The Role of Triage Liaison Physicians on Mitigating Overcrowding in Emergency Departments: A Systematic Review

Authors

  • Brian H. Rowe MD, MSc, CCFP(EM), FCCP,

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author
  • Xiaoyan Guo MSc,

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author
  • Cristina Villa-Roel MD, MSc,

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author
  • Michael Schull MD, MSc, FRCPC,

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author
  • Brian Holroyd MD, FRCPC,

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author
  • Michael Bullard MD, CFPC(EM), FRCPC,

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author
  • Benjamin Vandermeer MSc,

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author
  • Maria Ospina MSc,

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author
  • Grant Innes MD, FRCPC

    1. From the Department of Emergency Medicine (BHR, XG, CV, BH, MB), School of Public Health (BHR, CV, MO), University of Alberta, Edmonton, Alberta; the Division of Emergency Medicine Department Medicine, University of Toronto (MS), Toronto, Ontario; the University of Alberta/Alberta Health Services Evidence-based Practice Centre (BV), Edmonton, Alberta; the Institute of Health Economics (MO), Edmonton, Alberta; and the Department of Emergency Medicine, University of Calgary (GI), Calgary, Alberta, Canada.
    Search for more papers by this author

  • This study was funded by a grant from the Canadian Institutes for Health Research (CIHR; 200810KRS). Dr. Schull is supported by the CIHR as an Applied Chair in Health Services and Policy Research (Ottawa, ON). Dr. Rowe is supported by the 21st Century Canada Research Chairs program through the Government of Canada (Ottawa, ON).

  • Presented at the Canadian Association of Emergency Physicians annual scientific meeting, Montreal, Quebec, Canada, May 29–June 2, 2010.

  • Supervising Editor: Alan Jones, MD.

Address for correspondence: Brian H. Rowe, MD, MSc, CCFP(EM), FCCP; e-mail: brian.rowe@ualberta.ca.

Abstract

ACADEMIC EMERGENCY MEDICINE 2011; 18:111–120 © 2011 by the Society for Academic Emergency Medicine

Abstract

Objectives:  The objective was to examine the effectiveness of triage liaison physicians (TLPs) on mitigating the effects of emergency department (ED) overcrowding.

Methods:  Electronic databases (Cochrane Central Register of Controlled Trials, MEDLINE, EMBASE, Web of Science, HealthSTAR, Dissertation Abstracts, and ABI/INFORM Global), controlled trial registry websites, conference proceedings, study references, contact with experts in the field, and correspondence with authors were used to identify potentially relevant TLP studies. Intervention studies in which a TLP was used to influence ED overcrowding metrics (length of stay [LOS] in minutes, physician initial assessment [PIA], and left without being seen [LWBS]) were included in the review. Two reviewers independently conducted data extraction and assessed the citation relevance, inclusion, and study quality. For continuous outcomes, weighted mean differences (WMD) were calculated and reported with corresponding 95% confidence intervals (CIs). For dichotomous variables, individual and pooled statistics were calculated as relative risk (RR) with 95% CI.

Results:  From 14,446 potentially relevant studies, 28 were included in the systematic review. Thirteen were journal publications, 12 were abstracts, and three were Web-based articles. Most studies employed before–after designs; 23 of the 28 studies were considered of weak quality. Based on the statistical pooling of data from two randomized controlled trials (RCTs), TLP resulted in shorter ED LOS compared to nurse-led triage (WMD = −36.85 min; 95% CI = −51.11 to –22.58). One of these RCTs showed a significant reduction in the PIA associated to TLP presence (WMD = −30.00 min; 95% CI = −56.91 to –3.09); the other RCT showed no change in LWBS due to a CI that included unity (RR = 0.82; 95% CI = 0.67 to 1.00).

Conclusions:  While the evidence summarized here suggests that to have a TLP is an effective intervention to mitigate the effects of ED overcrowding, due to the weak research methods identified, more research is required before its widespread implementation.

Emergency department (ED) overcrowding is a complex and challenging issue facing health care systems worldwide.1 The importance of ED overcrowding cannot be under estimated, because it can lead to delays in time-sensitive diagnostic and treatment decisions, patients leaving without completion of care, patient and provider dissatisfaction, and poor health-related outcomes.2–4

The cause of ED overcrowding is multifactorial; however, it is generally considered to be a combination of input, throughput, and output stressors.5,6 Interventions have been designed to alleviate ED overcrowding at every level: input (e.g., ambulance diversion), throughput (e.g., rapid assessment zones/pods, clinical decision units), and output (e.g., full-capacity protocols, bed managers). While these last two are interventions designed to improve output for admitted patients, this is operationally outside of ED control and therefore requires systemwide interventions. One key area of delay is patient flow within the ED, or throughput. This is the period from patient arrival at the ED (either at triage or registration) to hospital admission or discharge home.1 A previous systematic review found that the majority of studies on ED overcrowding involved interventions directed to improve throughput.2 Recently, triage liaison physicians (TLPs) have been employed in several settings to address throughput factors that contribute to ED overcrowding.1 In the TLP role, physicians work with triage (a system of sorting patients based on acuity and risk) staff to expedite the care of patients, based on medical need, who are subject to unpredictable wait times due to lack of available ED treatment spaces.

To the best of our knowledge, no systematic review has specifically evaluated the effectiveness of a TLP on mitigating the effects of ED overcrowding. Such a review is crucial for policy makers and ED administrators, as it will summarize evidence on markers of timely access to health care, answer an important clinical question, and identify further directions for the research agenda.

The objectives of this study were to search for the available evidence on the effectiveness of a TLP to reduce ED overcrowding, to assess the quality of the research performed in this field, and to summarize ED overcrowding metrics used to evaluate this intervention.

Methods

Study Design

This was a systematic literature review, with a defined search strategy, study selection criteria, quality assessment and data extraction procedures, and analysis of the study results.

Search Strategy

A comprehensive literature search was conducted in seven biomedical electronic databases: Medline, Embase, EMB Reviews–Cochrane Central Register of Controlled Trials, HealthSTAR, Science Citation Index Expanded, Dissertation Abstracts, and ABI/INFORM Global. We used a vast number of keywords to identify relevant literature (see Data Supplement S1, available as supporting information in the online version of this paper). Clinical trial registries (ClinicalTrials.gov and controlled-trials.com) and Google Scholar Web searches were also explored. The search strategy developed for the Canadian Agency for Drugs and Technologies in Health (CADTH) report entitled “Interventions to reduce overcrowding in emergency departments” was updated for this study.2 The original CADTH report involved literature searches from 1966 to December 2005; we searched for citations between October 2004 and May 2009, without restrictions on language or publication status. The overlapping was necessary to ensure that all the new literature indexed after the CADTH report would be considered for inclusion.

Hand searches were performed to identify abstracts presented at the following major scientific conferences between October 2004 and May 2009: the American College of Emergency Physicians (ACEP), the Australasian College for Emergency Medicine (ACEM), the Canadian Association of Emergency Physicians (CAEP), the College of Emergency Medicine in the UK (CEM), and the Society of Academic Emergency Medicine (SAEM). In addition, the references of identified articles were manually searched. Primary authors and experts in the field were contacted to identify additional published, unpublished, or ongoing studies. The search results from the CADTH report were merged with those of the updated searches, resulting in a comprehensive search strategy to identify potentially relevant studies published from 1966 to May 2009.

Study Selection

Eligible studies were primary research that assessed the effect of a TLP to mitigate the effects of overcrowding in EDs serving adult (17 years or older) or mixed (i.e., child and adult) populations. Studies with one of the following designs were considered for inclusion in the review: parallel or cluster group randomized controlled trials (RCTs/cRCTs), controlled clinical trials (CCTs), prospective or retrospective analytical cohort studies, interrupted time series (ITS), case–control studies, and before–after designs. Studies were required to report numeric data on at least one of the following outcomes: ED length of stay ([LOS]; time from patient arrival/triage to physically leaving the ED), physician initial assessment (PIA; time from patient arrival to be seen by the emergency physician), proportion of patients leaving the ED without being seen (LWBS), and leaving the ED against medical advice (LAMA). While in many settings PIA would reflect the initial contact by physicians, advanced practice registered nurses, nurse practitioners, and/or physician assistants, as the TLP is largely a physician-based intervention, PIA seemed an appropriate outcome to report for this literature. Nonprimary research (e.g., editorials, commentaries, letters to the editor, narrative reviews, technology reports, and systematic reviews), studies conducted in pediatric EDs, multiple publications, and studies comparing two levels of the same intervention were excluded.

Three reviewers (BR, MO, XG) independently screened titles and abstracts of studies identified by the literature search. The full-text versions of articles deemed potentially relevant, as well as those that reported insufficient information to determine eligibility, were independently reviewed by two of four reviewers (BR, MO, CV, LW). Any disagreements were resolved by consensus. Non-English literature was translated by foreign language reviewers (SMH and DSF). Studies that met all inclusion criteria were eligible for quality assessment and data extraction.

Quality Assessment

A standard quality-rating tool developed by the Effective Public Health Practice Project (EPHPP)7 was used to appraise the quality of the evidence. This tool is based on guidelines set out by Mulrow and Oxman8 and Jadad et al.9 and has accepted validity and reliability.10 The tool rating is based on six criteria: selection bias, study design, confounders, blinding, data collection methods, and withdrawals and dropouts. Each criterion is rated as “strong,”“moderate,” or “weak” depending on information reported in the article. Once the ratings of characteristics are totaled, each study receives an overall assessment of strong, moderate, or weak quality based on nonnumeric criteria set out by the developers. Two assessors (BR, XG) independently rated the quality of included studies. The kappa statistic was calculated to measure the level of agreement between reviewers.11 Finally, discrepancies were resolved by consensus.

Data Extraction

Information regarding the study design and methods (e.g., year, country of origin, type of publication, study duration, number of participating centers), intervention characteristics and comparison groups, and measures and outcomes of interest (LOS, LWBS, PIA, etc.) was extracted using a pretested data extraction form. Finally, information was collected on study conclusions, as reported by the authors of the primary studies. Two reviewers (BR, XG) independently extracted these data. Any discrepancies in data extraction were resolved by consensus. If necessary, expansion of graphic representations of data from the manuscripts was used to obtain estimates of outcomes. Attempts were made to communicate with investigators for clarification or additional data.

Data Analysis

Characteristics of the included studies were summarized using descriptive statistics. Evidence tables were constructed to report information on each article’s source, study design, study population, treatment groups, and outcomes. Analyses were focused on ED LOS, PIA, and proportion of LWBS.

Meta-analyses were planned as part of the data analysis to derive pooled estimates from individual studies to support inferences regarding the effectiveness of TLPs. When studies were sufficiently similar after consideration of heterogeneity, effect sizes were combined and weighted using the Mantel-Haenszel variance method to produce an overall effect size for a given common outcome of interest. Meta-analyses used a random-effects model. The types of summary statistics considered in the meta-analyses were risk ratios (RRs) with 95% confidence intervals (95% CIs) for dichotomous outcomes and weighted mean differences (WMDs) with 95% CIs for continuous outcomes. Heterogeneity was tested using the chi-squared statistic and quantified using the I-squared (I2) statistic; I2 values of 25, 50, and 75% represent low, moderate, and high degrees of heterogeneity, respectively.12

If meta-analytic methods were not feasible, effect size estimates with corresponding 95% CIs were presented separately for each study. The median and interquartile range (IQR) improvement (by type of study) were also calculated for each of the study outcomes. Occasionally studies did not report standard deviations (SDs) for their estimates or had other missing data. In these cases, we determined the SD exactly from CIs or exact p-values, IQRs, or imputed values from other studies reporting similar outcomes in a similar population.13

A p-value of less than 0.05 was considered statistically significant in the analysis of data. All data were entered into Review Manager (RevMan, Version 5.0, The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark, 2008). Publication bias, or the selective publication of research depending on the results, was assessed using funnel plots.

Results

Search Results

The systematic search resulted in the identification of 14,446 potentially relevant citations, from which 3,615 studies clearly addressed the topic of ED overcrowding. After the studies titles and abstracts were screened, 354 full manuscripts were retrieved for further examination. The application of the selection criteria to the 354 studies resulted in 28 studies included. Figure 1 outlines the study flow for the review. The complete list of references of excluded studies is available upon request.

Figure 1.

 Literature search and study selection flow. MP = multiple publication; TLP = triage liaison physician.

Operational Issues

Study Characteristics.  Of the 28 included studies, 13 had journal publications,1,14–25 three Web page articles,26–28 and 12 abstracts,29–40 All studies were single-centered ED studies (Table 1).

Table 1. 
Descriptive Characteristics of Studies Included in the Review
Reference (First Author, Year)LocationSampleDurationStudy DesignTLP InterventionComparisonOverall quality
  1. CCT = clinical controlled trial; ENP = emergency nurse practitioners; IT = informatics technology; ITS = interrupted time series; NR = not reported; RCT = randomized controlled trial; RME* = Rapid Medical Evaluation Program (CEP America); SHO = senior house officer.

Holroyd,1 2007Canada5,7183 weeksRCTSenior physician triageNurse-led triageStrong
Levsky,14 2008United States79,3896 monthsBefore–afterTeam triageNurse-led triageWeak
Choi,15 2006Hong Kong2,6657 daysBefore–afterTeam triageNurse-led triageWeak
Chan,16 2005United States36,5226 monthsITSIT revisions, EP at triageNurse-led triageModerate
Han,17 2009United States17,2652 monthsBefore–afterEP at triageNurse-led triageModerate
Richardson,18 2004Australia17,2993 monthsBefore–afterSenior EP at triageNurse-led triageWeak
Travers,19 2006Singapore57610 daysCCTSenior EP at triageNurse-led triageWeak
Terris,20 2004United Kingdom2523 monthsCCTSenior EP at triageNo IMPACT teamWeak
Subash,21 2004United Kingdom1,0284 daysRCTEP at triageNurse-led triageWeak
Roger,22 2004United KingdomNR6 weeksBefore–afterTeam triageNurse-led triageWeak
Shrimpling,23 2002United KingdomNR2 monthsBefore–afterTeam triageNurse-led triageWeak
Partovi,24 2001United States1,7348 daysCCTSenior EP at triageNurse-led triageStrong
Grant,25 1999Australia21,1673 monthsBefore–afterSenior/junior EP at triageNurse-led triageWeak
RME,26 2006United StatesNRNRBefore–afterTeam triageNurse-led triageWeak
RME,27 2009United StatesNRNRBefore–afterTeam triageNurse-led triageWeak
RME,28 2009United StatesNRNRBefore–afterTeam triageNurse-led triageWeak
Ruoff,29 2004United StatesNR3 monthsBefore–afterSenior physician triageNurse-led triageWeak
Murrel,30 2009United States64,9076 monthsBefore–afterEP at triageNurse-led triageWeak
Schlicher,31 2009United States1,7891 weekBefore–afterEP at triageNurse-led triageWeak
Crane,32 2009United States4,3274 monthsCCTEP at triageNurse-led triageWeak
Gerton,33 2009United StatesNR5 daysCCTEP at triageNurse-led triageWeak
Porter,34 2009United StatesNR15 daysCCTJunior EP at triageNurse-led triageWeak
Sigal,35 2007United States121,89412 monthsBefore–afterMidlevel EP at triageNurse-led triageWeak
Baumann,36 2006United States243NRProspective cohort studyTeam triageNurse-led triageWeak
Chan,37 2005United States23,0866 monthsBefore–afterPhysician-directed ancillary testing at triageNurse-led triageWeak
Gray,38 2009Canada5,0203 weeksCCTEP at triageNurse-led triageStrong
Graham,39 2009Hong Kong1,1602 weeksBefore–afterSenior EP at triageNurse-led triageWeak
Crane,40 2007United States1436 monthsProspective cohort studyEP at triageNurse-led triageWeak

Based on the EPHPP tool, the overall quality of three studies1,24,38 was rated as strong due to the study design (i.e., RCT or CCT); two studies16,17 were of moderate quality; whereas the other 23 studies were rated as weak (Figure 2). Before the consensus process, the interrater agreement was good on the overall quality assessment (κ = 0.66). Most studies were rated as weak based on study design, data collection methods, and the failure to account for confounders.

Figure 2.

 Frequency of the six components of the EPHPP tool5 and Global* ratings. EPHPP = Effective Public Health Practice Project.

Primary Outcomes

ED LOS.  Nineteen studies1,14–17,21–25,29–33,35,36,38,39 reported outcome data on ED LOS. A pooled result from two RCTs indicated a significant reduction in ED LOS when comparing TLP interventions to nurse-led triage (WMD = −36.85 min; 95% CI = −51.11 to –22.58—note that a negative time is used to denote a shorter or improved time). All the non-RCT studies except for three25,32,33 reported statistically significant reductions in the individual estimates of ED LOS (Figure 3). A subgroup analysis of ED LOS by type of TLP (team vs. single-physician triage; Figure 4) as well as the median absolute improvement in ED LOS (–36 minutes; IQR = −46 to 21 minutes) yielded similar results. A subgroup analysis of ED LOS on Canadian Triage and Acuity Score (CTAS) level 3 or equivalent patients by type of study design, showed similar results (–39.00 minutes’ reduction; 95% CI = −61.92 to –16.08 in one RCT).1 The funnel plot of these studies showed asymmetry (data available upon request).

Figure 3.

 Effectiveness of TLP interventions on ED LOS in minutes for all patients. LOS = length of stay; RCT = randomized controlled trial; TLP = triage liaison physician.

Figure 4.

 Effectiveness of TLP interventions on ED LOS in minutes by type of intervention. LOS = length of stay; RCT = randomized controlled trial; TLP = triage liaison physician.

PIA.  Nine studies14,15,21,22,25,30,31,34,38 reported outcome data on PIA for all patients. One RCT showed a significant reduction in the PIA associated with TLP presence when compared to nurse-led triage (triage process with a TLP, –30.00 minutes; 95% CI = –56.91 to –3.09). Most of the non-RCT studies also showed a significant reduction in this indicator (Figure 5). The median absolute improvement in PIA was lower (–19 minutes; IQR = −26 to –11 minutes), but similar to the reduction in the PIA observed for CTAS Level 3 patients (–19.00 minutes’ reduction; 95% CI = −26.19 to –11.81 reported in one RCT).1

Figure 5.

 Effectiveness of TLP interventions on PIA for all patients. PIA = physician initial assessment; RCT = randomized controlled trial; TLP = triage liaison physician.

Secondary Outcomes

LWBS.  Twelve studies1,14,16–18,24,29,30,33,35,36,38 reported outcome data on LWBS. Findings with respect to patients LBWS were not statistically significant in the included RCT1 (RR = 0.82; 95% CI = 0.67 to 1.00); however, there was a significant decrease in the risk of LWBS among most of the non-RCT designs (Figure 6).

Figure 6.

 Effectiveness of TLP interventions on patients LWBS. LOS = length of stay; LWBS = leaving without being seen; RCT = randomized controlled trial; TLP = triage liaison physician.

LAMA.  Two studies1,29 reported outcome data on LAMA. Meta-analysis on this outcome was not conducted due to differences in study design between the studies. Individual study results on LAMA were inconsistent; a before-and-after study reported a significant reduction in the rates of LAMA after the implementation of TLP (8% vs. 12%, p < 0.001),29 and the RCT that compared TLP versus nurse-led triage reported nonsignificant differences in the rates of LAMA (0.63% vs. 0.69%).1

Due to high heterogeneity (I2 > 90%), we decided not to present pooled estimates for data resulting from non-RCT designs. When we subgrouped on type of intervention (team triage or single physician) some of the heterogeneity was explained, as the team triage group was quite homogeneous (I2 = 0%), but heterogeneity was still very high in the single physician study group (I2 = 94%), preventing us from generating a meaningful pooled estimate.

Discussion

This systematic review summarizes the best available evidence on the effectiveness of the TLP intervention in mitigating the negative effects of ED overcrowding. Moreover, it highlights the effect of this intervention on ED-overcrowding metrics, which in turn may guide the selection of services that may best be used to strengthen the health care system. While previous efforts to summarize the evidence on the benefits of interventions to reduce ED overcrowding have focused on other interventions,2 the current review examined the specific contribution of TLPs based on a comprehensive search, detailed data collection, and reproducible methods.

Most of the studies reported a prolonged LOS for patients presenting to the ED; the median LOS was 257 minutes in the nonintervention settings of these studies. Overall, the TLP interventions evaluated through RCTs were associated with an average 37-minute reduction in ED LOS; a lower, yet significant difference (23-minute reduction) was observed in the subgroup analyses by type of intervention. While speculative, this reduction could be related to the significant reduction in the time patients spent waiting to be seen (30-minute reduction in one RCT).21 To put this change into perspective, the savings of 30 minutes per patient using TLP in an average-sized ED (50,000 patients/year) could provide 75 hours of additional stretcher space per day for seeing additional patients. Regardless of the study design, when considering individual study outcomes (means), the median absolute improvement (36- and 19-minute reduction for ED LOS and PIA, respectively) was consistent with the results obtained from the conventional meta-analytic approach. Several Cochrane reviews41,42 of interventions to improve practice have considered this nonparametric approach when faced with small number of studies, potential confounding effects, and poor reporting.

A significant reduction in both ED LOS and PIA persisted when analyzing specifically for triage level 3 patients. Triage level 3 patients who require urgent care are stable on presentation, can be complex, can consume considerable ED resources, may have prolonged LOS, and often require admission.43 Our subgroup analyses by type of intervention could not explain the high heterogeneity observed in the non-RCT designs; other confounders such as the experience of the staff involved in these interventions and the hours in which they were operational may be influencing this finding.

A reduction in the risk of LWBS was nearly associated with TLP interventions in one RCT (RR = 0.82; 95% CI = 0.67 to 1.00). The results from RCTs and non-RCTs represent a 20% to 40% reduction in this important outcome in ED overcrowding.1,14,16,17,24,30,33 Evidence suggests that adult LWBS cases result from patients becoming frustrated with prolonged waits to be seen by a physician.44 These results further suggest that the TLP intervention mainly affected throughput by allowing earlier assessment of patients and facilitating management decisions.

The included studies showed diversity in their methodologies, with most of them being before–after studies. Only two RCTs assessing the effectiveness of TLP were identified. RCTs can reduce the effects of known or unknown confounders through allocation of treatments, concealment of allocation, and blinding; however, due to their complexity and costs, they are not usually applied to operations research conducted in ED settings.45 Health care interventions examined through observational designs (e.g., before–after studies, analytical cohort studies) are vulnerable to bias,46 and the low scoring in the quality assessment reflects this concern. Despite the extensive search, we were unable to identify more than a few high-quality studies in this field, further highlighting the need for operations researchers to use rigorous methods and report their findings fully.

The included studies differed somewhat with regard to study populations and types of interventions. The complexity of patients in each study appeared to vary due to different selection criteria or study settings. A previous study suggested that variation in ED LOS could be related to factors such as ED volumes and the geographic region of a given hospital.45 Currently, there is no universal or consistent guideline for the TLP function or interventions. In fact, investigators have applied different definitions to this role. Several studies advocated a consultant or senior physician as the TLP;1,18 some studies have employed junior physicians.25,34 Moreover, the responsibility of a TLP may vary from performing the same activities as a triage nurse,19 to guiding the adjudication of administrative issues, expediting the clinical evaluation/treatment, and in some cases speeding up disposition of less complex patients.1,38 These last activities have been related to achieving of the goal of “see and treat.”20,22,23,29 Most importantly, these differences may have critical implications during the implementation of such interventions into routine clinical practice. The different needs and TLP intervention roles should align with the characteristics of the individual ED and the factors contributing to the crowding. Despite the high heterogeneity (I2 > 90%), the majority of individual studies and the overall results demonstrated that TLP reduces LOS and PIA in the ED, hence improving ED throughput and partially mitigating ED overcrowding.

Future research is needed that is focused on more robust designs (e.g., interrupted time series or RCTs) of suitable duration (more than several days), which report pre- and postintervention patient comparisons, use valid/electronic data capture of all ED patients (or reports dropouts), ensure blinding of patients and outcome assessment, and adjusts statistically for confounders. Moreover, these studies must report comprehensive and detailed outcomes including patient times (with measures of variation), LWBS, patient and provider satisfaction, activities, and costs. Finally, the ED operations need to be clearly articulated, because the authors of one abstract concluded that the placement of physicians at triage might not be a viable model in community EDs with an average LWBS rate of 3.4%.33

One of the strengths of this review is that non-English language publications were not excluded from the review process. Likewise, we adopted a comprehensive strategy to appraise the methodologic quality of the included studies. Our approach to quality focused mainly on an assessment of the internal validity of the studies and their potential to bias the results due to deficiencies in the methods of participants’ selection, design, blinding, management of confounders, data collection, and handling of withdrawals and dropouts.

Limitations

The following limitations should be acknowledged when interpreting the results of this review. First, we were unable to report pooled results from the non-RCTs due to high heterogeneity, which limits the generalizability of the reported pooled results given the low number of studies involved. Based on this and on the fact that no multicenter study was identified by our search, our results may not be extrapolated to all ED settings. Our results were consistent, however, using both conventional and nonparametric meta-analytic approaches. Second, reports of important outcomes (e.g., times, LWBS, costs) were commonly missing, despite our efforts to contact the investigators for additional data; appropriate imputation techniques were used to estimate missing values in some outcomes and to ensure valid inference. Due to insufficient data, anticipated subgroup comparisons and sensitivity analyses were not always possible; however, subgroup analyses for triage level 3 patients and by type of intervention (team triage vs. single-physician triage) were completed.

Third, despite our comprehensive and detailed search strategy, our funnel plot suggested some publication bias. This suggests that small and nonsignificant trials may have been missed by this search strategy. We included unpublished and published studies in an effort to avoid this bias, and recent evidence suggests that publication bias is less pervasive in the ED literature;47 however, negative trials are less likely to be published and more likely to be excluded from a review of this nature, potentially biasing the study conclusions. None of the studies mention the potential negative effect of TLPs, including increased patient handovers, potential for lost information, and differing opinions on testing or diagnosis that may confuse patients. Finally, selection bias is possible; however, all abstracts and manuscripts were screened by at least two independent reviewers using standardized eligibility criteria in an effort to decrease the likelihood of this bias.

Conclusions

Emergency department overcrowding is a system-wide problem with no simple or immediate solutions. While triage liaison physician interventions may not address the leading factors accounting for this problem, studies with serious limitations suggest a positive influence on important indicators of the quality of care provided, such as ED length of stay and physician initial assessment, in adult or mixed (adult and pediatric) EDs. Because impressive human and financial resources are required whenever new policies and interventions are initiated to improve hospital performance, future research is required to assess the efficiency of triage liaison physician implementation and to better understand organizational factors for success before widespread use of this intervention can be recommended.

Acknowledgments

The authors thank the corresponding authors for their contributions on data and methodologic clarification: Yu Fai Choi, Sara Gray, Jin Ho Han, Theodore Chan, Michael Yeoh, Joanna Richardson, David Bryan, Sirous N. Partovi, Brent Ruoff, Phillip Asaro, David Spain, and Steven Grant. The authors are also grateful to Donna Ciliska and Donna Fitzpatrick-Lewis for their explanation of the EPHPP quality assessment tool and to Diana Satanovsky-Feldman and Siri Margrete Holm for their assistance in translation.

Ancillary