The Missing Piece in Today’s Opioid Conversation: Leveraging Data Science to Address the Crisis and Save Lives

By Anne Milgram and Brett Goldstein

Over the past decade, industries have used data to transform how we find jobs, buy groceries and get around. Companies like Lyft, Google and LinkedIn have used data science and big data analytics to find patterns, connections, efficiencies and ultimately answers to large problems. With that in mind, why aren’t we applying these powerful techniques to crises like the opioid epidemic?

CivicScape has spent the last three years focused on developing top-notch spatial analytics to understand where and when events will occur. To contribute to the scientific community’s push to mitigate the opioid epidemic, we spent substantial time applying our expertise in spatial analytics to understanding where and when fatal overdoses are most likely to happen. Every city, state, jurisdiction or township has finite resources but if we can help prioritize the hot spots of overdose events, public safety agencies have an opportunity to stage resources to respond in the most crucial first moments. Moreover, public health agencies have an opportunity to deploy resources to these locations even before an overdose occurs.

The Opioid Epidemic

In 2016, drug overdoses involving opioids resulted in more than 42,000 fatalities, making opioids the greatest drug epidemic in American history. Opioid overdoses now kill more Americans per year than automobile accidents and gun homicides. Worse, in the past two decades, the number of overdose deaths involving opioids have increased fivefold. This problem is large, and each day that passes without a solution carries with it, on average, the death of 115 Americans from opioid overdose.

In combating the opioid crisis specifically, public health and safety decision-makers lack reliable data to deploy resources. However, there is a critical piece of information that we should be harnessing to save lives: 911 calls for drug overdose. In many jurisdictions, drug overdose 911 calls are opioid-related. Understanding where the next opioid overdose is likely to occur can help to save lives. First, this can provide critical information for the police and other first responders, who can quickly administer Naloxone, an overdose reversal drug shown to be effective at reducing opioid-related deaths. While first responders are often equipped with the Naloxone, deploying these resources effectively is crucial to prevent overdose deaths. Second, for public health agencies working upstream to prevent the next overdose, this information is critical as it provides insight for the development of targeted interventions.

The CivicScape Approach

CivicScape combines the best of modern data and technology — using ensembled neural networks and patent-pending downsampling — to provide insight that traditional policing models are missing. Traditional methods of data analysis are useful for identifying common areas of historical opioid overdoses, but fail to identify patterns or to provide accurate, real-time information that can improve our response to this crisis. Tools like CivicScape can improve deployment of police, other first responders, and public health workers by helping them anticipate the most likely location of problems before they happen.

Case Study: Seattle

In an initial case study, CivicScape used publicly-available data from Seattle to anticipate where an overdose would occur. Starting with a diverse set of data that include 911 calls, weather, and spatial data, CivicScape identified the top locations that were most likely to be opioid hot spots within the next hour.

Critically, we found that 75% of all overdoses happen in just a third of the city. The remaining 25% of opioid deaths are spread out throughout the rest of the city. Using CivicScape, we were able to identify the neighborhoods throughout the entire city where opioid overdoses are likely to occur.

We also compared our work to the traditional way most police departments use data today. Many departments use a three-year statistical average of all reported opioid overdoses if they have quality overdose data. That traditional method, in our review, did not provide real-time information and also did not provide any insight into the 25% of deaths that are not concentrated in a small section of the city. CivicScape was nearly three times more accurate in anticipating opioid overdoses than traditional models in areas that are historically lower-risk. In other words, if we want to make the best decisions about how to deploy resources to prevent and treat opioid addiction, we need to use new technologies that capture the best science available today.

Open data, like the data CivicScape used from Seattle, has allowed us to illustrate a proof of concept for anticipating the occurrence of overdose; however, open data is limited and therefore limits the potential impact data and analytics can bring to this epidemic. Imagine what could be done if federal, state and local data and resources were brought together for this analysis. Using CivicScape forecasts, we have seen how first responders can better deploy officers and health care providers to proactively help in communities that are most at risk.