DOI: 10.1111/1475-6773.13147 Health Services Research RESEARCH ARTICLE Ambulance diversions following public hospital emergency department closures Charleen Hsuan JD, PhD1 Ninez A. Ponce MPP, PhD4 1 Department of Health Policy and Administration, Penn State University, University Park, Pennsylvania 2 Department of Emergency Medicine, University of California, San Francisco, Zuckerberg San Francisco General Hospital, San Francisco, California 3 School of Law, University of California, Los Angeles, Los Angeles, California 4 Fielding School of Public Health, Department of Health Policy and Management, University of California, Los Angeles, Los Angeles, California Correspondence Charleen Hsuan, JD, PhD, Department of Health Policy and Administration, Penn State University, University Park, PA. Email: chsuan@psu.edu Funding information University of California, Los Angeles; Agency for Healthcare Research and Quality, Grant/Award Number: R36HS024247-01; National Center for Advancing Translational Sciences, Grant/Award Number: UCLA CTSI #TL1TR000121   Renee Y. Hsia MD, MSc2   Jill R. Horwitz PhD, JD, MPP3    Thomas Rice PhD4  Jack Needleman PhD, FAAN4 Objective: To examine whether hospitals are more likely to temporarily close their emergency departments (EDs) to ambulances (through ambulance diversions) if neighboring diverting hospitals are public vs private. Data Sources/Study Setting: Ambulance diversion logs for California hospitals, discharge data, and hospital characteristics data from California's Office of Statewide Health Planning and Development and the American Hospital Association (2007). Study Design: We match public and private (nonprofit or for-­profit) hospitals by distance and size. We use random-­effects models examining diversion probability and timing of private hospitals following diversions by neighboring public vs matched private hospitals. Data Collection/Extraction Methods: N/A. Principal Findings: Hospitals are 3.6 percent more likely to declare diversions if neighboring diverting hospitals are public vs private (P < 0.001). Hospitals declaring diversions have lower ED occupancy (P < 0.001) after neighboring public (vs private) hospitals divert. Hospitals have 4.2 percent shorter diversions if neighboring diverting hospitals are public vs private (P < 0.001). When the neighboring hospital ends its diversion first, hospitals terminate diversions 4.2 percent sooner if the neighboring hospital is public vs private (P = 0.022). Conclusions: Sample hospitals respond differently to diversions by neighboring public (vs private) hospitals, suggesting that these hospitals might be strategically declaring ambulance diversions to avoid treating low-­paying patients served by public hospitals. KEYWORDS access to care, ambulance diversion, emergency department 1   I NTRO D U C TI O N particularly for patients with time-­sensitive conditions such as acute myocardial infarctions.6-8 Although many hospitals divert because In 2003, approximately one-­ third of U.S. hospitals temporarily their EDs are crowded,3,9,10 diversions may occur for other reasons.11 closed their emergency departments (EDs) to ambulances by de- Previous research suggests that hospitals may defensively divert 1 claring an ambulance diversion. Ambulance diversions delay emergency care, 2,3 and have been associated with increased mortality, Health Serv Res. 2019;1–10. 4,5 when neighboring hospitals declare diversions10 to prevent being overwhelmed by the neighboring hospital's patients.12,13 This paper wileyonlinelibrary.com/journal/hesr   © Health Research and Educational Trust   1 2       HSUAN et al. Health Services Research examines another cause for diversions: whether hospitals divert to they strategically declare diversions, since the patients never arrive avoid patients from public hospitals. at the first hospital (Appendix S1). Nonetheless, such strategic di- Although previous research modeled the economic benefits to versions would unnecessarily reduce or delay access to emergency hospitals of diverting uninsured and Medicaid patients,12 no study services to ED patients treated by public hospitals and undermine has examined whether hospitals actually do so. Here, we apply four EMTALA's purpose. tests for whether hospitals change their diversion behavior when a neighboring hospital on diversion is a public hospital, a behavior we term “strategic diversions.” First, we ask whether, after a neighboring hospital declares a diversion, hospitals are more likely to declare their own diversions if the neighboring hospital is public vs private. 3   M E TH O DS 3.1   Study design and data sources Second, we test whether hospitals divert at a lower ED census when We apply a retrospective analysis using 2007 California data. The neighboring public (vs private) hospitals go on diversion. The third units of analysis are hospital ED diversions leading to temporary and fourth tests focus on the potential costs of diversion—lower rev- emergency department closings in hospitals with neighboring pub- 14,15 —and suggest that a hospital diverting lic or private hospitals that had previously initiated an ambulance for strategic reasons may want to end its diversion sooner than a enue and gross margins diversion. We examine only complete hospital diversions. The ob- hospital diverting for capacity reasons. In the third test, we examine servation hospitals whose diversions are analyzed are private hospi- whether diversions are shorter when hospitals divert subsequent tals, both nonprofit and for-­profit hospitals, with a public hospital of to a neighboring public, vs private, hospital's diversion. Finally, we similar size located within 25 miles of the observation hospital and examine hospitals’ responsiveness to a neighboring hospital ending which are located in regions in which time-­stamped diversion data its diversion, examining whether responding hospitals end their di- are available. versions sooner after a neighboring hospital ends its diversion if the neighboring hospital is a public, vs private, hospital. We create our sample from hospitals listed in 2007 ambulance diversion logs that report start and end time for each hospital's diversion. We use 2007 data because it is the last year that Los Angeles 2   BAC KG RO U N D County (which accounts for 22 percent of all California hospitals) provided detailed data, so using more recent data would exclude a significant portion of the state and raise concerns about external It is unknown whether hospitals are more likely to defensively divert validity. However, we examine more recent aggregated data on di- after diversions by public hospitals compared to private (nonprofit or version hours and find that neither the percent of hospitals within for-­profit) hospitals. A hospital wishing to avoid low-­paying patients our sample that declared an ambulance diversion nor the amount has reason to avoid patients who are treated at public hospitals, in- of time that these hospitals were on diversion changed significantly cluding their EDs. Public hospitals are more likely to serve Medicaid from 2007 to the most recent year that these data were reported or uninsured patients, who are less profitable than others. Estimated (for most hospitals, 2015; 82 percent vs 71 percent, P = 0.342 and ED profit margins are −54.4 percent for uninsured and −35.9 percent 869 hours and 676 hours, P = 0.4454, respectively).* for Medicaid patients, compared with −15.6 percent for Medicare From the diversion logs, we identify 178 nonfederal, general hos- and 39.6 percent for privately insured patients.16 Adjusted mean ED pitals and match diversion log data with ED and inpatient discharge payments are lower for uninsured and Medicaid patients (30 and data from the State of California Office of Statewide Health Planning 50 percent, respectively) than for commercially insured patients.17 and Development (OSHPD) and data on hospital and financial char- Furthermore, uninsured and Medicaid patients are often more acteristics from OSHPD and the American Hospital Association's medically complex and consequently require more resources than Annual Survey. privately insured patients.18 Providers may also believe, perhaps in- There are 33 local EMS agencies (LEMSAs) in California, of accurately,19 that these patients increase their malpractice risk. 20,21 which 29 permitted ambulance diversions in 2007. 23 Researchers Moreover, laws intended to protect patients from being turned previously collected ambulance diversion logs from 15 of these away because of insurance status may not apply in the diversion LEMSAs (covering 61 percent of general medical hospitals in setting, leaving uninsured and Medicaid-­insured patients at risk. the state with EDs). We exclude data from five LEMSAs without For example, Medicare-­ participating hospitals are prohibited by identifiable hospital names or for which only aggregated data are the Emergency Medical Treatment and Labor Act (EMTALA) from available. denying patients’ emergency care based on their insurance sta- We exclude twenty hospitals (11.2 percent) consisting of (a) six tus. 22 However, hospitals wishing to avoid Medicaid and uninsured hospitals in LEMSAs that had fewer than three hospitals with diver- patients might strategically declare an ambulance diversion when sions, because such a small number would not allow us to measure neighboring public, vs private, hospitals are on diversion. Not only responsive diversion behavior; (b) two hospitals in diversion logs that would this likely not draw as much scrutiny as directly denying these could not be matched to OSHPD data; (c) five hospitals that partici- unprofitable patients’ emergency care, but under current regula- pated in a project to reduce diversion hours, 24 which may confound tions, hospitals might still be in compliance with EMTALA even if results; (d) six public hospitals that had fewer than 15 diversions in       3 HSUAN et al. Health Services Research 2007; and (e) one public hospital that did not have any neighboring where Y represents the outcomes described above (ie whether private hospitals. the observation hospital declares a diversion while the neighbor- Among the remaining 158 hospitals, we classify hospitals into ing diverting hospital is still on diversion status (a linear probabil- public and private hospitals. There are sixteen public hospitals and ity model); the duration of the diversion; and the time elapsed 142 private hospitals. between the neighboring hospital and observation hospital end We restrict the sample to 28 private hospitals that had public their respective diversions); neighbor ownership is whether the hospitals of similar size (as measured by annual ED visits and bed neighboring hospital is public or private; ED occupancy is the size) within 25 miles. Driving distance is generated using Google predicted number of ED patients in the observation hospital; ED Maps API. The sample is restricted to hospitals of similar size so crowding are variables that indirectly measure ED crowding; di- that emergency medical services would see the neighbor hospital version are diversion-­s pecific characteristics; hosp-­p air are char- as a reasonable alternative to the hospital initially on diversion. We acteristics specific to the observation hospital and neighboring do not restrict the number of private hospitals a priori, but find in hospital; and hosp represents hospital-­s pecific characteristics. All matching that markets do not include more than three private hospi- predictor variables are described below. Because the interaction tals of similar size and distance. terms were not statistically significant for the duration and timing outcomes, we report the main effects for those outcomes (ie Equation (1) minus the interactions) and the fixed parameters of 3.2   Outcome and explanatory variables the model. We use three outcome variables to identify responsiveness to diversions of neighboring hospitals. First, a dichotomous variable measures whether a hospital declares a diversion after a neighboring 3.3.1   Main predictor variables hospital declares a diversion, but while the neighboring hospital is The two main predictor variables are neighbor ownership and still on diversion status. the interaction of neighbor ownership × ED crowding. Neighbor Second, we use the number of minutes of the observation hos- ownership tests whether the responsiveness of observation hos- pital's diversion. A hospital diverting for strategic reasons may want pitals to neighboring hospital diversions depends on whether to end their diversions sooner than a hospital diverting for capacity the neighboring hospital is public or private. As discussed above, reasons. This is because diversions have been associated with lower if hospitals strategically divert, we expect the likelihood of de- revenue and gross margins,14,15 so hospitals that are strategically claring the diversion to be higher and the time to the end of the diverting may want to balance avoiding unprofitable patients being diversion to be shorter if the neighboring diverting hospital is diverted from public hospitals with losing profitable patients that are public. in their usual catchment areas. The length of the diversion has a long The second main predictor variable is the interaction of tail, so we exclude observations where the outcome is above the whether the neighboring hospital is public or private with ED oc- 95th percentile (>171 minutes), although we include these in sensi- cupancy at the observation hospital. ED occupancy is included as tivity analyses. a covariate because the likelihood of diversion should be higher Third, we use the number of minutes from the time the neigh- when ED census at the observation hospital is higher. Because boring hospital terminates its diversion until the observation hos- this is derived from discharge data, patients who are boarding (ie pital ends its own diversion. This outcome proxies how responsive admitted but waiting for a bed) are not included in this measure the observation hospital is to when the neighboring hospital ends its of ED occupancy. To test whether hospitals initiate diversions diversion. Cases in which the observation hospital terminates its di- at lower ED occupancy when a neighboring public hospital de- version before the neighboring hospital are excluded from the anal- clares a diversion, we interact the ED census with the indicator of ysis, but we include these in sensitivity analyses. The time elapsed whether the neighboring diverting hospital is a public hospital. As between the two hospitals ending their respective diversions has a discussed above, if hospitals strategically divert, we expect that long tail, so as with the second outcome, we exclude observations the interaction of whether the neighboring hospital is a public (vs where the outcome is above the 95th percentile (>119 minutes), al- private) hospital with the observation hospital's ED occupancy to though we also include these in sensitivity analyses. be negative. Emergency department occupancy is measured on a daily basis. A hospital that has been on diversion may have lower occupancy than 3.3   Statistical methods if it had not diverted, raising issues of reverse causality. To address We estimate the following equation as an instrumental variable lin- this, we use instrumental variable regression to generate a predicted ear probability model, with hospital-­level random effects: ED occupancy that is free of reverse causality. Our instrument is inpatient occupancy, which is related to diversions only through ED Y = fn(neighbor ownership + neighbor ownership × ED occupancy + neighbor ownership × ED crowding + ED occupancy + ED crowding + diversion + hosp-pair + hosp), (1) occupancy. For Equation (1), the interaction also includes an instrument for inpatient occupancy interacted with whether the neighboring hospital is a public hospital. 25 4       HSUAN et al. Health Services Research instrumented census; using clustered standard errors; and includ- 3.3.2   Other control variables ing outliers. We additionally test an additional hypothesis that relied As described in Equation (1), we additionally control for ED crowd- on slightly different data, whether hospitals changed the timing of ing, diversion-­specific characteristics, hospital-­pair-­specific charac- the beginning of their diversions if the first hospital in a market to teristics, and hospital characteristics for the observation hospital. declare a diversion was a public vs private hospital (Appendix S4). In this hypothesis, we examine whether hospitals in a market may 3.3.2.1   ED crowding “race” to declare a diversion when more than two hospitals are al- Emergency department crowding consists of three variables: the public, vs private, hospital. Thus, we measure whether the duration hour that the neighboring hospital declares its diversion; whether between the second and the third hospital to declare a diversion is the diversion occurred on a weekend; and average physician staff- shorter when the first hospital to declare a diversion is public vs pri- ing. We include hour as a measure of ED crowding to account for vate. Finally, we examine differences in severity of illness in diverting variation in the number of ED physicians on shift,9 nurse staffing, 2 hospitals by whether the neighboring diverting hospital is public vs and demand for ambulance and ED services. ready on diversion, and the first hospital to declare a diversion is a 26 We additionally in- private (Appendix S5). This proxies for whether patient acuity differs clude whether the diversion occurred on a weekend, as hospitals are on days when a neighboring hospital on diversion is public vs private. more likely to have crowding on weekends. 2,26 Physician staffing is the ratio of the average number of ED patients to the number of This study was approved by the UCLA and Penn State Institutional Review Boards. emergency medicine physicians with privileges (Appendix S2). 3.3.2.2   Diversion-­specific characteristics 4   R E S U LT S Following previous research, we include three independent variables Our study sample includes 28 private hospitals in seven California to control for factors associated with diversions other than census: LEMSAs, which were matched to 16 public hospitals of similar size the length of time that the neighboring hospital is on diversion;10 and driving distance. All but two of these private hospitals are the month of the diversion;4,27 and whether ED visits are extremely nonprofit hospitals. The majority of public hospitals in this study high in the LEMSA for that day. We adjust for length of time that the are teaching hospitals (62.5 percent), with a mean bed size of 490; neighboring hospital diverts in order to adjust for the influence of matched hospitals are significantly smaller (mean of 343 beds, the neighboring hospital's diversion on the hospital of interest. We P = 0.0081). Although public hospitals in our sample are larger than define whether ED visits are extremely high for that day as whether private hospitals, the median inpatient discharges and median ED the daily ED occupancy rate for EDs within the entire LEMSA is visits per year do not differ significantly between public hospitals above the 66th percentile. This variable, along with month, helps ac- and matched hospitals (inpatient discharges: 22 115 vs 16 681 count for external events that may increase demand. (P = 0.2416); ED visits: 50 618 vs 41 747 (P = 0.1073), respectively). Although there are some differences between public and private 3.3.2.3   Hospital-­pair-­specific characteristics hospitals in the matched sample, aside from patient characteristics, We also adjust for two factors specific to the hospital of interest significant difference in the number of annual diversions by public and the original diverting hospital which proxy for the influence of hospitals (median: 475 vs 277, P = 0.2089). most differences are not statistically significant (Table 1). There is no the neighboring hospital's diversion on the hospital of interest: the Public hospitals in our sample tend to treat relatively poorly in- relative driving distance between the two hospitals and the overlap sured, uninsured, and sick populations compared with their private in patient catchment on nondiversion days. Relative distance is the counterparts. Public hospitals treat about 1.5 times as many ED ratio of the driving distance between the hospital pair over the aver- patients (P = 0.0178) and 2.3 times as many inpatients (P = 0.0001) age driving distance of the closest five EDs. The overlap in patient with Medicaid; about 24 times as many ED patients (P < 0.0001) and catchment areas is calculated relying on Brooks and Jones28 method 3.7 times as many inpatients (P < 0.001) with no insurance; 60 per- for a “competitor market presence” (Appendix S2). cent as many ED patients (P < 0.001) and 65 percent as many inpatients (P = 0.0003) with Medicare; and 2.2 times as many ED patients 3.3.2.4   Hospital-­specific characteristics (P = 0.0034) and 2.4 times as many inpatients (P = 0.0055) with dual We control for hospital teaching status and hospital bed size. less likely to treat female patients in the ED (P = 0.0017) and as inpa- Medicare-­Medicaid eligibility. Public hospitals in our sample also are tients (P = 0.0023), but more likely to treat Native American/Alaska 3.4   Sensitivity analyses Native patients in the ED (P = 0.0160) and as inpatients (P = 0.0083). We find evidence of sequential diversions, with unadjusted anal- In sensitivity analyses, we examine alternate model specifica- yses finding that an average of 31.9 percent hospitals in our sample tions (Appendix S3), including using actual ED census rather than declaring a diversion after a neighboring hospital declares one (but       5 HSUAN et al. Health Services Research TA B L E   1   Hospital characteristics for public hospitals and matched private hospitals Public hospitals Matched private hospitals P-­value Ownership Public Nonprofit For-­profit Teaching (%) 16 (100.0%) 0 (0.0%) 0 (0.0%) 26 (92.9%) 0 (0.0%) 2 (7.1%) 10 (62.5%) 3 (10.7%) <0.001 <0.001 Median (IQR) ED visits per year 50 618 (28 323) 41 747 (19 446) 0.1073 Median (IQR) inpatient discharges per year 22 115 (14 398) 16 681 (10 310) 0.2416 Mean (SD) bed size 490 (187) 343 (159) 0.0081 Median (IQR) diversions 475 (1285) 277 (955) 0.2089 49.5 (5.0) 53.8 (3.4) 0.0017 White 38.9 (23.0) 53.9 (23.1) 0.0777 Black/African American 19.0 (15.6) 14.6 (13.4) 0.3980 6.6 (4.6) 6.3 (3.4) 0.8000 0.9 (1.3) 0.2 (0.2) 0.0160 30.4 (20.0) 19.9 (18.6) 0.1291 Characteristics of ED patients Mean (SD) % of ED patients who are female Mean (SD) % ED patients with a race of: Asian/Pacific Islander Native American/Alaska Native Other Missing 4.1 (10.6) 5.2 (15.4) 0.8136 34.1 (24.9) 25.0 (17.4) 0.1614 Medicaid 24.1 (11.0) 15.9 (10.5) 0.0178 No insurance 33.7 (20.4) 1.4 (5.9) <0.0001 Medicare 24.7 (12.1) 41.0 (9.1) <0.0001 1.1 (0.8) 0.5 (0.4) 0.0034 53.1 (6.4) 58.8 (5.0) 0.0023 White 58.2 (24.8) 64.8 (16.7) 0.3263 Black/African American 15.3 (14.4) 11.6 (9.7) 0.3450 Mean (SD) % of ED patients who were Hispanic Mean (SD) % of ED patients with: Dual-­eligible Characteristics of inpatients Mean (SD) % of inpatients who are female Mean (SD) % inpatients with a race of: Asian/Pacific Islander 7.5 (4.3) 9.9 (4.8) 0.1180 Native American/Alaska Native 0.9 (0.01) 0.3 (0.4) 0.0083 17.1 (15.5) 11.3 (10.6) 0.1714 1.0 (1.0) 2.1 (2.4) 0.1254 38.2 (22.7) 26.8 (15.2) 0.0520 Other Missing Mean (SD) % of inpatients who were Hispanic Mean (SD) % of inpatients with: Medicaid 33.8 (15.2) 14.8 (14.0) 0.0001 No insurance 14.8 (11.5) 4.0 (2.4) <0.0001 Medicare 23.0 (11.3) 35.4 (9.3) 0.0003 1.1 (1.1) 0.0055 Dual-­eligible 2.6 (2.2) Notes: The matched hospitals are private hospitals that are matched on driving distance, ED volume, and bed size. Descriptive statistics used chi-­square for categorical variables and t test and Wilcoxon-­Mann-­Whitney test for continuous variables. In the table above describing the race of hospital patients, we exclude seven hospitals from the ED rows and four hospitals from the inpatient rows that describe more than 50% of their patients’ races as “other” or more than 5% of their patients’ race are missing. “No insurance” includes patients whose expected payor is county indigent programs, other indigent, or self-­pay. Authors’ analysis of data from ambulance diversions logs and emergency department and inpatient discharge data from the State of California Office of Statewide Health Planning and Development data, 2007. 6       HSUAN et al. Health Services Research Whether neighboring hospital is a public hospital Coefficient 95% Confidence interval P-­value 1.15 [0.84, 1.46] <0.001 ED occupancy, log 0.21 [0.15, 0.26] <0.001 ED occupancy, log x whether neighboring hospital is a public hospital −0.25 [−0.31, −0.19] <0.001 N 38 371 TA B L E   2   Linear probability model regression with hospital random effects for whether a hospital declares a diversion Notes: A positive coefficient indicates that the variable is associated with an increased probability of declaring a diversion. Possible endogeneity for daily ED occupancy is addressed with instrumental variables, where the instrument is daily inpatient occupancy. Models control for teaching status, bed size, ratio of patients to emergency physicians with privileges, relative distance between hospital and neighboring hospital, overlap in patient catchment areas using the competitor market presence (Appendix S2), duration of the neighboring hospital's diversion, and whether the local EMS agency region experienced an unusually high ED volume, month, hour, whether the diversion is on a weekend. The model additionally controls for whether the neighboring hospital is a public hospital, interacted with ED occupancy, the hour, whether the diversion is on a weekend, and the ratio of patients to emergency physicians with privileges. Authors’ analysis of data from ambulance diversions logs and emergency department and inpatient discharge data from the State of California Office of Statewide Health Planning and Development data, 2007. Coefficient 95% Confidence interval P-­value Whether neighboring hospital is a public hospital −2.58 [−3.28, −1.89] <0.001 ED occupancy, log −5.22 [−8.30, −2.14] 0.001 N 11 641 TA B L E   3   Linear regression with hospital random effects for duration of diversion when a neighboring hospital is already on diversion Notes: A negative coefficient indicates that the variable is associated with a shorter diversion. Controlling for teaching status, bed size, ratio of patients to emergency physicians with privileges, relative distance between hospital and neighboring hospital, overlap in patient catchment areas using the competitor market presence (Appendix S2), duration of the neighboring hospital's diversion, whether the local EMS agency region experienced an unusually high ED volume, month, hour, and whether the diversion is on a weekend. The sample consists of hospitals that declare a diversion following a diversion by a neighboring hospital, and excludes diversions that last longer than 171 min. Authors’ analysis of data from ambulance diversions logs and emergency department and inpatient discharge data from the State of California Office of Statewide Health Planning and Development data, 2007. before the neighboring hospital ends its diversion; Appendix S6). In In adjusted analyses, sample hospitals are on average 1.2 per- adjusted analyses, consistent with crowding being a cause of diver- centage points (P < 0.001) more likely to declare a diversion if the sion, sequentially declaring a diversion in adjusted analyses is pos- neighboring diverting hospital is public rather than private, which itively associated with a hospital's ED occupancy (log-­transformed corresponds to a 3.6 percent increase in the probability of diversion. ED occupancy: 0.21, P < 0.001; Table 2). In unadjusted analyses, Furthermore, sample hospitals that divert following the diversion of which does not consider ED occupancy, diversion-­specific charac- a neighboring public hospital are more likely to have fewer patients teristics, hospital-­ pair-­ specific characteristics, or hospital charac- in the ED than those that divert following the diversion of a neigh- teristics, sample hospitals are more likely to declare a diversion if boring private hospital (log-­transformed ED occupancy x whether the neighboring hospital was private (34.4 percent, N = 5729) than the neighboring hospital is public, −0.25, P < 0.001). In other words, public (30 percent, N = 6516; P < 0.001; Appendix S6). However, this ED occupancy matters less when hospitals declare a diversion fol- relationship is opposite in adjusted analyses. This sign change be- lowing the diversion of a neighboring public, vs private, hospital. tween the unadjusted and the adjusted results occurs after adjusting In our sample, hospitals are on diversions for a mean of for the number of patients in the ED interacted with whether the 61.87 minutes (unadjusted; not shown). In adjusted analyses, when a neighboring hospital on diversion is public vs private (ie neighbor neighboring hospital is already on diversion, hospitals’ diversions are ownership x ED crowding from Equation 1). an average of 2.58 minutes shorter when the neighboring hospital is TA B L E   4   Linear regression with hospital random effects for time elapsed from when a neighboring hospital ended its diversion and hospital of interest ended its own diversion       7 HSUAN et al. Health Services Research Coefficient 95% Confidence interval P-­value Whether neighboring hospital is a public hospital −1.28 [−2.38, −0.19] 0.022 ED occupancy, log −1.85 [−6.64, 2.93] 0.448 N 11 163 Notes: A negative coefficient indicates that the variable is associated with a shorter time that elapses between when the neighboring hospital ends its diversion and the observation hospital ends its own diversion. Controlling for teaching status, bed size, ratio of patients to emergency physicians with privileges, relative distance between hospital and neighboring hospital, overlap in patient catchment areas using the competitor market presence (Appendix S2), duration of the neighboring hospital's diversion, whether the local EMS agency region experienced an unusually high ED volume, month, hour, and whether the diversion is on a weekend. The sample consists of hospitals that declare a diversion following a diversion by a neighboring hospital where the neighboring hospital ends its diversion first, and excludes diversions where the time elapsed between the neighboring hospital ending its diversion and the hospital of interest ending its own diversion is greater than 119 min. Authors’ analysis of data from ambulance diversions logs and emergency department and inpatient discharge data from the State of California Office of Statewide Health Planning and Development data, 2007. a public hospital, compared to a nonpublic hospital (P < 0.001), or an occupancy levels reinforces the evidence of strategic diversion. The average decrease of 4.2 percent (Table 3). probability of diversion is more significant to patients than the re- Finally, when a hospital goes on diversion following a neighboring duction in time to reopening; of the 6516 diversions in our sample hospital's diversion and the neighboring hospital ends its diversion that occurred when the neighboring diverting hospital was public, first, the second hospital ends its diversion a mean of 31.1 minutes 263 diversions might not have occurred had the neighboring divert- after the neighboring hospital ends its own diversion (unadjusted; ing hospital been a private hospital instead.† not shown). In adjusted analyses, hospitals end their diversions an There may be other explanations for these findings. One is that average of 1.3 minutes sooner when the neighboring hospital is a public hospitals are uniquely seen as “bellwethers” in some markets, public hospital, vs a private hospital 22 (ie, P = 0.022), corresponding such that hospitals are more likely to respond to their diversion ac- to a 4.2 percent decrease in time of diversion (Table 4). tivity as compared to diversions by other hospitals. However, it is Alternative model specifications (ie without instrumental vari- difficult to understand why sample hospitals might selectively treat ables; using clustered standard errors; and including outliers) have public hospitals this way, but not private hospitals that are matched similar results (Appendix S3). Using slightly different data, we find by size and distance. that the duration between the second and the third hospital to A second explanation is that hospitals in our sample declare diver- declare a diversion is shorter when the first hospital to declare a sions because they are being sent patients diverted from public hos- diversion is public vs private (P = 0.040; Appendix S4). There is no pitals, and these patients are of higher acuity. We think this unlikely significant difference in severity of illness (Charlson index) in ob- because we do not see any significant differences in severity of illness servation hospitals that divert after a neighboring public, vs private, in ED patients depending on whether neighboring hospitals are public hospital diverts (Appendix S5). vs private (Appendix S5). Furthermore, if patient acuity were affecting results, we would expect to see that sample hospitals’ diversions 5   D I S CU S S I O N would be longer when a neighboring hospital is public, vs private, so that the hospital would have time to treat the diverted patients with higher acuity. On the other hand, if hospitals were driven by strategic This study examines whether hospitals respond differently to di- diversion, we hypothesized that hospitals would want to be on diver- versions by neighboring public, vs private, hospitals, comparing di- sion for less time because they risk losing paying patients. Consistent version probability and timing when neighboring public vs private with our hypothesis, our results show that sample hospitals’ diversions hospitals of similar size and distance divert. These results provide are shorter when the neighboring hospital on diversion is public vs evidence that hospitals in our sample are strategically declaring and private. Thus, it seems unlikely that patient acuity drives our results. ending diversions to avoid patients that would otherwise have been A third explanation is that there is an unobserved cause for taken by ambulances to neighboring public hospitals. Given the sub- sample hospitals going on diversion when a neighboring public, vs stantial number of factors that can trigger a diversion, the magnitude private, hospital goes on diversion. We address this with a robust of the estimated effects (3.6 percent increase in likelihood of diver- set of controls drawn from the literature, including notably those sion, 4.2 percent shorter duration, 4.2 percent reduction in time to theorized in the Asplin et al's29 input-­throughput-­output model, but reopening) is material and the fact that diversions occur at lower ED as with any regression analysis, there may well be omitted variables 8       HSUAN et al. Health Services Research that explain the results. Our data do not permit us to measure the Building on prior research regarding defensive diverting, this number of patients, degree of crowding, physician and nurse staff- study finds that hospitals in our sample are more likely to defen- ing patterns, and number of ED boarders at the moment a hospital sively divert when the neighboring hospital is a public hospital, sug- declares a diversion. However, it is unlikely that these factors dis- gesting these hospitals may use diversions as a way to avoid treating proportionately dictate the observation hospital's decision to divert Medicaid and uninsured patients. Strategic diversions of this kind when the neighboring hospital that diverted was public vs private, delay access to emergency care for particularly vulnerable popu- particularly because patient acuity does not vary depending on lations—delays that may increase mortality.5 Strategic diversions whether the neighboring diverting hospital is public or private, as are particularly concerning given previous research suggesting described above. In other words, although these variables may be that minority patients are especially affected by diversions.8,23,38 important to understanding ED crowding, their omission likely does Vulnerable populations may be even further disadvantaged if hos- not bias our results. For instance, it is unlikely that the number of pitals strategically divert to avoid them. Thus, strategic diversions patients boarding at one hospital systematically changes depending undermine the goal of EMTALA to ensure access to emergency care on whether the neighboring diverting hospital is public vs private. for all patients, regardless of ability to pay. More research is needed This reasoning is supported by the literature. For instance, although to examine whether hospitals outside of this sample also engage in ED boarding plays a large role in ED capacity, and is an important strategic diversions. Unfortunately, this type of data is generally un- predictor of increased ambulance diversions, 2,30-32 studies of the available. Consistent with the goals of the Foundations for Evidence-­ predictors of ambulance boarding have focused on within-­hospital Based Policymaking Act of 2018,39 federal policy makers may wish differences as potential causes of boarding,33-35 rather than the to buttress EMTALA by collecting data on diversions. identity of neighboring hospitals that are diverting. If further research suggests that strategic diversions broadly Our estimates of strategic diversions may be conservative, since occur, what can be done? Federal policy makers may wish to audit we do not engage in further identification to select which hospitals reasons for diversions or amend the list of EMTALA violations to in- may be more likely to game. In addition, all but two of the private clude strategic diversions. It is also possible for state and local policy hospitals in this study are nonprofit hospitals; for-­profit hospitals makers to address strategic diversions. There are several examples may be more likely to strategically divert. of existing policies that might be adopted. Outright bans on ambu- This study is subject to some limitations beyond those noted lance diversions would decrease strategic diversions, but do not ad- above. We used 2007 data because this year was the last that Los dress the underlying reasons for capacity-­based diversions, that is Angeles County, which accounts for 22 percent of all California hos- ED crowding. Although studies on a ban on ambulance diversions pitals, provided detailed diversion log data. Despite the age of the adopted by Massachusetts in 2009 did not find negative conse- data, this study is still relevant given the continued importance of quences,40,41 these results may not be applicable to all jurisdictions if diversions in California. While some jurisdictions have moved to no-­ it were extended. A less sweeping approach would be to limit or pro- diversion policies since 2007, including Massachusetts, California mote standards governing the declaration of diversions. Hospitals has not. In fact, diversions continue to be extremely important in vary a great deal in terms of when they declare diversions, who California—in 2015, almost half (45 percent) of California hospitals is responsible for declaring them (ie physicians, nursing staff, and with ED visits declared ambulance diversions. While this is a signifi- hospital administrators), and the standards they use to make the de- cant decrease since 2007 (from 63 percent, P < 0.001), the large per- termination.11,42 Implementing a standardized procedure,43,44 such centage of hospitals declaring diversions suggests that diversions as using measures such as the Emergency Department Work Index remain an important issue to California hospitals. Furthermore, the (EDWIN)30 or the National Emergency Department Overcrowding mean number of diversion hours by hospitals that declare diversions Scale (NEDOC),45 might help reduce strategic diversions, although has remained high and has not changed significantly since 2007 standardizing procedures could be difficult to implement.9,31,46 * (2007: 688 vs 2015: 683 hours, P = 0.96). Finally, diversions will Alternatively, local EMS agencies might help the standardization likely continue to be an issue both in California and nationally, given process by requiring more information from hospitals before they research that suggests that ED use and crowding did not decrease declare a diversion. EMS agencies in some jurisdictions already do after Medicaid expansion36 and might have increased.37 this; for instance, Alameda County, California, requires that hospitals Our findings are based on data from California and in metropolitan or urban areas, which may be different from other markets. self-­report patient census, bed availability, number of patients in the ED waiting room, and number of boarded patients.47 Finally, our study design relies on matching nonsafety net hospitals by size and distance to safety net hospitals. This helps improve the comparability of diversions at a hospital of interest. However, this 6   CO N C LU S I O N design also limits the size of our sample. More study needs to be done to see if hospitals of varying size and in different states engage Our data show that hospitals in our sample may respond differently in similar behavior. In addition, further study needs to be done to to diversions of neighboring public, vs private, hospitals. Previous qualitatively assess what hospital and EMS agency policies (formal research suggests that minorities and low-­ income are particu- and informal) may be more likely to result in strategic diversions. larly adversely affected by diversions, with higher mortality than Health Services Research nonminorities and the higher-­income.8,23,38 This study suggests that not only might this population be adversely affected directly by diversions of public hospitals, but that they may also be more affected by sequential diversions, because private hospitals may be more likely to declare diversions after the neighboring public (vs private) hospital declares diversion. Furthermore, the results are consistent with our hypothesis that the difference in sequential diversion may be occurring because hospitals wish to avoid serving more uninsured and Medicaid patients. Given the potential impact of the observed differences on health outcomes, there may be opportunity for interventions at the local, state, and federal levels to improve the current structures influencing hospital diversion. AC K N OW L E D G M E N T S Joint Acknowledgment/Disclosure Statement: This study was supported by fellowships to Hsuan from the Agency for Healthcare Research and Quality R36 Grant (R36HS02424701), the NIH/ National Center for Advancing Translational Sciences (NCATS) UCLA CTSI Grant Number TL1TR000121, and a Dissertation Year Fellowship from the University of California, Los Angeles. None of the sponsors were involved in the study design, in the collection, analysis, and interpretation of the data, in the writing of the report, or in the decision to submit the article for publication. The content in this paper does not necessarily represent the official views of the Agency for Healthcare Research and Quality, the National Institutes of Health, or UCLA. The authors thank the National Bureau of Economic Research for providing data. Dr. Hsuan thanks the Penn State Department of Health Policy and Administration. Dr. Horwitz thanks the UCLA School of Law and University of Victoria Department of Economics. Dr. Ponce thanks the UCLA Center for Health Policy Research. E N D N OT E S * Authors’ calculations from the Office of Statewide Health Planning and Development, “Emergency Department Services—Ambulance Diversion Trend,” available at: https://data.chhs.ca.gov/dataset/emergencydepartment-services-ambulance-diversion-trend † In our sample, there are 21 712 observations where the neighboring diverting hospital is a public hospital (not shown). When the neighboring diverting hospital is a public hospital, the hospital of interest declares a diversion 30% of the time (equivalent to 6516 diversions; Appendix Exhibit S6). The adjusted results suggest that 1.2 percentage points of this 30% may be attributed to the fact if the neighboring diverting hospital was a public, vs private, hospital. This corresponds to 263 diversions (6516-­21 712 × (30-­1.2)/100 = 263). ORCID Charleen Hsuan  Renee Y. Hsia  Ninez A. Ponce  Jack Needleman        9 HSUAN et al. https://orcid.org/0000-0001-8720-2323 https://orcid.org/0000-0001-9819-6926 https://orcid.org/0000-0001-5151-6718 https://orcid.org/0000-0002-2875-0589 REFERENCES 1. Burt CW, McCaig LF. Staffing, capacity, and ambulance diversion in emergency departments: United States, 2003-­ 0 4. Adv Data. 2006;376:1‐23. 2. Schull MJ, Lazier K, Vermeulen M, Mawhinney S, Morrison LJ. Emergency department contributors to ambulance diversion: a quantitative analysis. Ann Emerg Med. 2003;41(4):467‐476. 3. Schull MJ, Morrison LJ, Vermeulen M, Redelmeier DA. Emergency department gridlock and out-­of-­hospital delays for cardiac patients. Acad Emerg Med. 2003;10(7):709‐716. 4. Sun BC, Hsia RY, Weiss RE, et al. Effect of emergency department crowding on outcomes of admitted patients. Ann Emerg Med. 2013;61(6):605‐611. 5. Shen YC, Hsia RY. Ambulance diversion associated with reduced access to cardiac technology and increased one-­year mortality. Health Aff (Millwood). 2015;34(8):1273‐1280. 6. Yankovic N, Glied S, Green LV, Grams M. The impact of ambulance diversion on heart attack deaths. Inquiry. 2010;47(1):81‐91. 7. Shen YC, Hsia RY. Association between ambulance diversion and survival among patients with acute myocardial infarction. JAMA. 2011;305(23):2440‐2447. 8. Hsia RY, Sarkar N, Shen YC. Impact of ambulance diversion: black patients with acute myocardial infarction had higher mortality than whites. Health Aff. 2017;36(6):1070‐1077. 9. McCarthy ML, Aronsky D, Jones ID, et al. The emergency department occupancy rate: a simple measure of emergency department crowding? Ann Emerg Med. 2008;51(1):15‐24, 24 e11-e12. 10. United States General Accounting Office. Hospital Emergency Departments: Crowded Conditions Vary Among Hospitals and Communities. Washington, DC: United States General Accounting Office; 2003. 11. Mihal N, Moilanen R. When Emergency Rooms Close: Ambulance Diversion in the West San Fernando Valley. Los Angeles, CA: Ralph and Goldy Lewis Center for Regional Policy Studies, UCLA School of Public Affairs, UC Los Angeles; 2005. 12. Hagtvedt R, Ferguson M, Griffin P, Jones GT, Keskinocak P. Cooperative strategies to reduce ambulance diversion. In: Rossetti MD, Hill RR, Johansson B, Dunkin A, Ingalls RG, eds. Proceedings of the 2009 Winter Simulation Conference. Austin, TX: Curran Associates; 2009. 13. Deo S, Gurvich I. Centralized vs. decentralized ambulance diversion: a network perspective. Manage Sci. 2011;57(7):1300‐1319. 14. Falvo T, Grove L, Stachura R, Zirkin W. The financial impact of ambulance diversions and patient elopements. Acad Emerg Med. 2007;14(1):58‐62. 15. McConnell KJ, Richards CF, Daya M, Weathers CC, Lowe RA. Ambulance diversion and lost hospital revenues. Ann Emerg Med. 2006;48(6):702‐710. 16. Wilson M, Cutler D. Emergency department profits are likely to continue as the affordable care act expands coverage. Health Aff. 2014;33(5):792‐799. 17. Hsia RY, MacIsaac D, Baker LC. Decreasing reimbursements for outpatient emergency department visits across payer groups from 1996 to 2004. Ann Emerg Med. 2008;51(3):265‐274, 274.e261-e265. 18. Scott JW, Havens JM, Wolf LL, et al. Insurance status is associated with complex presentation among emergency general surgery patients. Surgery. 2017;161(2):320‐328. 19. Studdert DM, Thomas EJ, Burstin HR, Zbar BI, Orav EJ, Brennan TA. Negligent care and malpractice claiming behavior in Utah and Colorado. Med Care. 2000;38(3):250‐260. 20. Deane L. The cost of lawsuit risks when treating the uninsured. Health Aff. 2010;29(11):2127; author reply 2128. 21. Burditt v. US Dept. of Health and Human Services, 934 F.2d 1362 (5th Cir. 1991). 10       HSUAN et al. Health Services Research 22. The Emergency Medical Treatment and Labor Act, 42 USC §1395dd et seq. 23. Hsia RY, Asch SM, Weiss RE, et al. California hospitals serving large minority populations were more likely than others to employ ambulance diversion. Health Aff. 2012;31(8):1767‐1776. 24. Abaris Group. California ED Diversion Project, Final Report. Walnut Creek 2009. 25. Wooldridge JM. Econometric Analysis of Cross Section and Panel Data, 2nd edn. Cambridge, MA: MIT Press; 2002. 26. Burt CW, McCaig LF, Valverde RH. Analysis of ambulance transports and diversions among US emergency departments. Ann Emerg Med. 2006;47(4):317‐326. 27. Sun BC, Mohanty SA, Weiss R, et al. Effects of hospital closures and hospital characteristics on emergency department ambulance diversion, Los Angeles County, 1998 to 2004. Ann Emerg Med. 2006;47(4):309‐316. 28. Brooks GR, Jones VG. Hospital mergers and market overlap. Health Serv Res. 1997;31(6):701‐722. 29. Asplin BR, Magid DJ, Rhodes KV, Solberg LI, Lurie N, Camargo CA Jr. A conceptual model of emergency department crowding. Ann Emerg Med. 2003;42(2):173‐180. 3 0. Bernstein SL, Verghese V, Leung W, Lunney AT, Perez I. Development and validation of a new index to measure emergency department crowding. Acad Emerg Med. 2003;10(9):938‐942. 31. Epstein SK, Tian L. Development of an emergency department work score to predict ambulance diversion. Acad Emerg Med. 2006;13(4):421‐426. 32. Institute of Medicine. Hospital-Based Emergency Care: At the Breaking Point. Washington, DC: Institute of Medicine; 2007. 33. Rabin E, Kocher K, McClelland M, et al. Solutions to emergency department ‘boarding’ and crowding are underused and may need to be legislated. Health Aff. 2012;31(8):1757‐1766. 3 4. Zhou JC, Pan KH, Zhou DY, et al. High hospital occupancy is associated with increased risk for patients boarding in the emergency department. Am J Med. 2012;125(4):416.e411‐416.e417. 35. Shaikh SA, Robinson RD, Cheeti R, et al. Risks predicting prolonged hospital discharge boarding in a regional acute care hospital. BMC Health Serv Res. 2018;18(1):59. 36. Pines JM, Zocchi M, Moghtaderi A, et al. Medicaid expansion in 2014 did not increase emergency department use but did change insurance payer mix. Health Aff. 2016;35(8):1480‐1486. 37. Nikpay S, Freedman S, Levy H, Buchmueller T. Effect of the affordable care act medicaid expansion on emergency department visits: evidence from state-­level emergency department databases. Ann Emerg Med. 2017;70(2):215‐225. e216. 38. Shen YC, Hsia RY. Do patients hospitalised in high-­minority hospitals experience more diversion and poorer outcomes? A retrospective multivariate analysis of Medicare patients in California. BMJ Open. 2016;6(3):e010263. 39. Foundations for Evidence-Based Policymaking Act of 2018, Pub. L. No 115-435, 132 Stat. 5529 (2019). 4 0. Burke LG, Joyce N, Baker WE, et al. The effect of an ambulance diversion ban on emergency department length of stay and ambulance turnaround time. Ann Emerg Med. 2013;61(3):303‐311. 41. Rathlev NK, Blank F, Osborne B, et al. No diversion in Western Massachusetts. J Emerg Med. 2013;44(2):313‐320. 42. Handel DA, Pines J, Aronsky D, et al. Variations in crowding and ambulance diversion in nine emergency departments. Acad Emerg Med. 2011;18(9):941‐946. 43. Hoot N, Aronsky D. An early warning system for overcrowding in the emergency department. AMIA… Annual Symposium proceedings. AMIA Symposium. 2006:339‐343. 4 4. Weiss SJ, Ernst AA, Nick TG. Comparison of the National Emergency Department Overcrowding Scale and the Emergency Department Work Index for quantifying emergency department crowding. Acad Emerg Med. 2006;13(5):513‐518. 45. Weiss SJ, Derlet R, Arndahl J, et al. Estimating the degree of emergency department overcrowding in academic medical centers: results of the National ED Overcrowding Study (NEDOCS). Acad Emerg Med. 2004;11(1):38‐50. 46. Hoot NR, Zhou C, Jones I, Aronsky D. Measuring and forecasting emergency department crowding in real time. Ann Emerg Med. 2007;49(6):747‐755. 47. Alameda County Emergency Medical Services. Ambulance Diversion. Oakland, CA; 2009. S U P P O R T I N G I N FO R M AT I O N Additional supporting information may be found online in the Supporting Information section at the end of the article.   How to cite this article: Hsuan C, Hsia RY, Horwitz JR, Ponce NA, Rice T, Needleman J. Ambulance diversions following public hospital emergency department closures. Health Serv Res. 2019;00:1–10. https://doi.org/10.1111/1475-6773.13147