Description
Wearable data are becoming an important source of health and disease data as they inform on a variety of personal, behavioral, social, contextual, and environmental health-relevant factors. Wearables have been primarily used for activity tracking and gained popularity with fitness applications; however, more recently, these devices have been used in an increasing number of health applications, including health monitoring, clinical-care, remote clinical-trials, drug delivery, and disease characterization to name a few.
We linked wearable data to clinical outcomes and we have found that data from wearable measurements are as good or better than clinical measurements for predicting adverse health events in cancer medicine, diagnosing neuromotor disorders in infants and for cardiac monitoring. We implement an open-source toolkit (W4H) to benefit the larger health community by working with our community of stakeholders comprised of the medical teams from our previous projects.
Description
- We first conceptualize wearable data as Geo-referenced Multivariate Time Series (GeoMTS) data, i.e., time series with geotagged information. This conceptualization allows us to represent different types of wearable data in a single form so we can separate the wearable type from its data management and analysis.
- Next, we develop these algorithms and models into a W4H Toolkit that we evaluate on three use cases with our community stakeholders. To facilitate the dissemination of our results and improve the sustainability of the W4H Toolkit we will implement our algorithms on Spark, a popular big data platform, and release an extensible W4H Toolkit software package that the larger health research community will be able to use with their data and sensors.
Case Studies
- Cancer Patient Monitoring: We analyze cancer patients under IRB protocol OS-16-2 at USC who were undergoing treatment with chemotherapy to study their fatigue level. Our results show that ECOG score did not predict the rate of adverse events, such as unexpected hospitalization and unplanned trips to the day hospital for hydration. By comparison, determination of the patient’s calorie expenditure during waking hours was highly predictive of serious adverse events over the course of the study.
- Cardiac Monitoring: Atrial fibrillation (AF) affects 1% of the general population and causes a third of all strokes. Smartphone ECG has the potential to overcome the shortcomings of traditional ECG recorders. The results have demonstrated the unique capabilities to expand the diagnostic scope, allowing patients to nimbly monitor for arrhythmia instantly on demand and outside of a traditional healthcare setting. Our further data analysis not only validates the ability of devices to accurately detect cardiac rhythms, but also demonstrates a consumer appetite for a device that enables continuous monitoring.
- Infant Neuromotor Monitoring: Early identification of impaired neuromotor control in infancy is necessary in order to provide early therapy intervention in accordance with known principles of experience-dependent neuroplasticity. Early intervention is often not initiated until an infant has demonstrated a consistent and severe delay in the acquisition of developmental milestones, in contrast with known principles of experience-dependent neuroplasticity. We have recently validated the use of wearable sensors for unobtrusive full-day, in-home movement monitoring of spontaneous infant leg movements.
References
- Luan Tran, Yanfang Li, Luciano Nocera, Cyrus Shahabi, and Li Xiong, MultiFusionNet: Atrial Fibrillation Detection With Deep Neural Networks, AMIA 2020 Information Summit , Houston, Texas, March 23-26, 2020.
- Tanachat Nilanon, Luciano P. Nocera, Alexander S. Martin, Anand Kolatkar, Marcella May, Zaki Hasnain, Naoto T. Ueno, Sriram Yennu, Angela Alexander, Aaron E. Mejia, Roger Wilson Boles, Ming Li, Jerry S. H. Lee, Sean E. Hanlon, Frankie A. Cozzens Philips, David I. Quinn, Paul K. Newton, Joan Broderick, Cyrus Shahabi, Peter Kuhn, and Jorge J. Nieva, Use of Wearable Activity Tracker in Patients With Cancer Undergoing Chemotherapy: Toward Evaluating Risk of Unplanned Health Care Encounters, JCO Clinical Cancer Informatics, Sep 24, 2020.
- Luan Tran, Manh Nguyen, and Cyrus Shahabi, Representation Learning for Early Sepsis Prediction, 2019 Computing in Cardiology (CinC), Singapore, 8-11 Sept. 2019.
- Mohammad Saeed Abrishami, Luciano Nocera, Melissa Mert, Ivan A. Trujillo-Priego, Sanjay Purushotham, Cyrus Shahabi, and Beth A. Smith, Identification of Developmental Delay in Infants using Wearable Sensors: Full-Day Leg Movement Statistical Feature Analysis, IEEE Journal of Translational Engineering in Health and Medicine, 25 January 2019.
People
Student(s)
 |  |  |
| Arash Hajisafi CS PhD Student, USC | Maria Despoina Siampou CS PhD Student, USC | Asmita Chotani CS Master Student, USC |
PostDoc(s)
 |
| Alireza Abdoli CS PostDoc, USC |
Collaborator(s)
 |  |  |
| Peter Kuhn Keck School of Medicine, USC | Jorge Nieva Keck School of Medicine, USC | Leslie Saxon Keck School of Medicine, USC |
 |  |  |
| Beth Smith Keck School of Medicine, USC | Luciano Nocera Viterbi School of Engineering, USC | Yao-Yi Chiang Department of CSE, UMN |
Principal Investigator(s)
 |
| Cyrus Shahabi Viterbi School of Engineering, USC |
