Article Text

Download PDFPDF

Predictive variables of an emergency department quality and performance indicator: a 1-year prospective, observational, cohort study evaluating hospital and emergency census variables and emergency department time interval measurements
  1. Enrique Casalino1,2,3,
  2. Christophe Choquet1,3,
  3. Julien Bernard1,2,3,
  4. Abigael Debit1,2,3,
  5. Benoit Doumenc1,3,
  6. Audrey Berthoumieu1,
  7. Mathias Wargon1,2,3
  1. 1Assistance Publique-Hôpitaux de Paris (AP-HP), University Hospital Bichat-Claude Bernard, Emergency Department, Paris, France
  2. 2Université Paris Diderot, Sorbonne Paris Cité, Paris, France
  3. 3Study Group for Efficiency and Quality of Emergency Departments and Non-Scheduled Activities Departments, Paris, France
  1. Correspondence to Professor Enrique Casalino, Service d'Accueil des Urgences, Hôpital Bichat-Claude Bernard, 46 rue Henri Huchard, 75018 Paris, France; enrique.casalino{at}bch.aphp.fr

Abstract

Objective Emergency department (ED) crowding impacts negatively on quality of care. The aim was to determine the association between ED quality and input, throughput and output-associated variables.

Methods This 1-year, prospective, observational, cohort study determined the daily percentage of patients leaving the ED in <4 h (ED quality and performance indicator; EDQPI). According to the median EDQPI two groups were defined: best-days and bad-days. Hospital and ED variables and time interval metrics were evaluated as predictors.

Results Data were obtained for 67 307 patients over 364 days. Differences were observed between the two groups in unadjusted analysis: number of daily visits, number of patients as a function of final disposition, number boarding in the ED, and time interval metrics including wait time to triage nurse and ED provider, time from ED admission to decision, time from decision to departure and length of stay (LOS) as a function of final disposition. Five variables remained significant predictors for bad-days in multivariate analysis: wait time to triage nurse (OR 2.36; 95% CI 1.36 to 4.11; p=0.002), wait time to ED provider (OR 1.93; 95% CI 1.05 to 3.54; p=0.03), number of patients admitted to hospital (OR 1.86; 95% CI 1.09 to 3.19; p=0.02), LOS of non-admitted patients (OR 9.5; 95% CI 5.17 to 17.48; p<0.000001) and LOS of patients admitted to hospital (OR 2.46; 95% CI 1.44 to 4.2; p=0.0009).

Conclusions Throughput is the major determinant of EDQPI, notably time interval reflecting the work dynamics of medical and nursing teams and the efficacy of fast-track routes for low-complexity patients. Output also significantly impacted on EDQPI, particularly the capacity to reduce the LOS of admitted patients.

  • Acute coronary syndrome
  • bacterial
  • cardiac care
  • efficiency
  • emergency care systems
  • emergency department management
  • HIV
  • infectious diseases
  • neurology
  • stroke

Statistics from Altmetric.com

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.

Introduction

Overcrowding is the most serious problem in emergency departments (ED) worldwide,1 affecting 10−74% of hospitals surveyed.2–8 The definition of ED overcrowding is unclear.9 Several definitions have been proposed and various crowding scales have been developed but these lack scalability and do not perform well in ED where crowding is less frequent.10 Crowding negatively impacts on quality of care, patient satisfaction and satisfaction of the emergency healthcare provider.11–15 A prolonged length of stay (LOS) in the ED is associated with a lower quality of care and worse outcome measures.16–22

Various time interval metrics have been used as quality indicators or crowding markers in ED.23 ,24 LOS has recently been proposed as a quality indicator,25 ,26 and as a valuable marker of overall ED efficiency and overcrowding.27 Excess LOS is defined as more than 4 h in the UK,28 ,29 4−6 h in Canada30 ,31 and 8 h in Australia.32 The ‘four-hour rule’ mandates that ED patients are discharged or admitted to hospital within 4 h of arriving in the ED.33 In the USA, a 4−6 h target has recently been used as a performance measure.34 In 2000, the NHS in England instituted a maximum length of ED stay of 4 h as a tool to improve ED quality, and a wait time to ED provider of 75 min, to eliminate the problem of corridors lined with sick patients waiting to be seen by a physician or be admitted to a hospital bed.28 Although there is no clear evidence that this target can improve quality of care,35 the introduction of the 4-h rule for ED patients has led to improvements in the proportion of patients managed within this timeframe, and its implementation has allowed hospitals and ED to set up organisational strategies to optimise the efficacy and performance of ED.33 ,36

Goals of this investigation

Several community, patient, ED and hospital-related causes of ED overcrowding, long wait times and long LOS have been identified.1–13 A national quality forum endorsed 10 voluntary consensus standards for ED quality, including measures of ED wait times and visit lengths for admitted and non-admitted patients.34 However, few data are available concerning their impact on ED, or on the interaction between different time intervals and ED quality and performance.

We investigated the association between the percentage of patients leaving the ED in less than 4 h (ED quality indicator) and independent variables including patient characteristics, ED census and organisational aspects, hospital census variables and time interval metrics in the ED.

Methods

Study design

This was a prospective, observational, cohort study measuring the daily percentage of patients with an ED visit of less than 4 h. In our unit this measure is known as the ED quality and performance indicator (EDQPI). This was the principal outcome measure.

The purpose of the study was to investigate the association between EDQPI and independent variables believed to be related to this outcome variable.

Two groups were defined according to the daily median EDQPI value:

  1. Best-days: days with a percentage of visits of 4 h or less equal or superior to the annual median value;

  2. Bad-days: days with a percentage of visits of 4 h or less inferior to the annual median value.

Setting

The study was carried out in the Bichat-Claude Bernard ED, a university unit located in the Paris metropolitan area. The annual expected volume of this ED, which receives only adult patients (≥15 years old), is between 60 000 and 65 000 visits. The study was carried out for the 12-month period from 1 April 2009 to 30 March 2010.

Selection of participants

All visits to the Bichat-Claude Bernard ED during the study period were included. Daily values for time interval measures and ED and hospital census variables were extracted from the electronic medical record database. This dataset was completely anonymous and did not contain any identifiable personal health information. The dataset is currently used as an ED quality and performance measure as part of an ongoing emergency activity and performance evaluation approved by the Assistance Publique-Hôpitaux de Paris committees on research and informatics.

Methods of measurement and outcome measures

Data about patient characteristics, time intervals and emergency census variables were obtained from the ED medical records (Urqual, France) and specific electronic database. Electronic medical records automatically record the time of arrival in the ED and the time when the different interventionists (triage nurse, ED provider) take over the patients, as well as the time of the final disposition decision by the ED physician and the time of departure from the ED. Hospital census variables were obtained from a specific electronic database (Gilda, Assistance Publique-Hôpitaux de Paris, Paris, France).

All independent variables were recorded as the number of daily occurrences and, for time interval metrics, measured as the mean±SD, median and IQR (25−75%).

The following variables were studied: patient characteristics, ED census variables and hospital census variables: daily visits; triage acuity level; type of visit (medical, surgical, psychiatry); patient's final disposition (not admitted, admitted (overall and admitted to the hospital, admitted to the observation unit (OU) and transferred patients)); patients aged 75 years or older and final disposition of these patients (not admitted, admitted (overall and admitted to the hospital, admitted to the OU and transferred patients)); admitted patients held in the ED at 08:00 h; 14:00 h and 18:00 h; available beds in the OU at 18:00 h.

Acuity was measured using a five-level scale (level I, resucitation; level II, emergent; level III, urgent; level IV, less urgent; level V, non-urgent).37 Similarly, senior nurses recorded the number of patients boarding in the ED (patients admitted to the hospital and held in the ED for at least 2 h) three times a day, at 08:00 h, 14:00 h and 18:00 h.

Mean time intervals were calculated automatically on a daily basis and were defined as follows: (1) wait time to triage nurse: number of minutes between the time the patient was identified in the ED and the time the patient was seen by the triage nurse; (2) wait time to provider (ED physician): number of minutes between the time the patient was identified in the ED and the time the patient was treated by the ED provider; (3) time interval from ED arrival to decision: number of minutes between the time the patient was identified in the ED and the time of disposition decision by the ED provider; (4) time interval from decision time to departure time: number of minutes between the time of disposition decision by the ED provider and the patient leaving the ED; (5) LOS: number of minutes between the time the patient was identified in the ED and the time the patient left the ED. Some of these time intervals were tracked for subsets of patients as a function of patient category, acuity level and patients' final disposition.

Primary data analysis

Data were extracted from the computerised ED system (Urqual, McKesson, Paris, France) and analysed using Statistica 10 software.

Standard statistics were used for the descriptive analysis: frequency, mean, median and IQR for time intervals and census parameters.

Spearmans' correlation coefficient was calculated between EDQPI and each continuous variable including census variables and time interval measurements; r>0.6 indicated a moderate to important correlation.

The measurements were compared for each variable in the best-days and bad-days groups. Statistical significance of observed differences in categorical outcomes was assessed using χ2 or Fisher's exact test, as appropriate. OR and 95% CI were calculated. For each studied variable, categories were defined by median value. For interval data, the non-parametric Kruskall–Wallis test or Mann–Witney U test was used. All statistical tests were two-tailed and p=0.05 determined statistical significance.

We used multivariable logistic regression analysis to investigate the predictive power of the variables in combination and their independent effect to test for differences between study groups constituted according to their median values. The presence of confounding was assessed empirically by entering potential covariates into a logistic regression model one at a time and by comparing the adjusted and unadjusted OR. Final logistic regression models included covariates that altered unadjusted OR by at least 10%. Redundant time interval measures (LOS as a function of different variables, for exemple final disposition decision and acuity level) were not included in the same model. Our analysis includes forward stepwise selection, forced entry of all potential risk factors and backward elimination. Our final model includes only variables statistically significant in one or more of these approaches, to reduce the chance of identifying spurious associations.38 Model fit was determined by Hosmer–Lemeshow statistics for homoscedasticity39 and the C-statistic to reflect overall fit.40

Results

Characteristics of study subjects

Data for 67 307 patients seen over 364 days were analysed (mean age 43.7±20 years, IQR 28−56, median 40). Patients arriving by ambulance represented 22 726 visits (33.8%). Visits on week and weekend days represented 303 days (83.2%) and 61 days (16.8%), respectively.

Table 1 summarises the baseline characteristics of the study cohort. There were between 130 and 238 visits per day (median 184). Triage assessment was as follows: level 1, 1%; level 2, 10.5%; level 3, 28.6%; level 4, 31%; level 5, 28.9%. Medical, surgical and psychiatric patients represented 58.4%, 35.9% and 5.7%, respectively.

Table 1

Baseline characteristics of the study cohort

Mean daily overall admission rate was 22.2±3.4% (IQR 19.8−24.4, median 22). Patients admitted to the OU represented 12.4±3% of daily visits, patients admitted to the hospital 7.3±2% and transferred patients 2.5±1.3%. Admission rates as a function of triage level were: level 1, 89.1%; level 2, 40.2%; level 3, 54.8%; level 4, 12.2%; level 5, 2.9%.

Patients aged over 75 years represented 9.6±2.4% of daily visits (IQR 4.3−16.5, median 9.6). The mean daily admission rate for patients aged over 75 years was 56.5 ± 13.1% (IQR 47.4−65.6, median 57).

Table 1 shows the mean daily time intervals as a function of patient characteristics, ED and hospital census variables. Some significant differences were found in daily time interval metrics as a function of categorical variables, notably time intervals from ED arrival to decision and from disposition decision to departure time, and for both as a function of final disposition decision. Similarly, significant differences were found for LOS according to patient category, acuity level and final disposition decision.

Main results

Interaction between EDQPI and time interval metric measurements

Spearman correlation coefficients between EDQPI and ED and hospital census variables are shown in table 2. None of the variables analysed had r>0.5. Table 2 also shows the results of Pearson's correlation analysis between EDQPI and time interval metrics. Wait time to ED provider, time interval from ED to a decision of non-admission, LOS of non-admitted patients, LOS for triage level 4 and LOS for triage level 5 all had r>0.6.

Table 2

Pearsons' correlation coefficient analysis

Quality and performance outcome measurements

The overall mean daily percentage of patients with a LOS of 4 h or less was 66±8% (IQR 62−72, median 67). Bad-days and best-days groups were defined according to EDQPI median value (bad-days, ≤67%; best-days, >67%).

Predictive factors for bad-days and best-days

Unadjusted analysis

Table 3 shows the results of the comparison bad-days versus best-days for continuous and categorical variables. Variables were dichotomised according to their median value.

Table 3

Predictive value of continuous and categorical variables. Unadjusted analysis

For daily visits, several comparisons were carried out: an analysis of bad and best-days versus two groups formed on the basis of the median number of daily visits and then analysed versus two groups formed on the basis of the 3rd percentile of the number of daily visits, and finally an analysis versus four groups formed on the basis of the 25th, 50th and 75th percentiles. All of these analyses were statistically significant (p<0.05). The median number of daily visits was retained as the value for analysis and only that result is shown in table 3 because it had the greatest statistical significance.

The overall admission rate (%) was significantly different between bad and best-days groups (23±3.3 vs 21.5±3.3, respectively; p=0.000002).

Adjusted logistic regression predictive factors associated with bad-days

After controlling for other variables, bad-days were more often associated with longer wait time to triage nurse and ED provider, with a large number of patients admitted to the hospital, and with a longer LOS for both admitted and non-admitted patients (table 4).

Table 4

Multiple logistic regression analysis predicting bad-days

Discussion

ED overcrowding is the result of a mismatch between the capacity to regulate and organise patient flow at three levels: input, throughput and output.4 Our results indicate that the daily percentage of patients leaving the ED in 4 h or less (EDQPI) is associated with ED operating characteristics measuring input to the ED, ED and hospital variables measuring throughput and ED and hospital-dependent variables measuring output.

Input variables indicating the demand for ED services were studied by evaluating the number of daily visits. In unadjusted analysis, significant differences were found between the two groups. Pearson's correlation analysis of the total number of daily visits and EDQPI gave r=0.268 demonstrating the weak impact of this variable on ED performance. Other authors have reported similar results, with the number of daily visits having only a moderate or non-significant impact on ED performance and crowding.41 ,42 Patients characteristics in terms of complexity and consumption of care should also be considered. In unadjusted analysis there was no significant difference between the two groups in the daily flow distribution as a function of patient type (medicine, surgery, psychiatry), or their acuity level. It has been reported that the number of admissions correlates with ED LOS in a variety of hospital settings.43 In our study, the total number of patients admitted was significantly greater on bad than on best-days. However, none of these input variables was significantly associated with EDQPI in multivariate analysis. Older patients are more often admitted and have a longer LOS compared with younger patients.44–46 Older patients are more complex clinically, require more time and attention, more often present with comorbidities, and are more often hospitalised.47 In unadjusted analysis, some associations were found between ED input variables focusing on elderly patients (number of daily visits, overall number of daily admissions, number of daily admissions to hospital, number of daily transfers) and EDQPI. However, none of these variables was a significant predictor in multivariate analysis.

Throughput indicators measure the performance of the ED in managing patients and the capacity of the ED to make decisions and orient patients towards other sectors of care. Patient flow has been proposed to measure throughput or process performance within the ED.48 In the present study, a number of time intervals was measured. Wait time from ED arrival to triage nurse was significantly longer on bad-days than on best-days (13±4 vs 10.9±3.2 min, respectively; p<0.000001). Triage nurse ordering has been related to enhanced patient satisfaction, improved care and improved team work, and appears to be an effective intervention to reduce ED LOS.49 This was the only variable in our study to evaluate nursing team performance in the process of care and, in our opinion, it probably reflects the work dynamics of all nursing teams beyond those of triage. Wait time from ED arrival to ED provider was also studied. Prolonged wait time decreases patient satisfaction, limits access, increases the number of patients who leave before being seen, and is associated with significant delays in care for patients with pneumonia, cardiac symptoms, or abdominal pain.14 ,15 ,50–55 In the present study, wait time to ED provider was significantly longer on bad-days than on best-days (78.1±18.3 vs 59.2±11.4 min, respectively; p<0.000001). Spearman's correlation analysis showed r=−0.72, indicating a strong association between EDQPI and wait time to ED provider. In multivariate analysis, wait time to triage nurse (OR 2.36; 95% CI 1.36 to 4.11; p=0.002) and wait time to ED provider (OR 1.93; 95% CI 1.05 to 3.54; p=0.03) remained significant predictors of bad-days. This shows the importance of the work dynamics of medical and nursing teams in reducing the wait time and the overall LOS.

Similarly, the time intervals from ED arrival to final disposition decision were also evaluated because these indicate the complete process of diagnostic and therapeutic evaluation and decision-making in the ED. In the present study, the time intervals from arrival to final disposition decision as a function of admission category were significantly longer on bad-days in unadjusted analysis. Pearson's correlation coefficient analysis gave r=0.72 between EDQPI and time to decision for non-admitted patients indicating a strong correlation between both variables. The capacity to organise adapted pathways of care for non-admitted patient thus has an impact on overall ED performance. Similarly, overall LOS, as well as LOS as a function of patient category (medicine, surgery, psychiatry), acuity level and final disposition decision, reflects the capacity of ED to set up adapted care pathways and to accelerate the flow of patients through the ED. For all of these variables, time interval measurements were significantly longer for bad-days than for best-days (table 3).

In the present study, most non-admitted patients were low-complexity patients (acuity levels 4 and 5). In multivariate analysis, LOS of non-admitted patients remained a significant predictor of bad-days and had the strongest predictive value (OR 9.5; 95% CI 5.17 to 17.48; p<0.000001), indicating that a reduction in time interval for these patients has a major impact on overall ED performance. Our results support the optimisation of specific pathways for patients of low-complexity and non-admitted patients in a strategy of overall improvement of ED quality and performance. Fast-track units should be set up in ED, as proposed previously.56 ,57

Management of ED output is complex. Overall ED LOS, even among discharged patients, is related to hospital bed occupancy.58–60 The availability of inpatient beds to which patients can be moved is an important factor in ED performance and LOS, particularly for severely ill patients.61 One of the most significant causes of ED crowding is hospital bed shortage.46 ,62 ,63 Here, we calculated the total daily number of admitted patients, patients admitted to the OU, patients admitted to the hospital and transferred patients, and the daily number of geriatric patients admitted to the hospital, to the OU, or transferred. In unadjusted analysis, all of these variables except the number of daily admissions to the OU differed significantly, with a higher number of admissions in the bad-days group. However, in multivariate analysis only the total number of patients admitted to the hospital remained a significant predictor of bad-days (OR 1.86; 95% CI 1.09 to 3.19; p=0.02). Surprisingly, our study showed that on the days when the hospital admitted a higher number of patients from the ED, ED performance was significantly less good. Several explanations can be proposed. The number of patients admitted from the ED does not necessarily reflect the number of beds available in the hospital. Other factors may intervene in patient admission from the ED such as transport availability, housekeeping practices, admission procedures and prioritisation of non-ED admissions.

To complete the evaluation of output, the LOS of admitted patients was also assessed as a function of their final disposition (hospital, OU, or transfer). In unadjusted analysis, LOS as a function of final disposition decision was systematically longer on bad-days than on best-days, notably LOS of patients admitted to the hospital (328.7±78.9 vs 286±88.8 min, respectively; p=0.000001). Unadjusted analysis also showed that the time interval from ED arrival to final disposition decision and the time interval from final disposition decision to patient departure were significantly longer on bad-days. In multivariate analysis, LOS of patients admitted to the hospital remained a significant predictor of bad-days (OR 2.46; 95% CI 1.44 to 4.2; p=0.009). These results suggest that on the days when the hospital admits more patients ED performance is not improved if the ED is not capable of reducing the time interval between arrival and final disposition decision, or if the hospital is not capable of reducing the delay between decision and admission to a service. These results demonstrate that the principal output predictive factor for bad-days is not the total number of patients admitted to hospital but the ease and speed with which a bed is identified and the patient is transferred to the medical or surgical ward.

As an indicator of output, our study also evaluated the number of ED boarding patients at three different times during the day, as well as the number of beds available in the OU at 18:00 h. The number of patients boarding in the ED has been proposed as a variable in a model for crowding score calculation.4 In unadjusted analysis, some of these variables were found to be significantly associated with the performance indicator, bad-days having a greater number of patients boarding in the ED at 08:00 h, 14:00 h and 18:00 h. However, none of these variables was a significant predictor in multivariate analysis.

Study limitations

Our study has several limitations. First, as no unique accepted quality or performance indicator of ED or of crowding exists,4 we chose to use the percentage of patients leaving the ED in less than 4 h.33 ,35 ,36 Second, the study design could not determine how changes in patient flow occurring during the day (input and output) might affect their LOS and the EDQPI. Third, crowding itself was not measured because crowding measures assess crowding as a static phenomenon, whereas crowding is dynamic and can fluctuate during the day.4 ,41 ,64 Fourth, the analysis relied on administrative data collected daily for pilotage purposes. Fifth, the study was conducted in a university-affiliated hospital in a Paris-Metropolitan area and the results cannot be generalised. Sixth, our data lack information about hospital occupancy rates, processes of care and organisational procedures. Finally, mean annual EDQPI in the present study was 67% while the reported value in UK was 97.4%.65

Conclusion

A comparison between best-days and bad-days demonstrated that five variables remained significant predictors for bad-days in multivariate analysis: (1) wait time to triage nurse (OR 2.36; 95% CI 1.36 to 4.11; p=0.002); (2) wait time to ED provider (OR 1.93; 95% CI 1.05 to 3.54; p=0.03); (3) number of patients admitted to the hospital (OR 1.86; 95% CI 1.09−3.19; p=0.02); (4) LOS of non-admitted patients (OR 9.5; 95% CI 5.17 to 17.48; p<0.000001); and (5) LOS of patients admitted to the hospital (OR 2.46; 95% CI 1.44 to 4.2; p=0.0009).

It can be concluded that input has little or no impact on EDQPI, even when the complexity of the patients is taken into account. Throughput is the major determinant of EDQPI. Time intervals measuring the work dynamics of medical and nursing teams are particularly important in measuring overall ED functioning. Finally, the efficacy of specific fast-track pathways for patients of low complexity who are not admitted to hospital is the most important factor in the overall performance of the service. Output has a significant impact on EDQPI, particularly ED processes allowing patients who have an indication for hospitalisation to be admitted as quickly as possible.

In summary, the largest contributors to ED crowding and delays in care are not ‘input’ factors, but rather ‘throughput’ and ‘output’ factors. Measurement of time interval metrics is a useful tool to evaluate ED processes, performance and quality of care.

References

Footnotes

  • Competing interests None.

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