mhealthtools: A Modular R Package for Extracting Features from Mobile and Wearable Sensor Data

Abstract

As life expectancy continues to rise with the increasing prevalence of modern medical practices, the costs associated with caring for an increasingly large elderly population have climbed as well. In response, there has been an evolving interest in using relatively inexpensive remote monitoring methods outside of the clinic to aid in the early detection and real-time assessment of Alzheimer’s disease (Kourtis, Regele, Wright, & Jones, 2019), Parkinson’s disease (Monje, Foffani, Obeso, & Sánchez-Ferro, 2019), mood disorders (Jacobson, Weingarden, & Wilhelm, 2019), and a wide array of other diseases (Majumder & Deen, 2019). Consumer-grade sensors embedded in smartphones and wearables have the potential to provide clinicians with remote digital biomarkers that can aid in the early detection and monitoring of disease via frequent, objective, and unobtrusive assessments (Coravos, Khozin, & Mandl, 2019). The extraction of relevant digital biomarkers from raw sensor data requires a combination of signal processing, data science and biological expertise in addition to extensive validation data (Goldsack, Chasse, & Wood, 2019; Monje et al., 2019). Software libraries to preprocess and extract features from such data are becoming increasingly numerous and varied (Czech & Patel, 2019; Saez-Pons, Stamate, Weston, & Roussos, 2019; van Gent, de Bruin, Kris, & Fernandes, 2019). But without an established set of preprocessing techniques and features for capturing disease signal, it becomes necessary to make use of a broad spectrum of methods and tools when exploring potential biomarkers (Kumar et al., 2013). To this end we have developed mhealthtools: an R package that aids in the construction of data pipelines for remote sensor data analysis.