Emergency Room Access in the United States: Rural Hospital Closures
Hospital closures have meant that 1.6 million individuals in communities affected by emergency room closures have to travel, on average, an additional 13.2 km (8.2 mi) to get to the next closest emergency room, making the average straight-line distance for affected populations 25.5 km (15.8mi). This has serious implications for the health of people in these communities.
This KFF brief highlights a big problem — closures of rural hospitals impacts a rural community’s access to healthcare & emergency care. There have been several news stories and such about this topic. However, there’s a distinct lack of quantitative data on the impacts of these closures. At least, a quick literature search for “hospital closures” doesn’t yield any recent, satisfying results. I thought I’d try to take a swing at this, and try to better understand what kind of effect hospital closures have had.
I’m using the term “hospital” here to mean a facility with emergency care/ acute care centers. Clinics and outpatient facilities aren’t taken into account here.
71 hospitals, 134 counties, 1.6 million individuals affected
400 census tracts in 134 different counties in the US were affected by 71 rural emergency room closures since 2010. This represents a change in access for 1.6 million people.
For these 1.6 million people, the average distance to the nearest emergency care center changed from 12.3 km (7.6mi) to 25.5 km (15.8 mi), in essence doubling travel distance to receive care.
(In case you’re not satisfied with the ‘naked eye’ approach, I also estimated effect size using Cohen’s d: we get a large effect size of d = 0.97)
For each individual community, hospital closures have meant that individuals had to travel an average of an additional 13.2 km (8.2 mi) to get to the next nearest acute care center, with a quarter of affected communities having to travel upwards of an additional 20 km (12.4 mi) to get emergency care.
Increases in distance to care are strongly associated with an increased risk of death.
While some of the communities may also be served by clinics and outpatient centers, as well as long term care centers, the loss of emergency rooms still represents a significant impact on care. Previous studies, like this one from Nicholl et al, (2007), have demonstrated that increases in distance to care are strongly associated with an increased risk of death. Nicholl et al found an increase of 1% in absolute mortality for every additional 10km that an individual lived from a hospital. Beyond just emergency rooms, additionally, having to travel farther to receive care means that some people will choose to forego medical care, including visits to primary care providers, leading people to be less likely to engage in preventative care (Countouris et al, 2014).
Hospital closures lead to delays in access to care and needed medical services among the elderly (Walker et al, 2011). Vulnerable populations experienced greater disruptions to their patterns of care (Bazzoli et al, 2012). Impacts aren’t limited to just increases in morbidity and mortality. Closing the sole hospital in a community leads to a 4% reduction in income, and an unemployment increase of 1.6 percentage points (Holmes et al, 2006). People also report feeling afraid, angry, and abandoned (Countouris et al, 2014). One hospital closing often means an increase in emergency room volume at other hospitals — though this doesn’t track 1:1 since some people choose to forgeo care instead (Lee et al, 2015).
The Centers for Medicare and Medicaid Services (CMS) designation for a critical access hospital (a designation for ‘lifeline’ hospitals in areas that would otherwise have no access) states that a critical access hospital must be more than 35 miles (56 km) away from the next nearest hospital. Using that definition, I decided that half of that (17.5 miles (28km)) would be a reasonable cutoff for a hospital to be considered “far away”. Here, we can see the impact on the communities most affected by hospital closures, where the travel distance after closures is now greater than 28km away.
Kings County: Coalinga Regional Medical Center
Kings County will have gone from the top 25th percentile of proximity to emergency room care, to the lowest 1 percentile.
One of the most heavily affected counties, Kings County in California, was the home of Coalinga Regional Medical Center. Kings County, a rural agricultural community, has, as of the last census, a 17.4% poverty rate, and 8.3% uninsured rate. The result of the closure means that individuals here will have to travel to Adventist Health in Hanford, 43 miles away, and, according to Google Maps, a 50 minute drive without traffic. It’s not hard to imagine that this drive would simply be unfeasible for some.
CRMC closed recently, in mid-June 2018. If the local government doesn’t figure out how to re-open the emergency room, Kings County will have gone from the top 25th percentile of proximity to urgent care, to the lowest 1 percentile. It is disconcerting to speculate on what this could mean for this community’s overall health. For example, hospitals near closed hospitals saw a 5% higher odds of mortality than other hospitals. A rough extrapolation of this to Kings County implies that there may be as many as 6 additional deaths per year in nearby emergency rooms, which doesn’t even begin to count the number of patients who forego medical care.
What does this mean?
Hospital closures exacerbate rural health challenges.
The scale of this problem is not insignificant; in general, rural communities have worse health than urban ones. These closures only serve to exacerbate rural health challenges. Things like Medicaid expansion appear to lead to substantially lower rates of hospital closures, but it does seem like an issue that’s flying under the radar. I’m hoping that some attempt at quantifying the problem here might make these issues seem a little more salient.
Although previous studies (Joynt et al, 2015) have found no impact on deaths among Medicare patients, they had no stratification on increase in travel distance, and relied on service areas calculated in 1990. Travel distance seems to be an important piece of the puzzle, so I’d love to repeat studies like Joynt et al’s, to try to tease out those factors.
Methods & Data Sources
A lot of inspiration was taken from this 2009 paper, Access to Emergency Care in the US, from Carr et al.
Homeland Infrastructure Foundation-Level Data (HIFLD) has a dataset containing the locations of hospitals within the US. This dataset was generated from state departments and federal resources, and was last updated in May of 2017. Sheps Center published a list of rural hospital closures, though this data is likely to be incomplete. HIFLD’s data also includes a column for the ‘status’ of a hospital being open or closed, but this list is also not complete.
I went through the HIFLD data, cross-referencing the Sheps Center’s list, then added status information for every one I could find on both lists, and confirm as closed. Hospitals that had reopened in the same location were excluded. This left me with 71 hospital closures.
The Centers of Population for all the census tracts in the US were used for distance calculations. Block groups were ignored for two reasons: computational capacity, and being able to potentially use American Community Survey data.
I used Python to take the list of population centers, and find the nearest hospital with an acute care center for each one. Second nearest hospitals were not taken into consideration. This took a long time.
Taking the Sheps Center closure data as the ‘validated’ data set, I then took communities that would have been affected by the closure, and calculated the next nearest hospital for those communities.
In order to map it, I ended up going back to R. Dr. Kyle Walker has created an R package, tidycensus, which made it really easy to get tract feature geometry. The bonus of tidycensus is that, for additional analysis, I can grab other demographic information to merge into the dataset.
You can find all my code in my github repo.
Stay tuned! Next update: fun with census data!