As emphasized by the most prominent health service researchers in emergency medicine, there is a need to document whether decreased access as measured by diversion affects the quality of care or outcomes and, if so, the extent of such effects. Specifically, we address the following research question. Is temporary ED closure on the day a patient experiences AMI, as measured by ambulance diversion hours of the nearest ED, associated with increased mortality rates among patients with AMI?
An ED on diversion can be considered as a signal that available resources are unable to match demand or a proxy albeit imperfect of crowding. For patients who had to be diverted elsewhere, ambulance diversion increases transport time, 8 likely causing delays in receiving treatment and potentially worse prognosis of AMI. Even if the increased transport time is trivial, the patients might end up in a less desirable setting eg, ED without catheterization capacity if the one ED with catheterization capacity is on diversion.
For nondiverted patients in an ED that is on diversion either because these patients were admitted before the status change, arrived by private vehicles, or were brought in under exception , their outcome could still be affected as they are in an ED during a time when clinicians or resources are limited in such a way to prevent optimal patient care.
Moreover, diversion in one hospital can potentially affect patients in nearby hospitals, as nearby hospitals would receive the diverted patients. This increased patient load could similarly cause treatment delays.
Many EDs are on diversion for short periods on a given day and in many instances have multiple episodes of diversion throughout a day. Our patient data contain date of admission, but not the exact time of admission. Although we cannot verify that a patient was diverted or not, the conceptual model described herein hypothesizes that longer exposure to diversion hours would be associated with worse outcome for both the diverted and nondiverted patients in the affected area.
We obtained detailed daily diversion logs for the years from each county by directly contacting their local emergency medical services agencies and securing permission.
All counties have daily logs available until November We only included patients from the relevant months or year when data for the corresponding county were available.
The local emergency medical services agencies govern and track diversion in all hospitals under each county's jurisdiction. The daily diversion log is specific to ED and trauma centers, and contains information regarding date and exact time diversion began and ended for every hospital as well as the reason for diversion in each instance ie, whether the ED diversion is due to ED saturation, if only trauma care is on diversion, lack of a neurosurgeon, equipment downtime.
During the study period, there were no policies to selectively divert patients with AMI to percutaneous coronary intervention—equipped hospitals in these 4 counties. For the purpose of our analysis, we excluded diversion that only applied to trauma center or psychiatric EDs and diversion due to lack of a neurosurgeon or computed tomographic scan downtime, because these types of diversion would not affect the admission of patients with AMI.
To capture the relevant hospital universe for matching patients to the correct EDs because hospitals not on diversion would not appear in the diversion logs , we merged the daily diversion logs with California Office of Statewide Health Planning and Development and Medicare Healthcare Cost Report Information System data sets to obtain additional facility data.
Patient data from the 4 California counties, including patients' mailing zip codes, were obtained from the Medicare Provider Analysis and Review. We linked each patient's zip code with longitude and latitude coordinates of each zip code using Mailer's software. In addition, we identified the diversion level of the nearest ED on the day a patient experienced AMI by merging the ED diversion data to the patient database on admission date and provider identification.
The study was approved by the Naval Postgraduate School Institutional Review Board and, regarding patient informed consent, a waiver was obtained as part of the institutional review board review because we used secondary data for analysis. These patients' Medicare records were linked to death certificates, if deceased, up until the end of March We applied several exclusion criteria to the patient sample.
First, we followed the exclusion criteria of McClellan et al 23 to minimize selection bias, which excluded patients who had a prior AMI admission within the past 12 months, patients who had a length of stay of 1 day because the patient might have been misclassified as AMI at the initial presentation , and patients without continuous Medicare part A coverage within the past 12 months. Our statistical model follows the same principle as the case-crossover design, while controlling for time-dependent variables.
We compared the percentage of patients with AMI who died within 7 days, 30 days, 90 days, 9 months, and 1 year when their nearest ED is in normal operation ie, no exposure to diversion [control group] and when the same ED is exposed to different levels of diversion ie, the same ED crosses over to higher exposure of diversion.
By using each ED as its own matched control, we can eliminate any inherent differences across EDs, such as possible differences in baseline mortality rates, quality of care, case-mix of the patient population, teaching status, or other unobserved characteristics that might be confounded with mortality rates.
We defined 4 diversion exposure levels as 0 hours reference group , less than 6 hours, 6 to less than 12 hours, and 12 or more hours. These cut points were determined before we linked the daily diversion data to patient outcomes by dividing the empirical distribution of the daily ambulance diversion hours into quartiles. The cutoffs for the quartiles are 3. We combined the first 2 quartiles because a priori we did not expect to see an association with inpatient mortality at lower levels of diversion and wanted to account for only practically meaningful thresholds.
We therefore used 6 and 12 hours instead of 6. The ED fixed effects removes any time-invariant unobserved differences across EDs, and the 3 diversion exposure indicators allow us to compare AMI mortality rates when the same ED is exposed to different levels of diversion.
Because each ED serves as its own matched control to compare mortality rates across different levels of diversion, we excluded patients from hospitals in which we observed fewer than 3 levels of exposure.
Although a logistic model is the natural choice for estimating a dichotomous dependent variable for cross-sectional data, it would result in an inconsistent estimator in a panel data setting because we are including a significant number of fixed effects. On the other hand, a linear probability model can provide consistent estimates.
We also included a list of disease-related risk adjustment following the work by Skinner and Staiger, 27 which uses the same patient data source. Specifically, risk adjustments were made if patients had peripheral vascular disease, chronic pulmonary disease, dementia, chronic renal failure, diabetes, liver disease, or cancer at the time of admission. We included hospital characteristics of the admitted hospital, including whether the hospital has catheterization capacity, hospital ownership for-profit, government , and size measured by log transformed total available beds.
In addition, we controlled for year trends overall mortality rates have decreased steadily over time and monthly seasonal trends within each year. For all models, we estimated heteroskedasticity robust standard errors, 28 which allow for intra-ED correlation among patients who lived closest to the same ED.
The final sample consisted of 13 patients from zip code areas whose admission date was within the relevant period in which ED diversion data were available. The Figure shows the mean hours of diversion per day between January and November among hospitals that reported positive diversion hours. The mean SD daily diversion duration was 7.
Merging the diversion information to the patient data, we excluded patients whose closest ED was not exposed to at least 3 levels of diversion and we excluded diversion logs from because the last matched admission date was December The multivariate analysis consisted of 11 patients.
Among these patients, , , , and patients were admitted for AMI when their closest ED was not exposed to diversion and exposed to less than 6 hours, 6 to less than 12 hours, and 12 or more hours, respectively. Patient demographics and comorbid condition characteristics generally do not differ by levels of diversion. Table 2 reports the hospital characteristics of admitted ED.
The number of patients who were admitted to their closest ED and the distance between admitted ED and closest EDs were similar across the 4 diversion categories.
The similar levels of travel pattern might suggest that distance is a minor factor in describing the relationship between diversion and mortality, and that other mechanisms discussed in the conceptual model section play a bigger role. Table 3 shows the multivariate results, focusing on the diversion variables only full regression results are shown in eTable 1.
The first column shows the mean mortality rates in our control group no diversion on day of admission. The next 3 columns show the regression-adjusted differences in mortality rates between each of the exposure groups and the control group. There were no statistically significant differences in mortality rates between no diversion status and when the exposure to diversion was less than 12 hours.
We performed several sensitivity analyses. First, to make sure that our results were not driven by the underlying differences across admitted hospitals, we estimated our model by replacing the nearest ED fixed effects with admitted ED fixed effects. Our results were similar and all conclusions remained the same. Second, our sample did not include patients who died on arrival or in the ED; those patients would have only had outpatient records.
We therefore obtained authorization to access 2 years of outpatient records and , resulting in 63 additional cases. When we added this group to our original sample, our conclusions on the key diversion variables remained the same. Third, we implemented an additional model by including an additional indicator for patients who bypassed their closest ED and interaction terms between the 3 diversion exposure categories and this bypass indicator.
However, the standard errors are too large to make definitive statements. Our study to our knowledge is the first multisite, multicounty analysis using daily ambulance diversion and patient-level data to evaluate the association between diversion and patient outcomes for patients experiencing AMI.
We showed that when the nearest ED is on diversion, a lower proportion of patients is admitted to hospitals with catheterization capacity, and a higher proportion is admitted to for-profit and government hospitals. Under a variety of specifications and sensitivity analyses, we found that lengthy periods of ED diversion are associated with higher mortality rates among patients with a time-sensitive condition such as AMI.
Specifically, when a patient's nearest ED was exposed to diversion for 12 or more hours on the day of admission, the patient experienced a higher death rate by about 3 percentage points than when that same ED was not on diversion.
This adverse relationship persisted even when we examined the 1-year mortality rate. When a hospital's ED is on diversion, it can affect different types of patients—those patients who were diverted, those patients receiving care or admitted while the ED is on diversion status, and those patients in nearby hospitals receiving the diverted patients.
Although we were able to examine patient and hospital interactions at a more precise level than the community-wide ecological analysis, we could not identify individual patients diverted from their ED of choice vs those who were not, or the mode of transportation those patients who arrived via private vehicles would be admitted.
Although our study design was advantageous in that it avoided confounding of patients who were or were not selected to be diverted, our results must be interpreted with caution because we cannot disentangle the precise mechanisms through which diversion affects patient outcomes.
Our results should not be interpreted as causal. Ambulance diversion is common and more likely to occur in urban settings—the National Center for Health Statistics estimated that hospitals divert more than 0.
Fortunately, we only observed the adverse relationship in hospitals that were on diversion for at least 12 hours on any given day. Notably, such long diversion hours are more likely to occur in winter and in densely populated metropolitan areas—both factors associated with increased ED demand.
These findings point to the need for more targeted interventions to appropriately distribute system-level resources in such a way to decrease crowding and diversion, so that patients with time-sensitive conditions such as AMI are not adversely affected.
It is important to emphasize that while demand on emergency care is increasing as evidenced by increasing utilization, supply of emergency care is decreasing. Our study has several limitations. First, we identified the nearest ED for each patient based on the longitude and latitude information of the patient's zip code 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 all 4 counties are in densely populated metropolitan statistical areas. Second, the patient's zip code on file is based on mailing zip code, which might not reflect the actual residence. We took the standard approach and applied exclusion criteria, dropping patients whose admitted hospital was more than miles away from their zip code.
Third, it is possible that some patients' closest EDs are out of the counties in which we can match diversion logs eg, a resident in San Francisco county might be closest to an ED in Alameda county. In our method that follows the case-crossover design, those patients would be excluded from the analysis, because we only included patients whose nearest ED experienced multiple levels of diversion.
Fourth, there might be reporting errors in the diversion daily logs. As long as the errors do not systematically differ by diversion duration ie, there are not more errors for log entries that record longer duration , we do not expect to have a bias in our estimates.
Therefore, our results should not be generalized to the younger population. Although these counties are demographically diverse, the proportion of black individuals is substantially lower and the proportion of other nonwhite minorities is substantially higher than individuals in the United States as a whole.
Also, these counties have few rural residents. Therefore, our findings may not be readily generalizable to other parts of the United States, particularly rural areas in which a single hospital is the only option for AMI care. In addition, the exclusion of patients who died before they could generate a hospital admission means our estimated mortality rate differences should be considered a conservative estimate.
Suppose we have a hypothetical patient who will die in either case, whether the ED is on diversion or not. In other words, the patient would contribute as 1 death under no diversion, but no deaths under diversion. The implication of this data limitation means the observed mortality rate is lower than the actual mortality rate when the ED is exposed to diversion, therefore, making our estimated difference in mortality rate between diversion and no diversion a conservative estimate. Diversion is a signal of a larger access problem in the health care system, representing resource constraints that are beyond patient factors and related to the hospital and health care system.
We show a strong relationship between prolonged ambulance diversion and increased mortality of patients with AMI. Although we cannot disentangle the precise mechanisms through which diversion affects patient outcomes, our results suggest that more integrated health care policies from the prehospital to in-hospital setting should include provisions that minimize instances in which hospitals are on diversion for prolonged periods.
Furthermore, restructuring of hospital and larger system-level resources to improve care delivery efficiency may be required to improve outcomes of patients with time-sensitive conditions, such as AMI. Possible policy options to improve such care could include patient flow initiatives that have been implemented in many counties and states with success.
In addition, it would be important for future analyses to disentangle the various mechanisms through which diversion might adversely affect patient care, so that policies targeting the right mechanisms may be adapted for better care that translates into better outcomes for patients in need.
It is also crucial to examine the relationship between ambulance diversion and the outcomes of nonelderly patients and patients experiencing other time-sensitive illness such as traumatic injuries. Author Contributions: Dr Shen had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design, acquisition of data, critical revision of the manuscript for important intellectual content, and administrative, technical, or material support : Shen, Hsia. Analysis and interpretation of data, drafting of the manuscript, statistical analysis, obtained funding, and study supervision : Shen. Role of the Sponsors: The sponsors had no role in the design and conduct of the study, in the collection, management, analysis, and interpretation of the data, or in the preparation, review, or approval of the manuscript.
Disclaimer: The article contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Institutes of Health or the Robert Wood Johnson Foundation. Moreover, all EDs must receive any patient diverted from other EDs, regardless of its own crowdedness and diversion status; that is, no "re-diversion" is allowed.
Patients are equally diverted only to EDs that are not in the diversion status i. If all EDs request diversion, then no patient will be diverted. The results of this simulation are shown in Fig 3. When following Rule 1, ED6 became excessively overcrowded because it accepted all diverted ambulance-transported patients. Fig 3 b shows the simulation results following Rule 2. ED1, which has the least resources in the region, becomes excessively overcrowded. A comparison of Fig 3 b and 3 a shows that ED6 becomes less crowded when following Rule 2 than when following Rule 1, but its CI is still over 1.
Fig 3 c shows the simulation results following Rule 3. Meanwhile, the CI of ED6 when following Rule 3 is slightly lower than that when following Rule 2 and is much lower than when following Rule 1. In the third study, we explored the effect of diverting ambulatory patients. In this study, we compared Rule 3 i. Ambulance-transported patients are diverted as in Rule 3. Ambulatory patients are diverted with no advice as to which ED to attend.
Thus, the diverted ambulatory patients may go to any other EDs with equal probabilities. In other words, the ambulatory patients are diverted according to Rule 2. Ambulatory patients are also instructed to go to EDs that are not in the diversion status. In other words, both ambulance-transported and ambulatory patients are diverted according to Rule 3. The simulation results are shown in Fig 4. Note that Fig 4 a using Rule 3 is exactly the same as Fig 3 c.
Fig 4 b and 4 c show the results when following Rule 4 and Rule 5, respectively. A comparison of Fig 4 a with Fig 4 b and 4 c indicates that there are some fundamental differences when EDs divert ambulatory patients in addition to ambulance-transported patients. Fig 4 b shows that Rule 4 results in ED1 becoming excessively overcrowded within half of a day.
Three sets of patient diversion strategies were evaluated via computer simulations based on a simplified ED model that represents the operations of various EDs using statistical processes.
While the simplified ED model and the assumptions made on the statistical processes may not entirely reflect real ED operations, these processes can nevertheless capture the average dynamics of patient flow in EDs and are widely acceptable. The key parameters of the model, including patient arrival rates, percentages of patients of different acuity levels, percentage of patients arriving by ambulance, and the total resources of the EDs, were assigned based on real data from six hospitals in the Tainan metropolitan region.
Study I evaluated the impacts of diverting all ambulance-transported patients A-AD and diverting only low-acuity ambulance-transported patients L-AD. According to the simulation results, A-AD is slightly more effective than L-AD, as expected, but this difference was insignificant.
The results of study I suggest that when ambulance-transported patients account for a small fraction of the total patients in the EDs of a region, diverting only ambulance-transported patients has limited impact on relieving ED crowdedness.
Moreover, diverting ambulance-transported patients of any acuity level has a similar outcome to only diverting those with low acuity levels. Assuming EDs are only allowed to divert ambulance-transported patients, study II evaluated the impact of three different diversion rules on the relief of overcrowding in several EDs in a region.
The simulation results indicate that diverting patients to EDs with more resources causes severe crowdedness in the largest ED in the region, while diverting patients equally to the other EDs tends to severely overcrowd the smallest ED in a region. Neither of these two strategies requires coordination among EDs, and clearly, neither is optimal.
The phenomena may be due to some EDs having idle capacity during periods that other EDs are full since they were not recipients of diverted patients. It also hints that regional ED utility could be optimal when diversion status of each ED is accessible to public. Study III assessed the effectiveness of the diversion of both ambulatory and ambulance-transported patients.
Based on the observed RCIs, we found that the outcome of arbitrarily diverting ambulatory patients is very similar to that of not diverting them because, in our simulations, the patients were unable to be re-diverted.
Hence, any patient that was diverted by one ED will consequently be accepted by another ED, regardless of whether the ED was overcrowded or not. The results of study III further suggest that the EDs with less resource tend to suffer when ambulatory patients were diverted arbitrarily. To improve ED efficiency, it is essential that EDs divert ambulatory patients appropriately. When ambulatory patients were diverted to EDs that were not in the diversion status, the RCI curve dropped significantly and was maintained below a certain level throughout the day.
The simulation showed that all EDs operated below or at their full capacities and that no overcrowded situations occurred throughout a day. Therefore, from a regional point of view, simply diverting ambulatory patients provides almost no benefit. However, if ambulatory patients are properly diverted to the EDs that are less crowded, the crowdedness of the EDs in the region improves significantly.
If such a mechanism is viable, our simulation results indicate that it would optimize the utilization of emergency medical resources in a region. To reduce the complexity of the simulations, we intentionally ignored the time required for certain ED operations, such as triaging patients, cleaning the treatment area, and any administrative processes.
Furthermore, many aspects of ED management are grouped into the treatment process. These include laboratory and radiological examinations, administration of medications, pending consultations, explanations to obtain patient consents for certain procedures and treatments, medical education prior to discharge, etc.
In the simulations, we assumed that hospital beds opened at a fixed rate. Although this assumption does not reflect the reality of hospital operations, it nevertheless does not largely change the outcomes in our simulations. In our simulations, we determined that ED input is the main contributor to crowding, as we deliberately overloaded ED capacity by setting an excessive patient influx. To assess ED output as the key source of crowding, it would be necessary to choose a more realistic statistical process to represent hospital bed availability.
Our model may accommodate EDs of different sizes. The impact of AD strategies may vary according to the number of hospitals in a single community as well as the treatment capabilities and capacities of each ED.
These are all parameters that can be adjusted in our simulation program. Based on a queuing model with parameters calibrated by real data, patient flows of six EDs in a region were simulated by a computer program.
The results indicate that with regards to minimizing the crowdedness of EDs in the whole region, the best strategy is to divert all ambulance-transported patients and ambulatory patients to the EDs that are not already crowded. The implication of the results of this study should be tailored and be validated in real practice.
The authors sincerely appreciate the contributions of the hospitals and emergency departments in the Tainan metropolitan region to this study.
Browse Subject Areas? Click through the PLOS taxonomy to find articles in your field. Abstract Emergency department ED overcrowding threatens healthcare quality. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited Data Availability: All relevant data are within the paper.
Funding: The authors received no specific funding for this work. Introduction Emergency department ED overcrowding threatens healthcare quality and is becoming a worldwide problem [ 1 ]. Download: PPT. Parameter settings in our simulations In our simulation studies, patient data from the EDs of six hospitals in the Tainan metropolitan region were used to derive the values of some key parameters of the queuing model.
Table 1. Table 2. Ethics This research was using computer simulation and adhered to the appropriate reporting guidelines and community standards for data availability. Question II When AD is initiated, should the ambulance-transported patients be diverted to EDs with more medical resources or be diverted to any other ED in the region?
For this study, we assume the following patient diversion rules: Rule 1. Fig 2. Simulation results of ambulance diversion AD strategy: Study I.
AD strategy study II The results of study I indicate that when patients are diverted from an ED with fewer resources to EDs with more resources, the ED with the most resources tends to be severely overloaded. In this study, we considered two other AD rules in addition to Rule 1 : Rule 2.
Rule 3. Fig 3. Rule 5. Fig 4. Simulation results of ambulance diversion strategy: Study III. Discussion Three sets of patient diversion strategies were evaluated via computer simulations based on a simplified ED model that represents the operations of various EDs using statistical processes. Limitations To reduce the complexity of the simulations, we intentionally ignored the time required for certain ED operations, such as triaging patients, cleaning the treatment area, and any administrative processes.
Conclusion Based on a queuing model with parameters calibrated by real data, patient flows of six EDs in a region were simulated by a computer program. Acknowledgments The authors sincerely appreciate the contributions of the hospitals and emergency departments in the Tainan metropolitan region to this study. References 1. Forecasting emergency department crowding: a prospective, real-time evaluation.
J Am Med Inform Assoc. Academic emergency medicine : official journal of the Society for Academic Emergency Medicine ; 38— View Article Google Scholar 3. Accid Emerg Nurs ; 18— Epstein SK, Tian L. Development of an emergency department work score to predict ambulance diversion. Academic emergency medicine : official journal of the Society for Academic Emergency Medicine ; — View Article Google Scholar 5.
A conceptual model of emergency department crowding. Ann Emerg Med ; — Effects of external and internal factors on emergency department overcrowding.
Siegel B. Triage for overcrowding. Hospitals should fix the emergency department problems they can control. Mod Healthc ; View Article Google Scholar 8. Anantharaman V. Impact of health care system interventions on emergency department utilization and overcrowding in Singapore.
Int J Emerg Med ; 1: 11— Using medical screening examinations to reduce emergency department overcrowding. View Article Google Scholar Impact of ambulance diversion policies in urban, suburban, and rural areas of Central Maryland.
Academic emergency medicine : official journal of the Society for Academic Emergency Medicine ; 8: 36—
0コメント