Predicting areas at risk for high numbers of drug-related deaths allows public health experts to develop targeted intervention policies. An international research team has designed a predictive statistical model that can predict drug fatality rates by county. The scientifically validated model suggests that data science could play a role in curbing the drug overdose epidemic in the United States.

More than 80,000 people died from drug-related causes in 2020, a 400% increase since 1999, when prescription opioid use caused the first wave of escalation in the US. Around 2010, illegal heroin use caused a second wave of overdose deaths. A third wave began around 2013 as illegally manufactured, synthetic opioids such as fentanyl hit the black market.

The team, lead by researchers from the University of California San Diego School of Medicine and San Diego State University, used third wave data from the CDC to train a retrospective model to identify patterns between specific characteristics of county and overdose deaths. The retrospective model then retrospectively predicted deaths in each county from 2013 to 2018. Compared to the actual number of drug-related fatalities, the model outperformed standard prediction tools currently in use, which only look at the previous year.

The researchers made their prediction model available as an open-source web tool, OD Predict Explorer. The tool’s interface lets other researchers explore the model’s predictions and the real-world data for each year. The team hopes the web tool will assist teams across the country in identifying areas where new outbreaks seem likely.

The model shows that fentanyl deaths were high in rural counties in the Midwest and Northeast as anticipated. But it also demonstrated that the number of overdoses continues to increase in Western states and urban areas, including San Francisco and San Diego. The team published their findings in Lancet Public Health in June.

A validated predictive model such as this will benefit from up-to-the-minute overdose fatality counts and precise information about drug availability in each county, both via prescription and illegal markets. The researchers hope that new partnerships which grant them access to restricted data will help them fine-tune the model. 

It may take up to two years for CDC data to power the model at the national level. But the researchers believe that states that share their data could start using the tool earlier. That means that policymakers could get an early start in targeting at-risk counties, implementing programs that improve access to treatment, facilitate harm-reduction, and reduce illicit drug exposure in communities. Experts agree that such measures will have an impact on the drug overdose crisis in America.