Classifying smartphone-based accelerometer data to obtain validated measures of subject activity status, step count, and gait speed

Abstract

Background: The ubiquitous spread of smartphone technology throughout global

Mobile Phone-Based Measures of Activity, Step Count, and Gait Speed: Results From a Study of Older Ambulatory Adults in a Naturalistic Setting

“…After determining that our questionnaire-based measures and physical performance battery successfully differentiated functionally intact from functionally impaired individuals, we examined whether our mobile-phone-based measures of physical activity did so as well. Active states were defined as periods where the participant was walking, climbing stairs, or otherwise active (high physical activity classification) [ 13 ]. Inactive states were defined as when the participant was resting or driving (low physical activity) [ 13 ].…”

“…Active states were defined as periods where the participant was walking, climbing stairs, or otherwise active (high physical activity classification) [ 13 ]. Inactive states were defined as when the participant was resting or driving (low physical activity) [ 13 ]. We noted significant differences in participant 24-hour and active state time budgets ( Figure 2 ).…”

“…For PROMIS measures, t scores were determined from raw scores by appropriate conversion tables. Raw acceleration data was low-pass filtered and baseline acceleration normalized to 1 g over the entire duration of data collection [ 13 ]. Our classification algorithm first identified epochs of “forgotten phone” versus epochs of participant carrying the phone.…”

“…Our classification algorithm first identified epochs of “forgotten phone” versus epochs of participant carrying the phone. For epochs of participant carrying the phone, we then classified behavior into active or inactive states, using a windowed (68 s long) Fourier analysis approach [ 13 , 31 ]. Active states were further differentiated into states with minimal locomotion, states with ongoing locomotion, and states where the participants were climbing stairs.…”

“…We have shown the feasibility of using mobile phone technology to measure an individual’s activity and lifespace (eg, the geographic expanse of an individual’s day-to-day travels) over prolonged periods of time in a non-invasive, near-continuous, robust, inexpensive, and user-friendly manner [ 11 ]. In order to extrapolate health parameters [ 12 ] from our participant-derived mobile phone data, we designed algorithms to measure clinically relevant aspects of activity, including gait bout duration, gait speed, and step counts [ 13 ]. Additional studies showed that for a broad group of individuals (ranging in age from 21 to 84 years), the activity metrics we measured by this approach strongly correlated with gait speed under controlled laboratory conditions [ 14 ].…”