If there are prediction models for who might be at risk of cancer or diabetes so that they can get attention before it's too late, why can't there be a predictor for overdoses? A new tool designed to help people who use stimulants like cocaine or methamphetamine answers that question, using demographics and other available information to identify who is at risk. The tool's developers at the Perelman School of Medicine at the University of Pennsylvania hope that it can be used to proactively offer resources and treatment to save lives, and their work is detailed in JAMA Health Forum.
"Substance use disorder, like other relapsing, chronic, or remitting disorders, will have 'flares,' so our work is meant to proactively offer resources and needed care to patients," said lead author Tuhina Srivastava, PhD, MPH, a former Epidemiology graduate student at Penn who is now a research scientist at the Institute for Health Metrics and Evaluation. "Too often, the response to people with substance use disorder is reactive or even punitive, so we believe this provides a potential step toward minimizing or eliminating that. It's classified as a chronic disease and should be treated as such."
A gap in public consciousness
While attention on the opioid crisis has surged over the years, the researchers knew that overdose deaths involving stimulants—especially those tied to cocaine and methamphetamine—account for 70 percent of all substance overdose deaths in Philadelphia and 60 percent nationwide. The toll has been on the rise since 2011, according to Centers for Disease Control (CDC) data, and even climbed by 85 percent between 2019 and 2023. More recent trends suggest hope, but the researchers hope to provide an enduring tool that might continue to reverse the tide.
"Although some progress been made in reducing opioid-related deaths in recent years, there is still a great deal of work to be done. At the same time, we are very concerned about the mounting harms of cocaine and methamphetamine use," said co-author Rebecca Arden Harris, MD, MSc, an assistant professor of Family Medicine and Community Health.
9 out of 10 for accuracy
The predictor tool was trained with de-identified data from the Medicaid program, which covers low-income people and other disadvantaged groups. This dataset was chosen because of the size of the population it covers (nearly 71 million people, or one in five people in the United States) and because it covers many who are, historically, at risk for stimulant overdose.
In tests, the researchers found the model to be extremely accurate at identifying those at risk of stimulant-involved overdose, effectively scoring above a 9 out of 10 on a common statistical accuracy scale. The tests also helped identify some common risk factors that are especially useful in identifying people who are at risk for future overdose, such as instances of previous overdoses, higher poverty levels in the area where they live, and factors in their living arrangement, such as how many people lived in one home.
The data included both people who had had a stimulant overdose and those who never had one. A variety of scenarios were considered, including cocaine overdose with and without opioids, and methamphetamine/ecstasy/psychostimulant overdose, also with and without opioids.
In addition to living in poverty and crowded housing, men were found to be much more likely than women to be at risk.
But the top predictor for all categories was prior substance use diagnoses or conditions.
"I think it's important to realize that the most predictive element is past history, and that combined with some of the other predictors may help providers identify who might especially benefit from extra resources, just as we would address a patient's history of heart attack," said co-author Cheryl Bettigole, MD, MPH, MA, a professor of Clinical Family Medicine and Community Health, as well as Medical Ethics and Health Policy.
Drawing clear conclusions
The team is hopeful their model will be used in population health settings soon to better direct resources such as cognitive behavioral therapy, the provision of naloxone, or offering incentive-based programs that reward individualized recovery goals.
For example, some Medicaid programs use prediction tools to identify people who are likely to have frequent hospital or emergency department visits. Through this, the patients' care teams are notified, allowing them to reach out early. Now that such a tool exists for stimulant overdose and has shown its strength, the team is hopeful that it will be implemented in a similar way.
"This is a transparent model," said senior author Sean Hennessy, PharmD, PhD, director and professor of Epidemiology. "It's an open algorithm. You can see exactly what's going on here, and that should build trust among clinicians and public health officials."
This research was supported by the U.S. National Center for Injury Prevention and Control (R01 CE003347).
