JMIR Publications

ABSTRACT

Background:

Older adults who engage in physical activity can reduce their risk of mobility and disability. Short amounts of walking can improve their quality of life, physical function, and cardiovascular health. Various programs have been implemented to encourage older adults to engage in physical activity, but sustaining their motivation continues to be a challenge. Ubiquitous devices, such as mobile phones and smartwatches, coupled with machine-learning algorithms, can potentially encourage older adults to be more physically active. Current algorithms that are deployed in consumer devices (e.g., Fitbit) are proprietary, often are not tailored to the movements of older adults and have been shown to be inaccurate in clinical settings. Few studies have developed step-counting algorithms for smartwatches – but only using data from younger adults and often validating them only in controlled laboratory settings.

Objective:

In this work, we sought to develop and validate a smartwatch step-counting app targeting older adults that has been evaluated in free-living settings over a long period of time (24 weeks) with a large sample (N=42).

Methods:

the steps of older adults. The app includes algorithms to infer the level of physical activity and to count steps. We validated the step-counting algorithm with a total of 42 older adults in the lab (counting from a video recording, N= 20) and in free-living conditions — one 2-day field study (N=6) and two 12-week field studies (using the Fitbit as ground truth, N=16). During system development, we evaluated four kinds of walking patterns: normal, fast, up and down a staircase, and intermittent speed. For the field study, we evaluated various values for algorithm parameters, and subsequently evaluated the method’s performance using correlations and error rates.

Results:

The results from the evaluation showed that our step-counting algorithm performs well, highly correlated with the ground truth and with low error rate. For the lab study, there was stronger correlation for normal walking R2=0.5; across all activities, the Amulet was on average 3.2 (2.1%) steps lower (SD = 25.9) than video-validated steps. For the 2-day field study, the best parameter settings led to an association between Amulet and Fitbit (R2 of 0.989) and 3.1% (SD=25.1) steps lower than Fitbit, respectively. For the 12-week field study, the best parameter setting led to an R2 of 0.669.

Conclusions:

Our findings demonstrate the importance of an iterative process in algorithm development in advance of field-based deployment. This work highlights various challenges and insights involved in developing and validating monitoring systems in real-world settings. Nonetheless, our step-counting app for older adults had good performance relative to the ground truth (a commercial Fitbit step-counter). Our app could potentially be used to improve the physical activity among older adults through accurate tracking of their step counts and in-app daily step-count goals.