Study #1: Anticipating Crowding*
We all know what overcrowding looks like when we see it. Patients parked in the hallways, full waiting rooms, upset patients, and staff stress levels on the rise. While we can all recognize it when it hits, wouldn’t it be great for a computer model to be able to predict that we are about to become overcrowded before we even know it?
A recently published article in Annals of Emergency Medicine from Vanderbilt University Medical Center attempts to answer this question. They tested five measures to determine reliability in measuring current ED crowding and predicting near-future ambulance diversion status.
The five tested formulae were the Emergency Department Word Index (EDWIN), the National Emergency Department Overcrowding Scale (NEMDOCS), the Demand Value of the Real-time Emergency Analysis of Demand Indicators (READI), the Work Score, and a straight-forward occupancy level. The first four of these are fairly complicated, particularly as compared with the occupancy level, which is simply the total number of patients in beds, hallways, and chairs over the number of licensed treatment beds. Over an eight-week period, a computer program calculated these scores every 10 minutes. The data was then pooled and analyzed with the results compared to the presumed gold standard for crowding—ambulance diversion. The authors acknowledge that there is no perfect standard for crowding, but note that ambulance diversion at their institution is instituted based on standard criteria and, perhaps more importantly, is monitored at a hospital level as a significant marker for crowding.
All five models could be made to predict ambulance diversion with 90% sensitivity, though with specificities varying from 32% to 70%. In other words, the worst system, in order to maintain a reasonably low rate of missing crowding, anticipated diversion three times for every actual event. The best formula—and we’ll reveal that in a moment—has only one false positive prediction for every two real ones. To trigger an intervention throughout the hospital, potentially entailing calling in staff and gearing up to a new level of activity and stressfulness, the model should be robust.
So which preformed best? Surprisingly, it was the simplest. Occupancy level outperformed all the other tools in tracking current crowding and it performed equally as well in forecasting near-term crowding. Unfortunately, none of the systems performed well in predicting crowding, as all returned significant numbers of false alarms whenever trying to predict more than about an hour in advance.
The authors noted that more objective outcome variables than ambulance diversion might help to better delineate this research question in the future. Further, this was a single institution study, so results may not be externally generalizable.
Technically, the neatest trick of this study may be that the authors measured complicated crowding statistics in real time. Had the measures worked, one could picture an overhead monitor flashing: “High risk of dangerous crowding in one hour.” Unfortunately, no model was able to accurately predict future crowding in the ED.
Perhaps the success of the simplest measure—the occupancy level—should not be surprising. Despite the use of guidelines to determine ambulance diversion, the gold standard used here, one tends to think of going on diversion when the department is obviously crowded. The occupancy measure—in calculating patients in beds, chairs, and hallways—is a bit of a back-of-the-envelope measure for what one is seeing. Despite the guidelines, there is likely some unmeasured interaction.
Still, the point remains: None of the computer models could predict either current or future ED crowding better than the far simpler occupancy level. Occupancy levels can be comfortably used to track ED crowding.
Study #2: ICU Delays**
With limited resources and higher nursing ratios, it makes intuitive sense that boarding ICU patients in the ED is not optimal. Still, there has been little evidence to date that boarding carries risk for the waiting ICU patient—until now.
Donald Chalfin and colleagues examined mortality risk of ED boarding and published their results in June’s Critical Care Medicine. They did not look at boarding overall, but at the subset of boarded patients who were headed to an ICU. They noted that in large hospitals with crowded EDs, patients waited a mean of almost six hours for transfer from the ED to an acute or critical care bed.
To examine this further, they looked at data from more than 50,000 patients and compared mortality rates for those who had more than a six hour wait to move from ED to ICU with those who had under a six hour wait. They compared in-ICU mortality, in-hospital mortality, and hospital length of stay. Importantly, this was six hours after the decision to admit to the ICU had already been made.
The data came from the Project IMPACT database, a voluntary, administrative database that includes data from approximately 120 ICUs spread across 90 hospitals. The study included all patients admitted to the ICU from the ED in 2000, 2001, 2002, and 2003. The database included a question targeted at uncovering boarding—did the patient spend six or more hours in a non-ICU setting immediately prior to ICU admission?
The number of patients who boarded for six hours or more was relatively small, only 1,036 of the 50,322 patients. The two groups were similar including age, gender, and APACHE II scoring—the last suggests similar severity of illness, though the case mix differed somewhat between the groups.
The results are reasonably impressive. Mortality in the group that went to the ICU expeditiously was 8.4% versus 10.7% for the group that waited. In-hospital mortality was 12.9% versus 17.4%. Median ICU stay was similar at 1.8 versus 1.9 days, but delay to ICU also correlated with a longer hospital stay: 6.0 days versus 7.0 days. A similar pattern held for a sub-group analysis of the most common diagnostic category, sepsis.
Two major differences between groups may confound these results—delayed patients were more likely to have sepsis and to receive both central venous access and mechanical ventilation; perhaps these procedures explain some of the additional time to transfer.
The authors conclude that ED boarding of ICU patients more than six hours is associated with increased mortality and that expeditious transfer may benefit these patients.
This is a tough study to consider from an ED perspective. One major goal of residency training in emergency medicine is to develop skills in critical care. Few would argue that the best care of the ICU patient is instantaneous transfer from the ED to ICU. So, where is the line? When does emergent care become critical care become floor care?
It is difficult to determine in a comparative study like this—there is no randomization here—just what is correlation and what is causation. Perhaps the delayed patients benefited from ED care and would have suffered even higher mortality had they been transferred sooner—though the authors attempt to solve this problem by controlling for acuity and diagnostic categories.
Furthermore, the comparison is somewhat uneven. This does not appear to be a comparison between a crowded ED and a crowded ICU but rather between a crowded ED and a controlled ICU. The authors did not have ED staffing or volume numbers available to them. A more apples-to-apples comparison would be patients who went to the ICU in less than 6 hours versus those who waited for six hours during a period when the ED was not crowded.
We might consider, too, what a similar study would look like from the perspective of patients transferred to the floor. There likely is a point of diminishing returns in ED care. We expeditiously conduct a work-up, enlist primary and consulting physicians, and initiate therapy. To a point, ED care improves patient care and hospital flow—but only to a point. Crowding sets in, delays build, and our area of expertise reaches its boundaries as the patient becomes a boarder and we—and our nurses and techs and supporting personnel—move on to the next patient. Defining this point of diminishing returns would provide a benchmark for quality care.
There is a reasonable take home point here. A delay in transfer to the ICU of six hours or more appears to be associated with increased mortality. Lacking better data, a time of six hours from admission decision to transfer appears to be a reasonable benchmark.
*Measuring and Forecasting Emergency Department Crowding in Real Time
Hoot, NR, Zhou, C, Jones, I, et al. Annals of Emergency Medicine 2007; 49:747-755.
**Impact of Delayed Transfer of Critically Ill Patients from the Emergency Department to the Intensive Care Unit. Donald B. Chalfin, MD, MS, FCCM; Stephen Trzeciak, MD, MPH; Antonios Likourezos, MA, MPH; Brigitte M. Baumann, MD, MSCE; R. Phillip Dellinger, MD, FCCM; for the DELAY-ED study group. Critical Care Medicine 2007; 35:1477-1483.