Data-driven profiling of accelerometer-determined physical behaviors and health outcomes in adults: A systematic review
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Objective
Data-driven, person-centered methods are analytical approaches applicable for profiling physical behaviors from wearable accelerometry data. These methods rely on data rather than predefined, knowledge-driven hypotheses to first identify multiple subgroups within a sample population and then evaluate their relationships with health outcomes. This systematic review aims to describe and summarize multidimensional physical behavior profiles identified from wearable accelerometry data across different adult populations using data-driven, person-centered methods, along with their relationships with health outcomes.
Method
Three electronic databases were searched for relevant articles published up to July 2025. Peer-reviewed journal articles that applied data-driven, person-centered approaches to accelerometer-monitored physical behaviors in adult participants (18 years or older) to create profiles of physical behaviors, and that examined the associations of the formed profiles with a health outcome, were considered.
Results
Of the 16,289 publications retrieved, 40 studies were included. The most commonly employed technique for physical behavior profiling was the machine learning K-means clustering algorithm (n = 18), followed by latent profile analysis (n = 8). A diverse set of descriptor variables was derived from accelerometer signals and utilized. Hourly metrics, calculated for each hour of a 24-hour day, were the most commonly utilized in the analysis. These included hourly average acceleration, hourly monitor-independent movement summary units, hourly activity counts, and hourly moderate-to-vigorous physical activity minutes. The review of derived profiles from accelerometer-measured physical behavior revealed several hypothesis-generating, preliminary evidence about how different components and/or aspects of physical behaviors could cluster together and synergistically influence health outcomes.
Conclusion
Data-driven, person-centered methods are viable analytical approaches increasingly employed in accelerometry studies to drive distinct physical behavior profiles. The application of these techniques to accelerometer-measured physical behaviors has generated data-driven findings regarding how various physical behavior profiles may differentially relate to health outcomes.
