Critically ill patients boarding in the emergency department and the association with intensive care unit length of stay and hospital mortality during the COVID-19 pandemic

Abstract

Boarding of critically ill patients in the emergency department (ED) has been associated with mortality and intensive care unit (ICU) length of stay (LOS). This study evaluated whether boarding time in the ED was associated with those outcomes. A retrospective analysis of patients admitted through the ED to the ICU was performed. Information on demographics, severity score, and diagnoses was collected. The continuous primary endpoint of ICU LOS was fitted by a log normal model on covariates, including ED LOS. A multivariate log normal model was also used to model covariates toward ICU LOS. The binary patient expiration status was modeled by univariate and multivariate logistic regressions to evaluate the association of mortality with covariates. ED LOS was not associated with ICU LOS (correlation with an estimate of −0.02 ± 0.06 [SE], P = 0.76). ED LOS was not associated with hospital mortality (estimate correlation of −0.07 ± 0.07 [SE], P = 0.33). Body mass index, APACHE IV score, mechanical ventilation, and diagnosis of COVID-19 were associated with LOS. Age, APACHE IV score, mechanical ventilation, sepsis, and COVID-19 were associated with mortality. In conclusion, ED LOS is not associated with ICU LOS or hospital mortality. These findings may be related to early therapeutic interventions applied in the ED.

The Institute of Medicine identified emergency department (ED) crowding as one of the most important challenges for public health.¹ Several publications demonstrated an association between ED crowding and an increase in hospital length of stay (LOS) and hospital mortality.^2–4 Prolonged stay of admitted patients in the ED is known as ED boarding. ED boarding is the primary reason for ED crowding. Although ED boarding exposes all patients to diminished quality of care, critically ill patients remain particularly vulnerable. Prior studies showed an increase in rates of ventilator-associated pneumonia and intensive care unit (ICU) mortality.^5,6 Specifically, within the group of mechanically ventilated patients boarding in the ED, those staying for >6 hours were almost 6 times more likely to die. Other studies revealed an increase of 1.5% in risk of ICU death for each 1-hour delay in admitting patients from the ED to the ICU.⁷ Despite the aforementioned data, many studies showed opposite results.⁸ The coronavirus disease 2019 (COVID-19) pandemic has caused a significant increase in hospitalizations and ICU bed demand, leading to a significant increase in the number of critically ill patients boarding in the ED. A recent publication showed that patients treated in the ICU during periods of peak ICU demand had a nearly twofold increased risk of mortality.⁹ However, whether those outcomes were related to an increase of critically ill patients boarding in the ED is unknown. To assess whether ED boarding is associated with hospital mortality and/or ICU LOS, we studied a consecutive series of ICU patients admitted at our institution through the ED.

METHODS

A retrospective analysis was performed of data collected from a consecutive series of patients admitted through the ED at Baylor University Medical Center to the medical or neurologic ICU from March 1, 2020, to October 31, 2020. Upon arrival to the ED, patients were deemed in “ICU status” (time zero) once an order to admit to the ICU was effectively placed in the electronic health record by a provider. ICU bedside nurses documented the actual time of arrival to the ICU once patients were physically located in an ICU bed. The LOS in the ED by ICU boarding patients was calculated as the time spent between time zero and ICU arrival.

Data were collected on demographic characteristics, the Acute Physiology and Chronic Health Evaluation (APACHE) IV severity score, need for invasive mechanical ventilation during the ED stay, and admission diagnosis. To standardize admission diagnoses throughout the dataset, patients were grouped according to compromised system as follows: 1) neurologic diagnosis, 2) cardiac, 3) respiratory, 4) gastroenterology and/or hepatology, 5) metabolic and/or toxicology, 6) sepsis, and 7) COVID-19. When more than one admission diagnosis was encountered, the electronic health record documentation was reviewed to assign a primary diagnosis motivating the ICU admission.

The goal of the study was to assess whether ED LOS by boarding ICU patients was associated with ICU LOS and hospital mortality. The ICU LOS was defined as the time between patient physical ICU arrival to physical ICU departure, including transfer to hospital floor or hospital discharge. In case of patient expiration, departure was computed accordingly based on time of death pronouncement. Due to the high prevalence of COVID-19 during the timeframe of the study, a further analysis was performed to assess the impact of this condition in clinical outcomes of interest. Furthermore, survival analysis for this diagnosis was performed. The study was approved by the Baylor Scott & White institutional review board (protocol #021-212).

The continuous primary endpoint of ICU LOS was fitted by a log normal model on each individual covariate, such as age, gender, body mass index (BMI), APACHE score, sepsis, COVID, ventilation, and ED LOS. A multivariate log normal model was also used to model these covariates jointly toward ICU LOS. The regression coefficients and associated P values were used to assess the direction of correlation and statistical significance between these covariates and the ICU LOS. The binary patient expiration status (mortality) was also modeled by univariate and multivariate logistic regressions to evaluate the association of mortality to the same collection of covariates using the odds ratio estimates and P values returned by the logistic regressions. Combined secondary outcomes of hospital LOS and the binary expiration status were illustrated by Kaplan Meier curves to visualize the survival probabilities. A multivariate Cox proportional hazard model was used to regress the time from hospital admission to death and the aforementioned covariates and return hazard ratio estimates and P values for evaluation.

RESULTS

A total of 1418 patients were admitted to the medical or neurological ICU through the ED during the study period. Demographic information, APACHE IV scores, and admission diagnoses are shown in Table 1. The average LOS in the ED, ICU, and hospital (with standard deviation [SD]) were 9 hours ± 6, 4.5 days ± 6, and 12 days ± 11.5, respectively. The overall hospital mortality was 20% (278 out of 1418). Of those patients discharged from the hospital, 34% went home, and 23% and 12% of them were discharged to long-term acute care or skilled nursing facilities, respectively. The histogram and scatterplot of the ED and ICU LOS data are shown in Figure 1.

Table 1.

Demographic information, severity score, admission diagnoses

Category	Variable	Result
Age (years)	≥65	666 (48%)
	<65	722 (52%)
	Mean ± SD	61 ± 16
Gender	Male	802 (57.8%)
	Female	586 (42.2%)
BMI (kg/m2)	≥30	520 (37.5%)
	<30	861 (62.5%)
	Mean ± SD	29.3 ± 9
APACHE IV score	Mean ± SD	21 ± 21.8
Admission diagnoses	Neurological	379 (26%)
	Cardiovascular	105 (8%)
	Respiratory	596 (43%)
	Gastrointestinal	55 (4%)
	Renal	44 (3%)
	Metabolic	49 (4%)
	Sepsis	165 (12%)
	COVID-19	376 (27%)
	Mechanical ventilation	465 (33%)

APACHE indicates Acute Physiology and Chronic Health Evaluation; BMI, body mass index; SD, standard deviation.

Figure 1.

Histogram and scatterplot of the emergency department (ED) and intensive care unit (ICU) length of stay (LOS) data. The upper row shows a right-skewed distribution. The lower row (log normal model) shows a normal distribution. ED LOS is measured in hours; ICU LOS is measured in days.

As data were skewed to the right, a log transformation was used to conform distribution to normality. A univariate log normal regression analysis between log ED LOS vs. log ICU LOS revealed a lack of correlation with an estimate of −0.02 ± 0.06 (standard error), P = 0.76. Furthermore, a multivariate log normal regression model was performed for the correlation of log ED LOS and log ICU LOS, but focusing on the diagnosis of COVID-19, sepsis, need for invasive mechanical ventilation, and demographic data, such as age, gender, BMI, and APACHE IV score. Table 2 shows data from multivariate log normal regression analysis.

Table 2.

Multivariate log normal regression analysis for intensive care unit length of stay

Variable	Estimate	SE	t value	P value
(intercept)	0.24	0.22	1.10	0.27
Age	−0.01	0.01	−2.07	<0.04
Men	0.02	0.07	0.27	0.79
BMI	0.01	0.01	4.09	<0.01
APACHE IV	0.01	0.01	2.86	<0.01
COVID-19 positive	1.16	0.07	15.4	<0.01
Ventilated	0.48	0.08	5.88	<0.01
Log ED LOS	−0.09	0.06	−1.74	0.08

APACHE indicates Acute Physiology and Chronic Health Evaluation; BMI, body mass index; ED, emergency department; LOS, length of stay; SE, standard error.

Among ICU patients boarding in the ED, a number of parameters were associated with prolonged ICU stay. Specifically, higher BMI and APACHE IV scores were directly associated with ICU LOS. Furthermore, the need for invasive mechanical ventilation and diagnosis of COVID-19 were associated with ICU LOS. Interestingly, patients’ age was inversely associated with ICU LOS. The assessment of ED LOS and hospital mortality revealed a lack of association. In fact, the univariate logistic regression showed an estimate correlation of −0.07 ± 0.07 (standard error), P = 0.33. A multivariate logistic regression model applying all aforementioned parameters is shown in Table 3. In this analysis, age >64 years, higher APACHE IV score, need for mechanical ventilation, diagnosis of sepsis, and COVID-19 were all directly associated with hospital mortality. A Kaplan-Meier plot depicting survival analysis for patients admitted to the ICU with COVID-19 is shown in Figure 2. Over time, the survival probability of patients with COVID-19 declined substantially compared with patients without COVID-19. Importantly, this difference became more evident after 30 days of hospitalization.

Table 3.

Multivariate logistic regression for hospital mortality

Variable	Estimate	SE	t value	P value
(intercept)	−3.10	0.22	−13.92	<0.01
Age (>64 years)	0.74	0.16	4.59	<0.01
Men	0.15	0.16	0.95	0.34
BMI >30 kg/m2	0.18	0.16	1.11	0.26
APACHE IV	0.70	0.08	8.86	<0.01
Sepsis	0.68	0.23	2.97	<0.01
COVID-19	1.18	0.17	6.79	<0.01
Ventilated	1.17	0.18	6.51	<0.01
Log ED LOS	−0.14	0.08	−1.75	0.08

APACHE indicates Acute Physiology and Chronic Health Evaluation; BMI, body mass index; ED, emergency department; LOS, length of stay; SE, standard error.

Figure 2.

Kaplan-Meier plot depicting survival analysis for patients admitted to the intensive care unit with COVID-19.

DISCUSSION

Our single-center retrospective study demonstrated that prolonged ED LOS for ICU boarding patients was not statistically associated with increased ICU LOS or hospital mortality. Hospital mortality was significantly increased in ICU-bound patients with COVID-19, sepsis, or age >65 years, as well as those requiring a mechanical ventilator and/or presenting with a high APACHE score.

The results of our study are in conflict with those of prior studies, which showed a direct association between ED boarding time and hospital mortality and LOS. Specifically, a retrospective analysis from Saudi Arabia that included 940 boarding ICU patients admitted to the ICU within 6 hours (group 1), between 6 and 24 hours (group 2), and later than 24 hours (group 3) revealed a direct association between boarding times and the aforementioned outcomes. Particularly, the hospital mortality rate within 6 hours was 22.5%, compared with 29.1% and 37.2% in those boarding in the ED between 6 and 24 hours or >24 hours, respectively.⁴ A prior study, which included 50,000 patients admitted to 120 ICUs located in the United States, revealed similar results, increasing hospital mortality and LOS for those patients boarding in the ED for >6 hours.¹⁰ Finally, a study that included four cohort studies from North America and Europe involving patients admitted with community-acquired pneumonia to the ICU demonstrated an increased odds ratio for 28-day mortality and hospital LOS for those patients admitted to the ICU with delay, compared with those rapidly transferred from the ED to ICU.¹¹

Despite the previously described information, other reports contradicted these findings. An observational study from the United Kingdom revealed that ICU patients boarding in the ED for >3 hours had similar ICU LOS and mortality rates compared with those boarding for a shorter amount of time.¹² Furthermore, a Finnish study revealed that such a delay was not associated with hospital outcomes or health-related quality of life outcomes 6 months after hospital discharge.¹³ Lastly, a US study showed that ED boarding for >24 hours was not associated with hospital mortality.¹⁴

Several explanations can be given for these discordant results. First, the patient populations included in each study might have affected the impact of ED boarding time on clinical outcomes. As an example, while 43% of the ICU patients included in our study were admitted due to respiratory conditions, only 8% were admitted due to a cardiac condition. Other studies included primarily ICU cardiac patients (up to 31%).⁴ In this group of cardiac patients, a delay in time-sensitive interventions, such as initiation of mechanical hemodynamic support or monitoring, may have affected outcomes in direct association with ED boarding time. Second, ED staffing models may have presented significant variability among studies. While general practitioners staffed some EDs, emergency medicine specialists (some of them trained in critical care medicine) staffed others. Therefore, this variability in training may have translated to variability in application of early interventions (i.e., sepsis bundle), affecting clinical outcomes. In our institution, emergency medicine and intensive care physicians collaborate in the development of protocols and standardization of practices. Hence, it is likely that ICU boarding patients may have received care similar to what they would have received in the ICU, not being significantly affected by ED boarding time. Lastly, the presence of ancillary services in the ED, such as respiratory care, pharmacy, and nutrition support, varies according to institutional practices. The presence of the aforementioned service lines in the ED allows the implementation of multiple therapies, such as advanced mechanical ventilation strategies, monitoring of drug interactions, and adequate nourishment, which could have diminished differences in outcomes between ED boarding ICU patients compared with those already admitted to the ICU. It is possible that variations in the availability of these services in EDs may have translated to differences in outcomes.

In conclusion, our study showed a lack of association between ED boarding time and ICU LOS or hospital mortality. Standardization of practices, implementation of protocols, and ED staffing models may have resulted in similar delivery of care for those ICU patients boarding in the ED compared with those already admitted to the ICU. Whether our findings can be extrapolated to other organizations remains to be studied.