Association of hospital closures with changes in Medicare-covered ambulance trips among rural emergency medical services agencies.

Abstract

The emergency medical services (EMS) system plays a critical role in the U.S. health care safety net, performing over 16 million transports per year.1 Although often treated as a publicly funded good, only two-thirds of EMS agencies are run by governments or municipal fire departments2 and many rely on volunteers. As a result, recruitment and retention of EMS personnel has become more difficult over time, especially in rural areas3 where EMS personnel report increasing stress and decreasing satisfaction.4 Closures of over 120 rural hospitals in the past decade5 could worsen strain on rural EMS providers. Existing research shows that closures negatively impact patients through increased time in ambulances, but closures may negatively impact EMS agencies too. If closures increase trip distance, agencies with fixed capacity have less time to prep vehicles and may be unable to respond to some calls, in particular nonemergent ones. Shifting away from nonemergent trips could have negative financial implications for agencies, as such trips may provide a more dependable revenue stream. We estimated the association of hospital closures with EMS agency miles, trips, and emergent status, stratifying by whether EMS agencies were publicly or privately supported. We linked annual, EMS-provider-level data on ambulance trips for patients covered by fee-for-service Medicare from the Physician and Other Supplier Medicare Public Use File (PUF) to hospital closures between 2012 and 2018 from the University of North Carolina Sheps Center (SC). The PUF data contain the annual count of service providers per year identified by Healthcare Common Procedural and Codes. We used these data to create our primary dependent variables: annual EMS-agency-level measures of total ground miles and trips, average miles per trip, and share of total trips that were “nonemergent” based on billing codes for nonemergent Basic Life Support, nonemergent Advanced Life Support 1, and specialty care transport (see Daa Supplement S1, Appendix S1 [available as supporting information in the online version of this paper, which is available at http://onlinelibrary.wiley.com/doi/10.1111/acem.14273/full]. For each EMS agency in the PUF we also extracted the agency's zip code and name. We used U.S. Housing and Urban Development zip-to-county crosswalk to assign each EMS agency to counties. We applied a text-based algorithm to agency names to assign the probable ownership type (see Appendix S2 for details and validation of this algorithm). We used the SC data to create our primary independent variable: a time-varying, county-level indicator of general acute care hospital closures, set to one in the year of hospital closures as well as all subsequent years. For counties with multiple closures, we used the earliest date of closure and we excluded from our sample EMS agencies located in counties that experienced a closure between 2007 and 2011, the 5 years before our data began. County-level controls included poverty rates, uninsured rates, population, demographic characteristics, and morality rates from injuries and substance abuse (controls are described in Appendix S2). After being limited to rural counties, our sample contained 104 EMS agencies in 53 counties affected by a hospital closure, and 5,338 EMS agencies in 1,117 counties not affected by closure between 2012 and 2018 (N = 32,644; Appendix S3 describes the sample.) Our difference-in-differences analysis compared outcomes for EMS agencies located in counties in which a hospital closed to the change over time among EMS agencies located in counties in which a hospital did not close. Our multivariable, ordinary least squares regression included the time-varying, binary indicator of closure, EMS-agency-level fixed effects and year-level fixed effects. Total trips, miles, and average miles per trip were logged because of the skewed nature of the data. Nonemergent trips were expressed as a share of total trips. All analyses were performed with STATA 16, and statistical significance was judged at the 5% level. Sensitivity analyses included defining EMS service areas as larger than the county by redefining hospital closures at geographic areas larger than the county; clustering the standard errors at the county level; reestimating the model using a negative binomial regression model to account for nonlinear outcomes6 including state-by-year fixed effects to account for differential Medicaid expansion across states over time; winsorizing outcomes to minimize the influence of outliers; and reestimating our difference-in-differences regressions as event studies, separately by year of closure to assess whether the results are due to specific hospital closures or whether outcomes were trending differently between agencies that were or were not affected by closures before the closure occurred. (Sensitivity analyses and results are provided in Appendix S4.) Figure S1 presents unadjusted means and 95% confidence intervals (CIs) for our main outcomes for EMS agencies in counties with a closure at any point over the 2012 to 2018 period and for those without a closure. Total ground miles increased by 800 miles among EMS agencies affected by closure while total number of miles traveled increased by only 400 over the same time period. Total ground trips decreased by 34 trips among EMS agencies affected by closures, while total ground trips increased by 20 trips among EMS agencies not affected by closure. Average miles per trip increased by 3 miles by 2018 among EMS agencies affected by hospital closures but did not change among EMS agencies not affected by closure. The nonemergent share of trips fell by 3 percentage points (PP) among EMS agencies affected by closure but remained the same among those that were not affected. Table 1 presents the results of our difference-in-difference analysis. After a hospital closes within its county, EMS ground miles increase by 16% (95% CI = 8% to 24%) relative to agencies in counties without a closure. Results are similar with controls (15%, 95% CI = 8% to 23%). Total trips do not change among EMS agencies affected by closures relative to those that were not affected with or without controls (95% CI = –10% to 3%). As a result, average miles per trip increased among affected EMS agencies relative to unaffected EMS agencies by 19% with or without controls (95% CI = 15% to 23%). After closure, the share of all trips that were nonemergent (either non-emergency or interfacility transfers) fell by 5 PP (95% CI = –7 PP to –3 PP), with or without controls. Relative to the nonemergent share of total trips in 2012 among hospitals not affected by closure (16%) this point estimate represents a 31% (5 PP/16%) decrease in nonemergent trips. Among EMS agencies assigned public support using our text-based algorithm, the average miles per trip increased by 22% (95% CI = 18% to 27%) among affected versus unaffected agencies. Average miles per trip increased by only 10% among EMS agencies that were privately supported (95% CI = 3% to 17%). Nonemergent trips decreased by 3 PP more among publicly supported EMS agencies affected by closures relative to control EMS agencies (95% CI = –4 PP to –2 PP) with no change among those that were privately supported (95% CI = –7 PP to –4 PP). Our results are consistent across all sensitivity analyses except for alternate clustering of standard errors. Our results suggest that hospital closures result in longer trips for EMS agencies, which is not surprising given that the average distance to the next hospital in many rural areas ranges from approximately 17 to 34 minutes.7 Increased distance per trip has both economic and operational significance because longer trips increase depreciation of vehicles, limit time available for stocking and cleaning, and increase technicians and paramedic staffing costs who must remain on duty for longer periods. Longer trips could also result in delays in time-sensitive care. Failing to provide timely care to acutely ill patients could result in “moral injury” for EMS workers, in which providers experience psychological distress from perceived patient suffering.8 Our results also suggest that hospital closures shift agencies toward less predictable sources of revenue. Given that nonemergent transport plays an important role in facilitating access to care for patients in rural areas where other transportation options may be limited, or unavailable altogether, our results suggest a previously unexplored aspect of hospital closures for patients. Finally, our results suggest that privately operated EMS agencies may have been less affected by hospital closures than publicly operated ones. Privately owned EMS agencies constitute approximately 20% of EMS agencies nationally2 and between 17% and 24% of our study sample. In light of increased rural provider burnout, Medicare reimbursement cuts to EMS agencies9 and an expected 5% decline in state and local budgets due to the coronavirus pandemic,10 public policy intervention may be warranted to provide support specifically to publicly operated EMS agencies affected by hospital closure. Our study had several limitations. First, we relied on fee-for-service Medicare claims data, which do not capture EMS use among agencies without the capacity to bill Medicare nor among EMS agencies that serve Medicare Advantage beneficiaries. Second, we assign public versus private support using a novel, text-based algorithm, which may produce measurement error, biasing our results toward zero. Third, we were unable to control for characteristics of EMS agencies such as capabilities, staffing mix, or number of emergency vehicles. Fourth, our analysis focused exclusively on ground EMS agencies and did not include air EMS. As rural hospitals close, EMS agencies—especially publicly supported ones—must travel farther and provide a higher share of emergent trips. The authors thank Hannah Geressu, Ruolin Lu, and India Pungarcher for research assistance. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.