This summer we are pleased to share research highlights from our Spatial Data Science Summer Fellows, showcasing their work on the US COVID Atlas and opioid risk environment research projects.
This week, Data Science Fellow Rachel Vigil (BA ’22) shares an update on her work on evaluating measures of access for opioid risk environment research projects.
This summer, as a HEROP Spatial Data Science Fellow, I have worked on data collection and curation for the Opioid Environment Policy Scan, a database project developed by HEROP to help characterize the multi-dimensional impact of opioid use across the United States. OEPS includes more than four dozen datasets available at multiple spatial scales, from Census tracts to ZIP codes to counties and states for most of the country. The HEROP team uses this data in research models and analyses for our opioid environment research projects, and because the project is hosted via GitHub, an open-source collaboration platform, it is publicly available to other researchers.
Measuring community access to substance use treatment (SUT) and medications for opioid use disorder (MOUD) is a core research topic that has emerged from our work with the Justice in Community Opioid Innovation Network (JCOIN), part of the NIH-HEAL Initiative. Access is an important component of work done to measure spatial public health, though the concept itself can be hard to define. What is access? And how do we measure it?
One definition of access comes from a landmark paper by Penchansky and Thomas (1981). They define access as “a concept representing the degree of ‘fit’ between the clients and the system.” They go on to identify four dimensions of access: accessibility, accommodation, affordability, and acceptability. It is essential to consider access as a multi-dimensional concept. A methadone clinic may be within walking or driving distance from a population (accessibility), but may be too expensive for that population to take advantage of (affordability), or socially frowned upon to use (acceptability). These dimensions work together to form a broader concept of access.
Spatially speaking, however, it is easiest to measure accessibility, which is how easy it is for a client or patient to physically get to the resource or service. This is what we focus on in our access data collection. Currently, we have three variables that measure accessibility: minimum distance to nearest resource, driving time to nearest resource, and count of resources within a 30 minute drive time.
Minimum distance is a commonly used measure of spatial access to health resources. For our calculations, minimum distance is measured as the nearest Euclidean distance between the centroid of either a census tract or ZIP code to the closest resource. Naturally, there are some limitations to this measure. Namely, it is not always a good reflection of how people actually travel to get to resources. Additionally, the centroids used in calculating the distance are geographic, not population based. Thus, this model does not accurately estimate the location of the population of interest, and therefore may not always be the best method for finding the “fit” between the client and the system.
Advanced Travel Metrics
This is where our other two methods of calculating accessibility come into play. Using a travel cost time matrix calculated by a CSDS Software Engineer Vidal Aguiano using OSRM, we can calculate the metrics “drive time to nearest resource” and “count of resources within 30 minutes drive.”The former is the minimum drive time from an area’s population-weighted centroid to the nearest resource, while the latter is the count of resources within a 30 minute drive of the population centroid. By using the population centroid, this model more accurately represents the location of the client. And by using drive time, this model is a much more accurate estimation of how people actually travel.
While a better measure of individuals’ travel behavior, this model also has its limitations. Notably, it rests on the assumption that the population has car access. And while it is true that the US has one of the highest car ownership rates per capita in the world, individuals experiencing homelessness are “nine times more likely to die from an overdose than those who are stably housed” according to the National Health Care for the Homeless Council, and may not have stable access to cars as a form of transportation. To this end, we want to add new access metrics to OEPS, metrics that measure access through alternate forms of transportation such as walking, biking and public transit.
These models all present unique challenges. Walking follows similar routes to car travel. Services like Google Maps, Grasshopper and Pandana provide models for walking travel times. However, these do come with challenges. For one, sidewalk grid networks showing where it is safe to walk (i.e. where sidewalks are available) are often incomplete or in disrepair. Incomplete sidewalk networks make it difficult to estimate travel times for populations with different abilities or limited mobility, since a wheelchair user, for example, would not be able to travel along a corridor with no sidewalk. Additionally, there is the question of what the maximum time it is reasonable to expect people to travel on foot to get to a resource. Former CSDS Data Engineer Dan Snow’s walking model stops measuring walking times after 90 minutes, but for our purposes, especially for a count within some amount of minutes, that is likely too long a trip to expect. A trip of that length, if it was just one way, could take up three hours of someone’s day.
Calculating travel times via bicycle can also be complicated. No city has a bike lane on every street, and even though many streets without bike lanes can be used by cyclists, they have been found to be more dangerous than streets without bike lanes. It isn’t surprising, then, that Uber found the streets most commonly biked on in urban areas are those with bike lanes. There also is often a disconnect between urban biking networks and rural ones, so it becomes difficult to connect them on a national scale. Google Maps does have methods of computing biking travel times, and work has been done on a city-by-city basis to measure travel times, the way forward using our national access datasets is unclear.
Finally, public transit presents its own set of issues. Though municipalities have well defined and scheduled public transit networks, transit times can vary drastically by method of transportation (bus vs. train), from route to route, and from city to city. In addition, there is no nationwide public transit system. And within certain transit systems, there exists a gap between the last stop in a transit journey and the client’s final destination. Models of transit times have been created, and some can be found through Dan Snow’s “Resources for Computing Travel Cost,” but are once again difficult to scale up nationally.
Even though these metrics come with their own sets of challenges, they are an important part of accurately and equitably measuring the multi-dimensional concept of accessibility. And in seeing multi-level accessibility modeling, we can better determine what next steps should be taken to address accessibility gaps through policy. Having a clear picture of accessibility also gives us a way to narrow down what is increasing or limiting access to certain systems or resources. For example, if there is a treatment provider or clinic that by our metrics is highly accessible, but is being underutilized in its community, we can start exploring the other three dimensions of access (such as affordability or acceptability) to see what limits clients from using the system. As we put more pieces of the access puzzle together, we can step closer to reducing harm associated with opioid use and increasing resources to help those in need in our communities.
This post was written by Rachel Vigil (BA ’22), HEROP Spatial Data Science Fellow.