© U.S. Cancer Pain Relief Committee, 2001 0885-3924/01/$–see front matterPublished by Elsevier, New York, New York PII S0885-3924(00)00260-8

Vol. 21 No. 3 March 2001 Journal of Pain and Symptom Management 197

Original Article

Differences Between Patients Referred to Hospice from Academic vs.Non-Academic Settings

David J. Casarett, MD, MA

Philadelphia Veterans Affairs Medical Center, and the Division of Geriatrics, the Institute on Aging, and the Center for Bioethics, University of Pennsylvania, Philadelphia, Pennsylvania, USA

Abstract

Hospice is designed to provide end-of-life care to patients who are terminally ill. Increasingly, hospices are developing affiliations with academic medical centers. However, little is known about this group of patients, and how their needs may differ from those of the general population of hospice patients. To identify differences between patients referred to an urban, non-profit hospice from academic vs. non-academic medical centers, a retrospective cohort study of 1,691 patients admitted to an inpatient and outpatient hospice program was conducted. Admission dates were between January 1997 and January 1999, and data were gathered until discharge or death. The major outcome of interest was the need for nursing interventions at the time of entry into hospice. Patients referred from academic medical centers were younger, had higher incomes, and were less likely to have Medicare or Medicaid. Patients referred from academic medical centers were less likely to have a Do Not Resuscitate order or a living will, and had more medical and nursing needs. Survival analysis revealed no difference in length of stay between patients referred from academic and non-academic medical centers. Patients referred to hospice from academic medical centers have greater needs for nursing and medical care than do patients referred from non-academic medical centers. The implications of these findings for policy are discussed.

J Pain Symptom Manage 2001; 21:197–203.

© U.S. Cancer Pain Relief Committee, 2001.

Key words

Hospice, palliative care, social support, Medicare

Introduction

The past ten years have seen a growing atten-tion to home hospice programs as a way ofimproving care for dying patients.

1,2

Hos-pice’s influence and presence has gradually ex-panded since its inception, and patients from a

variety of backgrounds and geographic regionsnow have access to the services that hospice of-fers.

3

One of the most recent developmentshas been the growth of affiliations betweenhospices and academic medical centers. Theseaffiliations should be a welcome opportunityboth for hospices seeking a broad base of com-munity support, and for academic medical cen-ters seeking to improve end-of-life care fortheir patients. However, although these affilia-tions hold promise, they also raise several im-portant questions.

For instance, it is possible that patients re-

Address reprint requests to:

David Casarett, MD, Insti-tute on Aging, Center for Bioethics, University ofPennsylvania, 3615 Chestnut Street, Philadelphia,PA 19104 USA.

Accepted for publication: April 18, 2000.

198 Casarett Vol. 21 No. 3 March 2001

ferred from academic medical centers may bereferred later in the course of their illness. Ifthis is the case, patients referred from academicmedical centers may have a length of stay that isshorter than that of other patients, perhaps withinsufficient time to meet patients’ needs forcare. This finding would provide reason for cau-tion among hospices that are considering affilia-tions with academic medical centers.

It is also possible that patients referred to hos-pice from academic medical centers may comewith greater needs for care. For instance, thecare of patients from academic medical centersmay have focused on life prolongation ratherthan on symptom control. Therefore, by thetime that they are referred to hospice, these pa-tients may have greater needs for pain andsymptom management, and perhaps greaterneeds for other nursing care such as the man-agement of intravenous catheters and gastros-tomy tubes. Patients from academic medicalcenters might also have greater needs for care ifthey suffer from more severe illnesses.

Finally, one important task in end-of-life careis that of anticipatory guidance, and attentionto the family’s needs for support. If academicmedical centers emphasize prolongation oflife, it is possible that other interventions likeanticipatory guidance may receive less atten-tion. If this is true, the families of patients re-ferred from academic medical centers mayhave a higher prevalence of anticipatory grief.

These differences, if they exist, have impor-tant implications for hospices that are consid-ering affiliations with academic medical cen-ters. For instance, short lengths of stay forpatients from academic medical centers maybe very difficult for many hospices to sustain,especially given lengths of stay that are alreadyless than one month.

3,4

Similarly, a finding thatpatients from academic medical centers havegreater needs for care would suggest that hos-pices that develop relationships with academicmedical centers should consider budget andstaffing plans carefully. An increased incidenceof anticipatory grief among the families of pa-tients from academic medical centers may sug-gest a need for more aggressive psychologicalsupport for these patients and their families. Fi-nally, all of these possibilities represent oppor-tunities for education of academic clinicians.

In order to determine whether such differ-ences exist between patients referred from aca-

demic and non-academic medical centers, thisretrospective cohort study was performed.

Methods

The patients in this retrospective study wereidentified in the database of a large non-profiturban hospice. This hospice provides home careas well as inpatient care at multiple institutionson a contractual basis. This hospice was chosenbecause it has a close relationship with an aca-demic medical center, and because it maintainsmeticulous computerized records of its patients.These records include clinical and demographicdata, as well as extensive clinical notes. Allrecords are used in daily patient care, whichhelps to ensure their accuracy and completeness.

Because of concern that the federal investi-gation of hospices that was begun in 1997might have affected patterns of care,

5,6

all datawere collected after January 1997. All patientswith a first admission to hospice between Janu-ary 1997 and January 1999 were included inthis study. All patients admitted during the pre-defined time period either died or had beendischarged at the time of data analysis.

Patients were defined as having been re-ferred from an academic medical center if theirprimary physician was based at the primaryteaching hospital of an allopathic medicalschool or at one of that hospital’s training sitesin Internal Medicine according the AmericanCouncil on Graduate Medical Education. Com-parisons were made between patients referredfrom academic and non-academic medicalcenters using data at the time of the patient’senrollment. Most variables were naturally codedin patient records as dichotomous (e.g., pres-ence of a Do Not Resuscitate order) or as ordi-nal (e.g., income ranges). Age was transformedinto an ordinal variable, in which one cutoffwas age 65, in order to account for the changein Medicare eligibility at this age. The smallnumber of patients with non-cancer diagnosesprecluded coding by diagnosis. Therefore, di-agnoses were coded as cancer, cancer with meta-static disease, and other diagnoses. This classi-fication system was chosen because it has beenfound to be relevant to prognostic predictionsin other settings.

7

In order to quantify the care needs of pa-tients at the time that they were referred, 11needs were identified in the computerized

Vol. 21 No. 3 March 2001 Hospice and Academic Medicine 199

record. These needs included opportunitiesfor interventions such as pain and other symp-toms. These needs also included other nursingneeds such as the presence of an indwelling in-travenous catheter or gastrostomy tube (Table1). It is important to note, however, that theseneeds did not include symptoms or problemsfor which a plan of management had been de-veloped. Thus, a patient with pain and nauseawho has an adequate therapeutic regimen inplace would not be counted as having needs re-quiring nursing intervention. Other needs,such as depression may be very important inthis population. However, the analysis pre-sented here is limited to those needs that areavailable in this database.

To determine whether observed differences inthese needs between academic and non-academicreferral sources were due to other confoundingvariables, a logistic regression model was con-structed. In this model, the outcome variable wasthe presence of at least one (of 11) needs requir-ing a nursing intervention. To build this model,univariate analyses of potential predictor variableswere conducted using odds ratios and chi-squaretests. For ordinal variables with greater than 2 cat-egories (e.g., age), dummy variables were created.Variables that reached the 0.25 level of signifi-cance were considered for inclusion into a pre-liminary regression model.

8,9

Variables were eliminated one at a time, andthe chi-square likelihood ratio test was used toevaluate the contribution of each variable. Aprocess of backward deletion was used to iden-tify independent predictors of withdrawal. Theremaining variables from the preliminary modelwere then restored, and were retained in the fi-nal model if they achieved statistical signifi-

cance or if they affected the significance of theother variables in the model.

Anticipatory grief, defined as the presenceof mourning symptoms before the patient’sdeath,

10,11

was compared between groups. Antic-ipatory grief was coded when the intake nurseand social worker agreed that an interventionsuch as psychotherapy or life review was indi-cated. Appropriate univariate tests were per-formed, followed by the construction of a logis-tic regression model using identical procedures.

To examine possible differences in survivalbetween patients referred from academic andnon-academic medical centers, Kaplan-Meier sur-vival functions were calculated and comparedwith the log-rank test.

Table 1

Needs for Intervention at the Time ofHospice Referral

SymptomsPainNauseaConstipationDyspneaDysphagiaAscites

Other nursing/social needsImpaired mobilityFeeding tube in useSeizuresIncontinenceIntravenous catheter in use

Table 2

Characteristics of All Hospice Patients (

n

5

1691)

n

%

Academic medical center

a

411 24.3Ethnicity

White 1055 62.4All other 610 36.1Missing 26 1.5

Marital statusFemale 937 55.4Married 686 40.6Other

b

1005 59.4Age

,

65 449 26.665–79 621 36.7

.

79 621 36.7Public insurance

c

419 24.7Annual income

,

$15,000 662 39.2$15,001–35,000 789 46.7

.

$35,000 194 11.5Missing 46 2.7

Do Not Resuscitate order 1177 69.6Advance directive 690 40.8Power of attorney (health care) 157 9.3Symptoms requiring intervention

d

981 58.0Primary diagnosis

Non-cancer 403 23.8Cancer (no metastases) 597 35.3Cancer (metastases) 691 40.9

Impaired mobility

e

301 17.8Living environmentHome alone 303 17.9Home with other 1215 71.8Nursing home 173 10.2

a

Primary physician was based at the primary teaching hospital of anallopathic medical school, or at one of that hospital’s designatedtraining sites in Internal Medicine.

b

Single, widowed, divorced/separated, living with partner.

c

Public insurance: Medicaid, Medicare, or no insurance.

d

Symptoms requiring intervention at the time of hospice enroll-ment (pain, nausea, dyspnea, constipation, dysphagia, ascites).

e

Impaired mobility defined as needing assistance with ambulationor transfers.

200 Casarett Vol. 21 No. 3 March 2001

STATA for Windows (version 5.0; Stata Co.College Station, TX, USA) was used for all sta-tistical procedures. The research committee ofWissahickon Hospice and the Institutional Re-view Board of the University of Pennsylvaniaapproved this study.

Results

Over a two-year period, 1,691 patients wereadmitted to this hospice program and were fol-lowed until discharge or death (Table 2). Thispatient population is diverse and in generalhas clinical and demographic characteristicsthat are typical of other hospices in this coun-

try.

3

However, one exception was noted. Thispopulation has a median length of stay of 13days, which is considerably less than the na-tional median of 25 days.

3

This difference sug-gests that these patients are referred to hospicelater in their disease course than are other pa-tients nationwide.

Of the 1691 patients in this cohort, 411 (24%)were referred from an academic medical center.When these patients were compared with allother enrolled patients over the same time pe-riod, several differences were found (Table 3).Overall, these patients tended to be younger(mean age 66 vs. 74) and were less likely to live ina nursing home (4% vs. 12%). Patients from aca-

Table 3

Demographic Characteristics of Patients from Academic and Non-Academic Settings

Academic

n

5

411 (%)Non-Academic

n

5

1280 (%) Odds Ratio 95% Confidence Interval

P

-value

Married 191 (46) 495 (39) 1.38 1.10–1.72 0.005Age

,

65 185 (45) 264 (21) 1.0065–79 132 (32) 489 (38) .39 .29–.50

,

0.001

.

79 94 (23) 527 (41) .25 .19–34

,

0.001Female 224 (55) 713 (56) .95 .76–1.19 0.67White 250 (62) 805 (64) .90 .72–1.14 0.39Public insurance 242 (59) 1030 (80) .35 .27–.44

,

0.001DNR order 261 (64) 916 (72) .69 .55–.87 0.002Power of Attorney 36 (9) 121 (9) .92 .62–1.36 0.67Living Will 134 (33) 556 (43) .63 .50–.80

,

0.001Income

a

,

$15,000 214 (52) 448 (35) 1.00$15,001–35,000 149 (36) 640 (50) .49 .38–.62

,

0.001

.

$35,000 42 (10) 152 (12) .58 .40–.84 0.005Nursing home resident 18 (4) 155 (12) .33 .20–.55

,

0.001

a

Comparisons use casewise deletion of missing data.

Table 4

Clinical Characteristics of Patients from Academic and Non-Academic Settings

Academic(

n

5

411 (%)Non-Academic(

n

5

1280 (%) Odds Ratio95%

Confidence Interval

P

-value

Primary diagnosisNon-cancer 76 (18) 327 (26) 1.00Cancer without metastases 135 (33) 462 (36) 1.26 .92–1.72 0.154Cancer with metastases 200 (49) 491 (38) 1.75 1.30–2.36

,

0.001Symptoms requiring intervention 266 (65) 715 (56) 1.45 1.15–1.82 0.002Pain 224 (54) 549 (43) 1.59 1.28–1.99

,

0.001Ascites 11 (3) 26 (2) 1.32 .66–2.67 0.44Constipation 30 (7) 63 (5) 1.52 .97–2.38 0.06Nausea/vomiting 18 (4) 55 (4) 1.02 .60–1.75 0.94Dyspnea 61 (15) 196 (15) .96 .71–1.32 0.82Dysphagia 5 (1) 31 (2) 0.50 .20–1.24 0.14Intravenous catheter 53 (13) 64 (5) 2.81 1.92–4.12

,

0.001Incontinence 47 (11) 104 (8) 1.46 1.02–2.10 0.04Impaired mobility 76 (18) 225 (18) 1.06 .80–1.42 0.67Feeding tube 29 (7) 53 (4) 1.76 1.10–2.79 0.02Seizures 9 (2) 15 (1) 1.89 .84–4.26 0.13Anticipatory grief 53 (12) 88 (7) 2.00 1.40–2.87

,

0.001

Vol. 21 No. 3 March 2001 Hospice and Academic Medicine 201

demic centers were also less likely to have a DoNot Resuscitate (DNR) order (63% vs. 72%) andless likely to have executed a living will (33% vs.43%). These patients were also more likely to bemarried (46% vs. 39%) and were less likely tohave public insurance (59% vs. 80%).

Several important clinical differences werefound as well. For instance, patients from aca-demic centers were more likely to have canceras their primary diagnosis (81% vs. 74%) (Ta-ble 4). Of those patients with cancer, patientsfrom academic centers were more likely tohave metastases at the time of referral (59% vs.51%). Patients from academic medical centerswere also more likely to have an intravenouscatheter at the time of hospice enrollment(13% vs. 5%) and were more likely to have agastrostomy tube in use (7% vs. 4%) (Table 4).

Overall, patients from academic centers werealso more likely to have needs that requirednursing interventions (78% vs. 69%). Specifi-cally, patients from academic medical centerswere more likely to have pain,(54% vs. 43%)and were more likely to be incontinent ofurine or stool (11% vs. 8%). Anticipatory griefwas noted by the assessment team more oftenin patients referred from academic centers(13% vs. 7%).

In a logistic regression model, referral froman academic medical center proved to be an

independent predictor of the presence of atleast one need requiring intervention (Table5). Other independent predictors were cancer,younger age, and residence outside of a nurs-ing home.

A logistic regression model was also con-structed to adjust for possible confounders ofthe association between referral source and thepresence of anticipatory grief (Table 6). In thismodel, referral from an academic medical cen-ter remained an independent predictor of thepresence of anticipatory grief in patients’ fami-lies. Anticipatory grief was also more likely ifpatients were married, and if they lived outsidea nursing home. Other variables, includingage, sex, the relationship of the next of kin,and the presence of a live-in caregiver, did notcontribute significantly to the model.

Although the mean length of stay wasshorter for patients from academic centers (34vs. 36 days) this difference was not statisticallysignificant (Wilcoxon Rank Sum test

P

5

0.79).In addition, Kaplan-Meier survival analysis didnot reach statistical significance (log rank test

x

2

5

0.30;

P

5

0.59).

Discussion

These data suggest that patients referred tohospice from academic medical centers may

Table 5

Logistic Regression Model of the Effect of Referral from an Academic Medical Center on the Presence of at Least One Need Requiring Nursing Intervention

Odds Ratio 95% Confidence Interval

P

-value

From academic medical center 1.35 1.03–1.78 0.03Primary diagnosis

Non-cancer 1.00Cancer 1.52 1.15–2.00 0.003Metastatic cancer 1.89 1.42–2.53

,

0.001Age

,

65 1.0065–79 1.04 .77–1.39 0.80

.

79 0.69 .52–95 0.02Nursing home resident 0.43 .30–.60

,

0.001

Chi square (6)

5

99.51;

P

,

0.001

Table 6

Logistic Regression Model of the Effect of Referral from an Academic Medical Center on the Presence of Anticipatory Grief

Odds Ratio 95% Confidence Interval

P

-value

From academic medical center 1.82 1.26–2.62 0.001Married 1.52 1.07–2.15 0.02Nursing home resident .14 .03–.58 0.006

Chi square (6)

5

35.40;

P

,

0.001

202 Casarett Vol. 21 No. 3 March 2001

differ from other patients in several importantways. For instance, several demographic differ-ences were noted, including differences inmarital status and health insurance. More im-portantly, however, several important clinicaldifferences were found. Overall, these datasuggest that patients referred to hospice fromacademic medical centers may have moreneeds for hospice care than do other patients.

First, these data suggest that patients fromacademic medical centers may be more likelyto have needs that require interventions bynurses or other care providers in the hospiceteam. Patients with symptoms will require addi-tional nursing contact and follow up. Patientswith intravenous catheters and gastrostomytubes will require additional nursing supervi-sion, and may require additional hospitaliza-tions of home procedures to maintain their pa-tency. These findings, taken together, suggestthat patients from academic centers may re-quire more intensive assessments and involve-ment by hospice nurses, home health aides,and other staff.

Second, in this study, the families of patientsreferred from academic medical centers weremore likely to exhibit anticipatory grief. Likepain and other symptoms, anticipatory grief re-quires assessment and interventions by the hos-pice team. Often, these interventions, such as alife review, are labor-intensive and time con-suming. Therefore, if anticipatory grief is morecommon among patients referred from aca-demic medical centers, hospices may wish toassess staffing needs accordingly for these pa-tients. In addition, hospices may also find ithelpful to consider early assessments and ac-tive anticipatory guidance through consulta-tion liaison teams based at the medical center.These sorts of interventions may be particu-larly important given the lower prevalence ofadvance directives in patients from academicmedical centers. Although the reasons for thisdifference are not clear, this finding suggeststhat hospice providers may have more oppor-tunities for psychosocial interventions and pa-tient education when patients are referredfrom academic centers.

Finally, these data indicate that patients fromacademic medical centers do not have shortersurvival than do other patients. This result wasunexpected, and is somewhat surprising, giventhat these patients are more likely to have meta-

static cancer. Moreover, these patients are alsomore likely to have other findings, such as painand other symptoms, which suggest advanceddisease. Confirmation of these results is neces-sary.

Although these findings are important, andshould be considered by hospices planning ac-ademic affiliations, this study has several limita-tions that should be noted. For instance, thesedata may not be representative of other hos-pices that are affiliated with academic medicalcenters. More specifically, these academic medi-cal centers may not be representative of others.Therefore replication of these findings in othersettings will be essential.

Another limitation of this study is its relianceon nursing assessments to identify symptomsrequiring intervention. Because these nursesdo not use a validated instrument to identifysymptoms, it is possible that some symptomswere overlooked. Moreover, the identificationof problems such as anticipatory grief may bequite subjective, and its identification in thesepatients may be somewhat inaccurate. Theseare significant concerns and the prevalence ofsymptoms reported here should be viewed withsome caution. However, there is no reason tosuspect that identification was either better orworse in patients from academic centers. In-deed, random misclassification would tend to re-duce differences between the two groups. There-fore, it seems unlikely that a systematic sourceof error biased the results reported here.

Despite these limitations, this study has im-portant implications for hospice planning andpolicy. If these findings are replicated in otherhospice settings, they suggest that hospices thatestablish ties with academic medical centersmay be caring for patients with more symp-toms, and more needs for care. Such differ-ences should be considered in drawing upplans for affiliations. Medicare and other pay-ers might also consider them in assessing reim-bursements to hospices.

Although these results are highly suggestive,further study will be essential. If these differ-ences are not better defined, and addressedthrough careful planning and perhaps throughcompensatory payment schemes, hospices mayexperience disincentives to develop close tieswith academic medical centers. This develop-ment would be unfortunate. One of the bestways to provide access to quality end-of-life care

Vol. 21 No. 3 March 2001 Hospice and Academic Medicine 203

is through hospice, and widespread access tohospice should be a priority.

12

In order toachieve this goal, it will be necessary to examinethese and other challenges that hospices mayface in enrolling groups of patients, in order toensure wide access to quality end of life care.

Acknowledgments

The author thanks Ms. Priscilla Kissick, Ms.Megan Bly, Dr. Janet Abrahm, and the staffand patients of Wissahickon Hospice. Dr. Cas-arett is the recipient of a Research Career De-velopment Award from the Department of Vet-erans Affairs and receives funding from theVistaCare Foundation, the Greenwall Founda-tion and the Commonwealth Fund.

References

1. Field MJ, Cassel CK. Approaching death. Im-proving care at the end of life. Washington, DC: Na-tional Academy Press, 1997.

2. Cassel CK, Vladeck BC. ICD-9 code for palliativeor terminal care. N Engl J Med 1996;135:1232–1234.

3. National Hospice Organization Fact Sheet.Available at http://www.nho.org/facts.htm. Re-trieved on November 28, 2000.

4. Christakis NA, Escarce J. Survival of Medicarepatients after enrollment in hospice programs. NEngl J Med. 1996;335:172-8.

5. Office of the Inspector General. Medicare hos-pice beneficiaries: services and eligibility. Washing-ton DC: Government Printing Office. Report OEI-04-93-00270.

6. Shapiro JP. Death be not swift enough. US Newsand World Report 1997:34–35.

7. Knaus W, Harrell FE, Lynn J, et al. The SUP-PORT prognostic model: objective estimates of sur-vival for seriously ill hospitalized adults. Ann InternMed 1995;122:191–203.

8. Hosmer DW, Lemeshow S. Applied logistic re-gression. New York: John Wiley and Sons; 1989.

9. Mickey J, Greenland S. A study of the impact ofconfounder-selection criteria on effect estimation.Am J Epid 1989;129:125–37.

10. Theut SK, Jordan L, Ross LA, Deutsch SI. Care-givers’ anticipatory grief in dementia: a pilot study.Int J Aging Hum Dev 1991;33:113–118.

11. Walker RJ, Pomeroy EC, McNeil JS, Franklin C.Anticipatory grief and AIDS: strategies for interven-ing with caregivers. Health & Social Work 1996;21:49–57.

12. Byock IR, Forman WB, Appleton M. Academy ofhospice physicians’ position statement on access tohospice and palliative care. J Pain Symptom Manage1996;11:69–70.