Article Text
PDF
Abstract
Background Prolonged emergency department (ED) stays make a disproportionate contribution to ED overcrowding, but the factors associated with longer stays have not been systematically reviewed.
Objective To identify the patient characteristics associated with ED length of stay (LOS) and ascertain whether a predictive model existed.
Methods This rapid systematic review included published, English-language studies that assessed at least one patient-level predictor of ED LOS (defined as a continuous or dichotomous variable) in an adult or mixed adult/paediatric population within an Organization for Economic Cooperation and Development country. Findings were synthesised narratively.
Results We identified 35 relevant studies; most included multiple predictors, but none developed a predictive model. The factors most commonly associated with long ED LOS were need for admission (10 of 10 studies) and older age (which may be a proxy for age-related differences in health condition and severity; 9 of 10), receipt of diagnostic tests or consults (8 of 8) and ambulance arrival (4 of 5). Acuity often showed a bell-shaped relationship with LOS (ie, patients with moderate acuity stayed longest).
Limitations Methodological choices made in the interests of rapidity limited the review's comprehensiveness and depth.
Conclusions Despite a sizeable body of literature, the available information is insufficiently precise to inform clinical or service-planning decisions; there is a need for a predictive model, including specific patient complaints. Deeper understanding of the determinants of ED LOS could help to identify patients and/or populations who require special intervention or resources to prevent a protracted stay.
- emergency care systems, emergency departments
- crowding
Statistics from Altmetric.com
Key messages
What is already known on this subject?
Numerous studies have explored the patient-level predictors of protracted emergency department (ED) stay, but the literature in this area has not been systematically reviewed.
It is well known that patients who require admission stay longer; other frequently studied patient variables include age, sex, severity/acuity and presenting complaint.
What might this study adds?
Our review confirmed the effect of age on length of stay (LOS), and suggested that this effect is largely explained by age-related differences in presenting complaint and need for admission.
Severity/acuity also predicts longer stays, although the effect of acuity is offset by the fact that higher-acuity patients are seen more quickly.
The pervasive problem of emergency department (ED) overcrowding has prompted efforts to better understand the drivers of ED usage. There are two types of potentially conservable usage: potentially preventable visits (frequent visits, non-urgent visits) and potentially reducible visit duration. Many studies and two recent systematic reviews have focused on the former, noting that preventable visits may worsen ED congestion by increasing patient volumes.1 ,2 However, evidence suggests that it is not frequent or non-urgent but lengthy visits that make the greatest contribution to ED congestion;3 ,4 therefore, it is at least equally important to examine the determinants of protracted ED length of stay (LOS). Upon discovering that the literature in this area had not previously been reviewed, we conducted a systematic review of the patient characteristics that predict long(er) ED LOS.
This work was undertaken to inform the development of our health region’s patient flow strategy. This region compared unfavourably with its peers on ED LOS, and had yet to make progress towards patient flow targets for which its Board had set a 2015 deadline. Our primary stakeholders were managers of the Emergency Program, who wished to identify priorities for using the extensive data collected on ED visits. Having collaboratively identified ED LOS as a priority outcome, we needed to discover what was already known about its determinants, whether there existed a predictive model we could use and if not, what variables should be included in developing such a model. Predictive models can be used by health professionals to flag patients who may require special attention or expedited linkage to other services to avoid a protracted stay; they may also enable real-time forecasting of the requisite level of ED resources. Modelling may also help planners identify high-risk populations who may benefit from targeted interventions or increased availability of certain services outside the ED. Our stakeholders were particularly interested in the determinants of extremely long stays (>24 h); the region had committed itself to eliminating these, but the prevailing rate hovered around 7%.
Methods
Given the urgency of moving forward, it was important to provide evidence to stakeholders within 4 months; therefore, we undertook a rapid review, a type of systematic review in which certain activities are streamlined or eliminated in the interests of speed.5 When developing our protocol, we made several methodological choices to ensure prompt completion: restricting the search to one database, having certain tasks performed by one rather than two reviewers, and using narrative rather than quantitative synthesis.
This article presents an updated version of the review (search update conducted December 2014); we note, however, that the nine articles added during the update did not alter the original conclusions. The protocol was not registered, but is available from the authors.
Inclusion/exclusion criteria were as follows:
Population: Adult or mixed adult–paediatric patient populations (not paediatric-only) in countries belonging to the Organization for Economic Cooperation and Development (OECD).
Outcomes: ED LOS, defined as either a continuous or a dichotomous variable (ie, ED LOS in minutes/hours or ED LOS >cut-off point).
Predictors: At least one patient characteristic; studies that assessed only non-patient-level factors (eg, hospital features, crowding, time of day) were excluded.
Study types: Bivariate, multivariate and/or cluster analysis to identify predictors of long(er) ED LOS, with or without the development of a predictive model (no qualitative studies, intervention studies, reviews or commentaries).
Source types: Published, peer-reviewed journal articles written in English (to limit the time required for searching and assessment).
The following search string was entered into PubMed (coverage 1947–2014): “(factor* OR predictor* OR predictiv* OR character* OR correlation OR demographic* OR determinants OR identification OR identifying) AND (“length of stay” OR “Length of Stay”[MeSH]) AND (duration OR longer OR prolonged OR extended OR length OR protracted) AND (“Emergency Medicine”[Mesh] OR “Emergency Medical Services”[Mesh] OR “Emergency Service, Hospital” [Mesh] OR “emergency department*” OR “emergency room*” OR ED[ti] OR ER[ti]).”
The search returned 2762 hits (see figure 1). Abstracts were screened by one reviewer (SAK or YC), yielding 47 full articles for assessment; these were read by both reviewers, who made inclusion decisions by consensus. We browsed records identified through PubMed’s ‘related citations’ function, but did not find any relevant reports that had eluded the search. One reviewer (SAK) extracted data on the study year, country, size, type(s) of patients and any subgroups included, definition of outcome variable(s), type of analysis, factors analysed and direction of any significant results. In the interests of rapidity, we did not contact authors for additional data, even when it was apparent that they had not reported results for all the predictors studied. (This sometimes occurred in studies that highlighted a specific predictor while controlling for patient demographics; it appeared to indicate only the authors’ focus of interest, not a deliberate decision to suppress certain outcomes).
Search and selection process. ED, emergency department; LOS, length of stay; OECD, Organization for Economic Cooperation and Development.
We did not conduct formal risk-of-bias assessment as most of the typical risk-of-bias considerations for observational studies do not apply to exploratory studies that seek only to identify predictors, not to confirm causal relationships. However, we did make note of any methodological weaknesses (eg, use of parametric tests for outcomes with a skewed distribution; absence of variables found to be important by other studies). In keeping with the rapid-review approach, we undertook a narrative rather than quantitative synthesis of findings.
Results
Study characteristics
We identified 35 relevant studies; of these, 15 included all patients in the ED, 2 included only non-admitted and 2 only admitted patients, 15 included only patients with a certain type of problem (eg, mental health, critical illness, trauma) and 1 included only older adults. Most of the studies came from the USA (18) or Canada (7), with 3 from Australia, 2 from France and 1 each from Germany, Ireland, Japan, Turkey and the UK. The characteristics of included and excluded studies are presented in tables 1 and 2.
Characteristics of included studies
Characteristics of excluded studies
Most studies treated ED LOS as a continuous variable; a few used a cut-off point (most frequently 4 h for the general population; 24 h in two studies of psychiatric and one study of critically ill populations). Few studies featured univariate analyses only; most presented some type of regression model assessing the impact of each factor while controlling for the others. To compensate for the fact that ED LOS typically shows a skewed distribution, most studies log-transformed this variable, used proportional-hazards regression or set a cut-off point and used logistic regression. Nearly all of the multisite studies that incorporated hospital characteristics correctly used a mixed model with a random intercept term for site; so did one study whose authors were concerned that site factors might confound the analysis as different sites served distinct patient populations. A few studies focused on only one predictor (eg, substance use) and did not analyse, or did not report results for, other factors. Unfortunately, no studies used cluster analysis or similar methods to identify subtypes of patients with long(er) LOS. Although all the studies were concerned with the predictors of ED LOS, none involved the development and evaluation of a predictive model to identify patients at risk of long(er) LOS. Indeed, many studies included patients’ disposition (admitted/non-admitted) as a variable, eliminating the possibility of using the findings for advance prediction.
Studies were heterogeneous in terms of the number of EDs studied (from 1 to over 300), the period of data collection and the number of predictors included in the analysis. However, as findings were generally consistent across studies, results were synthesised through simple ‘vote counting’, with no attempt to weight the findings by study size, quality or other characteristics.
Study findings
The following patient factors have been found to predict ED LOS (see table 3). While we will discuss each factor individually, the effects were observed in multivariate analyses unless otherwise noted.
Need for admission. Not surprisingly, admitted patients stay much longer than non-admitted ones (10 of 10 studies).4 ,6 ,11 ,12 ,15 ,21 ,22 ,27 ,34 ,38
Severity and acuity of condition. Patients who arrive by ambulance stay longer (four of five studies)11 ,15 ,22 ,39 and require more lengthy treatment;14 only one study, limited to mental health visits, found no such effect.34 Acuity tends to show a bell-shaped relationship with LOS: in five of nine studies, LOS was shortest for patients with the highest acuity (short waits) or lowest acuity (short treatment and/or fast-tracking), and longest for those in the middle (ie, three on the Canadian Triage Acuity Scale (CTAS)).6 ,14 ,17 ,22 ,39However, three studies found the longest LOS among CTAS 2s4 ,10 or at higher acuity levels in general,11 and one found a linear negative relationship between acuity and LOS.26 In four of six condition-specific studies, one or more indicators of severity predicted longer stay.12 ,13 ,25 ,38
Age. In general, older adults have longer LOS than younger adults (9 of 10 studies;6 ,11 ,14 ,15 ,18 ,20 ,21 ,22 ,24 1 study found no effect).39 However, three studies that analysed admitted and non-admitted patients separately found that only in the latter group did older patients stay longer;6 ,17 ,20 age effects were also detected in two studies of non-admitted patients10 ,19 but only one of admitted patients.26 Studies of condition-specific subpopulations tended not to show age effects (only one38 of nine studies 8 ,12 ,13 ,16 ,27 ,30 ,31 ,36 ,38) except among trauma patients (two of two studies).7 ,25 Within the population of older adults (usually defined as 65+years), one study found an age gradient21 while two others did not.18 ,26
Patient complaint. There is some evidence that visits related to mental health and substance abuse may be longer, although these effects appear to vary among EDs.9 ,14 ,19 ,33 Analysis of the impact of patient complaints in general was not done with sufficient frequency or consistency to support conclusions. No study of the general patient population considered specific physical diagnoses; one included broad diagnostic categories (which showed inconsistent effects across different segments of ED LOS),14 another reported that diagnoses were included in the analysis but provided neither definition nor results.10 Analyses of specific diagnoses or symptoms were more commonly undertaken in condition-specific studies,12 ,28 ,29 ,31 ,32 ,34 ,35 ,38 but the results may not be relevant to other populations.
Other characteristics. Of the seven American studies that considered insurance status, most found longer stays among uninsured patients8 ,14 ,17 ,34 ,38 or those without private insurance,12 and two also found longer stays among Medicaid patients;14 ,38 one study of non-admitted patients reported that only Medicare patients had longer LOS.19 One study found that neighbourhood-level economic deprivation predicted longer LOS15 while another did not16; two found longer stays among homeless persons8 ,38 while one found the reverse.4 Several American studies included race/ethnicity; some found longer LOS among patients who were black,27 ,19 Hispanic,17 ,19 Asian,38 or non-Caucasian in general,8 while others found no effect.16 ,34 Two studies suggested that language barriers could increase ED LOS;17 ,23 one found no effect.37 One study found that frequent users had shorter ED LOS; this was a function of their lower acuity (within each CTAS category, frequent and infrequent users did not differ in LOS).4 In contrast, another found that frequent users had similar acuity and longer LOS.24 Among studies of the general patient population, a minority (three of nine) found that females had longer LOS;4 ,14 ,22 one other obtained this finding in univariate but not multivariate analyses,10 and the rest found no significant difference.7 ,11 ,17 ,26 ,39 Likewise, none of the studies of condition-specific populations reported a significant effect for sex in multivariate analysis.
Tests and treatment. While the receipt of testing or treatment is not an inherent patient characteristic, it seems important to note that patients who receive physician consults, diagnostic tests (especially advanced diagnostic imaging and blood work) or procedures stay substantially longer than those who receive none (eight of eight studies).8 ,10 ,11 ,13 ,17 ,25 ,38 ,39 This finding also emerged strongly from two studies that we had excluded because they did not address patient characteristics per se.40 ,46
Predictors of ED LOS: summary of main findings
Rapid reviews invariably make trade-offs between thoroughness and speed, and ours was no exception. Its coverage was limited by certain decisions made in the interests of rapidity; we used a single database source, did not search specifically for studies focusing on a single factor and knowingly excluded non-English-language studies and those from non-Organization for Economic Cooperation and Development (non-OECD) countries. We do not believe that we missed a predictive model, but we may have missed relevant findings. Having a single reviewer screen abstracts and extract data can also lead to errors or omissions, although the data to be extracted were rather straightforward.
Discussion
Findings from a variety of countries confirm that ED LOS is meaningfully related to patient characteristics. The factor most commonly studied is patient age, which is frequently associated with long ED LOS. The age effect, given that it tends not to appear in condition-specific studies, may be largely explained by age-related differences in presenting complaint and need for admission. Severity/acuity is also associated with longer stays, although the effect of acuity is offset to some extent by the fact that higher-acuity patients are seen more quickly. The variability in findings related to acuity may reflect differences in practice patterns among hospitals and health systems. Social determinants of health, such as low socioeconomic status and minority race/ethnicity, may also predict longer LOS, although this too seems variable.
While these findings are valuable, they lack sufficient detail to inform the identification of at-risk patients or populations in a clinical or service-design context. Analysis of the specific patient conditions that may predict longer ED LOS has thus far been limited; further work in this area is needed. Such analysis might permit the development of a predictive model for long ED LOS, which is currently lacking.
A limitation of the literature is that large studies are restricted to the variables included in administrative data sources, which may not be fully comprehensive, precise or comparable across different health systems. However, we suspect that data on patient complaints may have been left unexplored even when available, due to the difficulty of meaningfully incorporating a categorical variable with so many categories. In future, it would be reasonable to begin with simple descriptive analysis of the most common complaints among all, short-stay and long-stay patients, in order to identify a manageable list of specific complaints to include in multivariate analyses (perhaps in addition to broad complaint categories).
A useful term for encapsulating what we know about the determinants of long ED LOS might be ‘complexity’. Part of the picture is patient complexity: The well-established effect of patient age (or rather, of the health conditions and other factors for which age is a proxy) and the observed bell-shaped relationship between LOS and acuity may suggest that long stays are especially likely when patients present with multiple or ill-defined problems that are therefore difficult and time consuming to address. Findings also point to the importance of treatment complexity, some of which may be a function of patient complexity, some of overuse of diagnostic tests or procedures. Indeed, an analysis of the dramatic rise in American ED occupancy over the years 2001–2008 concluded that the most responsible factor was an increase in practice intensity; population changes (ie, increasing age and burden of illness) also played a role, but a smaller one.52 A trend towards increased use of diagnostic tests, in particular imaging, has also been observed in Canada; this trend is not unique to EDs but has occurred throughout the health system.53
The idea of (ever-increasing) complexity may be valuable to bear in mind when appraising potential system responses to the problem of long ED LOS. Two types of response are possible: those in which clinicians target individual patients on the basis of screening (eg, case management) and those in which planners redesign services for all patients in a broad category (eg, care pathways, direct-to-treatment arrangements). Unlike service-redesign interventions, those that depend on screening demand a predictive model that has high sensitivity and specificity and is feasible to apply at the point of care; such a model does not yet exist, but may emerge from future analyses. A deeper question, however, concerns the extent to which individually directed solutions can suffice for system problems. If ED LOS is indeed a function of complexity, then trends in population characteristics and clinical practice have created a perfect storm: Patient problems are becoming increasingly complex, and the ED is increasingly a place to diagnose and manage complex problems. To go further, if the root cause of increasing LOS is that EDs are doing what they were never intended or designed to do, then attempting to address this issue patient-by-patient seems likely to prove cumbersome and inefficient.
Conclusion
Despite a sizeable body of literature on the patient-level predictors of long ED LOS, the available information is insufficiently precise to facilitate application by clinicians or service planners. There is a need for a more detailed understanding of the determinants of long ED stay, and an opportunity to develop predictive model(s), especially for extremely long stays (which have not yet been studied in the general patient population). Further research should incorporate previously studied variables—at minimum, age, sex, acuity and arrival by ambulance—as well as specific patient complaints. Such work will permit the identification of individuals at risk of protracted stay, supporting exploration of the prospects for patient screening; even more important, it will aid in determining how to develop better solutions for populations of patients with complex health needs.
References
Footnotes
Contributors SAK conceived the review; participated in screening, assessment and data extraction and drafted the article. CJM provided substantive feedback at all stages. YC was involved in article screening and assessment. MR conducted the literature search.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.