Physical Activity Monitoring in Adult US Patients with Multiple Sclerosis Using a Consumer Wearable Device

Abstract

Introduction: Wearable devices have been used to characterize physical activity in multiple sclerosis (MS). The objectives of this study were to advance the literature on the utility of free-living physical activity tracking from secondary analyses of a pilot study in MS patients. Method: The original observational study was conducted in participants with MS at PatientsLikeMe (www.PatientsLikeMe.com), an online network of patients with chronic diseases. Participants completed a baseline self-assessment, and received a Fitbit One^TM wearable device with instructions to upload data. Eligible participants (1) self-reported MS, (2) logged on to the PatientsLikeMe website 90 days prior to enrollment, and (3) consented to participate electronically. Participants (1) < 18 years, (2) living outside the United States, and (3) requiring wheelchair assistance for most daily activities were excluded. The secondary analyses were limited to participants with complete data on MS type, disease duration, and Multiple Sclerosis Rating Scale (MSRS) and at least 7 days of wearable data. Step count was used as a measure of physical activity. Results: The analysis cohort of 114 participants uploaded a mean of 20.1 days of wearable data over the 23-day study (87% adherence); participants averaged 4,393 steps per day. The mean age of participants was 52 years, predominantly female (75%), relapsing-remitting type (79%), with mean disease duration of 16 years. Mean MSRS score within 30-day of baseline was 32; 72% reported mild-moderate walking disability. The reliability of step count measured by intraclass correlation was 0.55 for a single day, ≥0.7 for 2-day average, and ≥0.9 for 7-day average. After controlling for covariates, self-reported disease severity (MSRS quartile) was an independent predictor of step count (p < 0.001). Least square means (LS means) for participants that were least disabled (lowest quartile) was 5,937 steps, which was significantly higher than participants in the second, third, and fourth quartiles (4,570, 3,490, and 3,272, respectively). Similarly, LS means of participants with no ambulatory disability (measured by MSRS walk component) was 6,931 steps, significantly higher than participants with greater disability (4,743, 4,394, 2,727 steps for symptomatic, mild, and moderate disability, respectively, p < 0.001). Discussion: Using an interactive platform, this study captured free-living mobility data in MS patients. Important metrics such as the use of a minimum of 2-day estimates and self-reported disability were found to be robust indicators and correlates, respectively, of participant activity levels. Further triangulation of such metrics may reduce the burden on patients, clinicians, and researchers when monitoring clinical status.

The Role of Wearable Devices in Multiple Sclerosis

“…However, this study is conducted in a relatively small sample of pwMS with low disability. Similar findings are reported in a secondary analysis conducted in pwMS enrolled from the PatientLikeMe platform: averaging 2 days of step counts provides an adequate level of reliability (intraclass correlation coefficient, ICC, ≥0.7), which becomes a very high level of reliability (ICC 0.9) considering averaging data for a whole week [ 87 ].…”

The Digital Neurologic Examination

“…Finally, activity trackers (e.g., smartbands), insole shoe sensors, and skin-mounted inertial wearable sensors have all been used to assess gait [58, 68, 69]. We are unaware of any published information on digital tools to assess the remaining cranial nerve systems (trigeminal sensory system; head, neck, and tongue movement; exocrine function), the somatosensory system, or reflexes.…”

The Contribution of Machine Learning in the Validation of Commercial Wearable Sensors for Gait Monitoring in Patients: A Systematic Review

“…of the papers collected their data in laboratory conditions (N = 53)[34-45,47-49, 54,56-58,60,61,63-65,67-73,75,76,78-85,87-92,95,97,99-102], while a smaller part collected data in free living conditions (N = 17)[46,50,51,59,77,85,86,96,103] (see Table1).Regarding the positioning of sensors and/or devices (Table4), 60% of the studies placed them on an inferior part of the body[35][36][37]40,47-49,52-58,60,62,63,67,70,71,73-76,78-81,83,85,87,88,90-92,95-100,102,103], generally on the feet (N = 14) or on the hips (N = 6). The chest was also widely used (49%) [34,37-39,44,48,50,54-56,59-61,64,65,67, 70,72,73,75,77,79,83,84,89,90,92-95,97,99,101,102]; 17% of the studies carried out sensor positioning on the hands and arms [38,40,46,48,52,63,66,67,77,80,82,90,102], while the other 17% used a trouser or jacket pocket [42,43,45,50,51,59,68,70,77,86,89,103] Frequency of sensor locations reported on the patient from the included studies.…”

“…[35][36][37][38][39]41,44,46,47,50,51,53,54,56,58,59,64,66,71,[75][76][77][78][79]81,[83][84][85][86][90][91][92][93][94]96,97,[101][102][103], others use annotations made by experts on data from videos or measurements during the experiment[37,38,40,43,48,52,55,63,67,70,74,80,…”

“…41,44,46,48,49,51,54,55,58,59,61,62,65-67, 71,72,74,76,78,82-85,87-92,94,98,99,101-103] where correlations, mean errors, or p-values are most commonly reported. The other evaluation methods present models either as a linear model (11.4%)[42,50,52,56,57,73,75,93,96,100] or as a machine learning model(17.2%) …”