JMIR Publications

ABSTRACT

Background:

The monitoring of free-living physical activity by wearable devices can offer real-time assessment of activity features associated with health outcomes or provide suggestions for treatment recommendations and adjustments. The conclusions of studies on physical activity (PA) and health depend crucially on reliable statistical analyses of the digital data. Data analytics, however, are challenging due to the various metrics adopted to measure PA, different aims of studies, and complex temporal variation within variables. The application, interpretation, and appropriateness of these analytical tools have yet to be summarized.

Objective:

This research reviews studies that adopted analytical methods for analyzing PA monitored by accelerometers. Specifically, this review addresses three questions: (1) What are the metrics used to describe an individual’s free-living daily PA? (2) What are the current analytical tools to analyze PA data, under the aims of classification, association with health outcomes, and prediction of health events? (3) What are the challenges in the analyses and what recommendations for future research are suggested regarding use of statistical methods in response to various tasks?

Methods:

This scoping review is conducted following the Arksey and O'Malley framework to map research studies by exploring the information about physical activity. Three databases, PubMed, IEEE Xplore, and the ACM Digital Library, were queried in February 2024 to identify related publications. Eligible articles were either classification, association, or prediction studies involving human PA monitored by wearable accelerometers.

Results:

After screening 1312 articles, 428 eligible studies were identified and categorized into at least one of the three thematic issues: classification (75/428), association (342/428), and prediction (32/428). Most articles (414/428) adopted a PA variable derived from three acceleration dimensions, rather than from a single-dimensional acceleration. All eligible articles considered PA metrics represented in the time domain (428/428); a small fraction also considered PA metrics in the frequency domain (16/428). Studies evaluating the influence of PA on health conditions have increased greatly. Among those in our study, regression-type models were most prevalent (373/428). The machine learning approach for classification research is gaining popularity as well (32/75). In addition to summary statistics of PA, several recent studies utilized tools to incorporate PA trajectories and account for temporal patterns, including longitudinal data analysis with repeated PA measurements or functional data analysis with PA as a continuum for time-varying association (68/428).

Conclusions:

Summary metrics can quickly provide descriptions of the strength, frequency, and duration of individuals’ overall physical activity. If the distribution and profile of PA are to be evaluated or detected, taking the PA metrics as longitudinal or functional data can bring in more information and improve comprehension of the role PA plays in health. Depending on the research goal, appropriate analytical tools can ensure the reliability of the scientific findings.