Abstract
Ambulance diversion, where emergency departments (ED) are temporarily closed to ambulance traffic, is an important system-level interruption that causes delays in treatment and potentially decreased quality of care. There is little empirical evidence investigating the mechanisms through which ambulance diversion might affect patient outcomes, however. We investigated whether ambulance diversion affects access to technology, likelihood of treatment, and ultimately health outcomes for patients with acute myocardial infarction. We found that patients whose nearest hospital experiences significant diversion indeed have reduced access to hospitals with cardiac technology. This leads to a 4.6% decreased likelihood of revascularization and a 9.8% increase in 1-year mortality. Policymakers may consider creating targeted policies to specifically manage certain time-sensitive conditions requiring technological intervention during periods of ambulance diversion.
Introduction
Ambulance diversion occurs when emergency departments (EDs) are temporarily closed to ambulance traffic due to a variety of reasons, such as overcrowding or lack of available resources, (1–7) and effectively creates a temporary decrease in ED access. While a few studies have found that overcrowding and ambulance diversion are associated with poor health outcomes,(8–10), the mechanisms through which diversion affects patients has been less well-studied. Proper identification of these mechanisms is critical as policymakers strive to implement solutions to improve quality of care for all populations as well as those experiencing the poorest outcomes. The potential value of exploring these mechanisms is to determine if exceptions to ambulance diversion for a small but extremely sick-yet-salvageable subset of patients could significantly improve outcomes.
Using 100% of Medicare claims and daily ambulance diversion logs from 26 California counties between 2001 and 2011, we investigated the potential mechanisms through which ambulance diversion leads to poorer patient outcomes. We analyzed changes in access, treatment, and outcomes when patients were exposed to different levels of diversion. Based on the conceptual pathway described below, we performed these analyses to understand the overall (i.e., net) effects, as well as to evaluate the contribution of the intermediary mechanisms.
Methods
Conceptual Model
In this section, we outline the potential mechanisms through which ambulance diversion can affect patient care, net of all the underlying causes. Most of the literature related to ambulance diversion focuses on the main cause of diversion: actions that occur at an individual hospital as a result of crowding. Overcrowding in one hospital can cause delays in receiving treatment. Empirical evidence has documented that overcrowding is associated with delays in administering thrombolytics for cardiac patients,(11) antibiotics for patients with pneumonia,(12) and pain medication for patients in severe pain.(13) More broadly, patients who need ED care during ambulance diversion periods, whether they have to be diverted elsewhere or receive treatment within the overcrowded ED, will likely experience delays to treatment time (Mechanism A in Appendix 1).(7)
Second, diversion could affect patients through routing them to settings that are less equipped technologically to handle complex conditions (Mechanism B in Appendix 1). The decreased access to cardiac technology in turn could decrease the likelihood of patients receiving needed treatment, and can have a direct consequence on their health outcomes.
Third, it is also possible that patients who need advanced cardiac intervention during ambulance diversion periods experience decreased likelihood of receiving treatment even in a hospital with cardiac technology capacity (Mechanism C in Appendix 1), if crowding and limited resources outstrip the capability of the staff to deploy their technology appropriately. Our study therefore explores the potential contribution of these three mechanisms on patient outcomes.
Data
Patient data were extracted from 100% Medicare Provider Analysis and Review (MedPAR), linked with vital files, between 2001 and 2011. We linked them with the Healthcare Provider Cost Reporting Information System and American Hospital Association annual surveys to obtain additional hospital-level information.
In order to identify the closest ED for a patient, we supplemented our hospital data with longitude and latitude coordinates of the hospital’s physical address or heliport (if one existed).(14) We obtained actual driving distance from the patient’s ZIP code centroid to the nearest hospital’s latitude and longitude coordinates based on Google Maps, using automation codes developed in Stata.(15)
To identify a hospital’s daily ambulance diversion hours, we used daily ambulance diversion logs from California local emergency medical services agencies (LEMSAs). Our logs contained data for 17 out of the 23 LEMSAs that did not ban diversion for the years of 2001–2011 (actual coverage dates vary by LEMSA). The 17 LEMSAs together represented 26 out of 58 counties and 88% of California’s population.
Patient Population
We identified the AMI population by extracting records between 2001 and 2011 that had 410.x0 or 410.x1 as the principal diagnosis, as done in previous studies,(8) and indicated residence in counties for which we had diversion data. We excluded all patients who were not admitted through the ED, as well as patients whose admitting hospital was >100 miles away from their mailing ZIP codes, since those patients likely did not reside at their mailing address or were admitted to hospitals while away from home. Last, for analysis of 30-day readmission, we used a smaller sample that excluded patients who were not eligible to be included in the analysis (for example, if a person died during initial admission, they would not be able to be readmitted to hospital) per CMS guidelines.(16)
Defining Access, Treatment, and Health Outcomes
We evaluated three dimensions of patient care: access, treatment, and health outcomes. We defined access as whether a patient was admitted to a hospital with the following technology for AMI (regardless of actual treatment received): cardiac care intensive unit, catheterization lab, and coronary artery bypass graft (CABG) surgery capacity.
We defined treatment received as whether a patient received a given procedure, identified by the ICD-9 procedure codes on the MedPAR. We examined three common treatments for AMI: percutaneous coronary intervention (PCI), thrombolytic therapy, and CABG.
Finally, we analyzed two sets of patient health outcomes: death (whether a patient died within X days from his ED admission, where X=30, 90, and 365 days) and readmission to the hospital within 30 days of the index discharge.
Defining Levels of Ambulance Diversion
We first calculated the total number of hours an ED was on diversion for a given day. Then, we identified the nearest ED on the day a patient suffered an AMI by merging ED diversion data with MedPAR, based on the admission date and provider ID of the nearest ED. Finally, using previously defined categories of diversion, (8) we classified patients into four categories based on the number of diversion hours of their nearest ED on the day of their admission: (1) 0 hours (reference group), (2) <6 hours, (3) 6–12 hours (not including 12 hours), and (4) ≥12 hours.
Statistical Models
We implemented the following multivariate models with the patient as the unit of analysis. In all models, the key variables of interest are the three dichotomous variables identifying the diversion level of the patient’s nearest ED (no diversion = reference group; <6 hours; 6–12 hours; ≥12 hours).
We implemented a linear probability model with fixed effects for each ED that was identified as the closest ED for each patient (this is equivalent to including indicators for each ED in the model) while controlling for time-dependent variables.(17) Using the 30-day mortality outcome as an example, the three diversion variables capture differences in the percent of AMI patients who die within 30 days when their nearest ED is in normal operation (i.e., no diversion) and when the same ED experiences different levels of diversion (i.e., the same ED crosses over to a higher level of diversion). By using each ED as its own matched control, we can eliminate any underlying differences across EDs,(18) such as possible differences in baseline diversion levels, baseline mortality rates, quality of care, case-mix of the patient population, or other unobserved characteristics that might be confounded with the outcomes.(19)
In all models, we controlled for patient demographics, specifically 5-year age groups, gender, minority and other non-white race, as well as 22 comorbid measures based on prior work.(20) We also included year indicators to capture the macro trends and took into account the hospital organizational characteristics of the admitting hospital, such as hospital ownership (for-profit, government), teaching status, size (measured by log transformed total inpatient discharges), occupancy rate, system membership, and Herfindahl index to capture the competitiveness of the hospital market within a 15-mile radius (0 being perfectly competitive and 1 being monopoly).
Model 1 described above captures the net effect of ambulance diversion (without differentiating the three mechanisms described in the conceptual model and Appendix 1) on the dependent variables. For treatment outcomes, we estimated an additional model (Model 2) that controlled for cardiac technology access. Results from these two models allowed us to evaluate the relative contributions of Mechanisms B and C on treatment received.
For mortality outcomes, we estimated an additional model (Model 3) that controls for both cardiac technology and actual treatment received. This final model allows us to control for contribution of Mechanisms B and C on patient outcomes and to attribute residual effects to Mechanism A.
Limitations
Our results should be interpreted in light of several limitations. First, our diversion data is self-reported by LEMSAs, so there is potential for errors and reporting bias. Given that it was directly obtained from the online reporting systems used by the LEMSAs, however, we believe the potential for significant bias is minimal.
Second, our patient data identify date but not time of admission. In addition, some patients can still be brought into an ED under exception even if the ED is already on diversion. While we cannot verify with absolute certainty that a patient was diverted, it is reasonable to assume that there is a negative relationship between the number of hours an ED is on diversion and the probability that a patient is admitted to this ED. In addition, our conceptual framework and hypotheses are based on the assumption that diversion can affect both diverted and non-diverted patients. Nonetheless, it is important to recognize that the inability to clearly identify the diverted and the non-diverted patients in our analysis implies that what we observe is the net effect of ambulance diversion. Because overcrowding is the main cause of ambulance diversion, we cannot separate out the differences in outcome that are purely due to overcrowding or purely due to diverted ambulances. This limitation also introduces measurement errors that may cause attenuation bias in our results, so that the estimated magnitude we observe is likely conservative relative to the true magnitude.
Third, we identify the nearest ED for each patient based on the longitude and latitude information of the patient’s ZIP code center and the hospital’s location. Two patients from the same ZIP code might have very different distances to the same ED. We believe the problem is minimized for our sample because most ZIP codes with adequate numbers of patients where we have available diversion logs are in densely populated Metropolitan Statistical Areas (according to Census data, 95% of CA population reside in urban areas). The imprecision in matching patients to their closest ED introduces measurement errors in our diversion indicators and likely makes our estimated results conservative.
Fourth, there is the possibility that the patient’s AMI did not occur at the patient’s ZIP code. Approximately 80% to 85% of AMIs, however, have been shown to occur at home.(21, 22) In addition, we are not aware of any empirical evidence to suggest that there would be any systematic misclassification in either direction across the diversion levels to bias our results. With the exclusion criteria we imposed in selecting our sample, we believe that this limitation should not affect our analysis.
Finally, our dataset was limited to Medicare beneficiaries in one state, so the results of our study may not be generalizable nationwide to patient populations under the age of 65 years. California, however, is a large and diverse state representing 12% of the U.S. population.
Results
Exhibit 1 illustrates the percent of patients who experienced diversion over time in our entire dataset. We plotted the trend separately for Los Angeles County and the rest of California, because LA has a much higher level of diversion and stopped providing diversion data after 2007. In general, we observe decreasing trends in diversion, which is consistent with county initiatives to decrease diversion. It is important to note that in our multivariate analysis, differences in level of diversion are being controlled for by the fixed effects, and that the regression compares differences in outcomes across diversion levels WITHIN each ZIP code.
EXHIBIT 1.
Percent of patients experiencing diversion in CA counties reporting ambulance diversion: 2001–2011.
Source: Authors’ tabulation of CA daily ambulance diversion data and American Hospital Association Annual Surveys.
Our final sample included 28,683 patients for our main analysis and 22,058 patients for the readmission analysis. Exhibit 2 (first row) shows that 14,628 patients (51%) experienced no diversion at their nearest ED on their day of admission; for 7173 patients (25%), their nearest ED was on diversion for <6 hours; for 4207 patients (15%), their nearest ED was on diversion for 6–12 hours; and finally for 2675 patients (9%), their nearest ED was on diversion for ≥12 hours of diversion. The rest of Exhibit 2 shows patient distribution for each diversion level. A notable difference in patient demographics across the diversion levels is that a larger proportion of non-white patients experienced a high level of diversion: among the patients in the ≥12 hours diversion category, 20% were nonwhite, while only 15% were nonwhite among patients who experienced no diversion (compared to 72% vs. 78% for their white counterparts). We observe this pattern for all non-white patients (black, Hispanic, and other non-white races). At the hospital-level, more patients were admitted to not-for-profit and fewer to teaching hospitals during periods of high diversion. Complete descriptive statistics of all patient characteristics are available in Appendix 2.
Exhibit 2.
Descriptive statistics of selected patient and hospital characteristics, by diversion level.
| Nearest ED was not diverted on the day of admission |
Nearest ED's exposure to diversion on the day of admission |
|||
|---|---|---|---|---|
| <6 hours | [6–12) hours | ≥12 hours | ||
| Number of patients in each diversion category | 14628 | 7173 | 4207 | 2675 |
| % or Mean | % or Mean | % or Mean | % or Mean | |
| Patient demographics | ||||
| Male | 51% | 50% | 50% | 48% |
| Female | 49% | 50% | 50% | 52%* |
| White | 78% | 73%* | 74%* | 72%* |
| Non-white | 15% | 17%* | 17%* | 20%* |
| Black | 5% | 5% | 6% | 7%* |
| Hispanic | 5% | 6% | 5% | 7%* |
| Other non-white race | 4% | 6%* | 5%* | 6%* |
| Age distribution | ||||
| 65–69 | 15% | 13%* | 13%* | 15% |
| 70–74 | 16% | 16% | 15% | 16% |
| 75–79 | 18% | 18% | 19% | 20% |
| 80–84 | 20% | 21% | 21% | 21% |
| 85+ | 30% | 31% | 31% | 29% |
| Other admission hospital characteristics | ||||
| Not-for-profit | 69% | 68% | 69% | 72%* |
| For-profit | 19% | 21%* | 21%* | 20% |
| Government | 13% | 11%* | 10%* | 8%* |
| Teaching hospital | 11% | 11% | 12%* | 8%* |
| Member of a system | 74% | 68%* | 68%* | 66%* |
| Admitted to closest ED | 51% | 48%* | 46%* | 45%* |
| Mean total beds in hospital | 268 | 280* | 295* | 286* |
| Mean occupancy rate | 0.68 | 0.68 | 0.70* | 0.70* |
| Mean HHI index | 0.19 | 0.12* | 0.09* | 0.10* |
Note:
Group mean is statistically different from the reference group (no diversion) at p<0.05. Complete descriptive statistics of other patient and hospital characteristics are in Appendix 2.
Exhibit 3 presents results based on Model 1. We found that patients exposed to the highest level of diversion (≥12 hours) have worse access to cardiac technology——by −2.57 percentage points for access to cardiac care intensive unit (p<0.05) compared with patients who were admitted on a day with no diversion, by −2.67 percentage points for access to catheterization lab (p<0.01), and by −2.30 percentage points for access to CABG facilities (p<0.01). Without controlling for hospitals’ cardiac technology, we observe decreased likelihood of receiving catheterization/PCI by −2.37 percentage points (p<0.05) and increased 90-day and 1-year mortality (by 1.83 and 2.83 percentage points, respectively, p<0.05). To put the coefficient’s magnitude in perspective, the baseline rate for receiving cath/PCI is 51%. A decrease of 2.37 percentage points represents a 4.6% reduction. Likewise, the baseline 1-year mortality rate is 29%. A 2.83 percentage point increase from this baseline rate represents a 9.8% increase in mortality.
Exhibit 3.
Regression-adjusted percentage point changes in outcomes relative to patients with no exposure to ambulance diversion, based on Model 1 specifications.
| Access (admitting hospital has:) |
Treatment | Outcomes | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
|
cardiac care unit |
cath lab | CABG | cath/PCI |
thrombolytic therapy |
CABG |
30-day mortality |
90-day mortality |
1-year mortality |
30-day readmission |
|
| Base rate (among patients in reference group) | (60%) | (73%) | (65%) | (51%) | (1%) | (6%) | (16%) | (22%) | (29%) | (34%) |
| Diversion status (Reference group: nearest ED not on diversion on the day of admission) | ||||||||||
| Nearest ED's exposure to diversion on the day of admission | ||||||||||
| <6 hours | −1.40* | −1.60** | −0.89 | −0.98 | 0.11 | −0.43 | −0.13 | 0.39 | 0.10 | 0.31 |
| [6–12) hours | −0.51 | −0.88 | −0.30 | −1.37 | −0.38 | −0.36 | 0.44 | 0.68 | 0.51 | 0.56 |
| ≥12 hours | −2.57* | −2.67** | −2.30** | −2.37* | −0.25 | −0.36 | 1.16 | 1.83* | 2.83** | 1.94 |
| Control for technology access? | N/A | N/A | N/A | No | No | No | No | No | No | No |
| Control for treatment received? | N/A | N/A | N/A | N/A | N/A | N/A | No | No | No | No |
| N | 28683 | 28683 | 28683 | 28683 | 28683 | 28683 | 28683 | 28683 | 28683 | 22058 |
Nearest ED Based on Google query of driving distance
Coefficients below represent changes in percentage point from the base rate.
Significant at +p<0.10 *p<0.05 **p<0.01.
Based on linear probability model with nearby-ED fixed effects. Other control variables include year indicators, patient demographics (5-year age groups, gender, black, Hispanic, other non-white race), 22 comorbid measures, hospital ownership (for-profit, government), teaching status, size, occupancy rate, system membership, and Herfindahl index.
Complete regression results are included in Appendix 3.
The middle panel in Exhibit 4 shows the regression results based on Model 2, which investigates the relationship between diversion and probability of receiving cardiac treatments when patients have comparable access to cardiac technology. We found no difference in the probability of receiving catheterization/PCI, thrombolytic therapy, or CABG once we control for cardiac technology availability. This result suggests that decreased likelihood of receiving catheterization/PCI is driven by the lack of physical access and not by other resource constraints.
Exhibit 4.
Regression-adjusted percentage point changes in outcomes relative to patients with no exposure to ambulance diversion, based on Models 2 and 3 specifications.
| Model 2 Treatment |
Model 3 Outcomes |
||||||
|---|---|---|---|---|---|---|---|
| cath/PCI |
thrombolytic therapy |
CAB G |
30-day mortality |
90-day mortality |
1-year mortality |
30-day readmission |
|
| Base rate (among patients in reference group) | (51%) | (1%) | (6%) | (16%) | (22%) | (29%) | (34%) |
| Diversion status (Reference group: nearest ED not on diversion on the day of admission) | |||||||
| Nearest ED's exposure to diversion on the day of admission | |||||||
| <6 hours | −0.64 | 0.09 | −0.39 | −0.24 | 0.26 | −0.08 | 0.36 |
| [6–12) hours | −1.23 | −0.38 | −0.36 | 0.27 | 0.45 | 0.20 | 0.57 |
| ≥12 hours | −1.60 | −0.29 | −0.23 | 0.90 | 1.50+ | 2.38* | 1.58 |
| Control for technology access? | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
| Control for treatment received? | N/A | N/A | N/A | Yes | Yes | Yes | Yes |
| N | 28683 | 28683 | 28683 | 28683 | 28683 | 28683 | 22058 |
Nearest ED Based on Google query of driving distance
Coefficients below represent changes in percentage point from the base rate.
Significant at +p<0.10 *p<0.05 **p<0.01.
Based on linear probability model with nearby-ED fixed effects. Other control variables include year indicators, patient demographics (5-year age groups, gender, black, Hispanic, other non-white race), 22 comorbid measures, hospital ownership (for-profit, government), teaching status, size, occupancy rate, system membership, and Herfindahl index.
Complete regression results are included in Appendix 4.
The results for health outcomes from Model 3 are shown in the third panel, where the patient outcomes are shown after hospital capability and treatment received are controlled for in the regression. We still observe that high levels of diversion are associated with an increased 1-year mortality rate by 2.38 percentage points (p<0.01), representing an 8.2% relative increase in mortality, when we compare patients across the diversion categories with comparable technology access and treatment patterns. We do not observe significant differences in other health outcomes.
Discussion
Our study provides a unique look at the mechanisms that explain the effect of ambulance diversion on outcomes. Our multivariate results indicate that patients are more likely to get admitted to hospitals with worse access to cardiac care technology when their nearest ED is on diversion. The lack of physical access to the technology is associated with a 4.6% relative reduction in the likelihood of catheterization/PCI, and a 9.8% increase in 1-year mortality.
At the same time, we also found that while ambulance diversion was not independently associated with decreased likelihood of receiving treatment when we controlled for a hospital’s available technology, patients admitted during periods of high diversion still experienced a statistically and clinically significant increase (8.2% relative increase) in long-term mortality. To place the magnitude of this effect size in context, very few pharmacological or clinical interventions have been proven to be associated with long-term outcomes such as mortality to such an extent.
Taken together, our results suggest that by directing patients to technologically appropriate facilities, we can potentially improve their probability of receiving necessary interventions even if the ED faces other resource constraints. But we also recognize that the improved access does not completely obliterate the disparity in health outcomes across diversion levels. The residual increase in 1-year mortality when we compared patients with comparable technology access and likelihood of receiving treatment, therefore, most likely points to the effects of delays and decreased quality of care as the result of overcrowding, as shown by other literature.
We recognize that the root cause of ambulance diversion is ED overcrowding, and strategies aiming to reduce overcrowding will also reduce the negative outcomes associated with ambulance diversion. Our findings suggest that, in light of the existing overcrowding that the ED system currently faces, it may be beneficial for local health systems to create policies to recognize that patients suffering from certain conditions such as acute myocardial infarction, which has been shown to benefit from revascularization, should still be guaranteed access to hospitals that are capable of delivering necessary interventions. This might be applied to other time-sensitive conditions that benefit from technological intervention, such as acute stroke or trauma. Such policies have been implemented for patients with severe trauma for this very reason, so that even when a hospital that is a trauma center is technically on diversion, exceptions are made for trauma patients.
Not all emergency conditions, however, require technological interventions (for example, patients with stable coronary disease(23)), and these patients may not suffer inferior outcomes due to barriers to technology. We also recognize that this policy would not completely eliminate the disparities in health outcomes among cardiac patients experiencing different levels of diversion, since delay in treatment is still an important factor influencing patient outcomes. It is, however, one potential way to reduce disparities under resource constraints. These policies should be carefully crafted based on evidence demonstrating benefit from technologically appropriate interventions. Ultimately, the most effective policies would be ones that resolve the underlying problems of resource constraints that contribute to ambulance diversion.
Conclusion
Our study shows that patients whose nearest hospital experiences significant diversion have poorer access to hospitals with cardiac technologies, which leads to a lower likelihood of receiving treatment with revascularization, and increased mortality. Policies that allow exceptions to a hospital’s ambulance diversion for cardiac conditions that have been shown to benefit from technological intervention may be one way to improve outcomes when resources are limited.
Supplementary Material
Acknowledgments
This research was supported by the National Heart, Lung, and Blood Institute, National Institutes of Health, Grant No. 1R01HL114822. The authors thank Jean Roth from the National Bureau of Economic Research (NBER) for assisting with obtaining and extracting the patient data, Nandita Sarkar from NBER for excellent programming support, and Julia Brownell and Sarah Sabbagh from the University of California, San Francisco, for technical assistance. None received additional compensation other than university salary for their contributions.
References
- 1.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. doi: 10.1111/j.1553-2712.2003.tb00647.x. [DOI] [PubMed] [Google Scholar]
- 2.Epstein SK, Tian L. Development of an emergency department work score to predict ambulance diversion. Acad Emerg Med. 2006;13(4):421–426. doi: 10.1197/j.aem.2005.11.081. [DOI] [PubMed] [Google Scholar]
- 3.Schneider SM, Gallery ME, Schafermeyer R, Zwemer FL. Emergency department crowding: a point in time. Ann Emerg Med. 2003;42(2):167–172. doi: 10.1067/mem.2003.258. [DOI] [PubMed] [Google Scholar]
- 4.Schull MJ, Slaughter PM, Redelmeier DA. Urban emergency department overcrowding: defining the problem and eliminating misconceptions. Cjem. 2002;4(2):76–83. doi: 10.1017/s1481803500006163. [DOI] [PubMed] [Google Scholar]
- 5.Yancer DA, Foshee D, Cole H, Beauchamp R, de la Pena W, Keefe T, et al. Managing capacity to reduce emergency department overcrowding and ambulance diversions. Jt Comm J Qual Patient Saf. 2006;32(5):239–245. doi: 10.1016/s1553-7250(06)32031-4. [DOI] [PubMed] [Google Scholar]
- 6.Pham JC, Patel R, Millin MG, Kirsch TD, Chanmugam A. The effects of ambulance diversion: a comprehensive review. Acad Emerg Med. 2006;13(11):1220–1227. doi: 10.1197/j.aem.2006.05.024. [DOI] [PubMed] [Google Scholar]
- 7.DeLia D, Cantor J. Research Synthesis Report. Princeton, NJ: The Robert Wood Johnson Foundation; 2009. Emergency department utilization and capacity. [PubMed] [Google Scholar]
- 8.Shen Y, Hsia RY. Association Between Ambulance Diversion and Survival Among Patients With Acute Myocardial Infarction. JAMA. 2011;305(23):2440–2447. doi: 10.1001/jama.2011.811. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Yankovic N, Glied S, Green LV, Grams M. The impact of ambulance diversion on heart attack deaths. Inquiry. 2010;47(1):81–91. doi: 10.5034/inquiryjrnl_47.01.81. [DOI] [PubMed] [Google Scholar]
- 10.Sun BC, Hsia RY, Weiss RE, Zingmond D, Liang LJ, Han W, et al. Effect of emergency department crowding on outcomes of admitted patients. Ann Emerg Med. 2013;61(6):605–611. e6. doi: 10.1016/j.annemergmed.2012.10.026. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Schull MJ, Vermeulen M, Slaughter G, Morrison L, Daly P. Emergency department crowding and thrombolysis delays in acute myocardial infarction. Ann Emerg Med. 2004;44(6):577–585. doi: 10.1016/j.annemergmed.2004.05.004. [DOI] [PubMed] [Google Scholar]
- 12.Pines JM, Localio AR, Hollander JE, Baxt WG, Lee H, Phillips C, et al. The impact of emergency department crowding measures on time to antibiotics for patients with community-acquired pneumonia. Ann Emerg Med. 2007;50(5):510–516. doi: 10.1016/j.annemergmed.2007.07.021. [DOI] [PubMed] [Google Scholar]
- 13.Pines JM, Hollander JE. Emergency department crowding is associated with poor care for patients with severe pain. Ann Emerg Med. 2008;51(1):1–5. doi: 10.1016/j.annemergmed.2007.07.008. [DOI] [PubMed] [Google Scholar]
- 14.Horwitz JR, Nichols A. Hospital ownership and medical services: market mix, spillover effects, and nonprofit objectives. J Health Econ. 2009;28(5):924–937. doi: 10.1016/j.jhealeco.2009.06.008. [DOI] [PubMed] [Google Scholar]
- 15.Ozimek A, Miles D. Stata utilities for geocoding and generating travel time and travel distance information. Stata Journal. 2011;11(1):106–119. [Google Scholar]
- 16.Center for Medicare and Medicaid Services. 2014 Measures Updates and Specifications Report Hospital-Level 30-Day Risk-Standardized Readmission Measures 2014. [[cited 2014 July 17]]; Available from: http://qualitynet.org/dcs/BlobServer?blobkey=id&blobnocache=true&blobwhere=1228890294248&blobheader=multipart%2Foctet-stream&blobheadername1=Content-Disposition&blobheadervalue1=attachment%3Bfilename%3DRdmsn_Updts_AMIPNCOPDSTK_032114.pdf&blobcol=urldata&blobtable=MungoBlobs. [Google Scholar]
- 17.Even though a probit or logit model is a natural choice for estimating a dichotomous variable in cross-sectional data, these models result in inconsistent estimators in panel data because we are including a large number of fixed effects (Greene, 2008). The linear probability models can consistently estimate the effect of STEMI on dichotomous outcomes (Buchmueller et al. 2006). One drawback of the model is that the predicted probability can be out of bounds. In our prior work analyzing a dichotomous outcome for AMI patients, we obtained virtually no out-of-bound predictions among the 1.49 million observations in our analysis. Another concern is that the error term is heteroskedastic. We correct this problem by estimating heteroskedasticity-robust standard errors that include adjustment for clustering within ZIP codes, and such an estimate is consistent in the fixed-effects model (Stock and Watson 2006).
- 18.The ED fixed effects also eliminate underlying differences across ZIP codes or LEMSAs since the nearest ED is the same for all patients from the same ZIP code.
- 19.Maclure M, Mittleman MA. Should we use a case-crossover design? Annu Rev Public Health. 2000;21:193–221. doi: 10.1146/annurev.publhealth.21.1.193. [DOI] [PubMed] [Google Scholar]
- 20.Elixhauser A, Steiner C, Harris DR, Coffey RM. Comorbidity measures for use with administrative data. Med Care. 1998;36(1):8–27. doi: 10.1097/00005650-199801000-00004. [DOI] [PubMed] [Google Scholar]
- 21.Nichol G, Thomas E, Callaway CW, Hedges J, Powell JL, Aufderheide TP, et al. Regional variation in out-of-hospital cardiac arrest incidence and outcome. JAMA. 2008;300(12):1423–1431. doi: 10.1001/jama.300.12.1423. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Norris RM, UK Heart Attack Study Collaborative Group Circumstances of out of hospital cardiac arrest in patients with ischaemic heart disease. Heart. 2005;91(12):1537–1540. doi: 10.1136/hrt.2004.057018. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Boden WE, O'Rourke RA, Teo KK, Hartigan PM, Maron DJ, Kostuk WJ, et al. Optimal medical therapy with or without PCI for stable coronary disease. N Engl J Med. 2007;356(15):1503–1516. doi: 10.1056/NEJMoa070829. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.