Article Text

Download PDFPDF

Realist analysis of streaming interventions in emergency departments
  1. Mohammed Rashidul Anwar1,2,
  2. Brian H Rowe3,4,
  3. Colleen Metge1,
  4. Noah D Star1,
  5. Zaid Aboud1,
  6. Sara Adi Kreindler1,5
  1. 1 Community Health Sciences, University of Manitoba, Winnipeg, Manitoba, Canada
  2. 2 Child Health Evaluation Sciences, The Hospital for Sick Children, Toronto, Ontario, Canada
  3. 3 Department of Emergency Medicine, University of Alberta, Edmonton, Alberta, Canada
  4. 4 School of Public Health, University of Alberta, Edmonton, Alberta, Canada
  5. 5 Health Systems Performance, George & Fay Yee Centre for Healthcare Innovation, Winnipeg, Manitoba, Canada
  1. Correspondence to Dr Mohammed Rashidul Anwar, Community Health Sciences, University of Manitoba, Winnipeg, MB R3T 2N2, Canada; anwarmr{at}


Background Several of the many emergency department (ED) interventions intended to address the complex problem of (over)crowding are based on the principle of streaming: directing different groups of patients to different processes of care. Although the theoretical basis of streaming is robust, evidence on the effectiveness of these interventions remains inconclusive.

Methods This qualitative research, grounded in the population-capacity-process model, sought to determine how, why and under what conditions streaming interventions may be effective. Data came from a broader study exploring patient flow strategies across Western Canada through in-depth interviews with managers at all levels. We undertook realist analysis of interview data from the 98 participants who discussed relevant interventions (fast-track/minor treatment areas, rapid assessment zones, diverse short-stay units), focusing on their explanations of initiatives’ perceived outcomes.

Results Essential features of streaming interventions included separation of designated populations (population), provision of dedicated space and resources (capacity) and rapid cycle time (process). These features supported key mechanisms of impact: patients wait only for services they need; patient variability is reduced; lag time between steps is eliminated; and provider attitude change promotes prompt discharge. Conversely, reported failures usually involved neglect of one of these dimensions during intervention design and/or implementation. Participants also identified important contextual barriers to success, notably lack of outflow sites and demand outstripping capacity. Nonetheless, failure was more commonly attributed to intervention flaws than to context factors.

Conclusions While streaming interventions have the potential to reduce crowding, a theory-based intervention relies on its implementers’ adherence to the theory. Streaming interventions cannot be expected to yield the desired results if operationalised in a manner incongruent with the theory on which they are supposedly based.

  • health policy
  • health system
  • performance management
  • total quality management
  • organisational effectiveness

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information. For confidentiality reasons, qualitative data collected for this study cannot be shared.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See:

Statistics from


Emergency department (ED) crowding is a common and potentially harmful phenomenon identified by several countries as a national crisis.1 Crowding in this setting is a multifactorial issue conceptualised using a model that includes input, throughput, output and system-wide factors. Numerous interventions have been directed at this complex problem. Some initiatives focus on improving ED throughput (eg, the time from arrival in the ED to disposition from the ED); others intervene elsewhere along the continuum of care in order to reduce input or facilitate output from the ED.2 Of those interventions implemented within the ED, several of the most common are based on the principle of streaming: directing different groups of patients to different processes of care. The practice of streaming is based on queuing theory; the underlying principle is that by creating separate queues on the basis of service characteristics (eg, anticipated service time, need for particular resources), clients can be served more efficiently.3–6 It is not unique to EDs, but may be used to promote efficiency and well-targeted care in many clinical settings.

Some ED streaming interventions occur at, or shortly after triage. They may separate out either low-acuity patients whose needs can be met quickly (eg, moving such patients into fast-track/minor treatment areas), or medium-acuity patients who may not require a bed for most of their stay (eg, directing such patients to a rapid assessment zone or intake model). Other interventions (including diverse types of short-stay units, such as observation units, diagnostic and treatment units, medical assessment units) stream patients later in their stay, segregating those who require specialised investigations, longer treatment and/or consultations so that their care does not interfere with efficiency of care for other patients. Short-stay units may operate within or outside the ED and may manage patients prior to disposition or after admission. There is no standard definition of a short-stay unit (nor of its specific variants), and such units serve a variety of functions: providing tailored care to patients with specific conditions; preventing brief hospitalisations; or simply moving patients out of a crowded ED.7 While some hospitals have only one such unit, serving a broad purpose, others offer an escalating sequence of such units for patients with different intensities of need.

Systematic reviews have concluded that minor treatment areas can reduce ED length of stay; however, reported effect sizes vary considerably.8 9 There is limited evidence on effectiveness of rapid assessment zones10 11; and evidence on short-stay units is inconclusive (perhaps unsurprisingly so, given the heterogeneity of interventions studied).7 12–14 Meanwhile, multisite studies consistently find that the same intervention may produce disparate results when implemented in different organisations and hospital EDs.15 16 This lack of clear direction presents challenges to health system leaders who may be struggling with the challenge of designing and implementing streaming interventions—be it in EDs or other healthcare settings. This multijurisdictional, qualitative study was intended to gain a deeper understanding of how, why and under what conditions streaming can be effective. It used a realistic evaluation17 lens to examine hospitals’ diverse experiences with ED-based streaming interventions.

Conceptual framework

Study design was informed by realistic evaluation,17 and analysis by the population-capacity-process model.18 Realistic evaluation is a type of theory-based evaluation that seeks to determine the causal mechanisms by which an intervention achieves its outcomes, and the context in which these mechanisms are able to operate.17 It is designed to determine the effectiveness of an intervention and to explore under what circumstances it works, and why it works (or does not work well). The realist approach begins by articulating the explicit or implicit theory by which an intervention is assumed to work. Then, what actually happens in practice is explored. An intervention may fail because there is a flaw in the programme theory—activities are not appropriately designed to trigger the intended mechanisms in the first place.19 In other cases, contextual factors may disrupt the posited causal chain that links intervention activities to outcomes via mechanisms. By looking beneath surface features of interventions in order to identify underlying mechanisms, realist analysis can generate ‘middle-range theory’ applicable to a broad family of interventions.

As streaming interventions are based on formal theory, it is possible to identify several potential mechanisms a priori. First, streaming interventions may ensure that patients wait for only those services they need; no patient must queue behind another who requires a different set of services.3 4 8 Second, they may reduce variability among patients, enabling them to flow more evenly, thus more efficiently, through the process of care.6 Third, by establishing these low-variability subgroups, interventions may facilitate the delivery of standardised care, thus promoting quicker recovery.5 13 The literature suggests two additional mechanisms specific to rapid assessment zones: (1) by establishing ‘one-piece flow’, they may eliminate lag time between the steps of assessment and treatment; (2) by keeping patients ‘vertical’ (ie, in chairs instead of beds or stretchers), they may optimise the use of space and physical resources.10 20

In practice, however, the extent to which streaming interventions reflect the official theory remains unclear. The empirical literature has focused on assessing whether streaming interventions work, rather than on probing why or under what conditions they work. While some authors have suggested success factors related to particular types of interventions, these have not been investigated systematically. Thus, little is known about either specific features of the intervention or of the external context that may facilitate or hinder the mechanisms of streaming interventions. Accordingly, this study was designed to determine: (A) how and why ED-based streaming interventions improve patient flow; and (B) what factors are perceived to affect such an intervention’s ability to achieve its desired impact.

To guide identification of relevant intervention design and context factors, we applied the population-capacity-process model of patient flow.18 This framework, generated from a case study of a poorly performing health system, was developed to explain why flow interventions fail. The study concluded that effective interventions link a defined population to appropriate capacity through an efficient process: ineffective interventions were found to have neglected one or more of these three crucial aspects. The model is gaining currency21–23 but has not yet been applied to in-depth analysis of a family of interventions.



Canadian healthcare is organised at the provincial level, and many provinces have devolved its administration to regional health authorities, which are disparate in size, demographics, service landscape and organisational structure.24 The problem of ED crowding appears particularly acute in Canada compared with other Organisation for Economic Co-operation and Development countries.25 Almost all Canadian jurisdictions have launched strategies to relieve crowding and improve flow; in the vast majority, however, substantial improvements either have not occurred or have not been sustained.26


This substudy is one component of the Western Canadian Patient Flow (WeCanFlow) study, which explored flow initiatives, in context, across 10 urban health systems spanning four provinces. The WeCanFlow study included in-depth interviews with 300 senior, middle and front-line managers purposively sampled for their involvement in flow, whether in the ED or elsewhere along the continuum of care; sampling, recruitment and data collection are fully described in a companion article.27 The interview guide featured questions about what had and had not worked to improve flow, yielding data on over 70 interventions spread across multiple domains (input/throughput/output/system-wide), each having been implemented by one or more sites in up to 10 regions. Following written informed consent, interviews were conducted in person or by telephone, audio recorded and transcribed verbatim.


In a preliminary round of analysis, coders (MRA, ZA, NDS) identified which interventions were mentioned by each participant, revealing that 98 of 300 participants discussed ED-based streaming initiatives. We then undertook a realist approach, as described earlier.

It is important to note that we were unable to quantitatively assess outcomes of all these interventions (and thus conduct a full realistic evaluation). Rich data on perceived outcomes from participant perspectives, however, enabled us to undertake a robust realist analysis.

After reading the 98 transcripts thoroughly for initial impressions, we carried out qualitative content analysis, a process led by one researcher (MRA) in frequent interaction with another (SAK). The two reviewers worked independently but connected regularly to debate alternative interpretations and reach consensus at each stage. We first inductively identified all explanations provided for success or failure; then categorised these as having to do with population, capacity or process; then paraphrased them as ‘because’ (it works because…) or ‘unless’ (it will/won’t work unless…) statements18 in order to identify them as pertaining to mechanisms or context. At this juncture, we discovered that many ‘unless’ factors were not true context factors but intervention factors, a point to which we will return. We also observed that some factors (eg, leadership support, clinician buy-in) constituted facilitators/barriers to the initial implementation of an initiative, rather than to its achievement of outcomes once implemented. In the interests of focus, implementation facilitators/barriers are excluded from further discussion.

Having revised the codes to ensure their accuracy and consistency, we clustered them into themes using Excel tables to facilitate iterative recategorisation and reorganisation of extracts. Interpretations were further refined through discussion with other members of the study team, which included both researchers and (clinician) managers.


Types of interventions

Three types of interventions were discussed by participants in sufficient detail to contribute to realist analysis: minor treatment areas (n=3), rapid assessment zones or ‘intake models’ (n=12) and short-stay units (eg, clinical decision unit, diagnostic and treatment unit, medical assessment unit, clinical assessment unit, rapid access and discharge unit) (n=22). Initiatives were spread across 26 hospitals. Of the 37 initiatives, 19 were described as effective and 14 as ineffective; in the remaining four cases, both benefits and limitations were reported by the same or different participants. Where more than one participant discussed the same intervention at the same site, they usually exhibited consensus on its overall effectiveness, although in a few cases some participant(s) emphasised the initiative’s benefits, other(s) its limitations.


Very few participants explicitly articulated the intervention mechanisms of successful streaming initiatives. Nonetheless, their accounts implied certain mechanisms, including the five we had identified from the literature (ensuring that patients wait for only those services they need; reducing variability among patients; facilitating protocol-driven care; eliminating lag time between steps; and promoting the efficient use of space). We also identified an additional mechanism: some participants suggested that streaming interventions fostered an ethos of efficiency and rapid discharge, which permeated the units involved and could also spread beyond them. Implied mechanisms and exemplar quotations are presented in table 1.

Table 1

Implied mechanisms

Intervention factors

The bulk of the data addressed features of the intervention thought to enable or hinder the effectiveness of streaming interventions. These were analysed using the three domains of the population-capacity-process model (see tables 2 and 3).

Table 2

Intervention features

Table 3

Intervention flaws


An essential design feature of streaming interventions identified was the separation of a particular kind of patient from the general ED population. While there was consensus on the nature of these subpopulations in the case of minor treatment areas (low-acuity cases) and rapid assessment zones (patients of moderate or indeterminate acuity), depictions of the intended population for short-stay units varied by site and intervention. Short-stay populations were variously described as patients requiring ‘short-term’ or ‘specialized’ care, and the time of the ‘short stay’ varied (from 24 up to 48 or 72 hours). Most participants agreed that such units were intended for complex or resource-intensive cases requiring a somewhat longer stay than the typical ED patient, although not so much longer as to preclude rapid turnover. Indeed, admission of long-stay patients was identified as a major flaw in the operationalisation of short-stay units; units that became occupied with frail elderly or alternate level-of-care patients lost the ability for rapid turnover and could no longer contribute to ED flow. This occurred when units either did not clearly define their intended population or intentionally admitted inappropriate patients in the attempt to free up space elsewhere. Only one participant argued that the practice of admitting long-stay patients to short-stay units was desirable (on the grounds that their site’s admission process was too time consuming to make short admissions worthwhile); all others characterised it as flaw in the intervention.


The cornerstone of streaming interventions is the provision of separate capacity—that is, physical and human resources—for each stream. Participants reported that all such interventions required dedicated space separate from the main ED (even if located in the waiting room), as well as dedicated physicians and other clinical staff. The most commonly reported intervention flaw was a lack of earmarked space and/or providers, which occurred either by policy (eg, some sites that did not believe they could resource an actual short-stay unit instituted a ‘virtual’ one) or as a result of the misuse of allocated space (eg, beds intended for high turnover being used for overflow of long-stay patients). When patients from multiple streams converged on the same resources, streaming interventions reportedly failed to improve flow.


Participants identified rapid cycle time (for minor treatment areas and rapid assessment zones), targeted discharge planning and maintenance of a strict time frame (for short-stay units) as process elements essential to the success of streaming interventions. They described how a consistent, disciplined process enabled the appropriate population of patients to flow rapidly through the designated capacity. The most commonly reported process-related intervention flaw was failure to maintain a strict time frame; short-stay units that failed in this regard soon became occupied with long-stay patients.

Context factors

Relatively few of the success/failure factors described were true context factors (ie, external to the intervention); however, participants did describe how certain population characteristics and capacity constraints could impede the functioning of streaming initiatives (see table 4).

Table 4

Context factors

Population characteristics

A few participants reported context factors related to the size of the eligible population, for instance, that streaming units had run out of space when ever-increasing patient demand ultimately outstripped capacity, or that reserving provider time for a low or variable volume of patients resulted in wasted capacity. The latter would be most likely to occur in hospitals that serve a small overall population—a context factor—although it could also reflect a flaw in the intervention itself (eg, overly strict admission criteria or failure to appropriately define the population). Moreover, this failure to anticipate the need (either by too few or too many referrals) suggested a lack of congruence between the intervention proposed and the site needs.

Capacity constraints

By far, the most widely reported contextual barrier to effective streaming was a lack of outflow capacity. This issue was primarily reported for units serving patients who could not be discharged home; while such units might manage patients efficiently, they failed to improve flow unless they were able to readily transfer patients to their next destination. Also, several participants stated that their site had inadequate resources to sustain streaming interventions—in particular, short-stay units, whose geographic footprint made them expensive.


This study used qualitative methods to conduct a realist analysis of streaming interventions designed and implemented to mitigate ED crowding. Using nearly 100 transcripts from 300 interviews and independent coding methods, this review provides insights into what factors may contribute to the effectiveness of several important throughput interventions.

While many participants reported effective interventions whose design was congruent with the theoretical basis of streaming, many others described interventions which—owing to their incongruence with this theory—could have been predicted to fail. These interventions either failed to clearly define separate streams (limiting their potential to reduce patient variability or promote standardised care); failed to provide each stream with separate capacity (forcing patients to wait behind individuals belonging to other streams); or allowed the designated capacity to fill up with long-stay patients. Such glaring flaws in intervention design and/or operationalisation suggest that not all planners clearly understood the theory underpinning streaming interventions. Some sites might have adopted such interventions imitatively, without understanding the theory behind them,28 or been compelled by higher level decision-makers to adopt a potentially inappropriate intervention. It seems noteworthy that participants offered very few explicit accounts of intervention mechanisms. Some participants, of course, might have merely neglected to mention mechanisms, but others might have neglected to consider them when choosing and adapting the intervention.

A realist approach, guided by the population-capacity-process model, helped to uncover why what was supposedly the ‘same’ intervention might work at one site but not another. Given increasing recognition of the importance of context, one might expect context factors to explain such differences—and indeed, the analysis revealed certain important external context factors, particularly the access block known to hinder patient outflow in many sites. However, variation in reported effectiveness seemed even more attributable to variation in intervention fidelity; that is, ineffective interventions lacked one or more core features of streaming. While it is perhaps unsurprising that low-fidelity interventions are perceived as ineffective, it does seem surprising that many sites are implementing such interventions and expecting them to work.

Of the external context factors we identified, one (inadequate or inconsistent demand for a particular stream) is well known to investigators of streaming, who have termed it the ‘anti-pooling’ or ‘carve-out’ effect.29 30 Another factor, lack of outflow sites, has previously been dubbed the ‘parking lot’ problem (ie, patients become ‘parked’ in capacity intended for short-term use because appropriate long-term capacity is unavailable).20 Greater familiarity with the theoretical basis of streaming might help planners to avoid introducing interventions into unsuitable contexts.

Our findings confirm the applicability of the population-capacity-process model20 to a range of diverse health systems; across numerous regions and sites. Findings also extend prior work by identifying common flaws in streaming interventions. This research also contributes to the literature by identifying an additional mechanism that may contribute to the effectiveness of streaming, namely the promotion of attitude change among staff. Unlike the other identified mechanisms, which are operational in nature, this one is psychological; future research might consider its potential role.

The idea of streaming, although not always identified as such, underpins diverse healthcare interventions: separation of emergency from elective general surgery,31 direct-entry subacute care,32 risk-stratified chronic disease management,33 stepped care for mental health,34 and so on. Our specific findings may not apply to other interventions, some of which serve easily defined groups; deliver care virtually (not requiring designated physical space); or stream patients to facilitate ongoing targeted care, not rapid discharge. However, we believe the following broad principles are transferable: prior to embarking on streaming, leaders should ensure that the intended populations are sufficiently clear and large; that enough capacity can be protected for the streams to run without mutual interference; and that the proposed process does not create new inefficiencies. Furthermore, our findings suggest two takeaway messages for health leaders that extend beyond streaming. First, in order to prevent design and operationalisation flaws, planners must clearly understand the underlying theory of the intervention. Second, it is important to ensure that both the external and internal contexts are favourable to the intervention before attempting implementation.


This study had several limitations. As objective data on outcomes of the 37 initiatives were unavailable, we were limited to analysis of perceived outcomes. Although we observed that participants rarely gave discrepant accounts of the outcomes of the same initiative (suggesting that such perceptions had some consistency), inferences must be drawn cautiously. No more than four people (and often only one person) discussed the same intervention at the same site, and even shared perceptions may be biased. In particular, responses may have been subject to social desirability bias, as managers may have been reluctant to admit the failure of their own interventions—indeed, most of the critical comments were applied to initiatives for which someone else was responsible. While we were able to glean valuable insights about potential drivers of both success and failure, we cannot establish the reliability of participants' observations, and our study must be regarded as hypothesis generating.

Data were drawn from interviews that covered a broad range of flow-related issues; some participants mentioned such interventions only in passing and might have provided more detail had such interventions been the sole focus of the interviews. There was particularly limited coverage of minor treatment areas; it may be that because these were a long-standing intervention at many sites, participants took them for granted or felt unable to comment on their impacts. This same limitation in design meant that our study may not have uncovered all the risks or potential inefficiencies associated with streaming. For example, if streams are defined too narrowly or cannot accommodate fluctuation in patients’ conditions, patients may have to be redistributed among streams. Participants did not report this problem, nor is it noted in the literature on ED streaming; however, it has been observed in relation to other kinds of streaming (eg, of subacute from acute inpatients20). Another limitation is that participants were limited to those in managerial roles at various levels and few front-line providers and did not include patients who might have offered broader perspectives on specific interventions. Despite these gaps, we were able to assemble a large data set comprising information from multiple and diverse sites.

Finally, this study was restricted to Canada; while there is no reason to expect that streaming interventions in other jurisdictions would operate via different mechanisms, context factors might vary widely, as might the prevalence of intervention flaws. Full-scale realistic evaluation of streaming interventions in different countries, as well as across clinical settings, would be a valuable direction for future research.


If the failure of a streaming intervention turns out to result from flaws in its own design or operationalisation, EDs may be able to address the flaws and make the intervention functional. Unfortunately, some intervention flaws may reflect deeper contextual factors that lie beyond the ED’s control. For example, short-stay units may persist in admitting long-stay patients due to an actual or perceived inability to access more appropriate destinations for them; small hospitals may hesitate to enforce clear admission criteria for fear of inducing a carve-out effect; or organisations may assign the same space and physicians to multiple streams because they lack resources to design the intervention optimally. These context factors are external to the intervention, thus resolving them would require collaborative commitment and/or environmental modification. Before introducing a streaming intervention, it is important to ensure that the context will permit such an intervention to be properly designed and executed. On the basis of our findings, we have developed a set of guiding questions for decision-makers who are contemplating such an intervention (table 5).

Table 5

Guiding questions


Interventions based on the principle of streaming have the potential to increase efficiency in EDs and in other areas.3–6 However, our findings provide evidence that a theory-based intervention is only as good as its implementers’ understanding of and adherence to the theory. If those designing and implementing streaming interventions do not follow the principles on which the intervention is premised, potentially valuable strategies are likely to fall short of their potential contributions.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information. For confidentiality reasons, qualitative data collected for this study cannot be shared.

Ethics statements

Patient consent for publication

Ethics approval

The project received approval from all relevant bodies for ethical and operational review in Manitoba, Alberta, Saskatchewan and British Columbia (University of Manitoba Health Research Ethics Board (HS 18666 (H2015:232)), University of British Columbia Providence Health Care Research Ethics Board (H15-02062), University of Calgary Conjoint Health Research Ethics Board (REB15-3026), University of Saskatoon Behavioural Research Ethics Board (BEH 15-377)).


This study used interview data from a parent study, and we express our gratitude to the interviewers, including Stephanie Hastings, Shannon Winters, Keir Johnson, Sara Mallison and Meaghan Brierley. We also thank Sarah Bowen for her insightful feedback and editing.



  • Contributors MRA and SAK conceived and designed this substudy, with guidance from BHR and CM. MRA, SAK, NDS and ZA conducted data analysis, and all authors participated in interpretation of findings. MRA wrote up the findings as a thesis; he and SAK drafted the article version, with all other authors providing critical review.

  • Funding This study was funded by the Canadian Institutes of Health Research (PHE-141802), with partnership funding from the Michael Smith Foundation for Health Research, Alberta Innovates, Saskatchewan Health Research Foundation and Research Manitoba. One of the author’s research was supported by CIHR through a Scientific Director Operating Grant (SOP 186483).

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.