MD2K, NIDA researchers find method for mining sensor data

One of the goals of mobile health, or mHealth, is to translate data from wearable sensors into useful information that can help people monitor and improve their health. One of the ways to do that is by providing a just-in-time intervention, such as a text message or prompt, that is issued based on sensor data collected in real time.

A major research challenge is determining when to intervene. Delivered too often or at the wrong times, an intervention might become ineffective or even counterproductive.

Researchers from the Center of Excellence for Mobile Sensor Data-to-Knowledge (MD2K), in collaboration with researchers at the National Institute on Drug Abuse (NIDA) Intramural Research Program have taken on this challenge.

The researchers developed a method to mine the data deluge produced by wearable sensors and identify the (rare, but) most appropriate moments at which to intervene. These new methods make it possible to use data from wearable sensors to monitor and manage stress, according to research presented recently at the 2016 CHI Conference on Human Factors in Computing Systems in San Jose, California.

The paper, “Finding Significant Stress Episodes in a Discontinuous Time Series of Rapidly Varying Mobile Sensor Data,” also lays the foundation for future work developing sensor-triggered just-in-time interventions for a variety of health issues that can be monitored via sensor data.

The researchers examined physiological data, GPS tracks, and activity data collected from 38 study participants in the NIDA Intramural Research Program over a period of 4 weeks as the participants went about their daily lives.

Sensors produce a continuous stream of data that must be reduced to identify health measures such as stress. Using the methods reported in the paper, including cStress, a recently validated model for stress detection, researchers were able to process the thousands of data points produced per minute by the sensors into a data stream of one value per minute and from that stream identify major (usually, rare) stress episodes.

The data used in the study had been collected as part of a larger outpatient study at the NIDA Intramural Program regarding the relationships among stress, addictive behavior, and daily activities. Data were collected from 38 opioid-dependent drug users receiving opioid agonist maintenance treatment. Participants wore wireless physiological sensors for about 10 or more hours each day. Study participants were compensated for their time.

Once stress episodes were identified, the researchers also compared them to geolocation data to examine the role of contexts such as noise, graffiti, drug activity, nearby businesses like bars, and the condition of properties in the area. The results suggest a disorderly environment – one with lots of graffiti, trash, broken windows or bars – elevate physiological signs of stress, while locations where more positive activities are occurring, such as people having pleasant and positive interactions or children playing, are associated with reduced signs of stress. The occurrence of a stress episode increased the likelihood of a subsequent stress episode.

The authors said their work creates new opportunities for future research to design interventions for dealing with daily stress in both work and personal lives. In addition, the methods used to identify and predict stress episodes may be useful in determining the timing and frequency of mHealth interventions based on other health measures obtained from sensor data.

For more details, including the list of authors, please see the full article. The research was supported by the National Science Foundation under award numbers CNS-1212901 and IIS-1231754 and by the National Institutes of Health under grants R01DA035502 (by NIDA) through funds provided by the trans-NIH OppNet initiative and U54EB020404 (by NIBIB) through funds provided by the trans-NIH Big Data-to-Knowledge (BD2K) initiative.

Full article:

Related article:

Copyright © 2017 MD2K. MD2K is supported by the National Institutes of Health Big Data to Knowledge Initiative (Grant #1U54EB020404)

Team: Cornell Tech, GA Tech, Harvard, U. Memphis, Northwestern, Ohio State, Open mHealth, UCLA, UCSD, UCSF, UMass, U. Michigan, Utah, WVU