Article, Psychiatry

Prolonged length of stay in ED psychiatric patients: a multivariable predictive model

a b s t r a c t

Background: We aimed to evaluate factors associated with prolonged emergency department (ED) length of stay (LOS) among Psychiatric patients and to develop a multivariable predictive model to guide future interventions to reduce ED LOS. Methods: electronic health records of ED patients receiving a psychiatric consultation and providing research au- thorization were reviewed from September 14, 2010, through September 13, 2013, at an academic hospital with approximately 73000 visits annually. Prolonged LOS was defined as >=8 hours.

Results: We identified 9247 visits among 6335 patients; median LOS was 4.1 hours, with 1424 visits (15%) with prolonged LOS. In the multivariable model, characteristics associated with an increased risk of a prolonged LOS included patient age 12 to 17 years (odds ratio [OR], 2.43; P b .001) or >=65 years (OR, 1.46; P = .007); male gen- der (OR, 1.24; P = .002); Medicare insurance coverage (OR, 1.34; P = .008); use of restraints (OR, 2.25; P = .006); diagnoses of cognitive disorder (OR, 4.62; P b .001) or personality disorder (OR, 3.45; P b .001); transfer to an un- affiliated psychiatric hospital (OR, 22.82; P b .001); ED arrival from 11 PM through 6:59 AM (OR, 1.53; P b .001) or on a Sunday (OR, 1.76; P b .001); or ED evaluation in February (OR, 1.59; P = .006), April (OR, 1.66; P = .002), and May (OR, 1.54; P = .007).

Conclusions: Many psychiatric patients had a prolonged ED LOS. Understanding the multiple, patient-specific, ED operational, and seasonal factors that predict an increased LOS will help guide allocation of resources to improve overall ED processes and patient care.

(C) 2015


Emergency department (ED) throughput challenges, including overcrowding, increasing patient length of stay (LOS), and the atten- dant negative impact on quality of care, have received considerable at- tention in the United States and elsewhere in recent years [1-3]. Overcrowding contributes to treatment delays, Ambulance diversion, elopement, undesirable medical events, introduction or exacerbation

Abbreviations: AUC, area under a receiver operating characteristics curve; ED, emer- gency department; IQR, interquartile range; LOS, length of stay; OR, odds ratio.

? Conflict of interest: None.

* Corresponding author at: Department of Psychiatry and Psychology, Mayo Clinic, 200 First St SW, Rochester, MN 55905.

E-mail address: [email protected] (G.J. Melin).

1 Dr Warren is now with the Department of Psychiatry, Boise Veterans Administration Medical Center, Boise, ID.

of existing Health care disparities, increased cost of Health care delivery, and even patient mortality [4-7]. Previous studies have analyzed factors associated with ED overcrowding and discussed individual patient characteristics associated with prolonged LOS, primarily focusing on the adult population [8-12]. Patients who seek emergency mental health care have an increasingly important role in increasing LOS, and specific factors influencing LOS in the psychiatric population have re- ceived considerable attention in the literature [13-17].

From 2000 through 2007, the percentage of all ED visits in the United States attributable to psychiatric care increased from 5.4% to 12.5% [16,18]. This dramatic increase occurred in the context of rising mental health diagnoses and burden, as noted by the World Health Organiza- tion [19], along with a decrease in inpatient US psychiatric bed availabil- ity, from 524878 beds in 1970 to 211 199 in 2002 [20]. The overall reduction of inpatient psychiatric beds without the creation of adequate community-centered mental health treatment facilities has contributed to a crisis in mental health care that often “requires seriously ill

0735-6757/(C) 2015

Table 1

Univariable associations with prolonged length of ED stay


All (N = 9247)

b8 hours (n = 7823)

>= 8 hours (n = 1424)

OR (95% CI)



Age, y, No. (%)

<= 11

364 (4)

311 (4)

53 (4)

1.08 (0.89-1.46)




1244 (13)

944 (12)

300 (21)

2.01 (1.74-2.33)



6628 (72)

5724 (73)

904 (63)

1.0 (reference)

>= 65

1011 (11)

844 (11)

167 (12)

1.25 (1.05-1.50)


Gender, No. (%) Female

4917 (53)

4220 (54)

697 (49)

1.0 (reference)


4330 (47)

3603 (46)

727 (51)

1.22 (1.09-1.37)



Race, No. (%) (n = 8764)


8400 (96)

6502 (88)

1161 (86)

1.0 (reference)


364 (4)

905 (12)

196 (14)

1.21 (1.03-1.43)



Insurance, No. (%)


3074 (33)

2652 (34)

422 (30)

1.0 (reference)


2312 (25)

1931 (25)

381 (27)

1.24 (1.07-1.44)




2246 (24)

1875 (24)

371 (26)

1.24 (1.07-1.45)


Other or self

1615 (17)

1365 (17)

250 (18)

1.15 (0.97-1.36)


Mode of arrival, No. (%) (n = 9226) private vehicle or walk-in

6027 (65)

5154 (66)

873 (61)

1.0 (reference)

Emergency medical service

2277 (25)

1916 (25)

361 (25)

1.11 (0.97-1.27)



law enforcement

893 (10)

713 (9)

180 (13)

1.49 (1.25-1.78)


Commercial vehicle or other

29 (b1)

21 (b 1)

8 (1)

2.25 (0.99-5.09)


Combined analysis of breath and blood alcohol

Not done or negative

9060 (98)

7677 (98)

1383 (97)

1.0 (reference)


187 (2)

146 (2)

41 (3)

1.56 (1.10-2.22)



Normal saline administered

341 (4)

277 (4)

64 (4)

1.28 (0.97-1.69)



Medical or physical restraints used

74 (1)

37 (b 1)

37 (3)

5.62 (3.55-8.89)



Haloperidol administered

294 (3)

162 (2)

132 (9)

4.83 (3.81-6.13)



Lorazepam administered

1107 (12)

795 (10)

312 (22)

2.48 (2.15-2.87)



Final diagnosis (n = 9246)


662 (7)

606 (8)

56 (4)

1.0 (reference)

Cognitive disorder or dementia

91 (1)

64 (1)

27 (2)

4.57 (2.70-7.73)



Disorder of childhood and adolescence

148 (2)

123 (2)

25 (2)

2.20 (1.32-3.66)



1951 (21)

1743 (22)

208 (15)

1.29 (0.95-1.76)



3142 (34)

2633 (34)

509 (36)

2.09 (1.57-2.80)



836 (9)

717 (9)

119 (8)

1.80 (1.28-2.51)


Personality disorder

63 (1)

49 (1)

14 (1)

3.09 (1.61-5.95)



716 (8)

545 (7)

171 (12)

3.40 (2.46-4.69)


suicidal ideation or homicidal ideation

1204 (13)

969 (12)

235 (17)

2.62 (1.93-3.57)


Substance use

433 (5)

373 (5)

60 (4)

1.74 (1.18-2.56)


Disposition type

Admit to affiliated psychiatric hospital

3341 (36)

2931 (37)

410 (29)

1.37 (1.18-1.58)



Admit to medical service

851 (9)

769 (10)

82 (6)

1.04 (0.81-1.34)



4178 (45)

3790 (48)

388 (27)

1.0 (reference)

Transfer to county detoxification center

166 (2)

137 (2)

29 (2)

2.07 (1.37-3.13)


Transfer to other (unaffiliated) psychiatric hospital

711 (8)

196 (3)

515 (36)

25.67 (21.12-31.19)


ED saturation status at hour of visit (n = 9246)


6181 (67)

5303 (68)

878 (62)

1.0 (reference)

Yellow or red

3065 (33)

2519 (32)

546 (38)

1.31 (1.17-1.47)



Maximum ED saturation on day of visit (N = 8610) Green

2817 (33)

2423 (33)

394 (30)

1.0 (reference)

Yellow or red

5793 (67)

4864 (67)

929 (70)

1.18 (1.03-1.33)



Psychiatric patients present at arrival


2600 (28)

2302 (29)

298 (21)

1.0 (reference)

>= 3

6647 (72)

5521 (71)

1126 (79)

1.58 (1.37-1.81)



hour of arrival

7 AM-2:59 PM

3711 (40)

3205 (41)

506 (36)

1.0 (reference)

3 PM-10:59 PM

4299 (46)

3596 (46)

703 (49)

1.24 (1.09-1.40)



11 PM-6:59 AM

1237 (13)

1022 (13)

215 (15)

1.33 (1.12-1.59)


Day of visit


1450 (16)

1196 (15)

254 (18)

1.35 (1.10-1.66)




1479 (16)

1244 (16)

235 (17)

1.20 (0.98-1.48)



1398 (15)

1208 (15)

190 (13)

1.0 (reference)


1436 (16)

1216 (16)

220 (15)

1.15 (0.93-1.42)



1327 (14)

1165 (15)

162 (11)

0.88 (0.71-1.11)



1055 (11)

912 (12)

143 (10)

1.00 (0.79-1.26)



Day of visita

1102 (12)

882 (11)

220 (15)

1.59 (1.28-1.96)


Weekday 6505 (70)

5539 (71)

966 (68)

1.0 (reference)

Weekend 2742 (30)

2284 (29)

458 (32)

1.15 (1.02-1.30)



Month of visit


787 (9)

645 (8)

142 (10)

1.75 (1.32-2.31)




652 (7)

525 (7)

127 (9)

1.92 (1.44-2.56)



751 (8)

626 (8)

125 (9)

1.59 (1.19-2.11)



727 (8)

577 (7)

150 (11)

2.06 (1.56-2.73)


Table 1 (continued)


All (N = 9247)

b8 hours (n = 7823)

>= 8 hours (n = 1424)

OR (95% CI)




859 (9)

701 (9)

158 (11)

1.79 (1.36-2.35)



825 (9)

728 (9)

97 (7)

1.06 (0.78-1.43)



849 (9)

754 (10)

95 (7)

1.0 (reference)


837 (9)

713 (9)

124 (9)

1.38 (1.04-1.84)



802 (9)

702 (9)

100 (7)

1.13 (0.84-1.53)



742 (8)

637 (8)

105 (7)

1.31 (0.97-1.76)



754 (8)

636 (8)

118 (8)

1.47 (1.10-1.97)



662 (7)

579 (7)

83 (6)

1.14 (0.83-1.56)


Abbreviations: AUC, area under the curve; ED, emergency department.

a Weekend arrival was defined as arriving from 5 PM Friday through 7 AM Monday.

individuals to deteriorate to dangerousness or grave disability before they can receive needed treatment” [21]. Increasing demand that outpaces resources has led to disproportionately increased ED LOS

[11] and considerable delays in boarding (ie, admission once a disposi- tion has been reached), and it is a problem in both the adult and pediat- ric psychiatric population [22-27].

To address this growing ED boarding of psychiatric patients, resources must be allocated toward patients with the highest risk of being boarded. To address this need, we sought to develop a multivariable predictive model that identifies patients at greatest risk for prolonged LOS.


Study design and setting

We conducted an observational cohort study of consecutive patients presenting to the ED of the Mayo Clinic Hospital, Saint Marys Campus (Rochester, Minnesota), from September 14, 2010, through September 13, 2013. Saint Marys Campus of Mayo Clinic Hospital includes a tertiary care academic ED with 73000 patients annually and has a dedicated psychiatric hospital with 73 beds. The Mayo Clinic Institutional Review Board approved the study protocol.

Selection of participants

All patients presenting to the ED during the study period were eligi- ble for our study. We identified all patients seen during the study period who received a psychiatry consultation and excluded those who refused research consent.

Data collection and outcome measures

Data abstracted from the electronic health record included time of pre- sentation, ED patient volume status at the time of presentation, emergen- cy severity index level, patient characteristics, chief concern, medication administration, final primary ED diagnosis, ED LOS, and final disposition.

We defined a prolonged LOS as >= 8 hours in the ED, based on a prior

study showing a median LOS just over 8 hours for psychiatric patients in 5 different EDs [13]. Weekend arrival was defined as arriving from 5 PM Friday through 7 AM Monday.

Final diagnosis was determined by the treating ED physician and not the consulting psychiatrist. Because of the broad range of final diagno- ses in this study, we grouped similar diagnoses together. For instance, a final diagnosis of anxiety could represent an anxiety spectrum disor- der (eg, panic attacks, obsessive compulsive disorder, trauma- and stressor-related disorders) for which anxiety was the prominent pre- sentation. Disorders of childhood and adolescence included but were not limited to reactive attachment disorder, oppositional defiant disor- der, attention-deficit/hyperactivity disorder, autism spectrum disor- ders, Depressive disorders, psychotic disorders, adjustment disorders, and parent-child relational problems. The diagnosis category of “medi- cal” included diagnoses related to conditions such as infection, and “other” included diagnoses that did not fit into any other category.

With regard to disposition, an admission to the “affiliated psychiatric hospital” signified patient admission to the psychiatric hospital directly affiliated with the ED.

To describe the ED capacity and saturation status, “green” status was defined as b 15 patients in the waiting room. “Yellow” or “red” status was defined as 15 or 30 patients, respectively, in the waiting room at a given time, regardless of age or chief concern. Collectively, yellow or red status is an indicator of ED overcapacity, and this status is triggered electroni- cally by the patient Tracking system. If the ED reaches the yellow or red status, the system automatically analyzes the waiting room status 2 hours later to determine whether the status can be downgraded.

Statistical analysis

Continuous features were summarized with medians, interquartile ranges (IQRs), and ranges. Categorical features were summarized with frequency counts and percentages. Univariable associations with a prolonged LOS were evaluated using logistic regression models and summarized with odds ratios (ORs) and 95% confidence intervals (CIs). A multivariable model was developed using stepwise selection, with the P value for a feature to enter or leave the model set to .05. Model discrimination (how well the features in the model separate pa- tients with and without a prolonged LOS) was summarized using the area under a receiver operating characteristics curve (AUC). The AUC can range from 0.5 to 1.0, with higher values indicating improved pre- dictive ability or improved discrimination. Model calibration (how well the predicted probabilities of the event estimated by the model agree with the observed event) was summarized using the Hosmer and Lemeshow goodness-of-Fit test. A statistically significant P value from this test would reject the null hypothesis that the features in the model fit the data well. Because some patients had multiple visits dur- ing the study period, the multivariable model was also evaluated using generalized estimating equations to verify that the parameter es- timates were similar after accounting for any correlation that might occur among visits from the same patient. Statistical analyses were per- formed using the SAS software package (SAS Institute Inc). All tests were 2-sided, and P b .05 was considered statistically significant.


We included 9247 ED psychiatric patient visits by 6335 unique pa- tients in our cohort. The median number of visits per patient during the study period was 1 (IQR, 1-1; range, 1-23). Among all study visits, the median LOS was 4.1 hours (IQR, 2.8-6.2 hours; range, 0.3-243.0 hours). To evaluate visits with a prolonged LOS, we separated study visits into 2 groups, with LOS b 8 or >= 8 hours. We identified 1424 visits (15%) with a prolonged LOS. Univariable comparison of features for visits with and without a prolonged LOS is shown in Table 1.

Patient characteristics associated with a prolonged LOS

Several patient characteristics were associated with longer LOS for behavioral health issues in a univariable setting. Patient age 12 to 17

Table 2

Multivariable model to predict prolonged length of stay

Feature OR (95% CI) P

Age, y

<= 11

1.58 (1.10-2.27)



2.43 (2.01-2.94)



>= 65

1.0 (reference)

1.46 (1.11-1.91)




1.0 (reference)



1.24 (1.08-1.41)




1.0 (reference)

1.17 (0.98-1.40)



1.34 (1.08-1.66)


Other or self

1.09 (0.89-1.33)


Medical or physical restraints used 2.25 (1.26-4.00) .006

Haloperidol administered 2.04 (1.46-2.85) b.001

Lorazepam administered 1.98 (1.62-2.41) b.001 Final diagnosis

Anxiety 1.0 (reference)

Cognitive disorder or dementia 4.62 (2.52-8.48) b.001 Disorder of childhood and adolescence 1.11 (0.60-2.05) .73

Medical 1.80 (1.26-2.56) .001

Mood 1.58 (1.14-2.20) .006

Other 1.39 (0.95-2.03) .09

Personality disorder 3.45 (1.67-7.11) b.001

Psychosis 2.50 (1.72-3.63) b.001 Suicidal ideation or homicidal ideation 2.10 (1.47-3.00) b.001 Substance use 1.69 (1.09-2.61) .02

Disposition type

Admit to affiliated psychiatric hospital 1.28 (1.08-1.51) .004

Admit to medical service 0.86 (0.65-1.15) .30

Discharge 1.0 (reference)

Transfer to county detoxification center 1.85 (1.18-2.89) .007

prolonged LOS. Patients with a documented blood alcohol level greater than 0.08 g/dL or a positive breath alcohol test were more likely to have a prolonged LOS (OR, 1.56; P = .01) compared with patients with neg- ative blood or breath alcohol test results or those who did not have ei- ther test performed in the ED. Prolonged LOS was also predicted by the use of restraints (OR, 5.62; P b .001) and administration of haloper- idol (OR, 4.83; P b .001) or lorazepam (OR, 2.48; P b .001) during ED management.

The final diagnosis was also significantly associated with prolonged LOS. For example, patients with a diagnosis of cognitive disorder or de- mentia were markedly more likely to have a prolonged LOS (OR, 4.57; P b .001) compared with patients with a final diagnosis of anxiety. In ad- dition, compared with patients discharged from the ED, patients were more likely to have a prolonged LOS if they were transferred directly to the county detoxification center (OR, 2.07; P b .001) or to another (unaffiliated) psychiatric hospital (OR, 25.67; P b .001).

ED operational and seasonal factors associated with a prolonged LOS

We also analyzed ED operational factors and factors associated with when patients presented to the ED. A patient’s ED visit had a greater chance of having a prolonged LOS if the ED reached overcapacity status at any time during the same day as the patient’s arrival (OR, 1.18; P =

.01). A patient had an even greater chance of having a prolonged LOS if the ED was already in an overcapacity status at the time of the pa- tient’s arrival (OR, 1.31; P b .001). Patients also had a greater chance of prolonged LOS if at least 3 other psychiatric patients were already pres- ent at the time of arrival (OR, 1.58; P b .001).

Specific times, days, and months also were associated with prolonged LOS. Patients arriving from 3 PM through 10:59 PM (OR,

Transfer to other (nonaffiliated)

psychiatric hospital

Emergency department saturation at hour of visit


1.0 (reference)

Yellow or red

1.31 (1.14-1.52)


Psychiatric patients present at arrival

b3 1.0 (reference)

>= 3

1.68 (1.42-1.98)


Hour of arrival 7 AM-2:59 PM

1.0 (reference)

3 PM-10:59 PM

0.98 (0.84-1.14)


11 PM-6:59 AM

1.53 (1.25-1.87)


Day of visit Monday

1.36 (1.07-1.73)



1.23 (0.97-1.57)



1.0 (reference)


1.24 (0.97-1.58)



1.04 (0.81-1.35)



1.30 (0.99-1.70)



1.76 (1.37-2.26)


Month of visit


1.26 (0.91-1.75)



1.59 (1.14-2.22)



1.30 (0.93-1.81)



1.66 (1.20-2.29)



1.54 (1.12-2.10)



1.00 (0.71-1.40)



1.0 (reference)


1.40 (1.01-1.93)



1.11 (0.79-1.55)



1.21 (0.86-1.70)



1.41 (1.01-1.97)



1.16 (0.81-1.65)


22.82 (18.44-28.24) b.001

1.24; P b .001) or from 11 PM through 6:59 AM (OR, 1.33; P = .001)

had a higher probability of having a prolonged LOS compared with those arriving from 7 AM through 2:59 PM.

In general, patients arriving during the weekend had a greater chance of prolonged LOS than those arriving on weekdays (OR, 1.15; P = .02). However, when we analyzed individual days of the week, the odds of having a prolonged LOS were higher if a patient presented on a Sunday (OR, 1.59; P b .001) or Monday (OR, 1.35; P = .004) com- pared with the reference day of Wednesday. The month that a patient presented was also significantly associated with prolonged LOS. Com- pared with the reference month of July, patients presenting in February, April, and May had the greatest risk of a prolonged LOS.

Multivariable analysis to predict prolonged LOS

years (OR, 2.01; P b .001) or 65 years and older (OR, 1.25; P = .01) were more likely to have prolonged LOS compared with patients aged 18 to 64 years. Male gender (OR, 1.22; P b .001), nonwhite race (OR, 1.21; P = .02), Medicaid (OR, 1.24; P = .005) or Medicare (OR, 1.24; P =

.005) insurance coverage, and those who were brought to the ED by law enforcement (OR, 1.49; P b .001) were more likely to have a

After analyzing the univariable results, we developed a multivariable model to predict a prolonged LOS for psychiatric patients in our ED (Table 2). The AUC of this model was 0.784. The P value from a goodness-of-fit test was .29, indicating that the features included in the model fit the data well. The associations of the features included in the multivariable model with prolonged LOS were similar after account- ing for any correlation among visits from the same patients by using generalized estimating equations. Demographic characteristics associ- ated with an increased risk of a prolonged LOS included patient age 17 years and younger (for ages <= 11 years, OR, 1.58 and P = .01; for ages 12-17 years, OR, 2.43 and P b .001) or 65 years and older (OR, 1.46; P = .007). Male gender was also a significant factor (OR, 1.24; P =

.002). Patients with Medicare insurance coverage (OR, 1.34; P = .008) and those who required a behavioral intervention (use of restraints [OR, 2.25; P = .006], haloperidol [OR, 2.04; P b .001], or lorazepam [OR, 1.98; P b .001]) also had increased risk of a prolonged LOS. The di- agnoses most associated with a prolonged LOS were cognitive disorder or dementia (OR, 4.62; P b .001), personality disorder (OR, 3.45; P b .001), and psychosis (OR, 2.50; P b .001). The factor most strongly as- sociated with a prolonged ED LOS was transfer to another (unaffiliated) psychiatric hospital (OR, 22.82; P b .001).

As shown in Table 2, several ED operational factors and factors asso- ciated with time of presentation were predictive of a prolonged LOS in a multivariable model. Increased risk of a prolonged LOS was predicted by ED overcapacity (yellow or red status) at the time of arrival (OR, 1.31; P b .001); 3 or more psychiatric patients already in the ED at the time of patient arrival (OR, 1.68; P b .001); and hour of arrival from 11 PM through 6:59 AM (OR, 1.53; P b .001). In addition, arrival on a Sunday (OR, 1.76; P b .001) or Monday (OR, 1.36; P = .01) increased the risk of a prolonged LOS. The months of arrival with the greatest risk of a prolonged LOS were February (OR, 1.59; P = .006), April (OR, 1.66; P = .002), and May (OR, 1.54; P = .007).


We analyzed factors associated with and predictive of a prolonged ED LOS for 9247 visits by 6335 unique adult and pediatric patients who presented to an academic ED and received a psychiatric consulta- tion during a 3-year period. Median LOS was 4.1 hours, and 1424 (15%) of visits had a LOS of 8 or more hours. Importantly, LOS was not analyzed as a continuous variable; it was analyzed as prolonged vs not prolonged to ensure that the data set was not skewed by only a few pa- tients with complicated psychiatric histories or presentations.

Multiple factors were significantly associated with a prolonged LOS in the multivariable analysis. Age (17 years and younger or older than 65 years), male gender, Medicare insurance, and use of behavioral inter- ventions (restraints, haloperidol, or lorazepam) increased the risk of a prolonged LOS. In addition, the ED diagnosis and disposition, most nota- bly transfer to another (unaffiliated) psychiatric hospital, increased the risk of a prolonged LOS. Finally, ED saturation; psychiatric patient saturation; and hour, day, and month of arrival were also predictive of prolonged LOS.

Although the overall median LOS of 4.1 hours in our cohort com- pares favorably to the median LOS of 8.2 hours reported in a recent study [14] of adult psychiatric patients presenting to hospitals in the greater Boston area, we found that patients at both ends of the age spec- trum were more likely to have an increased LOS. In a nationally repre- sentative sample of United States EDs from 2001 to 2008, among patients 18 years and younger, the median LOS for mental health- related visits exceeded that of visits for reasons unrelated to mental health by more than 1 hour [25]. However, the LOS for pediatric mental health visits was not compared with the analogous adult visits. In addi- tion, consistent with existing literature [14], we found that mental health patients 65 years or older had increased LOS compared with mental health patients younger than 65 years. Although specific reasons for increased delay were not directly analyzed, the importance of ex- cluding acute medical illness contributing to Mental illness exacerba- tion, the need for increased collateral information, and the possibility of reduced bed capacity in specialized pediatric and geriatric psychiatry are all plausible explanations. Our study adds to the growing literature describing the vulnerability of the psychiatric pediatric and Geriatric populations to prolonged LOS, and ours is the first study, to our knowl- edge, to directly compare pediatric to adult psychiatric LOS in the ED.

Of considerable concern, especially given the national discussion about access to care and disparities in health care, our study showed that Insurance type influenced LOS. When using Commercial insurance as the reference standard, those with Medicare were vulnerable to in- creased LOS. Herring et al [2] found that EDs primarily serving privately insured patients had a disproportionate increase in overall LOS from 2001 through 2005, but those serving primarily uninsured or Medicare patients still had a longer LOS. Our findings also are consistent with other study findings in which those with Public insurance were more vulnerable to a prolonged LOS [15]. Further analysis is needed to help elucidate this disparity.

The diagnosis that was most predictive of prolonged LOS in our co- hort was a cognitive disorder. This finding may reflect the potentially in- creased time for appropriate medical evaluation to exclude other

underlying causes. Personality disorder had the second strongest associ- ation with a prolonged LOS. Personality disorders are complex and may markedly exacerbate comorbid disorders such as mood and substance abuse disorders [28,29]; thus, they may require further medical evalua- tion and stabilization in the ED before a final disposition is determined. Our study analyzed the predictive value of time of year and found that the months of April, February, and May had the greatest associa- tions with a prolonged LOS. Miller et al [30] examined data on suicide completions and deliberate, nonfatal, self-harm admissions in 12 states in 1997. They reported that nonfatal admission rates were higher in April and May, whereas suicide rates were higher in February and March. Beauchamp et al [31] specifically examined suicide attempts and completions by poisoning from 2006 to 2010 in the United States and showed that poisoning attempts were more common in the spring and fall. A study of seasonal patterns of suicide in Finland from 1979 through 1999 [32] demonstrated a pronounced seasonal effect, with the greatest risk in the spring. In addition, Petridou et al [33] found peak incidence of suicidal behavior in June for the northern hemisphere and in December for the southern hemisphere. Although our study was not detailed enough to analyze specific spikes in suicidal behavior, our findings of February, April, and May being most associated with in- creased LOS is consistent with seasonal variations described for suicidality and suggests that seasonal variations in mental illness affects


We observed that ED presentation on a Sunday had the greatest as- sociation with increased LOS in our study. Associations between suicidality and day of the week have been evaluated previously. Miller et al [30] reported that both fatal and nonfatal self-harm attempts were more common on Monday and Tuesday and less common on Sat- urday. Likewise, Beauchamp et al [31] reported that suicide attempts and completions by poisoning among adults were most common on Sundays and Mondays. However, for those 19 years old and younger, the most common days were Mondays and Tuesdays. The fact that a Sunday or Monday presentation had the greatest association with in- creased LOS in our study may reflect inpatient psychiatric unit satura- tion and sluggish turnover during the weekend, or as the studies above have demonstrated, they could represent higher risk periods for fatal and nonfatal self-harm attempts.

In addition, arrival to the ED from 11 PM through 6:59 AM predicted an increased LOS. Presentation during the overnight hours was also as- sociated with an increased LOS in a pediatric study of psychiatric ED boarding [26]. This association may be indicative of decreased staffing overnight and the need for morning inpatient dismissals to occur before inpatient beds are available. In addition, overall ED saturation was asso- ciated with increased risk of a prolonged LOS. This association with in- creased ED saturation may be attributable to slower throughput during those times. The association with increased LOS when 3 psychi- atric patients were already present in the ED may also be reflective of an overall ED saturation or potentially a saturated statewide system of psy- chiatric beds, resulting in increased ED boarding.

The strongest predictive factor for a prolonged LOS in our study was the need for transfer to another (unaffiliated) inpatient psychiatric hos- pital, a finding consistent with literature describing both adult and pedi- atric populations [14,15,25]. The LOS associated with the need for transfer to an unaffiliated hospital is influenced by a number of factors, including a statewide psychiatric bed shortage (reflecting the nation- wide crisis of psychiatric bed shortage), geographic isolation, extensive medical clearance (including Laboratory analysis required by accepting outside institutions), and exclusion criteria at the affiliated psychiatric facility that necessitates Transfer of patients with a history of violence or sex offense.

Given the increasing need for emergent mental health care and shrinking inpatient psychiatric resources nationally, it is crucial to ex- amine ways to make safe, timely dispositions more feasible and to re- duce ED boarding. By analyzing specific patient characteristics, alternative treatment avenues can be designed and studied. Areas of

future research include the development of partial hospital programs as an alternative for patients whose illness demands a higher level of care than traditional Outpatient services but may not meet full threshold criteria for acute Inpatient hospitalization. In addition, strengthened al- liances between academically affiliated institutions and local communi- ty and county resources is worthy of further exploration and study. Our study also highlighted areas of potential seasonal, day-to-day, and hour- of-arrival vulnerabilities that may require enhanced staffing to mini- mize ED operational factors that contribute to prolonged LOS. As regula- tory bodies continue to scrutinize ED LOS, predictive models with wide national generalizability are increasingly important. This study supports prior studies elucidating patient characteristics that influence LOS and advances the discussion by developing a predictive model that can lead to a more nimble ED response in vulnerable situations.


Importantly, while these data included a large sample size over 3 years, it was a retrospective study that was limited to a single academic institution with an existing psychiatric hospital. We analyzed a number of factors associated with prolonged LOS, including some indicators of overall ED flow, but we did not specifically address times associated with input, throughput, and output, as other studies have done. For in- stance, our study did not separate times for medical clearance from psy- chiatric evaluation and disposition, so we are unable to comment on whether this was a confounding factor and, if so, for how many patients. Future studies could analyze which phase had the greatest contribution to overall LOS; targeted interventions could be developed at our study site and then disseminated to determine whether such interventions could be generalized to other care settings.

In addition, it would be helpful to know the impact, if any, of race, primary language, and other socioeconomic factors on LOS. Such factors should be considered in the important ongoing nationwide discussion of health care disparities.

Furthermore, the location of our ED (in a city of just over 100000 residents) may limit the generalizability of our study findings to EDs in more rural or more urban settings. There is nationwide variability in the number of available inpatient psychiatric beds, crisis beds, and local community shelters that can provide appropriate care. For those patients whose disposition is to somewhere other than home, this state-to-state variability in available next-step care resources is an im- portant consideration.


Significant numbers of psychiatric patients seeking emergency men- tal health care had a prolonged LOS (N 8 hours). The multivariable pre- dictive model showed the greatest risk of a prolonged LOS for patients

(1) at the ends of the age spectrum; (2) with public insurance; (3) need- ing behavioral intervention; (4) with a diagnosis of cognitive and per- sonality disorders; (5) needing transfer to an unaffiliated psychiatric hospital; (6) presenting at a time of ED saturation; (7) presenting in the months of February, April, or May; (8) arriving on a Sunday; and

(9) arriving during the overnight hours. Identification of factors associ- ated with an increased LOS adds to the discussion concerning delivery of scarce resources to increasing numbers of patients presenting with psy- chiatric illness. We add specifically to the discussion in our comparison of pediatric and adult patients and inclusion of time parameters related to ED LOS. This study highlights several areas where further research would help design targeted interventions. In an era of changing health care delivery, reimbursement, and insurance coverage and with the in- creasing prevalence of psychiatric illness, research aimed at improving the care, safety, and efficiency of the nation’s EDs continues to be highly relevant.


  1. Bullard MJ, Villa-Roel C, Bond K, Vester M, Holroyd BR, Rowe BH. Tracking emergen- cy department overcrowding in a tertiary care academic institution. Healthc Q 2009; 12(3):99-106.
  2. Herring A, Wilper A, Himmelstein DU, Woolhandler S, Espinola JA, Brown DF, et al. Increasing length of stay among adult visits to U.S. Emergency departments, 2001- 2005. Acad Emerg Med 2009;16(7):609-16 [Epub 2009 Jun 15].
  3. Forero R, McCarthy S, Hillman K. access block and emergency department over- crowding. Crit Care 2011;15(2):216 [Epub 2011 Mar 22].
  4. Hoot NR, Aronsky D. Systematic review of emergency department crowding: causes, effects, and solutions. Ann Emerg Med 2008;52(2):126-36 [Epub 2008 Apr 23].
  5. Henneman PL, Nathanson BH, Li H, Smithline HA, Blank FS, Santoro JP, et al. Emer- gency department patients who stay more than 6 hours contribute to crowding. J Emerg Med 2010;39(1):105-12 [Epub 2009 Jan 20].
  6. Liu SW, Thomas SH, Gordon JA, Hamedani AG, Weissman JS. A pilot study examining undesirable events among emergency department-boarded patients awaiting inpa- tient beds. Ann Emerg Med 2009;54(3):381-5 [Epub 2009 Mar 20].
  7. Hwang U, Weber EJ, Richardson LD, Sweet V, Todd K, Abraham G, et al. A research agenda to assure equity during periods of emergency department crowding. Acad Emerg Med 2011;18(12):1318-23.
  8. Moskop JC, Sklar DP, Geiderman JM, Schears RM, Bookman KJ. Emergency depart- ment crowding, part 1: concept, causes, and moral consequences. Ann Emerg Med 2009;53(5):605-11 [Epub 2008 Nov 22].
  9. Moskop JC, Sklar DP, Geiderman JM, Schears RM, Bookman KJ. Emergency depart- ment crowding, part 2: barriers to reform and strategies to overcome them. Ann Emerg Med 2009;53(5):612-7 [Epub 2008 Nov 22].
  10. Olshaker JS. Managing emergency department overcrowding. Emerg Med Clin North

    Am 2009;27(4):593-603 [viii].

    Slade EP, Dixon LB, Semmel S. Trends in the duration of emergency department visits, 2001-2006. Psychiatr Serv 2010;61(9):878-84.

  11. Yoon P, Steiner I, Reinhardt G. Analysis of factors influencing length of stay in the emergency department. CJEM 2003;5(3):155-61.
  12. Chang G, Weiss AP, Orav EJ, Jones JA, Finn CT, Gitlin DF, et al. Hospital variability in emergency department length of stay for adult patients receiving psychiatric consultation: a prospective study. Ann Emerg Med 2011;58(2):127-36 [e1, Epub 2011 Jan 12].
  13. Weiss AP, Chang G, Rauch SL, Smallwood JA, Schechter M, Kosowsky J, et al. Patient- and practice-related determinants of emergency department length of stay for pa- tients with psychiatric illness. Ann Emerg Med 2012;60(2):162-71 [e5, Epub 2012 May 2].
  14. Chang G, Weiss A, Kosowsky JM, Orav EJ, Smallwood JA, Rauch SL. Characteristics of adult psychiatric patients with stays of 24 hours or more in the emergency depart- ment. Psychiatr Serv 2012;63(3):283-6.
  15. Hazlett SB, McCarthy ML, Londner MS, Onyike CU. Epidemiology of adult psychiatric visits to US emergency departments. Acad Emerg Med 2004;11(2):193-5.
  16. Bachman D, King D, Robinson G, Moody J, Gibbs M, Krebs MJ, et al. Improvement Re- port: Reducing Length of Stay in the Emergency Department for Psychiatric Patients [Internet]. Cambridge (MA): Institute for Healthcare Improvement; 2010[cited 2015 Jul 31, Available from: MemberReportReducingLengthofStayintheEDforPsychiatricPatients.aspx].
  17. Owens PL, Mutter R, Stocks C. Mental health and substance abuse-Related Emer- gency Department Visits among Adults. Internet Rockville (MD): Agency for Healthcare Research and Quality (US); 2007[cited 2015 Jul 31, Available from:].
  18. Health statistics and information systems. Disease and injury country estimates [Internet]; 2015[Geneva (Switzerland), cited 2015 Jul 31, Available from: http://].
  19. Salinsky E, Loftis C. Shrinking inpatient psychiatric capacity: cause for celebration or concern? Issue Brief Natl Health Policy Forum 2007;823:1-21.
  20. Stone A, Rogers D, Kruckenberg S, Lieser A. Impact of the mental healthcare de- livery system on California emergency departments. West J Emerg Med 2012; 13(1):51-6.
  21. Alakeson V, Pande N, Ludwig M. A plan to reduce emergency room ‘boarding’ of psy- chiatric patients. Health Aff (Millwood) 2010;29(9):1637-42.
  22. Bender D, Pande N, Ludwig M. A Literature Review: psychiatric boarding [Internet]. Washington (DC): U.S. Department of Health and Human Services; 2008 [cited 2015 Jul 31]. Available from:
  23. Nicks BA, Manthey DM. The impact of Psychiatric patient boarding in emergency departments. Emerg Med Int 2012;2012:480 [360308, Epub 2012 Jul 22].
  24. Case SD, Case BG, Olfson M, Linakis JG, Laska EM. Length of stay of pediatric mental health emergency department visits in the United States. J Am Acad Child Adolesc Psychiatry 2011;50(11):1110-9 [Epub 2011 Oct 2].
  25. Wharff EA, Ginnis KB, Ross AM, Blood EA. Predictors of psychiatric boarding in the pediatric emergency department: implications for emergency care. Pediatr Emerg Care 2011;27(6):483-9.
  26. Waseem M, Prasankumar R, Pagan K, Leber M. A retrospective look at length of stay for pediatric psychiatric patients in an urban emergency department. Pediatr Emerg Care 2011;27(3):170-3.
  27. Hasin D, Kilcoyne B. Comorbidity of psychiatric and Substance use disorders in the United States: current issues and findings from the NESARC. Curr Opin Psychiatry 2012;25(3):165-71.
  28. Tyrer P, Mulder R, Crawford M, Newton-Howes G, Simonsen E, Ndetei D, et al. Personality disorder: a new global perspective. World Psychiatry 2010;9(1): 56-60.
  29. Miller TR, Furr-Holden CD, Lawrence BA, Weiss HB. Suicide deaths and nonfatal hospital admissions for deliberate self-harm in the United States: temporality by day of week and month of year. Crisis 2012;33(3):169-77.
  30. Beauchamp GA, Ho ML, Yin S. Variation in suicide occurrence by day and during major American holidays. J Emerg Med 2014;46(6):776-81 [Epub 2014 Jan 22].
  31. Partonen T, Haukka J, Nevanlinna H, Lonnqvist J. Analysis of the seasonal pattern in suicide. J Affect Disord 2004;81(2):133-9.
  32. Petridou E, Papadopoulos FC, Frangakis CE, Skalkidou A, Trichopoulos D. A role of sunshine in the triggering of suicide. Epidemiology 2002;13(1): 106-9.

Leave a Reply

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