Article Text

Download PDFPDF
An ED paradox: patients who arrive by ambulance and then leave without consulting an ED provider
  1. Malcolm B Doupe1,2,3,
  2. Suzanne Day4,
  3. Wes Palatnick5,
  4. Alecs Chochinov5,
  5. Dan Chateau6,
  6. Carolyn Snider5,
  7. Ricardo Lobato de Faria7,
  8. Erin Weldon5,
  9. Shelley Derksen8
  1. 1Faculty of Health Sciences, College of Medicine, University of Manitoba, Winnipeg, Manitoba, Canada
  2. 2Manitoba Centre for Health Policy, Winnipeg, Manitoba, Canada
  3. 3College of Medicine, Faculty of Health Sciences, University of Manitoba, Winnipeg, Manitoba, Canada
  4. 4Women's Xchange, Women's College Hospital, Toronto, Ontario, Canada
  5. 5Faculty of Medicine, Department of Emergency Medicine, University of Manitoba, Winnipeg, Manitoba, Canada
  6. 6Faculty of Medicine, Department of Community Health Sciences, Manitoba Centre for Health Policy (MCHP), University of Manitoba, Winnipeg, Manitoba, Canada
  7. 7Winnipeg Regional Health Authority, Seven Oaks Hospital, Winnipeg, Manitoba, Canada
  8. 8Faculty of Medicine, Manitoba Centre for Health Policy (MCHP), University of Manitoba, Winnipeg, Manitoba, Canada
  1. Correspondence to Dr Malcolm B Doupe, College of Medicine, Faculty of Health Sciences, University of Manitoba, 408-727 McDermot Avenue, Winnipeg, Manitoba, Canada R3E 3P5; malcolm_doupe{at}


Background Scientists have called for strategies to identify ED patients with unmet needs. We identify the unique profile of ED patients who arrive by ambulance and subsequently leave without consulting a provider (ie, a paradoxical visit, PV).

Methods Using a retrospective cohort design, administrative data from Winnipeg, Manitoba were interrogated to identify all ED patients 17+ years old as having zero, single or multiple PVs in 2012/2013. Analyses compare the sociodemographic, physical (eg, arthritis), mental (eg, substance abuse) and concurrent healthcare use profile of non-PV, single and multiple PV patients.

Results The study cohort consisted of 122 639 patients with 250 754 ED visits. Across all ED sites, 2.3% of patients (N=2815) made 3387 PVs, comprising 1.4% of all ED visits. Descriptively, more single versus non-PV patients lived in urban core and lowest-income areas, were frequent ED users generally, were substance abusers and had seven plus primary care physician visits. Multiple PV patients had a similar but more extreme profile versus their single PV counterparts (eg, 54.7% of multiple vs 27.4% of single PV patients had substance abuse challenges). From multivariate statistics, single versus non-PV patients are defined uniquely by their frequent ED use, by their substance abuse, as living in a core and low income area, and as having multiple visits with primary care physicians.

Conclusions PV patients have needs that do not align with the acute model of ED care. These patients may benefit from a more integrated care approach likely involving allied health professionals.

  • emergency department utilisation
  • mental health
  • research, epidemiology

Statistics from

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.

Key messages

What is known on this subject?

  • Defining potentially inappropriate ED users has proven to be controversial and several authors instead have stressed the importance of defining ED patients whose needs are not met by the acute model of ED care. To date however, the literature defining these latter patients is sparse.

What this study adds

  • This population-based study from Manitoba identifies a subgroup of ED patients who arrive by ambulance only to leave without consulting an ED provider. Such patients were defined uniquely by their substance abuse problems, by sociodemographic measures, and by their multiple visits with primary care physicians, emphasising the need for a more integrated care approach. Strategies are proposed to define additional groups of patients whose needs are not met by the acute model of ED care.


Amidst worldwide trends in ED crowding1 is increased interest in identifying patients whose needs are not met by the acute model of ED care, where short-term treatment is provided to patients with more severe injuries and medical conditions. To date, this literature has focused largely on frequent and non-urgent ED users, however with much controversy about the value of treating these patients elsewhere.2–4 Research has shown that frequent ED users often have multiple and complex chronic diseases,3 ,5 and that non-urgent patients contribute minimally to ED costs6 and negligibly impact patient flow.7 To date, scientists have yet to more clearly identify subgroups of ED patients whose needs extend beyond the acute model of ED care.

Studying the paradoxical, the seemingly contradictory, can provide an effective framework for examining healthcare challenges.8 ED patients who arrive by ambulance (emergency medical services, EMS) only to leave without consulting an ED provider, exemplify such a paradox. In this paper we profile these patients as one group of healthcare users with unmet needs. From a larger list of potential paradoxes (eg, patients with a non-urgent ED visit immediately after seeing a primary care provider), we selected this group for two reasons. First, the costs associated with ambulance transport and the related demands on EMS personnel are significant yet potentially wasteful when transported patients do not subsequently complete their ED care, with negative consequences if these limited resources are not readily available to acutely sick patients. Second, without receiving care from an ED provider, it is difficult to argue that a patient's needs have been met, especially if such ED visits are recurrent. It is important to note that while EMS transport costs and subsidy regulations vary across Canada,9 in general ambulance transport cannot be refused based on a patient's ability to pay.

Within a large Canadian urban environment, the goals of this paper are to: (1) describe the prevalence of paradoxical ED visits (PVs) overall and across ED sites, and; (2) profile the unique sociodemographic, health, concurrent healthcare use and subsequent death profile of patients with one and also multiple PV visits, as compared with patients without such visits.


Study design

This research uses a retrospective, population-based cohort study design. Data were captured using anonymised but person-level administrative records linked across multiple files (eg, primary care and specialist physician visits, hospitalisations), to provide a comprehensive picture of ED use, concurrent healthcare use and overall patient health status. As compared with the literature where several of these risk factors are measured using self-report techniques,5 ,10 administrative records provide these measures without recall and loss to follow-up biases.11 Potential limitations associated with using the ED data system in this study include the lack of physician-generated international classification of disease (ICD) diagnostic codes, required to clarify the actual reason for the ED visit. Also as a general rule, administrative records cannot be used to conduct causal research.

Setting and patients

Manitoba is 1 of 10 Canadian provinces and has a population of 1.2 million people. It is divided into five geographically diverse healthcare regions. The Winnipeg Health Region (WHR) is the largest urban region (population 720 000 people), while the remaining regions are either rural or remote. WHR is populated diversely and about 17% of its residents live in the core area (ie, the geographical centre of WHR, housing predominantly lower income and socioeconomic status people). Within Manitoba, the vast majority of tertiary care specialised services, such as trauma care and cardiac surgery, are provided in Winnipeg.

Electronic ED data in Manitoba are available only in WHR. This region contains six adult ED sites with a total of about 250 000 visits annually. One of these EDs is a tertiary teaching site (Health Sciences Centre) located in WHR core area, while a second is also a tertiary teaching site (St. Boniface Hospital) located just outside of the WHR core area. The remaining adult EDs are housed in non-tertiary hospitals in the four suburban WHR quadrants (Victoria, Concordia, Grace and Seven Oaks Hospitals) (see figure 1).

Figure 1

Map of ED locations in the Winnipeg Health Region (WHR), Manitoba, Canada.

Our cohort includes all registered Manitobans 17+ years old with at least one ED visit in WHR from 1 April 2012 to 31 March 2013. To allocate ED visits to each patient, we first identified his or her last (index) ED visit during this fiscal year and then captured all prior visits made by the patient in the previous 12 months. The start and end dates of our study period are therefore person-specific.

Each ED visit within the study period was defined as paradoxical if the patient arrived by ambulance to the ED, was triaged for care, but then left the ED prior to consulting with an ED provider. This definition matches the consensus definition of ‘left without being seen’ as reported by Welch et al.12 Using this strategy our cohort of ED patients was divided into three study groups defined as non-PV patients (had no paradoxical visits during their 1 year study period), single PV patients (ie, arrived by EMS and subsequently left without consulting an ED provider during one ED visit) and multiple PV patients (as above, for two or more ED visits).

Primary study measures

Similar to previous research,13 all ED patients were identified by their sociodemographic, mental and physical illness, concurrent healthcare use and subsequent death profile. Each of these variables was measured in relation to a patient's index ED visit. Strategies for doing so are provided in the following text, with further details available elsewhere.13

Sociodemographic factors include patient age, sex, income quintile and living in the WHR urban core area. Income quintile is an area-level measure developed using Statistics Canada Census Data, and assumes that people living in the same area have similar incomes.14 Census data were missing for 1.7% (N=2062) of all ED patients. As the distribution of these patients across study groups was most similar to our lowest income patients (data not shown), data for these income groups were combined. Person-level postal codes were used to define patients who lived in the WHR core area versus elsewhere.

Measures of chronic physical disease (arthritis, asthma, diabetes, ischaemic heart disease, stroke) and mental illness (anxiety, depression, dementia, personality disorder, schizophrenia, substance abuse) were measured using strategies developed for use with administrative data,15 which employ ICD code algorithms reported by physicians during ambulatory care visits and hospitalisations. As one example and as used by others,16 we defined patients as having a substance abuse problem if, in 5 years preceding their index ED visit, they were diagnosed at least once during a physician visit or hospital stay with any of alcoholic psychoses (ICD-9-CM code 291), drug psychosis (292), alcohol dependence (303), drug dependence (304) or non-dependent abuse of drugs (305).

During each patient's individualised study period, he or she was also coded as being a non-frequent (1–6 visits), frequent (7–17 visits) and highly frequent (18+ visits) ED user. These categories were developed using ‘break points’ in patient characteristics (eg, age, comorbid mental illness), with details provided in Doupe et al.13 Patients were also defined as having 0, 1–6 or 7+ visits with a primary care physician; as having 0, 1 or 2+ specialist physician visits; and as having 0 or 1+ inpatient hospital admissions. Patient death (yes, no) was also measured 180 days following each patient's index ED visit. Last, as the only visit-based result provided in this manuscript, study groups were also compared by ED visit acuity. This acuity is captured in the administrative data using the Canadian Triage Acuity Scale (CTAS) scale.17 This scale defines each ED visit with one of five levels of acuity, ranging from CTAS 1 (resuscitation, highest acuity level), CTAS 2 (emergent), CTAS 3 (urgent) and CTAS 4 and 5 (less and non-urgent).

Data analysis

Paradoxical visits were counted across all six ED sites in figure 1. Descriptive comparisons were made across non-, single and multiple PV patients. Logistic regression was then used to compare separately the profile of single with non-PV patients, and multiple with single PV patients. The unit of analysis in these models is the patient, and adjusted ORs are included for each covariate regardless of its statistical significance. Concordance (‘C’) statistics were used to measure how well each model discriminated across our study groups. All analyses were conducted using Statistical Analyses Software, V.9.4.

Select strategies were used to streamline our logistic regression models. First, variables with three or more categories were collapsed pending the results of interim analyses. As an example, table 3 describes the per cent of our cohort with 0, 1–6 or 7+ primary care physician visits. From our initial models, the adjusted odds of being a single versus non-PV patient did not differ statistically pending if people had 1–6 versus 0 primary care physician visits, but did differ significantly pending if people had 7+ versus 0 primary care visits. In these instances we collapsed the reference group to include people with 0–6 primary care visits, after checking to ensure that the size of the adjusted estimate for ‘7+ primary care visits’ did not change significantly (this was done by looking to see if the adjusted estimate for ‘7+ primary care visits’ in the reduced model remained within the 95% confidence limits of the adjusted estimate for this category reported in the original model). Variables were not collapsed in instances where all (or no) subcategories differed statistically from the reference group, or when β estimates changed statistically by collapsing reference groups, as described above. All predictors were fitted in the model.

Also from interim analyses, most types of mental illness were highly related to anxiety and depression (eg, across all patient groups combined, 90.6% of people diagnosed as having a personality disorder were also diagnosed with having anxiety and depression; data not shown). To minimise co-linearity and given the high prevalence of substance abuse in our cohort, we included only this measure in our final logistic regression models.


Our initial cohort consisted of 128 017 patients who made 256 388 ED visits during the study period (data not shown). Several tests were conducted to verify the completeness and accuracy of these data. As highlights, only 5378 of all patients (4.2% of the original cohort) with 5634 ED visits (2.2% of all visits) had to be excluded due to missing or non-matching patient identifiers to other administrative (eg, the hospital abstract) files. A similar proportion of these excluded patients (1.4%; 77 of 5378 people) were defined as paradoxical as compared with our final study cohort, thus minimising exclusion bias. Also, cross tabulations across different files show that 97.4% of all ED visits with a disposition of ‘hospital admission’ were captured in the hospital abstract file on the same day, and similarly 98.4% of all ED visits ending in death were found in the mortality file. Also as expected, a much greater proportion of ED visits triaged as ‘resuscitation’ (CTAS 1) arrived by ambulance (67.4%) versus visits triaged as ‘non-urgent’ (CTAS 5) (9.0%). Patients were reported as ‘left without being seen’ during only 0.3% of resuscitation visits as compared with 14.0% of non-urgent visits. These findings demonstrate the accuracy and overall value of using the ED data system for research purposes.

Our final cohort consisted of 122 639 patients who made 250 754 ED visits during our study period (table 1). Across all ED sites combined, 2815 of these patients made 3387 paradoxical visits during the study period, at an average of 9.3 visits per day. The proportion of total patients (2.3%) and visits (1.4%) defined as paradoxical was roughly similar across ED sites (data not shown). This ranged from a low of 1.1% of all patients (0.7% of visits) at the Concordia ED, to a high of 3.6% of all patients (2.5% of visits) at the Grace ED.

Table 1

Frequency of paradoxical use at all ED sites combined; patient-based and visit-based data

Further analysis of paradoxical use is provided in table 1. Across all sites combined, 5.9% (3387/56 996) of ED visits arriving by ambulance were coded as paradoxical, as were 14.3% of all visits where patients left without consulting an ED provider. Also, the majority (89.3%) of PV patients made only one paradoxical visit during the study period. Alternatively, 10.7% of PV patients had two or more paradoxical visits during the study period. This group of multiple PV patients accounted for 25.7% of all paradoxical visits.

Descriptively, single PV patients were often unique from their non-PV counterparts (table 2). These former patients were younger (ie, 82.6% were 17–64 years old vs 74.9% of non-PV patients), and a greater proportion lived in the Winnipeg core area (35.7% vs 16.9% of non-PV patients) and also in lower income areas generally. While the prevalence of chronic physical disease was similar between single and non-PV patients (eg, 15.0% of single PV patients were diagnosed as having diabetes vs 13.9% of non-PV patients), many more PV (27.4%) versus non-PV (9.2%) patients had substance abuse problems.

Table 2

Description of patient demographic, illness and concurrent healthcare use profiles by paradoxical use category

Patterns of healthcare use differed substantially between study groups. More single (13.6%) versus non-PV (2.5%) patients visited EDs frequently. Also concurrent with their ED visits, 43.6% of single PV versus 32.2% of non-PV patients had seven or more contacts with a primary care physician. Conversely, a similar proportion of these study groups used physician-referred types of medical care including hospitalisations (9.7% of single PV vs 8.1% of non-PV patients were hospitalised one or more times during the study period) and specialist physician visits (eg, 24.7% of single-PV patients had one or more visits to a specialist physician during the study period vs 27.2% of non-PV patients). Last, single PV patients were triaged as less or non-urgent (ie, a CTAS score of 4 or 5) during a greater proportion of their visits (51%) versus non-PV patients (41.0% of their visits) (data not shown).

Multiple PV patients are defined as an extended version of their single PV counterparts; 65.7% of multiple PV patients visited the ED frequently, 53.0% resided in the Winnipeg core area and 54.7% were diagnosed with substance abuse problems. Overall, between 4.3% (non-PV patients) and 7.0% (multiple PV patients) died within 180 days of their last index visit.

Multivariate results extend these findings (table 3). After adjustment for all other risk factors, having substance abuse challenges was reported to differentiate single from non-PV patients; the adjusted odds of being a single versus non-PV patient was 2.38 times greater (95% CI ranging from 2.16 to 2.63 times greater, p<0.0001) for people who had (vs had not) been diagnosed as a substance abuser. Similarly, the adjusted odds of being a single versus non-PV patient was 3.83-fold (95% CI 3.33 to 4.39, p<0.0001) greater for people who were versus who were not frequent ED users during the study period. Single versus non-PV patients were also more likely to live in an urban core (adjusted OR; AOR=1.77, 95% CI 1.61 to 1.90, p<0001) and low-income (AOR=1.56, 95% CI 1.43 to 1.71, p<0.0001) area, and concurrent with their ED visits, to have seven or more contacts with a primary care provider (AOR=1.31, 95% CI 1.23 to 1.48, p<0.0001). Single versus non-PV patients were also more likely to be younger versus older (AOR=1.31, 95% CI 1.16 to 1.48, p<0.0001), were less likely to have multiple contacts with specialist physicians (AOR=0.84, 95% CI 0.76 to 0.93, p=0.001) and were also less likely to die within 6 months of their index ED visit (AOR=0.70, 95% CI 0.55 to 0.89, p=0.004).

Table 3

Multivariate modelling; comparing single with non-paradoxical users, and multiple with single paradoxical users

Fewer factors uniquely differentiated multiple from single PV patients (table 3). After adjustment for all other measures, multiple versus single PV patients were 7.24 times (95% CI 5.23 to 9.84, p<0.0001) more likely to be frequent ED users in general, and were 34.58 times (95% CI 21.48 to 55.68, p<0.0001) to be highly frequent versus not frequent users. The adjusted odds of being a multiple versus single PV patient was also 1.79-fold (95% CI 1.08 to 2.96, p=0.025) greater for young versus older people.


Several authors have emphasised the benefits of defining ED patients with unmet needs versus by the perceived inappropriateness of their ED visit.2–4 As one example of patients with unmet needs, we use the concept of a PV to profile patients who arrive by ambulance to EDs and then subsequently leave without consulting an ED provider. Our analyses show that PV visits occurred on average of 9.3 times/day in our study cohort. While the overall rate of patients leaving without consulting an ED provider (9.5%, table 2) is higher in this cohort versus the national rate of 3% in Canada,18 the proportion of ‘left without being seen’ ED visits ranges from 0.4% to 15.0% worldwide.19 Collectively, this literature supports the notion that our findings can be generalised to other large metropolitan areas.

PV patients in this study are unique in key areas as compared with commonly identified ‘inappropriate’ ED users. Only 13.6% of single PV patients in this study (65.7% of multiple) were also defined as frequent ED users. Further, while some characteristics (eg, residing in a lower income neighbourhood, having a history of substance abuse) of our PV patients mimic frequent users,5 ,20 ,21 the literature defines these latter patients as having a disproportionally high prevalence of chronic physical disease3 ,5 ,20 which our PV patients do not. Similarly, we have shown previously that frequent ED users tend to have many hospitalisations and contacts with specialist physicians.13 In contrast, single versus non-PV patients in this study had fewer visits with specialist physicians. Last, our study demonstrates that paradoxical patients were less likely to die following their last ED visit as compared with non-paradoxical patients. Collectively, these results demonstrate that PV patients are defined uniquely by sociodemographic characteristics and not by having acute urgent medical needs.

As compared with non-PV patients in this study, both single and multiple PV patients are characterised by sociodemographic factors emphasising their need for alternate care strategies likely involving allied health providers. Much evidence demonstrates the effectiveness of such strategies. For example, Gordon (2001) shows that employing social service workers in large urban EDs helps to decrease repeat ED visits and hospitalisations made by frequent ED users.22 Similarly, Pope et al23 have shown that linking EDs with community-based care practices successfully improves the psychosocial health of frequent ED users and reduces their ED use rates. Employing social workers in EDs to facilitate postdischarge planning has also been shown to reduce subsequent ED and hospital use among chronically ill homeless patients.24 Implementing these types of strategies may hold great value for our PV patients, especially given that many of these patients were triaged as having less or non-urgent needs.

Concurrent with their ED use, single PV versus non-PV patients were more likely to have at least seven visits with a primary care physician. This result further supports our conclusions made elsewhere stating that EDs are typically not used as a surrogate for other healthcare services,13 and also supports others who have shown that simply providing more access to physician care minimally impacts ED usage. For instance, Hansagi et al25 used nurse advisors to refer non-urgent ED patients to a primary healthcare centre for follow-up care. While ED use patterns were reduced for the overall referral group, no reduction in ED use was reported for the subgroup of referred patients denoted as frequent ED users.25 Similarly, Scherer et al26 tested the impact of referring ED patients for follow-up care to a nearby primary healthcare centre, and report that only about 40% of patients followed through with their referral visit, even when it was made for the next day or during the same week. Physicians’ referral alone is thus unlikely a useful alternative care strategy for the paradoxical patients as defined in this research.

Given this information, in terms of future research, the question is less about creating effective alternate solutions to help care for various subgroups of ED patients, and more about on whom to focus these alternate solutions. In addition to the patients identified in this research, we recognise that there are several equally plausible patient groups who may similarly benefit from receiving alternate care, such as those defined as requiring general-practice type care only.27 Following others who have used consensus building techniques to develop ED care quality indicators28 and EMS research priorities,29 we propose that Delphi techniques may help to identify further ED patient groups who may benefit most from receiving alternate care strategies.


Three limitations exist in this research. First, our analyses were restricted to adult patients only, and hence our results cannot be generalised to include younger individuals. Second, electronic ED records in Manitoba exist for Winnipeg only, and thus our findings cannot be generalised to rural areas. While we also cannot determine the extent to which WHR residents visited an ED in another health region, the distance to these additional sites is considerable (at least 45 miles), and it is unlikely that patients travelled to these latter EDs using WHR ambulance services. Third, EMS transport data provide important details about who initiated the ambulance transport and defines the types of treatments that are provided during transport. While not yet linkable in Manitoba, the use of these data would help to further profile paradoxical patients, and would clarify the role of EMS personnel in helping to care for them. This information also has value for developing more broad protocols that define when paramedics should transport ED patients versus care for them on-site.


The literature to date has not clearly identified ED patients who may benefit from alternate models of acute ED care. We profile in this paper patients who arrive to an ED by ambulance and then subsequently leave without consulting an ED provider (PV). As compared with non-PV patients, PV patients were defined generally as having substance abuse problems and by their sociodemographic measures, and not by factors (ie, specialist physician visits, hospital admissions, proximity to death) demonstrating their need for more urgent acute care. This information, when combined with the literature to date, demonstrates the need to develop alternate care strategies for PV patients likely involving the community-based sector. Consensus-building techniques are also required to define additional patient groups who may benefit most from these types of alternate care strategies.



  • Contributors MBD planned the study, supervised data analyses, wrote drafts and approved the final paper. SD assisted with the writing of all drafts and the final version of the paper. WP helped plan the study and provided detailed revisions to the paper. DC helped plan the study and provided detailed input into statistical analysis. CS, RLdF, EW and AC each contributed to revising various versions of the paper. SD provided all data analyses and contributed to revising the paper.

  • Funding This research was supported by an unrestricted research grant provided by Manitoba Health (HIPC # 2005/2006-15).

  • Competing interests All intellectual properties of this manuscript are solely those of the lead and contributing authors.

  • Ethics approval University of Manitoba Health Research Ethics Board, Bannatyne Campus (file # H2005:153).

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