Several years ago, as a new mother, I started hearing about hypodermic needles showing up in local playgrounds that I had always considered safe. This was a couple of years before the opioid epidemic was national news but I wanted to understand whether this was a major issue in Massachusetts, where I live, or a series of isolated incidents. Because I work with maps every day, that was where I turned. In the years that have passed I’ve continued to gather data and use different geospatial tools to look at the opioid crisis. I’ve also been incredibly lucky that Massachusetts publishes a wealth of open/public data related to substance use disorder including fatal overdoses and treatment statistics.
In this project I’ve used Insights for ArcGIS to try to understand the epidemic over time as well as plan two different actions: One related to prevention and one related to treatment. After all, data is most useful when it’s helping people make better decisions.
Understanding Where the Problem Exists
If we compare maps of fatal opioid-related overdoses from 2000 and 2015, we see the drastic differences. As you can see in Figure 1, in 2000 Massachusetts 338 reported fatal overdoses from opioids. Only six towns or cities had more than 10 fatal overdoses with Boston having the highest number at 36. Jump ahead to 2015, shown in figure 2, and the image is much starker. In 2015, there were just shy of 1,600 deaths attributed to opioid overdose – an almost 5x increase! We also see in the heat chart that 39 towns or cities had ten or more fatal overdoses. The map tells us that these towns are spread out throughout the Commonwealth.
Figure 1. Fatal opioid-related overdoses, 2000.
Figure 2. Fatal opioid-related overdoses, 2015.
Like most chronic diseases, there is no silver bullet to fix the opioid crisis. However, there are several generally accepted helpful factors including prevention, treatment, and recovery. I am not an expert in substance use disorder, but I wanted to apply workflows from my experience working with private and public health organizations to look at potential prevention and treatment strategies. These decisions usually revolve around allocating resources, targeting the marketing or outreach for these resources, or understanding access to resources.
Figure 3. Targeting outreach for existing drop boxes.
Asking Questions to Identify a Resolution
Where would outreach be helpful?
The first question that I wanted to answer was how to increase use of existing prescription drug drop boxes. My first goal was to identify at-risk communities with existing drop boxes. I started by adding a spreadsheet of drop box locations and community-level treatment data to the map. I filtered the communities to include only those with admissions numbers for prescription drugs—at least 10 percent of treatment admissions listed another opioid, not heroin, as their primary drug. This allowed me to proxy risk for prescription drug misuse. These are the substances that we want properly disposed of in drop boxes. By then spatially filtering to only those towns with prescription drop boxes I have a short list of communities where outreach could be helpful.
Figure 4. High priority communities for drop box use.
Who I should target with a campaign?
Although you could use the same outreach tactics throughout the areas of interest, in many cases there are additional factors that make a targeted approach more effective (e.g., language or other demographic traits). We can get more of that information using segmentation data. When I enriched the list of communities with their dominant tapestry segment we see in the donut chart that several communities fall into the Green Acres, Front Porches, or Parks and Rec segments. I could try to find common behaviors between these segments to do an all-encompassing campaign but since my main goal is to reduce fatal overdoses, I’ve chosen to focus on areas with the highest numbers of fatal overdoses. By selecting the biggest community in the bubble chart, Taunton, which represents the highest number of fatal overdoses (14 in 2015) I can see the associated tapestry, Front Porches highlighted in the donut chart. People who fall into the Front Porches segment, among other things, use online gaming and dating, listen to hip hop and R&B, and watch Comedy Central—all information that could be used to tailor outreach efforts.
Figure 5. Dominant Tapestry for Taunton
Do affected communities have easy access to care?
I also wanted to understand access to care, specifically Medically Assisted Treatment (MAT), and whether that is affecting outcomes. There is evidence to show that MAT, the combination of medication and counseling/behavior therapies, can help sustain recovery. I wanted to understand whether lack of access to MAT was related to fatal overdose numbers. By using location analytics to calculate the population weighted average drive time from each town or city to the nearest MAT facility, regardless of whether it provided standalone methadone or office based suboxone/vivitrol, I could then use Insights to visualize possible relationships.
My initial findings were surprising but it led me to ask additional questions which did provide some interesting conclusions. The first thing that I determined was that there are close to 50 towns or cities in Massachusetts that are more than 30 minutes from the nearest MAT facility. As you can see on the map, this is shown with two distinct bands in the central and western regions as well as less accessible locations like the tip of Cape Cod, Nantucket, and Martha’s Vineyard. I expected to see a correlation between access to MAT and fatal overdose numbers, but I did not. As you see in the top scatter plot there is virtually no correlation.
Now, I decided to consider other factors that affect access such as public transportation. In many situations, patients are reliant on public transportation to get medical care so if it is difficult or impossible to get somewhere via public transportation the distance of that care is irrelevant. To model public transportation reliance, I enriched the towns/cities with the number of households that did not have access to a vehicle. The bottom scatter plot shows a strong correlation between the number of households without access to a vehicle and the number of fatal overdoses. A quick note, this analysis is not normalized by population so it’s possible that population is confounding. Nonetheless, I felt that this opens a door to additional questions about whether MAT facilities are really accessible to the populations that need them most.
Although I’ve been visualizing and analyzing variations of this data for several years I feel like I gained a new understanding using Insights for ArcGIS. Let’s start with the most obvious benefit. Being able to couple the map, charts, and tables in the same interface – and have them respond to each other – has let me explore the data more quickly and in more ways than I could have previously done. But beyond that, the ability to do ad hoc spatial analytics, from spatial filtering to enrichment, lets me do more than just visual analytics. It lets me ask questions to refine my analysis and manipulate the data to find answers that often lead to more questions. Insights for ArcGIS allows me to go as deep into the problem as necessary, giving me confidence in the analysis I share with others.
I encourage you to give Insights for ArcGIS a try with your data. Discover how quickly you can come to results you may have explored with other tools.