 And we are live. I'm Peter Higgins and I'm going to introduce our next speaker. Angela Lee is an R-spatial Research Analyst at the University of Chicago, working in Economics, Health, and Public Policy, known for our Lady Chicago and the R-spatial chat on Twitter. Also at CivicAngela, Marina Kozak is the Assistant Director of Health Informatics at the University of Chicago and a Health Geographer presenting Calculating Nationwide Access Metrics for Treatment of Opioid Use Disorder. Great. Thanks for having us, Peter. We're really excited to be here today talking about some of the work that we've been doing over the past year. I'm actually currently no longer at the University of Chicago, so I've moved on to a different institution, but Marina will be carrying on this work over the next few years. So this work has been sponsored by the NIH Heal Initiative, which is part of the J-Coin Network, which looks at opioid treatment for people in justice settings. And we did this as part of the Center for Spatial Data Science. And, Marina, I'll let you go ahead and introduce yourself as well. Sure. Yeah. So I'm Marina. I'm a Health Geographer, as well as Spatial Epidemiologist. I work primarily as a social determinants of health researcher in how SDOH impacts multiple different domains. And the opioid epidemic is one of those domains. So we'll be focusing on that today. Great. Cool. All right. So let me give a little bit of a background first on opioid use disorder. I'll refer to that as OUD for short. So OUD is a chronic and relapsing disease that affects the body and the brain. It can affect individuals of any background, which is something that's become increasingly known, hopefully, over the past few years. Also, it can result from taking prescription opioids after surgery or for pain. These are a unique set of circumstances kind of combined to drive what we now call the opioid epidemic. And that basically means in the past about 20 years, almost 450,000 people have died from an overdose involving opioids in 2018. There were 2 million people who had an opioid use disorder. So this gets further complicated, of course, as we talk about COVID. So while the data is still new and fresh, at the very minimum, we know that opioid-related mortalities have worsened during the COVID-19 pandemic. This is likely due to a variety of reasons. It could be from increased isolation that will drive relapse and overdose death. It can be related to limited treatment options as well as disruption to existing access, as well as exacerbated health and economic inequities. So at this point, we know that at least 40 states have reported increases in deaths as of this month. And at least there's been an 18% increase in related mortalities nationwide, though you'll see a lot of spatial heterogeneity, meaning in some areas it will actually be far, far higher percentages than 18. So something that's not as often talked about when we talk about the opioid epidemic is access to evidence-based treatment. So from this talk, we're really coming at this from a public health and kind of clinical science perspective. For a long-term recovery, we know that people are more successful in the recovery plans when they involve medications for opioid use disorder, specifically evidence-based ones. These not only help brain chemistry or normalize brain chemistry and curb withdrawal symptoms, but in obviously support recovery plans, but it also saves lives. So there's a huge reduction in mortality when MLUDs are included. And from a public health perspective, it also prevents downstream infections from injection drug use, like hepatitis C, HIV, and so on and so forth. There's three primary medications that we talk about in this area, methadone, buprenorphine, naltrexone, each work in very different ways. And then from a clinical perspective, it's really important to get the right treatment to the right person. And individuals will have preferences for different medications, and we find that success is really connected to whether or not they're able to get what their preference is. So next one. Just to give you kind of, you know, highlight the importance of these, we know that specifically methadone and buprenorphine are associated with reduced mortality, especially after non-fatal opioid overdoses. So the idea is to capture someone, if they've survived, you know, in this example, if they've survived an overdose to make sure that that doesn't happen again. Unfortunately, a minority of patients actually get access to MLUDs. And it's still not widely used in this, in the community for a variety of access stigma and other issues that we won't get into today. But to highlight, so next screen, just one kind of specific point here. Methadone has been associated with very high reduction in mortality specifically, which means less overdose deaths and followed by buprenorphine. And then there's still research being done on naltrexone. So that may be preferred in some areas for other complex reasons. So our central question is really, where can patients access evidence-based medications for OUD in the US? And then from a social determinants of health perspective, how does that kind of structural access mitigate health outcomes, right? And it ends up to be a very complicated topic because not all medications are used in the same way, and access will really matter. So different individuals need access to different medications in different ways. Methadone, you might need to use daily initially, and then weekly, and then monthly. So access is really critical, whereas buprenorphine can be picked up by pharmacy monthly, and then naltrexone would also be another monthly trip. So I'll switch over to Angela to delve into the details. Thanks, Marinette. So before I started on this work, I hadn't done that much in terms of access. And so one of my major goals for the first part of this was to sort of understand access and how we define access. So in terms of how we can measure access, there are many questions that we can think about when looking at this. So one of them, as we mentioned, was what resources being access and how frequently it doesn't need to be accessed. You can imagine that going to a methadone clinic regularly is very different from going to a specific clinic to get a prescription for buprenorphine. Additionally, how far away is the provider? And not only in distance, but also in time. So you can imagine in Chicago, a short distance that might be able to be traveled in rural areas might take a really long time to get to, depending on congestion and other factors. Additionally, how will the individual get to the resource? Do they have access to a car or do they have to rely on walking or public transit? And so adding these things into our understanding of access really helps us further our knowledge of how individuals can be treated. And this isn't only medications for opioid use disorder. This could be for primary care. This could be for, I think I talked to someone earlier today who was interested in liver transplants. So these are a set of methods and ways to think about the world that might be useful in other parts of the public health sphere or even beyond. So one of our first objectives was to define spatial access to medication. In the literature, you'll find some very common metrics. So basic ways of capturing spatial access include the count of resources per area. So within a given area, how many hospitals are within a zip code or looking at a count of resource buffers. So if we buffer hospitals by one mile, how many buffers intersect with this zip code? Often these methods are undertaken using proprietary or expensive software. But we also wanted to look at the sensitivity of boundaries. How are we certain that one mile is ideal for a specific area? Do we want to look at transit mode? Do we want to look at population density or re-running this analysis multiple times? For our access metric that we'll be talking about today, we are looking at minimum distance. So we took the centroids of zip codes and we measured from that centroid to the nearest MOUD resource. So in this case, you can see that we have two points and the point that is closest to the zip code centroid, we measure that distance. So it might be 1.2 miles. And we can do this at scale in R. So that brings me to my next point. We want to scale spatial access analysis. Point and click GIS has pitfalls. So you have to repeat your processes for each visualization of interest. This might also often be done by single analysts on costly licensed software in a separate research silo or department. We believe that using R for GIS makes spatial analysis scalable and meaningful to the larger research community. So you can imagine that geospatial data processing, access metric calculation and geospatial visualization, that might not be something that you initially do in-house. But with R, you might be able to remove that bottleneck and start looking at spatial access metrics and look at that in more detail. So the next bit I want to talk a little bit about the code that we used to do this. And R, I'm going to go through relatively quickly. We have written a tutorial with a lot of this information and it's online and we'll share the link later. So sit back and let the code wash over you. I'll be going through pretty quickly. The data that we're going to be using is the substance abuse and mental health services administration data from 2019. The sources are the National Survey of Substance Abuse Treatment Services and the Buprenorphine Waved Physicians Database. If you go to the SAMHSA website, there's this nice locator and you can actually check the specific types of substance use services that you're interested in, as well as a number of other things, healthcare centers, mental health. And you can download, I think, a CSV with the point locations of these facilities, as well as the addresses and other information. Note that this is a nationwide point location data set of substance abuse treatment centers prescribing MAUD. So you can actually get methadone at places that are not substance abuse treatment centers specifically. So anything that we show is probably even an undercount of the access, but it's the best data we have. We're also looking at the most recent year, so 2019. One of the projects we have underway is to get some historical data on the locations of these facilities and look at sort of more of a temporal analysis. To select our MAUD proxy variables, there are a number of MAUD relevant variables coded in the survey. So we relied on clinical expertise to select the proper variables relevant to long-term recovery and maintenance. You can see that from this juju plot that there are six methadone variables in the data set. We chose the one that was relevant to methadone maintenance specifically. This is really different from cases in which we might use methadone in an emergency context. The workflow that we used was first to process our data, second to calculate our access metrics, and third to visualize. I'm going to walk through each of these sections in R, and this is sort of the general data workflow that we used. And you can see that there's a lot sort of in the whole process. We have to combine for data processing. We have to look at addresses and a CSV, pull-in boundary data, and then take that as inputs to calculate access metrics in R. We can calculate a number of different types of access metrics. We'll be focusing on minimum distance. And then once we have that, we can put it into a visualization workflow with other boundary files, join attributes, and visualize this using we's the team app package in R. But you can throw it into a shiny app, and we've seen a lot of examples of great shiny apps lately. Okay, so I want to go through this quickly. But the first thing that we're going to do is prepare our origin or zip code data. We're going to be using the SF package, and the functions in that package really are very clear on what they do. So the first thing that we're going to do is read in our shape file using read SF. This is very similar to read CSV. But you can throw in any type of spatial file in there, and it'll read it and turn it into a spatial object. Then we'll transform our data. We're going to project it. So turn it from latitude and longitude, which is on the sphere of the globe to a flat surface. This is extremely important when you are measuring things like distance so that you get an accurate representation of distance. We're going to be using a projection that is specifically good for Chicago, which is the 3435 projection. And then we're going to calculate the centroids of those zip polygons using ST centroid. Finally, we'll take that centroid object and write it out using write underscore SF. And we can write to shape files or CSV. So we can take this point object in R and save it as a spatial thing, or we can save it as a CSV and share it with stakeholders down the road. We also, I don't have data here for calculating population-weighted centroids, but that is also something that you could do. Next, we'll prepare the destination or resource data. So these are where our MOUDs are. If you have solely addresses and if you just are working with addresses, you can first geocode those addresses. So what that will do is take an address 123 Elm Street and change it into latitude-longitude coordinates. Once you have those coordinates, you can use ST as SF and change that into a spatial object in R. Finally, you're going to project again using ST transform using the new coordinate reference system or CRS and put it into the same projection as your zip or boundary data. I'm going to take a look at the spatial data structure that we're working with in R because I know that not that many people have worked with spatial data structures in R. So if I go ahead and pull up the SF R object, what that looks like is something like this. SF is really good because it allows you to work with structures that look like data frames in R. If you've worked with the SP package, you'll notice that it's a little bit different in terms of the objects that are created. First, I want to point out that there's a header that gives you some geographical information about this data set. There are a number of things in here that are important, but the most important thing for you as you go forward and work with geospatial data is to look at the projection. 4, 3, 2, 6 means that the data is not projected and that it's in longitude and latitude and not in X and Y. You can see that in our geometry column, this is the last column in your spatial object that we have a point object and it has a longitude of 87, negative 87 and a latitude of around 41. These are in decimal degrees and they are not great for measuring distance. Okay, so to calculate minimum access and we can talk a little bit more about this, but I'll direct you to some resources that go into a bit more about spatial data structures. To calculate minimum access in R, we're going to be using a great function called ST nearest feature. This is built into the SF package and with point data, you can find the closest one to you. I think I talked to someone who didn't know about this function and had been calculating it by hand and running their own functions, but ST nearest feature is fast and it will allow you to find that resource. So we're going to find the indexes of the nearest clinics and then we're going to index the original methadone locations. And what that gives you is for each zip code, what is the address and geometry location of the nearest methadone provider? And you can see that in this case, the point data is the numbers are a lot bigger. This is because we've projected that data and the numbers are in X, Y, and I think it here they're in terms of feet instead of decimal degrees. Okay, for each origin code, we're going to calculate the distance the nearest destination. So now that you know what object is closest, you're going to calculate the distance between the center of your zip code and the nearest clinic. And then you're going to convert to miles. So when we do that, we get a nice vector of miles in R and this we're going to use as our access metric for now. So you can see for the first observation or data, the nearest methadone clinic is 6.98 miles away. Okay, and then finally visualizing access metrics in R. After we joined the distance value back to the original spatial file, we're going to make a thematic map with T map. And this is very similar. The syntax is very similar to GD plot two. However, it adds a few functionalities in there that allow you to work with spatial data specifically and then also visualize interactively. So I was in the geospatial mapping for the feather earlier and people asked a question about this. So not only can you make this map, but you can also run a function called T map mode view, then run the same code that you already typed in again. And what you'll get is an interactive version of the map, the static map that you've just created. Disclaimer, these map defaults are not great because the breaks are a little bit bigger than we want. I think carefully about the map breaks that you want. We also don't have all the resources that we need in here. So this is not a final visualization. This is just to show you what T map looks like and what it does. So T map mode view and T map mode plot will allow you to go between static and interactive mapping and T map. So I really, really like this package. Okay, I sped through all of that really quickly but I'm gonna go ahead and direct you to these free high quality resources for learning more. These are both books that have come out within the last two years. And they are just phenomenal. If you want like a overview of sort of geocomputation in art or geospatial health data and how to build shiny apps for health and ethnic specifically, I think purposes, go ahead and take a screenshot of these links. I really recommend these resources for learning how to do this. Okay, I'm gonna jump to the results. So we are still at the beginning of some of this analysis. So there's definitely more to be done but I'm gonna go ahead and show you the US wide access metrics that we calculated for each of the medications. These are the access metrics for a buprenorphine based on minimum distance. You can see that access is actually, it looks pretty good, not terrible. However, I will point out that these access metrics are to potential prescribers. So they're all the positions that could potentially prescribe you buprenorphine. We have no way of recording actual prescribing. So it's the best we have. This is not realized access. This is sort of potential access. Naltrexone looks a little bit worse and like the access is not as comprehensive. You may also say, hey, like this looks like it's just an urban rural map and that's potentially true but when we jump to methadone, you'll see that sort of looking at buprenorphine and naltrexone methadone is a lot sparser. So this is really problematic because methadone has been shown to be one of the most effective medications in reducing mortality. And it's one of the medications that needs to be accessed most frequently. So every day. And you can see that the access for this is just way worse than buprenorphine and naltrexone respectively. So these are just preliminary results. But I think this is really, really telling in terms of methadone access and the sort of sparseness of that access across our nation. We also took a look at sort of the median distance to the nearest resources versus rural and urban areas. So you can see that the rural distances are a lot higher than the suburban and urban distances. Obviously this doesn't take into account congestion, travel, like time to distance. But you can see, and I think in here we actually threw in dialysis. Dialysis is very similar to methadone in terms of how frequently you have to access it. And so if we use dialysis as sort of a baseline for what access should look like, you can see that methadone is way, way farther away than dialysis centers. Okay, extensions and next directions again or sort of at the beginning of this whole research process. So we wanted to present some of the work we are doing just to get it out there but we are working with three MOUD treatments, methadone, buprenorphine and deltruxone. We're doing it for 50 states. We're also going to add in more access metrics. So you can see that we are looking at distance nearest and we're also interested in the count within a certain threshold, the time to the nearest location, a specific access score. And as I mentioned before, we're looking at time periods. So historical data and trying to digitize a lot of those directories from SAMHSA. And then finally, we want to add in more resources like primary care to better understand sort of the landscapes surrounding care for individuals who are suffering from opioid use disorder. Thank you, so we're at the end of the talk. Finally, translating everything that we've done, we're writing an opioid toolkit to explain to our partners what we're doing and give them the resources to do it themselves. So we're going to share this with our cohort and Jcoin. We're preliminary hoping to release this in fall 2020 and share it across our network and community groups. Right now there's a tutorial for access metric calculation, sort of I sped through really quickly up there and some visualization and mapping. We plan to add a lot more in the next bit of time, but if you are curious in doing this process for your data, I will go ahead and give you the link I think in the next slide. And thank you so much for having us. I'm really excited that we got to share this work, even though I think we're still at the outset. It's really important to work and I'm really grateful to Marunia for giving me a chance to work on this. And here are the two links that helped me a lot when I was getting started. So I'm going to go ahead and share those with you. Thanks so much. Super, and if you could drop those links into the chat, that would be awesome when you get a chance. We're coming up on a break. So if folks need a bio break, go ahead. But Marunia and Angela are obviously here if you want to pop a few questions. I will not feel badly if people need to take a break. One question that got a lot of votes was, is methadone versus the other forms of access divided by socioeconomic status? It seems like they're very different access. Yes, definitely, definitely. There's a kind of a long historical and complex reasoning behind that. And it's also kind of related to the different types of what the medication will actually do. So for example, naltrexone seems to be preferred in areas that might have a more punitive approach to opioid use disorder. Because with naltrexone, you're blocked from feeling the effects of opioids. So it's very complicated, but yes, definitely. And another question, it's a little bit slightly different topic, but I think an interesting question. Do you know anything about the opioid related mortality rate, or do you have data on that during the pandemic compared to before? And have you seen increases since March or April? Yes, definitely. I know that in some parts of the countries, it's doubled, we had a slide at the very beginning that highlights some of the most recent research that came out of this, but for example, there's been at least an 18% increase in mortality. And that's globally, that's the average. So that doesn't even include areas that had it even more, where it was even more severe. And the concern again, is that a lot of the areas that had very few resources, like they might have had just one medication. Again, the disruption to that access is even more problematic. There is a website on the national health statistics. I think it's in our references that has drug over those deaths by I think date. And so we pulled sort of this from that dashboard. I think the 18% in mortality rates is sort of preliminary analysis that has come out looking at sort of extrapolating it from some other things. I don't think there's any national collecting of opioid related mortality specifically. So it's mostly just drug overdose. Great. And it looks like Murni has already started to respond to this, but Trang Lee asked, rather than distance, should you do travel time? Because it's so variable in urban versus rural. Exactly. Yeah, I think this is sort of the next stage. And we've actually gotten to generate, and we put this in the appendix because we wanted to, I realized that, let's see, someone might ask about it, but... Clearly the fans are sticking around during the back. Okay. So yeah, we've done it for Illinois. In terms of travel times, it's a lot more difficult computationally because you have to look at street networks, think about like how you're gonna route from destination A to destination B. But our research center has actually put together, it's a pipeline package about calculating access metrics from census tracks to the other. And we can drop the link to that chat in the chat as well. So let's see. Yeah, so these are what we've done in Illinois so far, but we haven't scaled them up yet to the nations. That's very cool. Awesome. Well, I'm sure we will try to figure out a way to curate all these links and share them with folks, but it's just great to see so much spatial work in the health space. Awesome. Thank you very much. Thank you for having us. Thanks Peter. Thanks.