Q&A: Universities, tech collaborate to predict opioid overdose with wearables
Earlier this month, news broke that the Digital Medicine Society launched a partnership with developers and academic organizations, including Duke University, Google Fitbit, UNC Chapel Hill, and others. The stated aim of the collaboration is to use technology to reduce deaths associated with opioid relapse and overdose by leveraging data from consumer wearables, both dedicated devices and health-monitoring applications on smartphones.
Over the next five months, the collective will conduct a pilot study—led in part by Duke’s Big Ideas Lab—to train an AI-based tool for early intervention, marking a significant step toward integrating digital health technologies into routine opioid use disorder (OUD) care.
HealthExec interviewed Candice Taguibao, associate program director at the Digital Medicine Society, to gather more details on the partnership, including particulars on how the project intends to identify patients at risk of overdose—and to discuss the privacy implications of wearable technology that monitors health.
Taguibao, a health researcher, leads efforts to develop a sensor-based relapse prevention tool for OUD patients, specifically the use of wearables to assess overdose risk.
Editor's Note: The interview transcript has been edited for clarity and concision.
HealthExec: For this project, what health and lifestyle data are the Digital Medicine Society and its collaborators collecting to enable wearable devices to predict overdose risk?
Taguibao: Based on formative research—including a systematic review and stakeholder interviews with clinicians and patients—various measures potentially associated with opioid relapse risk have been identified. Physiological signals and behavioral traits have been collected that can one: potentially detect relapse risk and two: be measured using wearable technologies like the Oura Ring. Metrics to be tracked include heart rate, physical inactivity, and social isolation. Additionally, smartphones will be used to capture mental health characteristics such as self-reported stress, anxiety and depression.
Is it imminent detection, or does it look for warning signs?
The data will be used to train a tool designed to prevent opioid relapse. A prospective study is being prepared to obtain a representative training dataset by enrolling approximately 1,000 participants with opioid use disorder. The study will collect around 30 predictor variables, primarily captured from continuous, sensor-based data gathered through consumer wearables—such as the Oura Ring—and smartphones.
The primary clinical outcome is relapse, which will be measured through urine drug screenings and self-reported data. This dataset will then be used to develop a risk detection model.
Is this data anonymized? How is the user identified without sacrificing privacy?
All data analyzed from our prospective studies to develop the training dataset are de-identified. Multiple precautions are taken to protect study participants’ data, including the use of unique participant IDs instead of names or any contact information to maintain confidentiality.
Confidentiality is further safeguarded through techniques such as timestamp shifting, which makes it more difficult to link data to individuals based on specific dates.
Location data will not be continuously tracked. Instead, the study will use geofences to monitor whether participants enter or exit specific 'risky' locations that they voluntarily share within a defined time period. All data collected from sensor-based digital health technologies, electronic patient-reported outcomes, and clinical records will be securely stored in Health Insurance Portability and Accountability Act (HIPAA)-compliant platforms, with encryption during and after transmission.
All identifiable data will be handled in strict accordance with HIPAA requirements. When study results are reported, no personal identifying information will be disclosed.
Collaboration with key stakeholders and individuals with OUD will guide the ethical framework to ensure that the implementation of this tool is responsible and trustworthy.
Is data shared with healthcare providers who attend to these patients, such as through an EHR integration?
Data is not currently being shared directly with providers through EHR systems. However, studies are underway to develop a tool that can provide this capability. As an incentive for completing study activities, participants may keep the wearables at no cost and may choose to share their data with their respective care teams.
Is data shared with law enforcement? Can law enforcement access data from an individual, with or without a warrant?
No, data is not being shared with law enforcement and law enforcement does not have access to the data.
What about for individuals who may be developing a problem. Can the AI potentially detect emerging signs of dependency?
Not currently. For now, the only assessment being made is whether the identified potential predictor variables are associated with the risk of opioid relapse.
Is there any kind of direct patient support?
Patients involved in the study are currently seeking treatment. Participants are being recruited from clinical sites that have already offered treatment and support services, including the Morse Clinic and Alcohol & Drug Services. Ideally, this is something that will be used alongside a treatment plan, for patients who are already seeking support for OUD.