INTRODUCTION
Emergency department (ED) crowding occurs when demand for emergency services exceeds care capacity. When EDs cannot deliver timely and patient-centered care, quality of care suffers.Reference Derlet and Richards 1 - Reference Bond, Ospina and Blitz 3 The negative impacts of ED crowding are well known, including prolonged wait times,Reference Derlet and Richards 1 , Reference Bond, Ospina and Blitz 3 - 5 increased hospital admission rates,Reference Forster, Stiell and Wells 6 increased health-care costs,Reference Krochmal and Riley 7 and mortality.Reference Bernstein, Aronsky and Duseja 8 - Reference Johnson and Winkelman 10 The problem of crowding is prevalent internationally and continues to be recognized as a health care priority.Reference Bernstein, Aronsky and Duseja 8 , Reference Forero, Hillman and McCarthy 11 - Reference Hughes 13 In Ontario, the Pay-for-Results program was established in 2008 to provide financial incentives for improving measures of ED quality and efficiency. 14 Our hospital, Southlake Regional Health Centre, started participating in the Pay-for-Results program in 2008.
Interventions to decrease incidence of ED crowding are well known and have been shown to decrease wait times to varying degrees.Reference Wiler, Gentle and Halfpenny 15 - Reference Liu, Hamedani and Brown 23 However, many of these interventions involved adding additional ED resources.Reference Wiler, Gentle and Halfpenny 15 , Reference Pines, Pilgrim and Schneider 19 , Reference Handel, Epstein and Khare 21 , Reference DeFlitch, Geeting and Paz 24 - Reference Han, France and Levin 28 We sought to improve wait times through process improvements (i.e., without new staff, ED renovation, or operational costs). After re-evaluating the typical sequence of front-end operations,Reference Wiler, Gentle and Halfpenny 15 we hypothesized that a multifaceted process of reorganization designed to target triage, patient inflow, and ED staffing patterns would lead to significant improvements in patient flow and decrease wait times.
The objective of this study was to examine the impact of simultaneous process improvements on length of stay (LOS) according to patient acuity and disposition. Secondary objectives included examining the impact of the intervention on physician-initial-assessment (PIA) times, left-without-being-seen (LWBS) rates, and left-against-medical-advice (LAMA) rates.
METHODS
Setting and population
Southlake Regional Health Centre (Southlake) is a 400-bed community hospital in Newmarket, Ontario. It provides regional tertiary care programs, including cardiac, child and adolescent mental health, oncology, and thoracic surgery. From January 1, 2010 to December 31, 2012, the ED received over 251,000 patients, with an admission rate of 11.2%. We conducted a retrospective 36-month pre- and post-intervention study to examine the impact of our process improvements. The Southlake Research Ethics Board provided ethics approval.
Design
Process improvements without new staff, ED renovation, or operational cost were implemented at Southlake on June 6, 2011. The overall intervention comprised 1) streamlining triage, 2) implementing parallel processes, 3) flexible nurse-patient ratios, 4) flexible exam spaces, and 5) flexible physician scheduling. Each process will be described briefly. Further details and rationale behind the design have been described in the Supplementary Material, and elsewhere.Reference Whatley, Leung and Duic 29 , Reference Whatley 30 During the month prior to the intervention start-date, meetings between physicians and other ED health care providers were held to describe and review what impact the intervention would have on the ED, and to address any questions. Formal training was not required.
Streamlining triage
Triage remained the first step for all arriving patients. Nurses determined the presenting complaint, assigned a Canadian Triage and Acuity Scale (CTAS) level, and determined the most appropriate location for care within the ED. However, all patients were immediately accepted inside the ED, and there was no “waiting room” at the front of the ED. Accordingly, triage nurses were re-deployed into the ED to provide direct patient care, and subsequent ED processes occurred in parallel.Reference Whatley, Leung and Duic 29
Parallel processes
In the traditional ED, patient flow followed a sequential process.Reference Wiler, Gentle and Halfpenny 15 , Reference Morgan 31 In contrast, parallel processing would allow multiple processes to occur simultaneously to reduce waits and bottlenecks.Reference Wiler, Gentle and Halfpenny 15 For example, this occurs when CTAS 1 (resuscitative) patients arrive. Physicians and nurses immediately converge to provide care, and registration occurs later. We applied this principle to all patients, regardless of acuity, and eliminated sequential processes which lead to bottlenecks.Reference DeFlitch, Geeting and Paz 24 There was no restriction on when registration occurred or which health care provider that a patient saw first.Reference Whatley, Leung and Duic 29
Flexible exam spaces
Stable patients without threat to life or limb comprise the majority of patients in the ED.Reference Saunders 32 , 33 This was the case at our institution, and we believed that these patients could sit inside the ED. With flexible exam spaces, health care providers called these patients into nearby exam rooms. Patients were in exam rooms only if required for examination. This facilitated a rapid turnover of rooms and maintained patient flow. Patients moved to chairs for IV treatments, or waited for imaging and test results. As a result, non-acute patients collected in areas that in practice could hold a large number of patients.Reference Whatley, Leung and Duic 29
Flexible nurse-patient ratios
There was no fixed nurse-to-patient ratio in the care areas described. Because patients were always accepted into the ED after triage, nurse-to-patient ratio could rise above the traditional 4:1 or 5:1.Reference Chan, Killeen and Kelly 34 By having flexible nursing teams without rigid assignments, resources could be moved to areas with the highest demand.Reference Whatley, Leung and Duic 29
Flexible demand-based scheduling
Rigid physician scheduling with defined start and stop hours was changed to a meet unscheduled and fluctuating demand. This demand-based scheduling incorporated the efficiency of each physician, and estimated patient-volume per day to provide approximate shift start times. Prior to the shift, physicians would call and speak to the charge ED physician to negotiate start times based on current wait times. This method prevented overstaffing the ED and did not require the addition of any physician shifts.Reference Whatley, Leung and Duic 29
Outcome measures
ED performance measures according to provincial standards and the literature were not identical. 14 Therefore, we summarized ED performance measures in multiple ways. The primary outcome measures were mean and 90th-percentile LOS for CTAS 2-5 patients, and mean LOS for non-admitted (i.e., discharged) patients according to acuity. LOS was calculated as the difference between the earlier of triage or registration time and the time that the patient physically left the ED or was admitted and physically left the ED.
Secondary outcome measures included 90th-percentile PIA, mean PIA, LWBS, and LAMA. PIA was calculated as the difference between the earlier of triage or registration time to the start of emergency physician assessment. LWBS, a common measure of safety and satisfaction, was calculated as a percentage of daily ED volume.Reference Mohsin, Forero and Ieraci 35 We also studied LAMA patients, because they have been associated with increased re-admission rates and risk of emergent hospitalization.Reference Ding, Jung and Kirsch 36 , Reference Jerrard and Chasm 37 LAMA patients were calculated as a percentage of daily ED volume.
Data extraction
The study used a computerized ED information system, which integrates patient tracking and patient charted data (STAR McKesson, McKesson, San Francisco, CA, and Med2020, MED2020 Health Care Software Inc., Orleans, ON). Database queries with Microsoft Access (Microsoft Corporation, Redmond, WA, USA) collected patient demographic information, clinical data, patient flow time stamps, and disposition for all patients received in the ED from January 1, 2010 to December 31, 2012.
The following patients were collected but excluded from regression analysis: critically ill or required resuscitation (CTAS 1), died or were dead on arrival in the ED, direct admissions to a service, and transferred to another facility. These patients were excluded because they either bypass or are accelerated through the usual ED intake processes, and would not be impacted by our intervention. If they were included in aggregated metrics, it would artificially decrease wait times. Furthermore, we knew that the excluded patients usually do not represent a substantial burden to ED throughput at our institution. Administrative records of ED physician staffing schedules were used to determine the total number of physician hours per day, including the activation of any ED backup physicians. For each patient, LOS and PIA were calculated. These values were used to produce monthly values for each metric and determine the primary and secondary outcome measures.
Statistical analysis
We divided our data into pre-intervention and post-intervention intervals starting on June 6, 2011. T-tests were used to compare descriptive statistics for patients before and after the intervention. Segmented regression of interrupted time series (ITS) analysis was performed to assess whether our intervention had an impact on primary and secondary outcomes immediately, and over time. The outcomes of ITS analysis were changes in the outcome level immediately after the intervention (step-change), and changes in the pre-intervention and post-intervention slopes (trend-change) to determine whether the impact of the intervention was sustained with time.Reference Wagner, Soumerai and Zhang 38 ITS analysis was performed on secondary data sets generated by bootstrap simulations, and to calculate standard error and confidence intervals. Additional details regarding the statistical analysis are provided in the Supplementary Material.
The level of statistical significance was α=0.05. A Bonferroni correction was made to control for type 1 error inflation. Missing data were excluded from statistical analysis. ITS analysis was performed using R version 3.2.1 (R Foundation for Statistical Computing, Vienna, Austria). All other analyses were performed using IBM SPSS version 20.0 (IBM Corp., Armonk, NY, USA).
RESULTS
The ED received 251,899 patients during the 36-month study period. The following met exclusion criteria: 1468 CTAS 1 patients, 621 transfers to another facility, 329 direct admissions, and 263 dead on arrival or expired in the ED. During the post-intervention period, there was no clinically meaningful change in the mean age or daily proportion of patients assigned CTAS 2 to 5 acuity scores. There was a significant increase in daily patient volume (210.66 v. 247.20, p<0.003), and mean number of patients to physicians (32.33 v. 37.78, p<0.003). No other descriptors differed significantly. The final sample included 249,258 patients over 521 days pre-intervention, and 575 days post-intervention (Table 1).
CTAS=Canadian Triage and Acuity Scale; ED=emergency department; SD=standard deviation.
* Statistical significance at p<0.003 level after Bonferroni correction for multiple comparisons.
Prior to the intervention, there were significant trends for overall mean LOS (-0.011 hours/month, p<0.005), and 90th-percentile LOS (-0.033 hours/month, p<0.005) (see Supplementary Material). After the implementation of process improvements, there were step-changes or reductions of 0.64 hours in mean LOS (p<0.005), and 0.81 hours in 90th-percentile LOS (p<0.005) for CTAS 2 to 5 patients. When separated by acuity and disposition, there were significant reductions for mean non-admitted LOS for CTAS 2 to 4 patients (Figure 1). After Bonferroni correction, there was no significant reduction for non-admitted CTAS 5 patients. There were small and significant increases in trend of 0.012 hours/month (p<0.005) for non-admitted CTAS 2, and 0.014 hours/month (p<0.005) for non-admitted CTAS 3 patients (Table 2). The resulting post-intervention segment trend for non-admitted CTAS 2 and 3 patients were 0.003 hours/month and 0.002 hours/month, respectively (see Supplementary Material). These findings were similar to the results from a separate analysis that examined patients separated by acuity only (data not shown).
CI=confidence interval; CTAS=Canadian Triage and Acuity Scale.
* Statistically significant at p<0.005 level after Bonferroni correction for multiple comparisons.
† Statistically significant at p<0.05 level.
For secondary outcomes, there were significant reductions in mean PIA for CTAS 2 to 5 patients by 43.81 minutes (p<0.005, Figure 2), and 90th-percentile PIA by 91.39 minutes (p<0.005). LWBS and LAMA patients decreased by 1.27% (p<0.005) and 0.91% (p<0.005) of daily patient volume, respectively (Figure 3). This corresponds to a 35.3% and 61.9% decrease from pre-intervention LWBS and LAMA patients, respectively. There were no significant trend changes for secondary outcomes. The full results of ITS analysis are provided in the Supplementary Material.
DISCUSSION
This study shows that an intervention comprising simultaneous ED process improvements can significantly improve overall measures of ED quality and efficiency. This was achieved without cost or additional resources, and sustained despite increases in daily patient volumes by 17.3%. After the process improvements, the average ED physician was more productive and attended to 15% more patients each shift.
For CTAS 2 to 5 patients, we observed a 38.4 minutes (11.1%) reduction in mean LOS, and 43.8 minutes (36.2%) reduction in mean PIA immediately after our intervention was implemented. When patients were separated by acuity and disposition, there were reductions in LOS up to 15.5% of pre-intervention LOS times. For non-admitted CTAS 2 and 3 patients, the reductions in LOS exceeded the 30-minute reduction threshold for clinical significance, a level which has been associated with decreased 7-day mortality,Reference Krochmal and Riley 7 re-admission of discharged patients,Reference Krochmal and Riley 7 and LWBS.Reference Mohsin, Forero and Ieraci 35 For secondary outcomes, the reduction in 90th-percentile PIA times by over 90 minutes (35.7%) is clinically significant because serious illnesses presenting to the ED, such as myocardial ischemia, require efficient diagnosis and timely care.Reference Schull, Hatcher and Guttmann 39 LWBS, commonly associated with prolonged wait times,Reference Kennedy, MacBean and Brand 40 decreased by 35.2%. The post-intervention mean LWBS was 0.91% of daily volume, and below the provincial mean of 3%.Reference Baibergenova, Leeb and Jokovic 41
We did not expect to find clinically significant trend-changes in ED quality metrics. The process improvements were one-time changes meant to improve and maintain patient flow. As a result of the intervention, there were statistically significant increases in trend for LOS. However, the trend-changes and resulting post-intervention trends were small (<1 minute/month). We consider them clinically insignificant and likely caused by increased patient volumes. The absence of clinically significant upward trend-changes indicated that the clinical benefits of our process improvements were sustained during the post-intervention period.
Our results are comparable to similar interventions affecting multiple aspects of ED organization. A multifaceted redesign by Spaite et al. (2002) resulted in LOS reductions of 27% for admitted patients and 31% for non-admitted patients.Reference Spaite, Bartholomeaux and Guisto 42 Chan et al. (2005) studied a rapid entry and accelerated care at triage process that facilitated immediate bedding and accelerated test ordering, which resulted in a less than 9% (31 minutes) reduction in LOS and 58.7% reduction in LWBS.Reference Chan, Killeen and Kelly 43 However, both of the models cost over $1 million to implement.Reference Spaite, Bartholomeaux and Guisto 42 , Reference Chan, Killeen and Kelly 43 Ng et al. (2007) applied Lean manufacturing principles to improve the flow of patients throughout the ED. There were reductions in overall LOS for discharged patients from 3.6 to 2.8 hours (22.2%). However, the improvements were mainly limited to low-acuity (CTAS 4 and 5) patients. Mean PIA times decreased from 111 to 78 minutes (29.7%), and mean LWBS from 7.1% to 4.3% (39.4%).Reference Ng, Vail and Thomas 25 Our study demonstrates how clinically significant changes impacting the wait times of higher and lower-acuity patients could be achieved without massive expenditures.
Our intervention did not require the fixed reallocation of physician or nursing resources, a common occurrence in interventions that institute a dedicated area for low and/or medium-acuity patients. These studies have reported reductions in wait times, LOS, and LWBS.Reference Wiler, Gentle and Halfpenny 15 However, a mid-track intervention for medium-acuity patients by Soremekun et al. (2013) found that siphoning off medium-acuity patients and assigning physician resources from the main ED to the mid-track area contributed to increases in the LOS of high-acuity patients.Reference Soremekun, Shofer and Grasso 44
Streamlining triage and accepting all triaged patients into the ED reduced triage complexity. This released triage nurses from triage to provide direct patient care within the ED. By having patients wait inside of the ED, health care providers immediately noticed any increase in patient volume and were able to respond faster to mid-acuity patients who deteriorated while waiting. In addition, traditionally sequential ED processes that occurred in waiting rooms could unfold in parallel within the ED. This approach was distinct from “immediate bedding and bedside registration” interventions which have been implemented only when the ED is not at capacity (i.e., “when possible”).Reference Morgan 31 , Reference Spaite, Bartholomeaux and Guisto 42 , Reference Chan, Killeen and Kelly 43 Therefore, despite the significant costs to implement and maintain immediate bedding and bedside registration interventions,Reference Spaite, Bartholomeaux and Guisto 42 , Reference Chan, Killeen and Kelly 43 the impact on ED crowding was likely minimal.Reference Wiler, Gentle and Halfpenny 15 Similarly, our concept of simplifying triage went against other operational interventions, which expanded triage, such as physicians-in-triage. Although physician-in-triage interventions have been associated with reductions in overall LOS by 11 to 56 minutes, they required up to 16 hours/day of additional physician coverage, and increases in nursing hours.Reference Holroyd, Bullard and Latoszek 26 - Reference Han, France and Levin 28 This led us to question whether the resources could have been better utilized by allowing them to flexibly operate within the ED and respond to fluctuations in patient volumes. In fact, some physicians-in-triage trials were terminated if the main ED was overwhelmed and the triage physician was required for patient care within the department.Reference Chan, Killeen and Kelly 43 , Reference Subash, Dunn and McNicholl 45
We recognized that health care providers needed to be flexible to meet fluctuations in ED demand throughout the day. This involved deploying nurses as flexible teams that did not adhere to traditional ratios of nurses to patients.Reference Chan, Killeen and Kelly 34 Likewise, abandoning rigid clock-based scheduling for demand-based scheduling allowed physicians to adapt to changing patient volumes without overstaffing. By using real-time data to make decisions, we felt that it empowered health care professionals to manage patient flow in real-time. The changes empowered frontline staff to take responsibility for managing demand for ED services. To our knowledge, this the first study to describe modifications on physician scheduling as an intervention on patient flow. Indeed, as opposed to creating an additional layer of ED complexity through rigid implementation of novel systems during hours of peak ED demand, our study has shown that clinically significant reductions in LOS and PIA times could be achieved by providing health care providers the flexibility to become responsive to patient demand for timely assessments. 5
The idea of flexible exam spaces is similar to the more recent concept of “vertical patient flow.” Vertical patient flow eliminates the unavailability of ED beds as a key bottleneck in patient flow by having patients sit in upright chairs while awaiting treatments or tests.Reference Liu, Hamedani and Brown 23 In contrast to other studies that limited the idea of being “vertical” to lower-acuity patients,Reference Liu, Hamedani and Brown 23 , Reference Soremekun, Shofer and Grasso 44 we applied it to all patients who were clinically stable and ambulatory. Our intervention of having internal waiting areas leverage upon this idea, thereby reserving ED beds for clinically appropriate cases only.
Because our process improvements were made without additional investments or significant re-organization, the implementation of these changes did not require institution-wide support or political cooperation from other services. This is in contrast to other operational interventions that increased ED process complexity.Reference Wiler, Gentle and Halfpenny 15 , Reference Handel, Epstein and Khare 21 , Reference DeFlitch, Geeting and Paz 24 - Reference Han, France and Levin 28 Moreover, research on the cost-benefit of interventions is universally lacking.Reference Wiler, Gentle and Halfpenny 15 , Reference Oredsson, Jonsson and Rognes 16 , Reference Handel, Epstein and Khare 21 Although these factors may limit the generalizability and implementation of other interventions,Reference Wiler, Gentle and Halfpenny 15 , Reference Handel, Epstein and Khare 21 they do not apply to our study.
LIMITATIONS
Prior to the intervention, our ED hosted two learning events on Lean principles, but there were no attempts to formally implement a Lean intervention. During the first month of the intervention, all ED staff took time to adapt to the new system. Nonetheless, we were fortunate for the lack of resistance. This was attributed to the preparatory meetings and the belief that our changes would improve quality of care. The generalizability of our results is limited by the single-site, non-randomized, and retrospective study design. Nevertheless, it was impractical to host a randomized controlled trial in a complex ED setting where improving clinical outcomes is the priority.Reference Asplin and Yealy 46 From available 2013-2014 data, although the ED admission rates for Canadian hospitals (est. 10.0%) and ≥200-bed “large” community hospitals in Ontario and Canada (11.5% and 11.9%, respectively) 47 were similar to our institution’s rate of 10.9% that year, our study intervention and findings may not apply to Canadian<200-bed community hospitals and teaching hospitals with markedly different ED admission rates (<8.1% and 14.3%, respectively) 47 and organization. We were not able to isolate the impact of individual components within our intervention. During the study period, we did not exclude the potential of seasonal trends and other unmeasured changes as confounding factors. We did not measure other metrics of ED efficiency, including 72-hour returns of discharged patients, patient satisfaction, or patients returning to a different hospital. Although the study occurred in a fee-for-service remuneration paradigm, we believe that our model could be applied to institutions under alternative funding models.
A combination of process improvements meant to optimize flow in the ED without the addition of resources was associated with clinically significant reductions in LOS, PIA, LWBS, and LAMA for non-resuscitative patients. Future research is required to determine how components from various initiatives to address ED crowding can be integrated.
Acknowledgements
The authors acknowledge and thank Dr. Grant Innis for his contributions to the manuscript.
Competing interests: None declared.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/cem.2016.382.