Article, Emergency Medicine

The effects of emergency department crowding on triage and hospital admission decisions

a b s t r a c t

Background: Emergency department (ED) crowding is a recognized issue and it has been suggested that it can af- fect clinician decision-making.

Objectives: Our objective was to determine whether ED census was associated with changes in triage or disposi- tion decisions made by ED nurses and physicians.

Methods: We performed a retrospective study using one year of data obtained from a US academic center ED (65,065 patient encounters after cleaning). Using a cumulative logit model, we investigated the association be- tween a patient’s acuity group (low, medium, and high) and ED census at Triage time. We also used multivariate logistic regression to investigate the association between the Disposition decision for a patient (admit or dis- charge) and the ED census at the disposition decision time. In both studies, control variables included census, age, gender, race, place of treatment, chief complaint, and certain interaction terms.

Results: We found statistically significant correlation between ED census and triage/Disposition decisions. For each additional patient in the ED, the odds of being assigned a high acuity versus medium or low acuity at triage is 1.011 times higher (95% confidence interval [CI] for Odds Ratio [OR] = [1.009,1.012]), and the odds of being assigned medium or high acuity versus low acuity at triage is 1.009 times higher (95% CI for OR = [1.008,1.010]). Similarly, the odds of being admitted versus discharged increases by 1.007 times (95% CI for OR

= [1.006,1.008]) per additional patient in the ED at the time of disposition decision. Conclusion: Increased ED occupancy was found to be associated with more patients being classified as higher acu- ity as well as higher Hospital admission rates. As an example, for a commonly observed patient category, our model predicts that as the ED occupancy increases from 25 to 75 patients, the probability of a patient being triaged as high acuity increases by about 50% and the probability of a patient being categorized as admit increases by around 25%.

(C) 2019

Introduction

Emergency Departments (EDs) are busy places. In 2015 there were

136.9 million ED visits in the United States [1]. This high volume often leads to ED crowding that has been associated with numerous negative patient outcomes including delays in lifesaving care that result in in- creased mortality and Low patient satisfaction [2-5].

It has been suggested that crowding of the emergency department can lead to difficulties with clinician decision-making and potentially impact equity in care [6]. Two such vital decision points that are tied

* Corresponding author.

E-mail address: [email protected] (B. Linthicum).

to care quality and equity are the triage level assignment decision made by nursing staff and the disposition decision made by providers. Nationally, emergency departments represent a significant source of hospital admissions accounting for nearly all the growth of hospital ad- missions in recent years [7]. The decision to admit a patient is made by emergency providers based upon available Individual patient data, however recent research suggests that this decision may also be influ- enced by crowding of the ED itself [8]. This recently published study at a single academic medical center finds a statistical association between the likelihood of hospital admission and increased ED census. It was suspected that as EDs become busier there is a cognitive offloading that occurs for the physician by admitting patients rather than spending time and mental energy arranging safe discharges for patients who may

be in a “gray area.”

https://doi.org/10.1016/j.ajem.2019.06.039 0735-6757/(C) 2019

Making a disposition decision sooner during an individual patient’s visit rather than waiting to see if a patient improves during the ED stay allows physicians to move on to see the next patient or complete the next task. There is some evidence from literature that as load in- creases in a system, workers speed up their service rate [9] and this ef- fect may be what is being observed during times of high ED volume. Physicians may be, in effect, speeding up their services and increasing their “productivity” by choosing admission over discharge for patients who are in the gray area and for whom the right decision is not clear. Another study found that as the ED becomes more crowded the number of patients who are admitted to the hospital and have less than a 24-h hospital stay increases; suggesting that some of these admissions that occur during times of high census may be avoidable [10].

In other areas of healthcare, this relationship between decision mak- ing and crowding has also been found. One study found a correlation be- tween ICU occupancy level and the rate of ICU discharges [11]. Another study found a similar relation in obstetrics, where midwives were more likely to refer high complexity patients to obstetricians at times of in- creased congestion as opposed to when census levels are much lower [12].

This change in decision-making seems to occur even though it fur- ther contributes to system congestion. Ironically, boarding of admitted patients is thought to be a sizable contributor to crowding itself resulting in throughput delays of both admitted and discharged patients at an ED. [13,14] Understanding the relationship between ED census and individual provider and nurse decision-making may provide oppor- tunity for operational changes in workflow to prevent decision fatigue at times of high census. Previous work has demonstrated the existence of a safety tipping point [15]. Knowing that such a point exists and where it lays can aid in operational planning.

In addition to the admission decision, another critical decision that is made during a patient’s ED visit is the triage classification. This is often the first important decision made during a patient’s ED visit affecting how quickly the patient is evaluated by a provider. Only one other study has investigated the relationship between ED crowding and triage decisions and they concluded that there was no association [16]. Note that this study used the Australasian National Triage scale at a single ter- tiary care hospital in Australia. Furthermore, it treated patient census as a binomial categorical factor of “busy” or “non-busy” utilizing a single value to separate the two. A “busy” weekday in this study was defined as N140 visits whereas 139 visits would constitute a “non-busy” weekday.

The aim of our study was to use statistical methods to test the hy- potheses that ED census was associated with changes in triage and dis- position decisions at an academic hospital in Southeastern US. To the best of our knowledge, our study is the first to look at ED census and tri- age assignment decisions by using the census level directly in the anal- ysis rather than introducing arbitrary binary classifications (e.g., busy vs. non-busy) for the census level. Therefore, our modeling framework supports the exploration of how census count is associated with triage or admission decisions along the complete range of observed census levels.

Materials and methods

Study design and setting

Following approval from the institutional review board, we per- formed a retrospective study using a data set of patient visits collected at the ED of an academic hospital in the Southeastern US. During the study period, which covered the year 2012, this ED received approxi- mately 184 patient arrivals per day (67,203 patient visits per year). This is similar to the mean (61,447 visits per year) and median (60,639 visits per year) patient volumes from a survey of 75 academic emergency departments across the U.S. during the same year [17]. The triage system in place was the 5-level Emergency Severity Index

triage system, with levels from ESI 1 (patient dying) to ESI 5 (no ED re- sources needed) [18]. At the time of the study the ED had 59 beds spread across five adult pods: A, B, C, D, and a behavioral health ED (BHED), as well as a pediatric pod. Pods A and B operated 24 h a day seeing acute adult patients while pod D operated during peak hours and cared for primarily lower acuity patients. Pod C and BHED were dedicated to be- havioral health patients although occasionally other patients were housed in these areas. Due to the non-homogeneity and inconsistent nature of their visits to the ED and hospital, behavioral health patients were excluded from our statistical analysis.

Data analysis

The data available for each patient included demographic informa- tion (age, gender, and race), clinical information (triage acuity/ESI and chief complaint), disposition category (admit or discharge), and place of treatment (pod). Our goal was two-fold, to investigate the association between census and nurses’ triage decision, and similarly the associa- tion between census and physicians’ admission decision. We also con- sidered other available variables as potential control variables in the model (e.g., a patient’s age may impact either the triage nurse’s assess- ment or the admission decision by the provider) with reference to the relevant literature.

The data were cleaned before use in the statistical models. We de- leted questionable data elements including but not limited to obvious erroneous entries, patient walkouts, behavioral health visits, or time el- ements that occurred in non-chronologic order. Additionally, we ex- cluded patients with invalid or missing Acuity scores. Duplicate records and those with missing or insufficient entries for the variables of interest were also excluded from the study. Whereas the original data had approximately 67,203 entries, after cleaning the data set contained 65,065 validatED patient encounters eligible for statistical modeling.

Patient age was categorized into 8 Clinically meaningful groups: b3 month(m) old, 3 m to 3, 3 to 8, 8 to 18, 18 to 40, 40 to 55, 55 to 70, and >=70. These age groups were included as the levels of a categorical variable in subsequent statistical modeling. All other variables were also treated as categorical with the exception of census level, which was included in all models as a continuous variable, enabling us to asso- ciate any observed census count with the likelihood of admission or tri- age decisions. For race and pod, we combined categories that have b10 outcomes of each type of response (according to the criterion suggested in Agresti [19]) to a single category named “Other”.

Exploratory analysis confirmed that a patient’s chief complaint could be highly predictive of admission and hence was a desirable component to include in the model. To control the complexity of the model, we se- lected the 45 most common chief complaints (out of 8000), which had sufficient numbers of occurrences as to be informative. These 45 chief complaints were included explicitly in the model as levels of the “chief complaint” factor. (For a list of these 45 chief complaints, see Table S1 in Supplemental Material.) All other chief complaints were included in the “Other” category. This way, we retained much of the information contained in the chief complaint data while limiting the complexity of the model.

Census, which was our primary control variable of interest, refers to the total number of patients in the ED, i.e., the number of patients in the waiting room and those occupying a bed. For our analysis of triage deci- sions, the census level used for each triage decision was the census level at the time of the corresponding patient’s arrival, whereas for the anal- ysis of disposition decisions, the census level was computed at the dis- position decision time of the corresponding patient. In addition to the overall ED census, we also considered boarder census, which is the total number of Boarding patients in the ED, as a potential control vari- able for our Statistical models to see if the number of boarders could be correlated with provider decisions.

Table 1

Breakdown of patient characteristics for variables of interest.

Characteristic Percent in data set

Disposition

Admit 29.6

Discharge

70.4

ESI

1

0.9

2

13

3

57

4

24.9

5

4.2

Gender

Female

54.6

Male

45.4

Race

African American

30.0

Asian

1.1

Caucasian

53.8

Native American

0.4

Other

12.3

Unknown

2.4

Age

Below 3 m 0.8

3 m to 3 5.2

3 to 8 4.7

8 to 18 7.5

18 to 40 34.3

40 to 55 21.6

55 to 70 15.3

Over 70

10.6

Pod

A

27.8

B

23.4

C

2.8

D

27.2

Pediatrics

15.7

BHED

3.1

Table 1 illustrates the breakdown of characteristics of all the patients in the cleaned data set with the exception of chief complaints (due to its large number of categories) and census (because it is treated as a continuous variable). Prior to model fitting, we performed an explor- atory data analysis to assess the univariate association between the con- trol variables and the outcomes, i.e., triage level/ESI and disposition (admit and discharge). Also, we have not found any significant multicollinearity among control variables as we explain in more detail in Supplemental Material. All data and statistical analysis in this work was performed in R [20].

Statistical modeling

Association between census and triage decision

To investigate how census might impact Triage nurses‘ assignment of Acuity levels, we fit a cumulative logit model [19]. We collapsed the five level ESI scale into three acuity groups: low (ESI 4/5), medium (ESI 3) and high (ESI 1/2). This reduced the complexity of the response var- iable in the model (acuity assignment) without losing much informa- tion as relatively few patients in the data set were assigned an ESI 1 or ESI 5 score. This resulted in a three-level cumulative logit model with low, medium or high acuity group as the response variable, which depended on census and the other relevant independent variables discussed previously. Specifically, the cumulative logit modeling ap- proach enabled us to understand how an independent variable (such as census) may be associated with the likelihood of a patient being placed into each of the categories of interest (such as low, medium or high acuity).

After the exploratory analysis, we conducted likelihood ratio tests between several candidate models (with different sets of independent variables) to identify a final model sufficient for testing the following

hypothesis: ED census count has an impact on the likelihood of a patient being triaged in the low, medium or high category by the triage nurse. Table 2 provides the control variables of the resulting cumulative logit model for acuity group (low, medium, high) as the dependent variable and the p-value results of the likelihood ratio tests for each control var- iable. Note that all independent variables included in this cumulative logit model are significantly associated with the dependent variable at a 0.01 level of confidence. (The p-value result of the likelihood ratio test for boarder census was 0.41, which indicated that including boarder census in addition to the overall census does not statistically improve the model.)

Association between census and admission decision

In this part of the study, we fit a multivariate logistic regression model to assess the association between the disposition decision and census, which is calculated at the time a disposition decision is made for the corresponding patient. The logistic regression model is similar to the cumulative logit model, but only has two categories (admit or dis- charge) for the dependent variable. We considered multiple models and conducted likelihood ratio tests to identify which control variables to in- clude in the final model. The control variables in the final model and the corresponding p-value results of the likelihood ratio tests for model se- lection are provided in Table 2. Note that all independent variables in- cluded in the final logistic regression model are significantly associated with the dependent variable at a 0.05 level of confidence. (The p-value result of the likelihood ratio test for boarder census was 0.78, which indicated that including boarder census in addition to the overall census does not statistically improve the model.)

Results

To estimate the impact of census on triage acuity assignment and disposition decision, we calculated odds ratios (ORs) [19] for both statis- tical models discussed in the statistical modeling section above. Specif- ically, in this case, the OR indicates how changes in a control variable (such as census) may increase or decrease the likelihood (odds) being assigned to a higher acuity level or being admitted. We next discuss our findings from each model separately.

Association between census and triage decision

We found by fitting the cumulative logit model with partial propor- tional odds that the relationship between nurses’ triage decision and census (at time of arrival) was statistically significant. The OR for a

Table 2

p-values from likelihood ratio tests for all independent variables included in the selected cumulative logit model for triage decisions and multivariate logistic regression model for disposition decision.

Cumulative logit model for triage decision

Control variables p-Value

Race b0.01

Gender b0.01

Age group b0.01

Chief complaints b0.01

Census b0.01

Multivariate logistic regression model for disposition decision

Race b0.01

Gender b0.01

Age group b0.01

Acuity b0.01

Pod b0.01

Census 0.014

Chief complaints b0.01

Interaction between age and acuity b0.01

Table 3 Odds ratios of Prob(high acuity) versus Prob(low or medium acuity) and Prob(medium or high acuity) versus Prob(low acuity), and corresponding 95% confidence intervals for in- tercept, census, race, gender, and age.

Association between census and admission decision

In the multivariate logistic regression model fitting, we found that

there was a statistically significant association between providers’ ad-

Intercept

Census

Prob(high acuity)/Prob(low or Prob(medium or high medium acuity) acuity)/Prob(low acuity)

0.057 [0.052,0.063] 1.403 [1.315,1.496]

1.011 [1.009,1.012] 1.009 [1.008,1.010]

mission decision and census at the time when disposition decisions are made. The OR for admission per patient increase in census was

1.007 (95% CI = 1.006 to 1.008). ORs from the multivariate logistic re- gression analysis are reported in Table 4 except for chief complaints and interaction terms, which are provided in Tables S2 and S3, respec- tively, in Supplemental Material. For an example of the logistic regres-

Race (contrast: Caucasian)

African American

0.699 [0.661,0.739]

0.693 [0.665,0.722]

Asian

0.792 [0.628,1.002]

0.898 [0.759,1.062]

Native American

1.219 [0.847,1.752]

1.387 [0.998,1.928]

Other

0.540 [0.493,0.592]

0.778 [0.735,0.822]

Unknown

0.994 [0.852,1.160]

0.898 [0.800,1.007]

Gender (contrast: Female)

Male 1.345 [1.282,1.410] 0.901 [0.869,0.935]

Age group (contrast: 18 to 40)

Below 3 m

2.143 [1.683,2.729]

0.970 [0.799,1.178]

3 m to 3

0.644 [0.554,0.749]

0.422 [0.390,0.457]

3 to 8

0.794 [0.687,0.918]

0.462 [0.426,0.500]

8 to 18

1.741[1.591,1.905]

0.812 [0.760,0.868]

40 to 55

1.165 [1.088,1.247]

1.401 [1.334,1.470]

55 to 70

1.551 [1.445,1.664]

2.494 [2.343,2.655]

Over 70

1.705 [1.577,1.844]

5.601 [5.076,6.181]

patient being triaged as high acuity versus low or medium is 1.011 times greater when census is increased by one unit (95% CI = 1.009 to 1.012). We also found that for triaging a patient as medium or high versus low acuity is 1.009 times higher when census is increased by one unit (95% CI = [1.008, 1.010]). Results on odds ratios for all variables are reported in Table 3 except for chief complaints, which are provided in Table S1 in Supplemental Material. Using the cumulative logit model, we also calcu- lated the marginal probabilities of being assigned each acuity level (low, medium, and high) at different census levels for a common group of pa- tients (Caucasian females aged between 18 and 40 who had abdominal pain as their chief complaints); see Fig. 1. Such a framework is useful for interpreting results for key patient subpopulations.

sion model, we computed the probability of admission for a common group of patients: Caucasian females who are aged between 18 and 40, categorized as ESI3, with a chief complaint of abdominal pain and treated in Pod A, at different levels of census. The result is shown in Fig. 2. The slope of the line is the same for all patients in the model how- ever the probability of admission is higher or lower based on individual patient characteristics.

Discussion

To the best of our knowledge, there is only one other study that in- vestigated the relationship between nurses’ triage decision and ED cen- sus at the decision time and we are the first to consider census as a continuous variable (as opposed to a binary variable as in the prior work) and to use a cumulative logit modeling to do so. In contrast to that previous study from Australia [16], we found a statistically signifi- cant association between ED census and nurses’ triage decisions. Specif- ically, as can be seen from Fig. 1, as census increases from 25 to 70 patients in the ED (representing, respectively, 10% and 90% quantiles of census from the data set), the probability of a patient being triaged as high acuity increases by about 50%, while the probability of a patient being triaged as low acuity decreases by approximately 25%. On the other hand, the probability of a patient being triaged as medium acuity (ESI 3) seems to change only slightly with census.

The relationship between physicians’ admission decision and ED census at the decision time was observed in a prior work: Gorski et al.

[8] performs a retrospective analysis using 18 months of all adult patient encounters seen in the main ED of an academic tertiary care center, and finds that there is a positive association between the likelihood that a patient would be admitted and the waiting room census and physician

Fig. 1. Marginal probabilities of different acuity levels versus census for a patient subgroup: Caucasian female, aged between 18 and 40, with abdominal pain.

Table 4 Odds ratios of Prob(admit) versus Prob(discharge) and corresponding 95% confidence in- tervals for intercept, census, race, gender, acuity, age group, and pod.

Prob(admit)/Prob(discharge) Intercept

3.188 [2.763,3.679]

Census

1.007 [1.006, 1.008]

Race (contrast: Caucasian)

African American

1.033 [0.985,1.084]

position is in doubt. The same may hold true for triage nurses. As deci-

Asian

0.892 [0.729,1.093]

sions become more pressured triage nurses may err on the side of

Native American

2.138 [1.556,2.938]

caution and triage the patient a higher acuity than they otherwise

Other Unknown

0.823 [0.764,0.887]

0.807 [0.695,0.938]

would have. Work outside of health care has found similar decision fa-

tigue in parole hearings [21]. Parole decisions made late in the day or

Gender (contrast: Female)

long after a meal are more likely to result in the parolee staying in

Male

1.218 [1.167,1.271]

prison, the decision that is viewed as more cautious. As more and

Acuity (contrast: ESI3)

more decisions are made a decision maker tends to pick what is consid-

ESI1

20.891 [12.519,34.861]

ered the less risky of two choices even though this may not always be

ESI2

3.687 [3.313,4.104]

the best decision for the directly affected individual or others in the

Establishing an association does not prove cause and effect. Never- theless, the correlations we found support what ED providers, nurses, and managers have suspected all along: As the ED becomes more crowded, there may be a tendency among providers and nurses to change their behavior in decision making towards being more risk averse. It may be that as the executive and Cognitive function is taxed by the load, the clinicians of care make the decision that appears to be the safest choice for the individual patient. In the case of providers, they may opt for admission over a discharge in cases where the best dis-

ESI4 0.115 [0.095,0.139]

ESI5 0.018 [0.006,0.055]

Age Group (contrast: 18 to 40)

Below 3 m 3.179 [2.358,4.285]

3 m to 3 1.279 [1.072,1.525]

3 to 8 1.199 [0.999,1.439]

8 to 18 1.252 [1.077,1.456]

40 to 55 1.697 [1.587,1.816]

55 to 70 2.913 [2.714,3.125]

Over 70 4.325 [4.002,4.676]

Pod (contrast: BHED)

A 0.661 [0.587,0.744]

B 0.561 [0.498,0.631]

C 4.381 [3.680,5.217]

D 0.216 [0.190,0.247]

Pediatrics 0.397 [0.339,0.465]

load census. Our results firmly support this earlier study in that we found a similar odds ratio for admission that increases as census does. From Fig. 2, we can see that as census increases from 25 to 75 patients in the ED, the probability of a patient being categorized as admit in- creases by around 25%. Note that our study includes pediatric patients in addition to adults unlike Gorski et al. [8] that only considered adults and yet we still observed similar results.

system.

Limitations

This study includes data from a single academic center with average patient volume. The findings on relation between census and disposi- tion are similar to a previous study at an academic center with smaller volume but it may be that academic centers have unique patient popu- lations or organizational structures differing from Community settings. Processing of admitted patients does tend to provide a greater challenge in academic centers [22]. Also, our findings on relation between census and triage decisions should not be generalized to EDs that use a triage system other than ESI. Finally, a prospective case-control study would allow better identification of factors that affect nurses’ triage and pro- viders’ admission decisions in the ED.

Conclusions

In this study, we found a correlation between overall ED census and likelihood of admission as well as changes in triage decisions that result in more patients being triaged to higher acuity levels. This supports a growing body of evidence that situational stressors such as high census may influence decisions made by nurses and physicians in the ED.

Fig. 2. Probability of admission versus census (with 95% CI) for Caucasian female patients aged between 18 and 40, categorized as ESI3, presented with abdominal pain, and treated in Pod A.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi. org/10.1016/j.ajem.2019.06.039.

References

  1. Centers for Disease Control and Prevention. NHAMCS: 2015 emergency department summary tables. https://www.cdc.gov/nchs/data/nhamcs/web_tables/2015_ed_ web_tables.pdf, Accessed date: 18 September 2018.
  2. Johnson KD, Winkelman C. The effect of emergency department crowding on patient outcomes: a literature review. Adv Emerg Nurs J 2011;33:39-54.
  3. McCarthy ML, Zeger SL, Ding R, et al. Crowding delays treatment and lengthens emergency department length of stay, even among High-acuity patients. Ann Emerg Med 2009;54:492-503.e4.
  4. Richardson DB. Increase in patient mortality at 10 days associated with emergency department overcrowding. Med J Aust 2006;184:213-6.
  5. McCuster J, Vadeboncoeur A, Levesque JF, Ciampi A, Belzile E. Increases in emer- gency department occupancy are associated with adverse 30-day outcomes. Acad Emerg Med 2014;21:1092-100.
  6. Hwang U, Weber EJ, Richardson LD, et al. A research agenda to assure equity during periods of emergency department crowding. Acad Emerg Med 2011;18:1318-23.
  7. Morganti KG, Bauhoff S, Blanchard JC, et al. The evolving role of emergency depart- ments in the United States. RR-280-ACEP. Santa Monica, CA: RAND Corporation. Available at https://www.rand.org/pubs/research_reports/RR280.html; 2013 [Accessed 23 April 2018].
  8. Gorski JK, Batt RJ, Otles E, Shah MN, Hamedani AG, Patterson BW. The impact of emergency department census on the decision to admit. Acad Emerg Med 2017; 24:13-21.
  9. Diwas SK, Terwiesch C. Impact of workload on service time and patient safety: an econometric analysis of Hospital operations. Manag Sci 2009;55:1486-98.
  10. Freeman M, Robinson S, Scholtes S. Gatekeeping under congestion: An empirical study of referral errors in the emergency department. INSEAD working paper. Avail- able at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3036999; 2017, Accessed date: 1 May 2018.
  11. Diwas SK, Terwiesch C. An econometric analysis of patient flows in the cardiac inten- sive care unit. Manuf Serv Oper Manag 2012;14:50-65.
  12. Freeman M, Savva N, Scholtes S. Gatekeepers at work: an empirical analysis of a ma- ternity unit. Manag Sci 2016;63:3147-67.
  13. Fogarty E, Saunders J, Cummins F. The effect of boarders on emergency department process flow. J Emerg Med 2014;46:706-10.
  14. Kang H, Nembhard HB, Rafferty C, DeFlitch CJ. Patient flow in the emergency depart- ment: a classification and analysis of admission process policies. Ann Emerg Med 2014;64:335-42.
  15. Kuntz L, Mennicken R, Scholtes S. Stress on the ward: evidence of safety tipping points in hospitals. Manag Sci 2014;61:754-71.
  16. Richardson D. No relationship between emergency department activity and triage categorization. Acad Emerg Med 1998;5:141-5.
  17. Association of Academic Chairs of Emergency Medicine. ED benchmarking survey summary results. https://www.saem.org/aacem/resources/benchmarking-survey/ ed-benchmarking-survey-summary-results; 2012.
  18. Emergency severity index. In: Gilboy N, Tanabe P, Travers DA, Rosenau AM, Eitel DR, editors. Version 4: Implementation handbook 2012 edition. AHRQ publication no. 12-0014. Rockville, MD: Agency for Healthcare Research and Quality; 2011.
  19. Agresti A. An introduction to categorical data analysis, volume 135. New York: Wiley; 1996.
  20. R Core Team. R: A language and environment for statistical computing. Vienna, Austria: R Foundation for statistical computing; 2016https://www.R-project.org/.
  21. Danziger S, Levav J, Avnaim-Pesso L. Extraneous factors in judical decisions. Proc Natl Acad Sci 2011;108:6889-92.
  22. Horwitz LI, Green J, Fau-Bradley EH, Bradley EH. US emergency department perfor- mance on wait time and length of visit. Ann Emerg Med 2010;55:133-41.

Leave a Reply

Your email address will not be published. Required fields are marked *