Researchers Combine Wearable Biosensors and Machine Learning to Combat the Opioid Crisis

Image of the E4 wristwatch. A single user input button is available to toggle through different operation modes. You can also observe two electrodes on the wrist strap for measuring electrodermal activity. Image courtesy of Empatica.

Researchers have combined wearable biosensors with machine learning to aid in treatment for substance abuse.

Wearables have been finding increasing utility in the healthcare industry, from step counters to incentivize fitness to heart rate monitors for preventative healthcare applications. These low-power, small-footprint devices are tricky to design, but their potential to improve health and fitness applications has been inarguable.

Now, in research settings, these little devices are finding a calling in combatting a grave issue that was declared a public health emergency by the Department of Health and Human Services in 2017: substance abuse.

A Growing Substance Abuse Pandemic

The Centers for Disease Control and Prevention (CDC) has reported an increasing number of opioid-overdose related deaths across the United States with 47,600 reported deaths in 2017 up from 18,515 in 2007. Unfortunately, the substance use epidemic has not only affected the United States, but has been reported in many countries around the world, including Canada and France.

Statistics from 1999 to 2017 showing an increasing trend in opioid overdose-related deaths. Image courtesy of CDC WONDER via the National Institute on Drug Abuse.

Members of the scientific community have endeavored to combat this growing epidemic with the help of technology. Of the different technological platforms being explored, wearables present a particular set of advantages, namely the ability to continually track recovering substance abuse disorder patients in their daily lives. Such tracking would assess treatment adherence, facilitating intervention if necessary, as well as alert ambulatory services in times of need.
To pursue this possibility of remote monitoring, a group of researchers and physicians at the University of Texas at Tyler and the University of Massachusetts, Medical School have been investigating using a wearable biosensor to monitor treatment adherence in substance use disorder patients. Their 2019 paper entitled, “A Machine Learning-based Approach for Collaborative Non-Adherence Detection during Opioid Abuse Surveillance using a Wearable Biosensor” details their study.

Detecting Opioid Use with Wearable Biosensors

The research group utilized the E4 wristband wearable biosensor from Empatica, the same company that makes the Embrace2 Watch, a wearable biosensor for detecting epilepsy in children.

Image of the E4 wristwatch. A single user input button is available to toggle through different operation modes. You can also observe two electrodes on the wrist strap for measuring electrodermal activity. Image courtesy of Empatica.

Unlike the Embrace2 Watch, the E4 has not been FDA cleared, but is instead intended for research purposes. The E4 is packed with a suite of sensing capabilities including heart rate, skin impedance, temperature, and locomotion.
In this study, the research group use the sensor capabilities of the E4 to collect data on active drug users, then applied machine learning techniques to develop predictors for non-adherence of treatment regimens.

Building on Previous Studies

Members of this research group have been working in this technology space for a number of years now as the drug epidemic has, unfortunately, continued to grow. Their earliest study dates back to 2015 in which they employed the Q Sensor, the now-discontinued predecessor of the E4 wearable, to monitor changes in electrodermal activity, skin temperature, and locomotion. The group recorded data from a single patient during his daily activities.

The Q Curve and Q Sensor Pod from Affectiva. Image courtesy of MIT Affective Computing Group.

The Q Sensor was able to detect changes in the recorded physiological signals and such changes were corroborated with self-reported instances of drug use. As one would imagine, retroactively detecting drug use is not particularly amenable to real-time monitoring.

Detecting Opioid Use with Machine Learning 

In 2018, the group demoed an algorithm for automatically detecting opioid use again by measuring changes in electrodermal activity, skin temperature, and locomotion. In their 2018 study, titled “Automatic Detection of Opioid Intake Using Wearable Biosensor”, they improved upon their 2015 paper by utilizing machine learning and pattern recognition to automatically differentiate baseline readings of different physiological signals from abnormal readings elicited by substance use.

Possible Chipsets for Wearable Biosensors

Not many details on the specific chipsets used in the E4 or Q-Sensor wearable are available. However, given the popularity of biometric monitors, we can speculate the use of any number of popular chipsets available for heart rate, such as the AFE4490 from Texas Instruments or the more recent MAX3010X series of ICs from Maxim Integrated, which have been popularized by the heart rate monitor function in the Samsung Galaxy series of phones.

Maxim Integrated has been placing emphasis on healthcare applications in many of the hardware components and tools they release, including a wearable platform for remote biometric monitoring for developers. Maxim, along with Omron and ROHM, also offers biometric monitoring wearables as end devices.

Functional block diagram of the MAX30102. Image from the MAX30102 datasheet. Click to enlarge.

Certainly the MAX30205 human body temperature sensor from Maxim Integrated would be appropriate here or the MLX90632 from Melexis if a designer needed a non-contact IR thermophile route, instead. If you'd like to learn more about this subject, we briefly discussed a few chipsets for EDA in a previous article on wearable static exercise monitors.

From examining the scientific literature, we can clearly see the increasing need for improved processing capabilities on embedded hardware to handle the ever-increasing amount of sensor data required to make intelligent algorithms.