Effects of Initiating Palliative Care Consultation in the Emergency Department on Inpatient Length of Stay

Abstract

Background/Objective:

Increased attention has been directed at the intersection of emergency and palliative medicine, since decisions made in the emergency department (ED) often determine the trajectory of subsequent medical treatments. Specifically, we examined whether inpatient admissions after palliative care (PC) consultation initiated in the ED were associated with decreased length of stay (LOS), compared with those in which consultations were initiated after hospital admission.

Methods:

Education and training on PC and the consultation service were provided to ED physicians and nurses. The PC service evaluated patients in the ED during weekdays and provided telephone consultation nights and weekends with postadmission follow-up. We compared the outcomes of these patients with those whose PC needs were identified and addressed through consultation postadmission. PC consultation data between January 2006 and December 2010 were retrospectively collected from the administrative records system and analyzed using propensity scores within multivariate regression.

Results:

Included in the analysis were 1435 PC consultations, 50 of which were initiated in the ED across the 4-year study period. Propensity scores were calculated using patient-level characteristics, including All Patient Refined Diagnostic Related Group (APRDRG) risk of mortality (ROM) and severity of illness (SOI), age, gender, readmission status, facility, and insurance type. Regression results showed that consultation in the ED was associated with a significantly shorter LOS by 3.6 days (p<0.01).

Conclusions:

Early initiation of PC consultation in the ED was associated with a significantly shorter LOS for patients admitted to the hospital, indicating that the patient- and family-centered benefits of PC are complemented by reduced inpatient utilization.

Introduction

Clinical and financial benefits of inpatient palliative care (PC) consultation services include improved care quality in the areas of symptom and pain management, patient and family satisfaction, and advance care planning due to improved clarification of treatment goals and adherence to patient preferences.^1–3 Financial benefits such as lower utilization and decreased daily costs and average length of stay (LOS) have also been found.^4,5

Increased attention has been directed at the intersection of emergency and palliative medicine, because decisions made in the emergency department (ED) often determine the trajectory of subsequent medical treatments. In addition, the ED frequently cares for PC-appropriate patients, including those at the end of life or with life-limiting illnesses. A number of issues have been recognized as barriers to providing PC services in the ED, including differences between the palliative and emergency medicine cultures, inadequate training of ED clinicians, and logistical barriers.^6,7 Still, there is a growing recognition that the PC model fits well in the ED and can provide benefits that may outweigh the barriers.^8,9 Early anecdotes point toward improved patient and clinician satisfaction, prevented hospitalizations, and preferential admission to inpatient PC units as benefits of PC in the ED. However, to our knowledge, no previous empirical study in the peer-reviewed literature has evaluated the effect of a PC intervention in the ED. In this study we examined whether hospital LOS was affected by PC consultation initiation in the ED versus after hospital admission.

Conceptual framework

The ED is a setting for triage, treatment, and determination of subsequent course of care. Patients with advanced illness are generally triaged by acuity level and then based on their condition admitted to the hospital or discharged. Admissions may go to the hospital floor, the intensive care unit, or a lower intervention care setting, such as a dedicated PC unit. Pathways vary based on multiple factors, the most prominent being level of acuity, presence of advanced illness, and bed acuity.^10,11 Patient preferences are generally not as highly considered in the ED due to requirements for timely disposition.¹² To better match patient wishes with the care received, we hypothesized that creating a PC pathway in the ED may lead to an improvement in patient-centered care and a decrease in the intensity and invasiveness of care when appropriate (see Fig. 1). Specifically, we hypothesize that hospitalized patients who receive PC consultation in the ED will have a lower LOS than those whose consultation occurs after admission.

FIG. 1.

Potential pathways for palliative care–appropriate patients.

Methods

California Pacific Medical Center (CPMC) in San Francisco, California, is a not-for-profit academic medical center affiliate of Sutter Health, a Northern California-based health system. CPMC consists of four campuses. The intervention was initiated at the Davies campus of CPMC in 2008, and then spread to the Pacific campus the following year. The Davies campus is a smaller, nonteaching community hospital with an average annual ED volume of 13,000 visits and an admission rate of 11%. The Pacific campus is a larger, teaching hospital with 28,000 annual visits in the ED and an admission rate of 24%. The CPMC PC team, consisting of two physicians and a nurse practitioner, all of whom are full-time and PC board-certified, circulates between both campuses during business hours and has access to chaplains and social workers on an as-needed basis. The PC team also has an inpatient service at the Davies campus whose patients are housed on swing beds in an acute care unit with specially trained registered nurses. Other details regarding the service have previously been described.⁴

PC was integrated into the ED in the following four ways: First, we identified one ED physician and two ED nurse champions to be the primary in-unit boosters. They attended the Education in Palliative and End-of-Life Care for Emergency Medicine and End-of-Life Nursing Education Consortium Critical Care programs, respectively. After completion, the clinicians worked with the study team to implement the new PC pathway in the ED through peer education and reinforcement using case studies of current patients. Second, they conducted a more formal in-service at the ED physician and nursing staff meetings at each site regarding PC service offerings, targeted patient groups for PC referral (see below), the referral process, and basic tenets about PC and symptom management. All ED attending physicians and nurses were in attendance, though residents who circulated through the Pacific ED did not attend. The PC pathway focused on four patient groups that were most identifiable to ED clinicians with the highest potential for improved care: 1) metastatic cancer; 2) advanced dementia; 3) congestive heart failure with two or more ED visits at a Sutter Health facility in the last 6 months; and 4) chronic obstructive pulmonary disease with two or more ED visits at a Sutter Health facility in the last 6 months. Clinicians were encouraged to request PC consultation in the ED for patients with other conditions as needed. Third, ED clinicians received a laminated info card regarding PC and the four high-risk patient populations. Fourth, PC team members made rounds to the ED to remind clinicians of these categories and discuss available services.

After identifying patients appropriate for PC consultation, ED clinicians placed them within the PC “pathway.” Nurse-initiated requests required physician approval. Consultations requested during available hours (Monday-–Friday, 7am to 7pm) were executed immediately; for off-hour requests, the PC team was paged and if it was determined appropriate by the PC team member on call, patients were admitted to the acute PC unit by the evening/weekend hospitalists, who had previously received specific training on admitting patients to the acute PC unit. The PC team would then assume the attending role on the next working day. If not appropriate for the acute PC unit or a bed was not available, the patient was admitted to the medical service, and a PC consult was initiated on the current unit on the next working day.

Data collection

Preintervention data collection for both campuses started in January 2006. Postintervention data were collected over a 28-month period from August 2008 to December 2010 and over a 14-month period from November 2009 to December 2010, for the Davies and Pacific campuses, respectively. CPMC's administrative records system included both ED and palliative medicine service data. Intervention group patients received or were referred to PC consultation in the ED. Referral for PC consultation occurred after hospital admission in the comparison group.

Data analysis

Data were analyzed using an ordinary least squares (OLS) multivariate regression. In addition, we calculated a propensity score based on several observable variables, which included the following patient-level characteristics: age, gender, All Patient Refined Diagnostic Related Group (APRDRG) risk of mortality (ROM) and severity of illness (SOI), 90-day readmission status, facility, and insurance type.⁵ APRDRG is a severity coding methodology developed by 3M Health Information Systems that assigns an SOI and ROM score in four categories, used to evaluate resource utilization and predict inpatient mortality.¹³ The propensity score approach creates a predicted probability of treatment using a logistic regression to “balance” the groups based on observable characteristics.^14,15 We used the flexible logit specification to estimate the propensity score and then included the propensity score as a covariate in our regression.¹⁶

Our assumption is that outcomes are independent of assignment to the intervention group. In the study, we compare the estimates of two key models: 1) multivariate regression not including a propensity score and 2) multivariate regression including covariates and propensity score. The last is the “doubly robust” model, illustrated by the following equation for each individual i: \documentclass{aastex}\usepackage{amsbsy}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{bm}\usepackage{mathrsfs}\usepackage{pifont}\usepackage{stmaryrd}\usepackage{textcomp}\usepackage{portland, xspace}\usepackage{amsmath, amsxtra}\pagestyle{empty}\DeclareMathSizes{10}{9}{7}{6}\begin{document} \begin{align*}{\rm y}_{\rm i} = & \alpha + \beta_{1}{\rm x}_{{\rm i}1} + \beta_{2}{\rm x}_{{\rm i}2} + \beta_{3}{\rm x}_{{\rm i}3} + \beta_4{\rm x}_{{\rm i}4} + \beta_{5}{\rm x}_{{\rm i}5} \\& + \beta_{6}{\rm x}_{{\rm i}6} + \beta_{7}{\rm x}_{{\rm i}7} + \beta_{8}{\rm x}_{{\rm i}8} + \beta_{9}{\rm x}_{{\rm i}9}+ \beta_{10}{\rm x}_{{\rm i}10} + \varepsilon _{\rm i} ,\end{align*} \end{document}

where ɛ_i ∼ N (0, σ)

and

x_i1=palliative care consultation status

x_i2=patient age

x_i3=patient gender

x_i4=APRDRG risk of mortality

x_i5=APRDRG severity of illness

x_i6=facility

x_i7=readmission status

x_i8=insurance type

x_i9=study period (pre- or post-intervention start)

x_i10=propensity score of the ith patient

y_i=the length of stay for the ith patient.

Data analysis was performed with Stata statistical software (StataCorp., College Station, TX).

Missing/Excluded data

There were 85 observations with missing APRDRG ROM and SOI occurring when the procedures and diagnoses are unclassifiable; these observations, all from the control group, were excluded. In addition, outlier observations with total LOS values greater than 3 standard deviations [SD] above the mean were excluded. The final sample included 1435 observations.

Results

At the Davies and Pacific campus, 215 and 1220 patients, respectively, were admitted to the hospital through the ED and received a PC consultation at some point during their stay. Of the 215 cases at the Davies campus, 34 of the PC consultations were initiated in the ED and 181 after admission. For the 1220 cases at the Pacific campus, 16 of the PC consultations were initiated in the ED and 1204 after admission. Summary statistics for the intervention and control groups are shown in Table 1, broken down by hospital. The mean LOS for the intervention and control groups were 4.32 and 8.29 days, respectively (p<0.01), and LOS was consistently lower in the intervention group regardless of whether they were in an acute PC swing bed or on a non-PC unit. Table 2 shows the patient volumes meeting the four screening criteria as discussed above. A substantial number of patients not meeting the screening criteria received consultation in the ED by a joint decision of the ED attending and the PC team based on the clinical picture of the patient. These patients had varying diagnoses of low frequency, the most common diseases of which were kidney failure (6%), pneumonia (4%), and intracerebral hemorrhage (3%). Additionally, a few patients had cancer but did not meet criteria for metastatic cancer.

Table 1.

Summary Statistics of Patients with ED-initiated versus Postadmission PC Consultations

	Control		Intervention
	(n=1385)		(n=50)
Age, years	Pacific (n=1204)	Davies (n=181)	Pacific (n=16)	Davies (n=34)
Mean	76.7	76.0	81.1	69.4
Median	81.0	82.0	87.0	70.5
Interquartile range	66.0–88.0	65.0–89.0	74.0–91.5	58–80
Gender, n (%)
Female	651 (54.1)	94 (51.9)	11 (68.8)	12 (35.3)
Risk of mortality, n (%)^**
Minor	27 (2.2)	5 (2.8)	1 (6.3)	1 (2.9)
Moderate	268 (22.3)	43 (23.8)	3 (18.8)	9 (26.5)
Major	512 (42.5)	75 (41.4)	8 (50.0)	23 (67.7)
Extreme	397 (33.0)	58 (32.0)	4 (25.0)	1 (2.9)
Severity of illness, n (%)^**
Minor	18 (1.5)	2 (1.1)	1 (6.3)	2 (5.9)
Moderate	188 (15.6)	37 (20.4)	4 (25.0)	11 (32.4)
Major	609 (50.6)	84 (46.4)	7 (43.8)	19 (55.9)
Extreme	389 (32.3)	58 (32.0)	4 (25.0)	2 (5.9)
Readmission, n (%)
% Yes	155 (12.9)	24 (13.3)	2 (12.5)	9 (26.5)
Study period, n (%)^**
% Postintervention	502 (41.7)	159 (87.9)	4 (25.0)	30 (88.2)
Insurance, No. (%)
Medicare	490 (40.7)	48 (26.5)	4 (25.0)	10 (29.4)
Medicaid	64 (5.3)	16 (8.8)	1 (6.3)	5 (14.7)
Commercial	645 (53.6)	115 (63.5)	11 (68.8)	19 (55.9)
Other	5 (0.4)	2 (1.1)	0 (0.0)	0 (0.0)
Mean length of stay (Standard error)	8.16 (0.18)	9.15 (0.54)	3.69 (0.71)	4.62 (0.65)

p<0.05, ^**p<0.01 between intervention and control groups.

Table 2.

Number of Patients Meeting Screening Criteria

		Control (n=1385)		Intervention (n=50)
	Total (n=1435)	Pacific (n=1204)	Davies (n=181)	Pacific (n=16)	Davies (n=34)
Dementia,^a n (%)	86 (6.0)	65 (5.4)	18 (9.9)	0 (0.0)	3 (8.8)
Cancer, n (%)	373 (26.0)	319 (26.5)	34 (18.8)	3 (18.8)	17 (50.0)
Advanced COPD, n (%)	29 (2.0)	24 (2.0)	4 (2.2)	1 (6.3)	0 (0.0)
Advanced CHF, n (%)	46 (3.2)	42 (3.5)	3 (1.7)	0 (0.0)	1 (2.9)

Patients with dementia (not advanced dementia) are represented.

CHF, congestive heart failure; COPD, chronic obstructive pulmonary disease.

In the control group, 6 of the 1204 patients and 1 of the 181 patients who received a PC consultation after admission at Pacific and Davies, respectively, were admitted to the inpatient PC unit. Of the 34 patients seen by the PC team in the ED at Davies, 2 were admitted to the intensive care unit. An additional 15 patients were admitted directly to acute PC swing beds on the Davies campus. At the Pacific campus, none of the 16 patients were admitted directly to the acute PC swing beds at the Davies Campus, although patients are frequently transported between the two campuses.

For the model used to derive the propensity scores, the area under the receiver operator curve (ROC) was C=0.81, signifying it adequately predicted the probability of PC in the ED. SOI and facility were significant at the 0.05 level. In the baseline regression model (Model 1), PC initiated in the ED compared with after hospital admission was associated with a mean decrease in total LOS by 3.53 days (p<0.01). All covariates were significantly associated with total LOS, except patient gender and insurance type (Table 3). In addition, admission to an inpatient PC unit and whether the referral and consult occurred on the same day were not significantly associated with LOS and were not included in the model. Model 2 included covariates and propensity scores and showed that the intervention was associated with a mean decrease in LOS by 3.63 days (p<0.01). A unit increase in each of APRDRG ROM and SOI (between the categories of minor, moderate, major, and extreme) was associated with an average increase in LOS by 0.79 and 2.21 days, respectively, in the direction as predicted. Facility, patient sex, and insurance type were not significantly associated with LOS. Finally, both an increase in age and positive readmission status were associated with a significant decrease in LOS (−0.74 and −2.23 days respectively, p<0.01 for both).

Table 3.

Factors Associated with Total Inpatient Length of Stay, Regression Coefficients (SE) for Models 1 and 2 (n=1435)

	1	2
Palliative care consultation in the ED	−3.53 (0.88)^**	−3.63 (0.89)^**
Risk of mortality	0.82 (0.28)^**	0.79 (0.29)^**
Severity of illness	2.12 (0.31)^**	2.21 (0.35)^**
Age	−0.75 (0.12)^**	−0.74 (0.12)^**
Readmission status	−2.17 (0.46)^**	−2.23 (0.47)^**
Study period	−1.14 (0.33)^**	−1.12 (0.33)^**
Facility	1.59 (0.48)^**	1.14 (0.93)
Gender	0.54 (0.32)	0.56 (0.32)
Insurance type	0.27 (0.16)	0.26 (0.16)
Propensity score		3.22 (5.75)
Constant	2.38 (0.91)^**	2.07 (1.10)

p<0.05, ^**p<0.01.

SE, standard error.

Discussion

We found in our study that early initiation of PC consultation in the ED was associated with a significantly lower LOS (–3.63, p<0.01). The average time to consult was 0.26 days, for ED initiated consultations, as compared with 6.5 days for referrals occurring after hospital admission. Although the numbers are small, these results illustrate the utility of a PC pathway in the ED to appropriately triage patients based on their conditions and treatment goals and improve patientcentered care. In addition, we found that age and prior admission were associated with a significant decrease in LOS, pointing to the possibility that these are factors associated with patients or family caregivers opting for less intensive treatment.

A recent study¹⁷ found that among patients who visited the ED during the last month of life, a disproportionate number were subsequently hospitalized and of those, a disproportionate number died in an acute care setting (77% and 68%, respectively). Hospice use was found to be the strongest (negative) predictor of ED use at the last month of life. Hospice care is not appropriate for all patients, but this illustration of health care utilization at the end of life speaks to the potential of PC to redirect patients to alternative care that may better meet treatment goals and potentially lower the cost of care. PC planning in the outpatient setting is preferred but for unavoidable ED visits, PC in the ED shows great potential.

Very recent studies have found relatively long inpatient delays for PC consultations (around 6.0 days).^18,19 These clearly illustrate the necessity for moving PC consults to occur sooner after admission and also the clear opportunity of PC consults even earlier in the ED to potentially improve quality of care while simultaneously reducing utilization and costs. From a clinical perspective, there is a growing realization that patient treatment goals may not align well with the traditional emergency medicine model.¹⁷ Some may argue that a chaotic ED is not an appropriate setting to discuss goals of care, and that PC consults provide expertise in areas such as pain management and family emotional/bereavement support that an emergency clinician simply does not have the capacity or capability to provide given time pressures and training.²⁰ In the current study we do not examine the impact of the PC service on ED clinician satisfaction, but this should be an area for further research. Additional demonstration of benefits to patients and their families, education, training, and facilitated consultation with PC teams can support this care delivery innovation. Reduced costs from unwanted and unnecessarily extended hospital admissions provide the financial incentive for hospitals to appropriately staff PC programs, especially given the implications of the Affordable Care Act and accountable care organizations on hospitals and health systems.

Limitations

A clear limitation of the analysis is nonrandom patient assignment to control or intervention group. As a result, there are inherent challenges to accurately determining the intervention effects, including estimation bias due to selection, which randomization would otherwise have accounted for, and endogeneity due to unobserved characteristics in patients referred in the ED versus those who were not. We attempt to decrease biased estimates by utilizing propensity scores; however, we are unable to test whether the technique yields reasonably accurate estimates.

In addition, the small number of patients in the intervention group results in low power to detect an effect on LOS. Although the number of patients receiving the intervention was low in both campuses, the Pacific campus had a substantially lower uptake of the intervention, which may be a result of different implementation and ED processes between the two campuses. The Pacific ED has greater patient volume, greater number of clinicians including residents, and underwent a shorter implementation period as compared with the Davies campus. The study team is following up with ED staff in both campuses to understand barriers to referring appropriate patients to the palliative medicine team in the ED.

Third, the lack of PC swing bed use at Davies from patients admitted through the Pacific campus ED is an organizational barrier. Although it is fairly easy to transfer the patient by ambulance the short distance to a PC swing bed, it is easier to admit to the Pacific campus as an inpatient and have the PC team serve as a consult service, thus limiting access to specially trained staff on the Davies PC swing bed unit. This could potentially have affected LOS among this subset of patients where this portion of the intervention is less available, although facility was not associated with LOS in our final model.

Finally, although we believe that the patients identified in this analysis primarily used a Sutter Health facility for their care, it is possible that some patients had ED visits at other non-Sutter Health facilities that were not captured by these data.

Conclusion

This study is a first step toward examining the benefits of PC in the ED on hospital utilization.

Although our study showed that patients with life-limiting illness seen by the PC team in the ED have an associated decrease in inpatient LOS upon admission, further study is needed to examine why changes in LOS occurred, as well as whether there are potential differences in intensive care unit LOS, costs, clinical outcomes, and patient satisfaction outcomes. Additional study is also needed to examine the best methods for implementation of PC intervention in the ED, especially given the time sensitive nature of ED care and the staffing required to provide PC in such a setting.

Footnotes

Acknowledgments

The authors thank the Metta Fund for its support. Linda Blum, RN, NP, and Catherine Seeley, MD are part of the PC team at CPMC. Finally, we would like to acknowledge emergency medicine practitioners at CPMC: Robin Serrahn, MD, Holly Bennett, RN, and Claudia Colley, RN.

Author Disclosure Statement

No competing financial interests exist.

References

Brody

, Ciemins

, Newman

, Harrington

. the effects of an inpatient palliative care team on discharge disposition. J Palliat Med, 2010; 13:541–548.

Hanson

, Rowe

, Wessell

, Caprio

, Winzelberg

, Beyea

, Bernard

. Measuring palliative care quality for seriously ill hospitalized patients. J Palliat Med, 2012; 15:798–804.

Temel

, Greer

, Admane

, Gallagher

, Jackson

, Lynch

, Lennes

, Dahlin

, Pirl

. Longitudinal perceptions of prognosis and goals of therapy in patients with metastatic non-small-cell lung cancer: Results of a randomized study of early palliative care. J Clin Oncol, 2011; 29:2319–26.

Ciemins

, Blum

, Nunley

, Lasher

, Newman

. The economic and clinical impact of an inpatient palliative care consultation service: A multifaceted approach. J Palliat Med, 2007; 10:1347–1355.

Morrison

, Penrod

, Cassel

, Caust-Ellenbogen

, Litke

, Spragens

, Meier

. Palliative Care Leadership Centers' Outcomes Group: Cost savings associated with US hospital palliative care consultation programs. Arch Intern Med, 2008; 168:1783–1790.

Chan

. End-of-life and palliative care in the emergency department: A call for research, education, policy and improved practice in this frontier area. J Emerg Nurs, 2006; 32:101–103.

Grudzen

, Richardson

, Hopper

, Ortiz

, Whang

, Morrison

. Does palliative care have a future in the emergency department? Discussions with attending emergency physicians. J Pain Symptom Manage, 2011; 43:1–9.

Quest

, Asplin

, Cairns

, Hwang

, Pines

. Research priorities for palliative and end-of-life care in the emergency setting. Acad Emerg Med, 2011; 18,18:e70–e76.

Grudzen

, Stone

, Morrison

. The palliative care model for emergency department patients with advanced illness. J Palliat Med, 2011; 14:945–950.

10.

Calder

, Forster

, Stiell

, Carr

, Perry

, Vaillancourt

, Brehaut

. Mapping out the emergency department disposition decision for high-acuity patients. Ann Emerg Med, 2012; 60:567–576.

11.

Nugus

, Holdgate

, Fry

, Forero

, McCarthy

, Braithwaite

. Work pressure and patient flow management in the emergency department: Findings from an ethnographic study. Acad Emerg Med, 2011; 18:1045–1052.

12.

DeLia

, Cantor

, Brownlee

, Nova

, Gaboda

. Patient preference for emergency care: Can and should it be changed? Med Care Res Rev, 2012; 69:277–293.

13.

Lagman

, Walsh

, Davis

, Young

. All patient refined-diagnostic related group and case mix index in acute care palliative medicine. J Support Oncol, 2007; 5:145–49.

14.

Rosenbaum

, Rubin

. The central role of the propensity score in observational studies for causal effects. Biometrika, 1983; 70:41–55.

15.

Dehejia

, Wahba

. Propensity score-matching methods for nonexperimental causal studies. Rev Econ Stat, 2002; 84:151–161.

16.

Austin, PC:. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behav Res, 2011; 46:399–424.

17.

Smith

, McCarthy

, Weber

et al. Half of older Americans seen in emergency department in last month of life; most admitted to hospital, and many die there. Health Aff, 2012; 31:1277–1285.

18.

Gruzden

, Hwang

, Cohen

, Fischman

, Morrison

. Characteristics of emergency department patients who receive a palliative care consultation. J Palliat Med, 2012; 15:396–399.

19.

Pantilat

, Kerr

, Billings

, Bruno

, O'Riordan

. Characteristics of palliative care consultation services in California hospitals. J Palliat Med, 2012; 15:555–560.

20.

DeVader

, Jeanmonod

. The effect of education in hospice and palliative care on emergency medicine residents' knowledge and referral patterns. J Palliat Med, 2012; 15:510–515.