Emergency department crowding predicts admission length-of-stay but not mortality in a large health system

Abstract

Background: Emergency department (ED) crowding has been identified as a major threat to public health.

Objectives: We assessed patient transit times and ED system crowding measures based on their associations with outcomes.

Research design: Retrospective cohort study.

Subjects: We accessed electronic health record data on 136,740 adults with a visit to any of 13 health system EDs from January 2008 to December 2010.

Measures: Patient transit times (waiting, evaluation and treatment, boarding) and ED system crowding [nonindex patient length-of-stay (LOS) and boarding, bed occupancy] were determined. Outcomes included individual inpatient mortality and admission LOS. Covariates included demographic characteristics, past comorbidities, severity of illness, arrival time, and admission diagnoses.

Results: No patient transit time or ED system crowding measure predicted increased mortality after control for patient characteristics. Index patient boarding time and lower bed occupancy were associated with admission LOS (based on nonoverlapping 95% CI vs. the median value). As boarding time increased from none to 14 hours, admission LOS increased an additional 6 hours. As mean occupancy decreased below the median (80% occupancy), admission LOS decreased as much as 9 hours.

Conclusions: Measures indicating crowded ED conditions were not predictive of mortality after case-mix adjustment. The first half-day of boarding added to admission LOS rather than substituted for it. Our findings support the use of boarding time as a measure of ED crowding based on robust prediction of admission LOS. Interpretation of measures based on other patient ED transit times may be limited to the timeliness of care.

FIGURE 1

A, ED LOS and inpatient mortality, not adjusted for patient characteristics. B, ED LOS and inpatient mortality, adjusted for patient characteristics. Note that these measures correspond with the Center for Medicare and Medicaid Services (CMS) ED-1b Measure. For clarity, these graphs show up to the 99.9th percentile of LOS (x-axis). The y-axis denotes the predicted probability at any given hour divided by the predicted probability at the median number of hours. All models were adjusted for ED site of care (a dummy variable for site and year). The model adjusted for patient characteristics included age strata (18–39, 40–49, 50–59, 60–69, 70–79, 80 plus), sex, race/ethnicity, and preexisting comorbidities (Elixhauser category), ambulance versus nonambulance arrival, triage blood pressure and pulse, ESI triage score, preexisting comorbidities (Elixhauser category), primary discharge diagnosis (expanded CCS multilevel categories), shift (12 am–8 am, 8am–4 pm, 4pm–12 am), weekend versus weekday, and month CCS indicates Clinical Classification Software; ED, emergency department; ESI, Emergency Severity Index; LOS, length-of-stay.

FIGURE 2

A, Boarding time and inpatient mortality, not adjusted for patient characteristics. B, Boarding time and inpatient mortality, adjusted for patient characteristics. Note that these measures correspond with the Center for Medicare and Medicaid Services (CMS) ED-2b Measure. For clarity, these graphs show up to the 99.9th percentile of boarding time (x-axis). The y-axis denotes the predicted probability at any given hour divided by the predicted probability at the median number of hours. All models were adjusted for ED site of care (a dummy variable for site and year). The model adjusted for patient characteristics included age strata (18–39, 40–49, 50–59, 60–69, 70–79, 80 plus), sex, race/ethnicity, and preexisting comorbidities (Elixhauser category), ambulance versus nonambulance arrival, triage blood pressure and pulse, ESI triage score, preexisting comorbidities (Elixhauser category), primary discharge diagnosis (expanded CCS multilevel categories), shift (12 am–8 am, 8am–4 pm, 4pm– 12 am), weekend versus weekday, and month. CCS indicates Clinical Classification Software; ED, emergency department; ESI, Emergency Severity Index.

FIGURE 3

A, Boarding time and admission LOS, adjusted for patient characteristics. B, Time-averaged occupancy and admission LOS, adjusted for patient characteristics. Note that a reference line is used to aid interpretation. As admission LOS included boarding time, an increase in admission LOS >1 hour for every hour of boarding suggests wasted time (when the curve rises above the reference line). Below the reference line, admission LOS increases <1 hour for every hour of boarding. A horizontally moving curve indicates complete substitution of inpatient time by boarding time. Graphs show up to the 99.9th percentile of each predictor (x-axis). The y-axis denotes the predicted probability at any given hour divided by the predicted probability at the median number of hours. All models were adjusted for ED site of care (a dummy variable for site and year). The model adjusted for patient characteristics included age strata (18–39, 40–49, 50–59, 60–69, 70–79, 80 plus), sex, race/ethnicity, and preexisting comorbidities (Elixhauser category), ambulance versus nonambulance arrival, triage blood pressure and pulse, ESI triage score, preexisting comorbidities (Elixhauser category), primary discharge diagnosis (expanded CCS multilevel categories), shift (12 am– 8 am, 8am–4 pm, 4pm–12 am), weekend versus weekday, and month. CCS indicates Clinical Classification Software; ED, emergency department; ESI, Emergency Severity Index; LOS, length-of-stay.