NSF Grant to Use Data-Driven Machine Learning to Combat Opioid Crisis
By David Slipher
How can machine learning help stop the opioid epidemic? It begins with gathering enough large-scale data to develop predictive treatment policies that can be supported across communities in a targeted approach.
That’s the aim of a nearly $149,000 National Science Foundation grant awarded to Virginia Commonwealth University researchers in the College of Engineering, the Wilder School of Government and Public Affairs, and the VCU School of Medicine Department of Family Medicine & Population Health.
A deadly lack of information
According to the National Institutes of Health, in 2018, opioids were responsible for nearly 70% of drug overdoses in Virginia, or nearly three deaths each day. A major challenge in combating this crisis is a lack of treatment information being collected and shared across municipalities in the state.
“We are trying to help address this in part with new technology to gather and analyze data to identify contributing factors,” said principal investigator Sherif Abdelwahed, Ph.D., professor of electrical and computer engineering. “These could include economic demographics, emergencies like the pandemic, natural disasters and other circumstances.”
One of the biggest challenges in developing addiction treatment approaches at the community level comes from a lack of data indicators to explain what is happening at the population level.
“Besides the overdose cases reported to emergency medical services and aggregated overdose deaths reported by state agencies, neighborhood-level opioid abuse information is difficult to come by,” said co-principal investigator Sarin Adhikari, Ph.D., a research economist with the Center for Urban and Regional Analysis at the Wilder School of Government and Public Affairs. “Lack of detailed information reduces the ability to identify high-risk neighborhoods and plan for preemptive strategies to reduce opioid-related deaths.”
Innovating data collection
VCU researchers plan to address this data disparity by establishing and leading a team of local stakeholders to broaden their understanding of effective, community-involved solutions, beginning in the Richmond area. Using smart technologies and data analytics, the researchers hope to develop efficient and scalable decision-making methods that will reduce overdoses and health inequities in the Richmond region and across Virginia.
Data collection will come from multiple sources, some quite unexpected. One unique method will be the deployment of robots in municipal sewage systems to test wastewater for traces of drug metabolites. Researchers will also conduct social network data mining to track the geographic concentration of drug-related keywords used on social media and internet searches. More traditional emergency medical services response data, like overdose location and patient outcome will help round out these novel approaches. Each of these methods will help serve as macro barometers to indicate overall opioid trends across communities.
Through collecting these comprehensive drug use statistics, the researchers seek to utilize machine learning to derive predictive models for forecasting opioid-related overdoses at the population level. The hope is that development of real-time assessment and computerized decision making tools can enable rapid analysis which may help predict the impacts of different community health solutions. The team’s goal is that better analytics will help leaders to make more effective decisions on how to best allocate limited resources to fight opioid addiction.
“The team expects to develop a methodology to integrate these various data sources to identify geographic hotspots and high-risk zones, understand the characteristics of the target population, identify areas for deploying appropriate community-based services, and also understand the root causes of the increase, or decrease, in drug abuse,” said Adhikari.
In the coming year, the research team will present its findings to community partners through a series of workshops. They’ll incorporate the feedback to improve further data collection and analysis.
From population-level solutions, the team is optimistic that they’ll be able to make impacts that save lives and improve health for Virginians.
“We hope to uncover new insights in opioid abuse in the Richmond region through this study, Adhikari said. “Identifying the socio-economic and demographic determinants of opioid abuse at the neighborhood-level could help local governments and nonprofits to reallocate resources, provide customized services, and respond with appropriate policy interventions.”