Setting
Sunnybrook is a 1200-bed academic tertiary level hospital in central Toronto. It is a trauma, regional stroke, interventional cardiology, neurosurgical and oncology center. During the study period, Sunnybrook Emergency Department received 45000-patient visits per year with an admission rate of 22%.
Design
We conducted a cluster, randomized-controlled trial of the impact of a Physician-Nurse Supplementary Triage Assessment Team (MDRNSTAT) on EDLOS at Sunnybrook Emergency Department over a 26-week period (October 1, 2009 - April 1, 2010). Clusters were ED patients arriving from 8:00–14:30. During control clusters, ED patients were triaged by a standard nurse for acuity scoring, registered, and assigned an ED stretcher (Figure 1 light arrows). If no stretcher was available, patients stayed in the waiting room until one became available. Afterwards, an emergency physician would assess, order investigations, manage and determine disposition. During intervention clusters, we added a MDRNSTAT without changing time to triage (Figure 1 - bold arrows). Both ambulatory and ambulance patients could be seen by the MDRNSTAT. After being assigned an acuity score by the standard triage nurse, the patient was assessed by the MDRNSTAT physician in a room behind the triage bay where orders were administered instead of waiting for a stretcher in the ED. The MDRNSTAT nurse completed lab and medication orders; whereas, the radiology service completed diagnostic imaging orders. The MDRNSTAT nurse could assist triage or emergency if required. The MDRNSTAT physician could also request consults and, possibly, discharge the patient. If not discharged, the patient would return to the waiting room to wait for a stretcher where the usual ED physician would take over and follow up on the results or consultations. If an ED stretcher was immediately available, patients bypassed the MDRNSTAT and were seen by the usual ED physician.
Inclusion and exclusion criteria
We included weekday ED patients arriving from 8:00–14:30. We excluded critically ill (CTAS 1) patients requiring immediate resuscitation (i.e. cardiac arrest) since any delays for MDRNSTAT assessment at triage would be unethical and harmful. Patients directed (“directs”) to the ED for an admitting service (eg. oncology) assessment and not an emergency physician, were also excluded from the study.
Randomization and allocation
Randomization was conducted centrally, before study initiation, using a computer-generated algorithm by the study statistician. The unit of allocation was the ED shift. The randomization schedule ensured 2 control days between interventions to minimize any carryover MDRNSTAT effect. To anticipate predictable surges in the number of ED patients arriving at 11:00, the MDRNSTAT was assigned on weekdays from 8:00–14:30. Our rationale for this timing was to prevent queuing rather than reacting to increased waiting times.
In the daytime weekday study period, there were 65 MDRNSTAT shifts divided amongst 14 physicians and 14 nurses who were currently on staff. There were 66 usual nurse-triage alone (control) shifts. Physicians and nurses voluntarily signed up for paid MDRNSTAT shifts in addition to their regularly scheduled ED shifts. The MDRNSTAT added one physician and one nurse to the 3 physicians and 15–19 nurses typically working each dayshift.
The usual emergency physicians were instructed to not work at triage. The MDRNSTAT could not be blinded to staff; however, patients were unaware of the MDRNSTAT because its schedule was not publicized. Time recordings of primary outcomes were uninfluenced since they were automatically collected by a computerized patient-tracking emergency department information system.
Data collection procedure
Time data, CTAS levels, and patient disposition were retrieved from two different computerized information systems: Electronic Patient Records and Emergency Department Information Systems. Electronic Patient Records collected diagnostic imaging and laboratory order times. The Emergency Department Information System collected the rest of the data. Ethics approval was obtained from the Sunnybrook Hospital Research Ethics Board (August 2009). The requirement for formal informed consent from patients was waived.
Outcome measures
The primary cluster outcome measure was median EDLOS for discharged (non-admitted) patients directly seen by emergency physicians (MDRNSTAT or usual ED physician) with no consultation (“non-consulted”) because emergency department interventions could not influence consultation duration or hospital bed availability for admissions. According to the provincial thresholds, EDLOS was the interval between triage time to ED discharge. We analyzed EDLOS according to patient acuity, disposition, and consultation.
Secondary outcome measures were: triage time to start of emergency physician or assessment, EDLOS among patients seen by emergency physician and referred for consultation/admission or left-without-being seen, percentage of CTAS 2–5 patients who were seen within Ontario’s Pay for Results wait-time target thresholds, and left-without-being-seen rate, a widely used measure of safety and satisfaction. [2, 5, 27, 28] Other secondary outcome measures were triage time to: laboratory, diagnostic imaging, consultation, and bed request order time.
To measure unintended harm, we searched our database for patients who returned to our hospital within 48 hours after being discharged from triage by MDRNSTAT. These patients’ charts were reviewed (IC) for a change in management, admission or death. Management change occurred if diagnosis or treatment was different between the first and second visit. No time parameters or external hospital data could be collected for patients who left-without-being-seen. All outcome measures were assigned a priori.
Sample size and statistical analysis
We estimated that a sample size of 50 clusters (shifts) per group, with 32 individuals per cluster, would have greater than 90% power to detect a difference of 30 minutes between the group medians. We used 30 minutes as a clinically significant reduction, because previous trials determined this difference was associated with decreased 7-day mortality, re-admission of discharged patients [4] and left-without-being seen rate [27]. The Ontario Ministry of Health’s target was a 10% reduction [14] of Sunnybrook’s baseline physician initial assessment time of 324 minutes, which also equals 30 minutes. We used a standard deviation of 70 minutes, intra-cluster correlation of 0.1 [29] using a two-sided t-test, and a significance level of 0.05.
The analyses were carried out comparing patients in groups, control versus treatment, adjusting for the clusters (shifts) in which the patients appeared. This resulted in 66 control clusters (for a total of 3163 visits) compared to 65 intervention clusters (for a total of 3137 visits) and therefore the visits were not treated as independent observations for analysis purposes. The number of clusters was increased to 65 in order to compensate for the variable number of individuals per cluster.
Descriptive statistics were calculated for variables of interest with categorical measures summarized using counts and percentages. The primary outcome, EDLOS, as well as the other secondary time variables were summarized using medians. For analysis purposes, the outcomes were log transformed and the mean group differences of EP’s, MDRNSTAT, the combination of the two and the control group were compared using linear models adjusting for the correlation among observation taken from the same cluster (shift). Confidence intervals for medians were determined with 1000 bootstrap simulations. The 2.5th and 97.5th percentile respectively were reported as the 95% confidence interval. For variables with no events, such as rate of harm we calculated one-sided 95% confidence intervals using the Hanley estimate of 3/n where n is the total sample size of interest [30]. All analyses were carried out using SAS Version 9.1 (SAS Institute, Cary, NC, USA).