Door-to-Imaging Time for Acute Stroke Patients Is Adversely Affected by Emergency Department Crowding
Background and Purpose—National guidelines call for door-to-imaging time (DIT) within 25 minutes for suspected acute stroke patients. Studies examining factors that affect DIT have focused primarily on stroke-specific care processes and patient-specific factors. We hypothesized that emergency department (ED) crowding is associated with longer DIT.
Methods—We conducted a retrospective investigation of 1 year of consecutive patients in our prospective Code Stroke registry, which included all ED stroke team activations. The registry and electronic health records were abstracted for 27 potential predictors of DIT, including patient, stroke care process, and ED operational factors. We fit a multivariate logistic regression model and calculated odds ratios and 95% confidence intervals. Second, we constructed a random forest recursive partitioning model to cross-validate our findings and explore the proportional importance of each category of predictor. Our primary outcome was the binary variable of DIT within the 25-minute goal.
Results—A total of 463 patients met inclusion criteria. In the regression model, ED occupancy rate emerged as a predictor of DIT, with odds ratio of 0.83 (95% confidence interval, 0.75–0.91) of DIT within 25 minutes per 10% absolute increase in ED occupancy rate. The secondary analysis estimated that ED operational factors accounted for nearly 14% of the algorithm’s prediction of DIT.
Conclusions—ED crowding is associated with reduced odds of meeting DIT goals for acute stroke. In addition to improving stroke-specific processes of care, efforts to reduce ED overcrowding should be considered central to optimizing the timeliness of acute stroke care.
Stroke is a leading cause of death and long-term disability.1,2 Timely intervention in acute ischemic stroke is essential, prompting national guidelines calling for intravenous administration of recombinant tissue-type plasminogen activator within 60 minutes of patient arrival to the emergency department (ED) and timely endovascular treatment when appropriate.3 A key driver of timely definitive management is ED door-to-imaging time (DIT), with guidelines calling for 25 minutes or less.4
Several stroke-specific care processes are known to improve DIT, prompting national recommendations for adopting the following evidence-based processes: emergency medical services prenotification, rapid triage protocols, single-call stroke team activation, toolkits, rapid laboratory testing, team-based approach, and prompt feedback of data.5,6 Mode of patient arrival (emergency medical services versus private vehicle) is another care process known to affect DIT.7,8 In addition, patient-specific factors, including sex, symptom severity, symptom duration, history of diabetes mellitus, and initial blood pressure, also are reported to affect DIT.8–10
Although research related to DIT in stroke is extensive, it has focused primarily on process-specific and patient-specific factors. There has been little focus on how operational environments of care—more specifically, ED crowding—may affect the management of acute stroke patients. ED crowding deleteriously affects other time-sensitive treatment strategies, including definitive treatment of ST-segment elevation myocardial infarction and antibiotic therapy for pneumonia, so it stands to reason that it also may affect the timeliness of stroke care.11,12 We are aware of only 1 investigation of the effects of ED crowding on stroke care.13 In that study, Chatterjee et al13 concluded that ED crowding did not affect DIT for patients with symptoms for <3 hours. However, the hospitals in that study exhibited only modest crowding, raising the possibility that under conditions of more intense ED crowding, DIT may be affected adversely. We hypothesized that ED crowding is associated with longer DIT for patients with suspected acute stroke.
Study Design and Setting
We retrospectively queried the prospective stroke registry of our urban, regional referral stroke center hospital for consecutive patients who presented to the adult ED and met criteria for Code Stroke (stroke team activation) between June 15, 2014 and June 15, 2015. Routine registry monitoring showed that ≈60% of stroke activations achieved DIT within 25 minutes. The sample size required to demonstrate a 2-sided difference in proportions of 10 percentage points with 80% power was 191, achievable using 1 year of registry data.
During the study period, the hospital admitted ≈27 000 adult inpatients and was the primary teaching site for multiple residencies, including neurology and emergency medicine. The adult ED cared for ≈65 000 patients. The hospital had a 24/7 stroke team, which worked closely with emergency medicine. All American Heart Association/American Stroke Association Get With The Guidelines recommendations were implemented.5,6 Point-of-care testing was in place, and chest x-rays were performed after computed tomographic (CT) scans unless emergently needed. Electrocardiograms were performed after CT or in parallel with essential initial actions. The ED had a dedicated, in-department CT scanner, and Code Stroke patients were prioritized for imaging in the following manner: the CT scanner was held open when a Code Stroke was activated (CT technicians were included in the team page). If another patient was being imaged at the time of a Code Stroke activation, their CT was completed, and the Code Stroke patient was imaged immediately next.
By protocol, Code Stroke was activated when any patient presented with symptoms or findings consistent with an acute stroke and symptom onset was within 12 hours. Our multidisciplinary stroke committee previously established the 12-hour window accounting for 3 key considerations: prioritizing sensitivity over specificity for the mobilization of the stroke team and resources, availability of resources enabling possible treatment beyond 4.5 hours of symptoms in select cases, and institutional research protocols. The committee felt that the potential patient benefits gained from this expanded window outweighed the potential inefficiencies it may have caused.
The institution maintains a prospective registry of all patients for whom Code Stroke is activated, which includes patient/visit identifiers, sex, age, and time stamps for care events, including ED arrival, Code Stroke activation, and CT completion. A nurse coordinator maintains the registry and verifies its accuracy.
Trained research assistants, blinded to the ED operational metrics included in the study, retrospectively reviewed the electronic health record (EHR) for each patient to validate the registry data and abstract the following information using standardized abstraction forms: mode of arrival (emergency medical services or private vehicle), triage emergency severity index score,14 location of initial ED care (dedicated resuscitation area versus main acute care area), initial vital signs (heart rate, respiratory rate, systolic blood pressure, diastolic blood pressure, and pulse oxygen saturation), supplemental oxygen use and delivery method (none, nasal cannula, face mask, bag-mask ventilation, or intubated), Glasgow Coma Scale score, assessment of alertness and orientation (person, place, and time—range of 0–3), National Institutes of Health Stroke Scale score, initial blood glucose value, elapsed time since the patient was last known to be at their baseline neurologic condition, previous history of stroke or transient ischemic attack, previous history of diabetes mellitus, and previous history of hypertension. The EHR also was reviewed to determine whether the ED team documented treatment of another emergent, life-threatening condition before obtaining the CT. Such conditions were classified as airway/breathing, hypertension, hypotension, hypoglycemia, electrolyte abnormality, or multiple above conditions. Rare missing values in the registry were obtained from the EHR by the abstractor. A second investigator independently searched the EHR for values missing after initial abstraction. Missing values not available in either the registry or the EHR (vital signs, n=1; glucose, n=4; and Glasgow Coma Scale, n=75) were replaced with the corresponding median value for the remaining data set. One entry in the registry was a duplicate; the affected patient was analyzed only once.
Investigators not involved in patient data abstraction queried the EHR to determine the following ED occupancy metrics at the time of arrival of each registry patient: number of patients in ED treatment areas, number of patients in the ED waiting room, and number of admitted inpatients boarding in the ED (defined as patients who remained in the ED after the decision to admit).15 The ED occupancy rate (EDOR) was calculated as the total number of ED patients, regardless of ED location (including the waiting room and hallway locations), divided by the number of licensed ED beds.16 Although we recognize the potential limitations with all quantitative measures of ED crowding,17 we selected EDOR and patient occupancy counts based on their previous use in the stroke literature and their face validity as instantaneous measures of ED demand–capacity mismatch.
Our primary outcome was the binary variable of DIT within the national goal of ≤25 minutes. We estimated that our projected sample size would support a model containing a maximum of 29 predictor variables.18 We considered 27 candidate predictors, including the patient, process of care, and operational factors listed in Tables 1 and 2 and the following time variables: arrival hour of day, day of week, and month/year. Recognizing that a delay in stroke identification may strongly influence DIT, we also included the binary variable of door-to-activation time (DAT) ≤15 minutes (national goal) as a predictor. We fit a multivariate logistic regression model using stepwise backward elimination to minimize the Bayesian information criterion with maximum likelihood estimation. Candidate models were validated using 75% of the data set for training and 25% for validation, with the validation column randomization stratified on final diagnosis of stroke or intracranial hemorrhage and DIT as a continuous variable, to assure a balanced validation set. Odds ratios (OR) and 95% confidence intervals were calculated using exponentiation from the regression parameter estimates.
Given the inherent threshold effect of DAT on DIT (it is impossible to achieve DIT within 25 minutes if the DAT is >25 minutes) and high probability that DIT and DAT share codependent predictors, we secondarily also sought to model DAT time as a function of the same predictor variables included in the primary analysis. We used the same logistic regression technique described above, except that we used DAT as the dependent variable, so there were 26 candidate predictors.
We performed another secondary analysis using an alternative statistical technique, constructing a random (bootstrap) forest recursive partitioning model, to cross-validate our primary analysis findings and explore the proportional importance of each category of predictor. Details of this methodology are available in the online-only Data Supplement.
Analyses were conducted using JMP Pro 12 (SAS Institute Inc, Cary, NC). The University of Massachusetts Medical School Institutional Review Board approved the study.
Characteristics of Study Subjects and Outcome Data
We identified 490 consecutive Code Stroke patients in the registry during the study period. In 1 case, the stroke team activation was immediately cancelled, and CT was never completed, so that patient was excluded from analysis. Twenty-six additional patients were documented as requiring intervention for another emergent, life-threatening condition before CT (airway/breathing intervention, n=18; hypertension, n=6; hypotension, n=1; and multiple, n=1) and were excluded. Clinical outcomes of included and excluded patients are shown in the study flow diagram (Figure 1). Table 1 reports the baseline characteristics of included patients.
DIT ranged from 4 minutes to 3 hours 41 minutes (median, 21 minutes; interquartile range [IQR], 20 minutes) and was within the 25-minute goal for 281 patients (60.7%). DAT ranged from <1 minute to 3 hours 37 minutes (median, 5 minutes; IQR, 17 minutes). Among the 334 patients (72.1% of total) with DAT within 15 minutes, DIT ranged from 4 minutes to 1 hour 46 minutes (median, 18 minutes; IQR, 10 minutes) and was within 25 minutes for 275 patients (82.3%). Figure 2 illustrates the cumulative distribution functions of DIT for each population.
The final regression model performed well with area under the curve of 0.93 in the training set and 0.90 in the validation set. Predictors included in the final model are shown in Table 3 with their multivariate OR of DIT within 25 minutes. An increase in EDOR across the entire range of the study population (47%–242%) corresponded to a multivariate OR of DIT within 25 minutes of 0.025 (95% confidence interval, 0.003–0.172).
The logistic regression model of DAT within 15 minutes exhibited good overall fit with area under the curve 0.92 in the training set and 0.90 in the validation set (P<0.0001). The predictors in the final model are shown in Table 4 with their OR of DAT within 15 minutes.
The random forest model also performed well with area under the curve of 0.98 in the training set and 0.92 in the validation set, with the final forest containing 42 classification trees. The proportional importance of ED operational factors in predicting DIT was 13.8% in the model. Results of the proportional importance of each predictor variable in the forest are available in the online-only Data Supplement.
Our results reveal that ED crowding can adversely impact DIT for acute stroke patients. Traditionally, ED crowding has not been considered in acute stroke management research, with most studies focusing on stroke-specific processes of care and patient-specific factors. To our knowledge, this is only the second investigation to consider the effects of the ED work environment on the care of stroke patients, and it is the first to show an adverse impact on timeliness of care in patients who may be potential candidates for acute intervention.13
In the only previous study to consider ED operational factors, Chatterjee et al13 found crowding not to be associated with increased DIT in patients who had symptoms fewer than 3 hours, leading to the conclusion that ED crowding did not affect timeliness of care in thrombolysis-eligible patients. At first, this discrepancy might prompt concerns about the generalizability of both studies, however, there are differences in the study designs and populations that warrant consideration. Chatterjee et al13 evaluated DIT among patients with symptoms for fewer than 3 hours as a function of quartiles of ED crowding metrics and then contrasted the results with a random sample of patients with symptoms >3 hours. In that data set, ED crowding did not adversely affect DIT for patients who presented within 3 hours, but there was an adverse effect for patients with symptoms longer than 3 hours. Because the accepted intravenous thrombolytic eligibility period at the time was 3 hours, it was presumed that the 2 groups were treated with a different sense of urgency, thus contributing to the difference in susceptibility to the adverse effects of ED overcrowding. In other words, the sense of urgency for patients who met criteria for thrombolysis may have been enough to overcome the ED crowding factors that adversely affected DIT in the comparison group. Our study design differed in that all patients with symptom onset within 12 hours were treated as acute stroke activations and included in our analysis, and we controlled for the time elapsed since the patient was last known to be at their baseline neurological status as a continuous parameter. Ultimately, onset time did not contribute to the final analysis when the best model was selected on the basis of whole-model performance. Presumably, this difference was because of evolving expectations that patients within 12 hours of symptom onset be treated with urgency given advances in endovascular therapeutic options since 2008.
However, we think that the key difference between the Chatterjee et al13 findings and our own relates to the magnitude of ED overcrowding. In the previous study, the median EDOR at the academic site comparable to our own was 78% (IQR 22). At our site, the burden of ED overcrowding was greater, with a median EDOR of 122% (IQR 53). Our analysis considered ED operational factors as continuous predictors in the regression model, rather than using quartiles, permitting a more granular view of the exposure–response relationship between crowding and DIT. Using this methodology, EDOR did emerge as a significant predictor, and the random forest approach estimated that ED operational factors accounted for nearly 14% of the algorithm’s prediction of DIT.
In considering both investigations in totality, we feel the results may actually be consistent. It is likely that the urgency for patients who met stroke activation criteria was similar in both studies, but trends in therapeutic options have expanded the time window for that urgency. At lower levels of overcrowding, as in the Chatterjee et al13 population, the barriers to timely DIT may be more easily overcome. However, as the burden of crowding increases, as in our data set, the ED environment may be less able to support timely stroke care. Alarmingly, it seems that high EDORs are associated with a decreased ability to deliver timely DIT for patients who may be intervention eligible, regardless of duration of symptoms.
With regard to process of care and patient factors that may affect timeliness of care, our findings were consistent with previous studies. Arrival by emergency medical services was associated with improved DAT and DIT,7,8 and patient factors such as blood pressure, blood glucose, National Institutes of Health Stroke Scale score of >2, intubation/high oxygen requirement, and sex were found to be significant factors that predicted activation and imaging times.8–10 Interestingly, time cycle factors were significant in our study, contrasting previous research showing no association between DIT and time of day or day of week.19 The underlying mechanism is not clear, but time may be a surrogate for ED crowding, given the cyclic nature of ED patient arrivals and patient volumes. Demand–capacity mismatch—influenced by factors such as staffing levels and hospital/ED bed availability—may differ by time of day and day of week. Of note, we are not aware of any day or time differences in stroke-specific care processes at our institution.
It should also be noted that during the study period, our ED implemented a process improvement related to stroke care in October 2014. Accordingly, we included month/year as a candidate predictor to account for temporal trends, and patient presentation after October 2014 ultimately was included in the final model. Other than potentially increasing stroke awareness among staff indirectly, the core intervention was to initially evaluate all suspected stroke patients in a dedicated ED resuscitation room. We were unaware of this strategy having been rigorously tested, but it had been reported as part of a portfolio of stroke-specific care interventions at other centers.6 This practice was occasionally tried before formal implementation in October 2014. Thereafter, it occurred more frequently but did require ramp-up in staff adoption consistency. Therefore, in addition to the month/year variable, we included initial assessment in the resuscitation room as a separate predictor variable. Even after accounting for temporal trends, we were struck by its magnitude of effect on achieving DIT within 25 minutes (OR 10.7) and DAT within 15 minutes (OR 42.9). Despite controlling for multiple patient and process factors, the underlying factors leading to patients being initially assessed in the resuscitation area do remain unclear. It is possible that higher odds of timely activation and imaging may represent some transposition of cause and effect. An unmeasured heightened focus may have played a role, or perhaps triage was more sensitive, resulting in resuscitation area utilization being a surrogate marker for triage accuracy.
The strongest predictor of DIT was DAT. Although this is intuitive for several reasons, it warrants consideration as a potential limitation of the study. There is a threshold effect when considering binary variables because it is impossible to meet the DIT goal if DAT is >25 minutes. Furthermore, some factors that could plausibly influence DIT may also influence recognition and activation. Our secondary DAT analysis highlights some of these factors, but we caution against combining the results of both analyses because of the risk of overestimating the effect magnitude for codependent variables. Our random forest analysis suggests that DAT alone may account for nearly 31% of the model’s classification ability for predicting DIT, despite the variables considered in building the regression models of DAT and DIT being identical. This suggests that there may be hidden factors influencing DAT not accounted for in our data set. We considered using activation-to-imaging time as our primary outcome, but this was unsupported by available literature, and our preliminary analyses suggested that doing so would not provide advantage in dimensional reduction compared with DIT. We also considered analyzing DIT as a continuous variable but determined that this would have been error prone because of the large number of nonlinear associations and interaction terms. Another potential limitation of our study may have been that, because of EHR limitations, it was not feasible to determine if a patient was occupying the ED CT scanner at the time of each Code Stroke activation and, if so, the amount of time remaining to complete that scan. The median time required to complete a CT among all ED patients and CT types (stroke and nonstroke) was 16 minutes. Because of our institutional protocol of not initiating other CT scans when Code Stroke was activated, only a portion of an ongoing CT scan potentially could have affected the DIT for a Code Stroke patient. Presuming a normal distribution, the potential overlap of an ongoing CT and the DIT of a Code Stroke patient would be, on average, half of the median total CT length or just 8 minutes. In addition, essential pre-CT actions, such as intravenous catheter placement and transport would need to be performed regardless of whether the CT scanner was occupied, thereby reducing the potential impact of an ongoing CT to even <8 minutes. Therefore, although it stands to reason that there may have been an increased likelihood that another patient would have been occupying the CT scanner at the time of a Code Stroke activation during times of overcrowding, we think the unmeasured effect of CT occupancy on DIT likely was minimal. Other potential study limitations included the small absolute number of patients who received thrombolysis or interventional procedures, which precluded a robust analysis of that subpopulation and the single-center, retrospective study design.
Notwithstanding these potential limitations, this investigation is the first to show that ED crowding can adversely affect timeliness of stroke care. These findings imply that to optimize stroke care, focusing on improving stroke-specific processes of care alone is insufficient; ED crowding also must be addressed. A common misconception is that ED crowding is an ED-centric issue, however, numerous studies have shown that hospital-wide flow problems lead to boarding of admitted inpatients in the ED, which is a key driver in ED overcrowding.20–24 Several strategies to reduce ED crowding have been cited in the literature, including co-ordinated hospital bed management systems, timing elective admissions to smooth hospital census, temporary inpatient admission and discharge units, hospital full capacity protocols, financial incentives, boarding admitted patients in inpatient hallways when the ED is beyond capacity, and multidisciplinary hospital flow management teams.22,23,25,26 The results of this investigation suggest that it would be prudent for leaders responsible for the care of acute stroke patients to join ED providers in their ongoing call for large-scale, systemic improvements shown to reduce ED overcrowding.
After controlling for patient-specific and stroke process of care factors, increased ED overcrowding is associated with reduced odds of meeting established DIT goals. This finding, previously unreported, has substantial implications for acute stroke patients and those responsible for their care.
We would like to acknowledge Paula Paige RN, BSN for assistance with data collection.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.116.015131/-/DC1.
- Received August 17, 2016.
- Revision received September 25, 2016.
- Accepted October 10, 2016.
- © 2016 American Heart Association, Inc.
- 2.↵Centers for Disease Control and Prevention. Prevalence of stroke—United States, 2006–2010. MMWR Morb Mortal Wkly Rep. 2012;61:379–382. https://www.cdc.gov/mmwr/preview/mmwrhtml/mm6120a5.htm. Accessed November 11, 2016.
- Jauch EC,
- Saver JL,
- Adams HP Jr,
- Bruno A,
- Connors JJ,
- Demaerschalk BM,
- et al
- Kelly AG,
- Hellkamp AS,
- Olson D,
- Smith EE,
- Schwamm LH
- Fonarow GC,
- Smith EE,
- Saver JL,
- Reeves MJ,
- Hernandez AF,
- Peterson ED,
- et al
- Ruff IM,
- Ali SF,
- Goldstein JN,
- Lev M,
- Copen WA,
- McIntyre J,
- et al
- Ekundayo OJ,
- Saver JL,
- Fonarow GC,
- Schwamm LH,
- Xian Y,
- Zhao X,
- et al
- Haršány M,
- Kadlecová P,
- Švigelj V,
- Kõrv J,
- Kes VB,
- Vilionskis A,
- et al
- Sauser K,
- Bravata DM,
- Hayward RA,
- Levine DA
- Chatterjee P,
- Cucchiara BL,
- Lazarciuc N,
- Shofer FS,
- Pines JM
- 15.↵National Quality Forum. National Voluntary Consensus Standards for Emergency Care: A Consensus Report. 2009. http://www.qualityforum.org/Publications/2009/09/National_Voluntary_Consensus_Standards_for_Emergency_Care.aspx. Accessed August 14, 2016.
- Rodríguez-Rivera IV,
- Santiago F,
- Estapé ES,
- González-Sepúlveda L,
- Brau R
- 20.↵Institute of Medicine (U.S.). Committee on the future of emergency care in the United States health system. Hospital-Based Emergency Care: At the Breaking Point. Washington, D.C.: National Academies Press; 2007.
- Rabin E,
- Kocher K,
- McClelland M,
- Pines J,
- Hwang U,
- Rathlev N,
- et al
- Warner LS,
- Pines JM,
- Chambers JG,
- Schuur JD