Support Care Cancer (2006) 14: 7177 DOI 10.

1007/s00520-005-0819-2

ORIGINA L ARTI CLE

Marylou Crdenas-Turanzas Richard M. Grimes Eduardo Bruera Beth Quill Guillermo Tortolero-Luna

Clinical, sociodemographic, and local system factors associated with a hospital death among cancer patients

Received: 16 February 2005 Accepted: 6 April 2005 Published online: 21 April 2005 # Springer-Verlag 2005 We want to state that no financial support was provided for this study M. Crdenas-Turanzas (*) Section of Health Services Research, Department of Biostatistics and Applied Mathematics, The University of Texas M. D. Anderson Cancer Center, 1515 Holcombe Blvd., P.O. Box 196, Houston, TX 77030-4009, USA e-mail: mcardena@mdanderson.org Tel.: +1-713-5634303 Fax: +1-713-5634246 R. M. Grimes . B. Quill The University of Texas School of Public Health, 1200 Herman Pressler Dr., Houston, TX 77030, USA E. Bruera Department of Palliative Care and Rehabilitation Medicine, The University of Texas M. D. Anderson Cancer Center, 1515 Holcombe Blvd., P.O. Box 8, Houston, TX 77030, USA G. Tortolero-Luna The University of Texas School of Public Health, 7000 Fannin St., Houston, TX 77030, USA

Abstract Objective: The study was conducted to examine factors associated with hospital deaths among a group of cancer patients. Patients and methods: A retrospective chart review of the M. D. Anderson Cancer Center Tumor Registry was conducted. Participants were all adult cancer patients, residents of the State of Texas diagnosed and treated since January 1, 1990, and who died during the years 1999 and 2000. The study outcome was the site of death. Main results: The inclusion criteria were met by 866 patients of whom 504 (58%) died in a hospital. The group included 489 (56%) men. A number of 641 (74%) were White, 104 (12%) Hispanic, 92 (11%) Black, and 29 (3%) of other origin. The majority, 501 (58%), had been diagnosed with stage IV disease, and the median survival time was 14 months. Multivariate logistic regression analysis showed patients diagnosed with hematologic cancers to be significantly more likely (p<0.001) of dying in hospitals, odds ratio [OR 2.88] and confidence interval [95% CI 1.79 4.63], women diagnosed with breast and gynecological cancers were significantly less likely (p=0.03) of dying at hospitals odds ratio [OR

0.64] and confidence interval [95% CI 0.420.96], when compared with patients diagnosed with other cancers. Lower household income per zip code of residency was marginally associated (p=0.06) with hospital deaths. Conclusions: The study identified groups of cancer patients at risk of hospital death. These results should account when planning the allocation of hospital palliative care services as well as when informing policy decisions about health care financing and delivery of these services. Keywords Place of death . End of life . Cancer

Introduction
The care of dying patients was recognized as a national health care issue in the 2001 report of the Institute of Med-

icine and the National Research Council [14]. The report identified wide gaps in knowledge about end-of-life care and the need for attention from researchers in the biomedical and social sciences and in health care services.

72

Most terminally ill patients prefer to die while cared for at home [11, 13, 17, 20]. Although hospice care is the kind of care people want at the end of life, only 25% of the patients who died in the USA during the year 2000 used hospice or home-based hospice services [15]. During the same year, according to the Centers of Disease Control and Prevention (CDC) out of 2,403,351 deaths reported in the USA, some 1,200,000 (50%) occurred in hospitals [2]. However, a large proportion of deaths are the result of long-term chronic illness. Malignancies, cerebrovascular disease, chronic lower respiratory disease, diabetes mellitus, chronic liver disease, and cirrhosis accounted for 39% of all deaths in the USA during 2002 [23], a number that does not account for deaths due to other chronic illnesses for which hospice care would be appropriate, i.e., some types of cardiovascular disease, nephritis, nephrotic syndrome, and nephrosis. Most cancer patients tend to die in short-stay hospitals and hospices, which makes characterization of the dying, as well as recognition of the determinants of in-hospital death important tools in planning and implementing quality endof-life services for cancer patients [8, 19, 24, 25]. Researchers have identified several factors associated with the place of death of cancer patients. These factors can be classified as clinical, patient, and local health care system factors. Clinical factors associated with a hospital death are the type of cancer [6, 9], extension of the disease (stage) [1, 5], the presence of comorbidities [7, 22], and survival time [16]. Sociodemographic factors include race [9, 18], socioeconomic status [16], age [10], gender [10, 12, 16], marital status [10, 16], and type of insurance [8, 30]. The influence of the local health care system resources on health outcomes has already been established by several researchers [16, 26, 27] Thus, if system factors influence patients place of death, one would expect that at least end-of-life services would be available for patients dying in hospitals. According to a survey of 3,939 hospitals who responded to the 1998 American Hospital Associations annual survey [31], 44% of the hospitals did not provide general end of life, pain management, and hospice services, and only one third provided one of the three services. End-of-life care is not likely to be available to patients dying in hospitals. Because the quality of care for the dying is a public health concern, this study examines the clinical, sociodemographic, and local health care systems factors associated with cancer patients deaths. In the long-term, research on factors influencing the place of death may help to inform policy decisions about health care financing and delivery of endof-life services.

registry includes all patients registered at this cancer center since 1944. As a section of the Medical Informatics Department, the registry maintains a database of 330,000 patients with the purpose of facilitating clinical cancer research. Patient information is abstracted from charts, validated by quality control procedures, and entered into the computerized patient information system by department personnel. The main source of information on patients vital status is mail and phone contact, but when no status can be established, vital statistics are used. Death certificates information from the Bureaus of Vital Statistics of Texas, Oklahoma, and New Mexico are compared monthly with the names of patients with cancer who are not known to have died. Patients are considered inactive if they have not been seen during the last 12 months. In an attempt to contact these patients, the Tumor Registry sends a letter to establish their vital status. On 2003, the response rate was about 80%. Study sample Cases were included if the following criteria were met: diagnosis of cancer between January 1, 1990 and December 31, 2000; death during the period of study, January 1, 1999 December 31, 2000; residence in the state of Texas; age 18 or more years; received primary diagnosis and treatment at MDACC; and place of death reported to the Tumor Registry. Patients diagnosed with in situ carcinomas, basal cell carcinomas, referred for a second opinion, or those who came only for screening or diagnostic procedures were excluded. Patients who came in an emergency situation for specific therapy or to undergo a diagnostic test not available to their referring physician were also excluded from this study. The total eligible sample of adult patients referred to this center for complete diagnosis and treatment during the study period was 2,734. Out of them, 911 had information on place of death, and 15 were excluded because of diagnosis of basal cell carcinoma. Thirty cases were eliminated because of missing data on marital status (n=8), ethnicity (n=1), and zip code of residency (n=21). Data on zip code of residency was used to evaluate family income and local health care resources available in the patients community of residence. Measurement Outcome variable The outcome variable was the place of death. Cases who died at MDACC or another hospital were classified as hospital deaths, while those who died at home, nursing home, or hospice were classified as dying elsewhere. To reflect the likelihood of hospital deaths, place of death was dichotomized giving the value 1 to a hospital death and the value 0 to a death occurring elsewhere.

Materials and methods

Data sources We conducted a retrospective chart review of the M. D. Anderson Cancer Center (MDACC) Tumor Registry. The

73

Independent variables These variables included three main groups: clinical, sociodemographic, and local health care system variables. Clinical variables, which were obtained from the MDACC Tumor Registry, included type of cancer, stages IIV, presence of diabetes or hypertension, and survival. Sociodemographic variables obtained from the registry were age at death, type of insurance (Medicaid/Medicare or other), marital status, race/ethnicity, and gender. The data on the median family income per zip code of residence, used to evaluate socioeconomic status, was obtained from the U.S. Census 2000 [3]. The type of cancer was grouped as leukemia and lymphoma, lung and bronchial, upper gastrointestinal, female cancers, head and neck carcinomas, etc. This classification was based in information provided by the Health Care Utilization Project Nationwide Inpatient Sample (HCUP-NIS) [21]. Percentages of in-hospital deaths per discharges were obtained and ranked by principal diagnoses (data not presented). In this data set, malignant neoplasm without specification of site, cancer of the liver and intrahepatic bile duct, leukemias, pancreatic, bronchial, and lung cancer were ranked among the first ten causes of in-hospital deaths. Female cancers, including breast and gynecological cancers, were included as a separate category because of the inconclusive evidence, suggesting that women are less likely than men to die at home [28]. The stage of the cancer was grouped as local (stage I), regional (stages I and II), and distant (stage IV); this variable was excluded later from the multivariate analysis because of a previous description of collinearity with survival [16]. With the aim of evaluating the influence of comorbidities, data on the presence of diabetes or hypertension was collected as a categorical variable. Sociodemographic variables such as age at the time of death were analyzed as continuous variables, while the marital status was classified as married, single, divorced, and widower. The type of insurance was dichotomized as having Medicare/Medicaid or not, and race/ethnicity was classified as White, Hispanic, Black, Other. To measure local health care system factors, information on registered short-term hospital beds/1,000 habitants per county, number of hospices per county, number of home health care agencies per county, and the type of county (urban/rural) was obtained from the Texas Almanac 2002/ 2003 [29] and the Texas Cancer Data Center [4]. Counties were classified as urban if located in a Consolidated Metropolitan Statistical Area (CMSA), Primary Metropolitan Statistical Area (PMSA), or a Metropolitan Statistical Area (MSA), according to the U.S. Census Bureau 2000 [3]. Statistical analysis Description of the study sample is presented by using measures of central tendency and frequencies. Bivariate analysis was accomplished by using Chi-square statistics

for categorical and ordinal variables and the t test and MannWhitney test for continuous variables. Logistic regression was used for the multivariate analysis, and covariates were selected for inclusion in the model if p value obtained in the bivariate analysis was less than 0.25. As it is considered a potential confounder, age at the time of death was included a priori in the multivariate model. All statistical analysis was conducted using SPSS release 11.01 (2001).

Results
The median age at diagnosis of this group of patients was 61 years, and the median age at death was 63 years. Of the cases, 56% were male and 68% were married. About three quarters were White (74%), Hispanic and Black participants were only 12 and 11% of the sample, respectively. More than half (58%) died in a hospital. A number of 146 (17%) were lymphoma or leukemia patients, 179 (21%) were diagnosed with lung and bronchial cancers, 113 (13%) had upper gastrointestinal cancers (including esophagus, stomach, and duodenum), 47 (5%) female patients had breast cancer, 78 (9%) had gynecological cancers (such as ovary, uterus, and cervix uterus), 71 (8%) had head and neck carcinomas, and 232 (27%) were categorized as having other types of cancers (such as prostate, colorectal, pancreatic, brain, and sarcomas). The majority (83%) were residents of a county located in a CMSA, PMSA, or MSA, and more than half (60%) were residents of the Greater Houston Area. This area includes Harris, Brazoria, Chambers, Fort Bend, Galveston, Liberty, Montgomery, and Waller counties. Most of the residents of the Greater Houston Area (72%) were residents of Harris County. Ninety-five percent were living in counties with at least one home health agency, almost 25% of them were living in a county lacking of a hospice (Table 1). The bivariate analysis showed clinical, patient, and local health care system factors were associated with hospital deaths (Table 2). The highest percentage of deaths in hospitals occurred amongst leukemia and lymphoma patients (79%) and the lowest was among upper gastrointestinal cancer patients (48%, p<0.0001). Overall, a marginal statistical association (p=0.06) was found between place of death and stage of disease, but when only patients with local and distant stage disease were compared, a statistically significant difference in the place of death was found (p=0.02). Patients with distant disease were more likely to die in hospitals (61%) than those diagnosed with local disease (53%). Sixty-nine patients (8%) had diabetes or hypertension at the time of diagnosis. A nonsignificant difference with hospital deaths was observed between patients with diabetes or hypertension (67%) and those without these diseases (57%, p=0.13). Of the patients in the sample, 28% survived only 6 months, and 47% had died

74

Table 1 Demographics of cancer patients with known place of death treated at MDACC and who died in 19992000 Variable Site of death Hospital Elsewhere Total Clinical factors Cancer site Leukemia and lymphoma Lung and bronchial Upper gastrointestinal Breast cancer Gynecological cancers Head and neck Other cancers Extension of disease Local (stages I and II) Regional (stage III) Distant (stage IV) Coexisting diabetes or hypertension No Yes Survival <1 year 12 years 25 years >5 years Patient factors Age at death <65 years 6570 years 7180 years >81 years Race White Hispanic Black Other Gender Male Female Marital status Married Not married Median household income <US$32,310 US$32,310US$37,972 US$37,973US$50,940 >US$50,940 Medicare and/or Medicaid at diagnosis Yes No Number (%)

Table 1 (continued) Variable Number (%)

504 (58) 362 (42) 866 (100)

146 179 113 47 78 71 232

(17) (21) (13) (5) (9) (8) (27)

Local health system factors Short-term hospital beds/1,000 habitants per county High (4.38)a 520 (60) Low (<4.38) 346 (40) Hospices per county of residence High (3)a 442 (51) Low (<3) 424 (49) Home health agencies per county of residence 442 (51) High (11)a Low (<11) 424 (49) Urbanb county of residence Yes 720 (83) No 146 (17)
a b

204 (23) 161 (19) 501 (58) 797 (92) 69 (8) 390 190 164 122 (45) (22) (19) (14)

Cutoff point is median value Includes all counties located in Consolidated Metropolitan Statistical Areas (CMSA), Primary Metropolitan Statistical Areas (PMSA) and Metropolitan Statistical Areas (MSA)

468 147 199 54 641 104 92 29

(54) (17) (23) (6) (74) (12) (11) (3)

489 (56) 377 (44) 592 (68) 274 (32) 233 189 221 223 (27) (22) (25) (26)

363 (42) 503 (58)

after 1 year of the initial cancer diagnosis. A significant (p=0.02) association was found between place of death and medial survival time, meaning that patients who died in hospitals survived less time than those who died elsewhere. When taking into account age at diagnosis, the association between survival and place of death did not remained significant. The meanSD age at death was 61.713.8 years with a median of 63 years (range 2094). Quartiles showed that 25% of these patients were diagnosed with cancer by the age of 51 years and 75% by the age of 69 years. However, no differences in mean age at diagnosis was observed with place of death (p=0.15). The averageSD household income zip code of residence of all patients was $40,769.097,905.48 (median $42,598.00; range $16,504.00$70,835.00). An inverse relationship between median zip code household income and place of death was observed. Patients living in areas of low median income (<$32,310.00) were more likely to die in hospitals (65%) compared to those living in areas of high median (>$50,940.00) income (56%, p=0.02). There was large variation in the number of short-term hospital beds available to cancer patients. While only 26 (3%) patients in the sample were living in counties without hospital beds at all, 377 (43%) residents of Harris County had 14,914 short-term hospital beds registered in their county. The average number of short-term beds per 1,000 county inhabitants in the sample was 3.73 (median 4.38, range 015.71). Although Harris County cases had the highest total number of hospital beds registered compared to the rest of counties in the sample, when beds per 1,000 county inhabitants were estimated, Harris County had only 4.39 registered beds per 1,000 county inhabitants. Of the patients in the sample, 16% had access to more beds per

75

Table 2 Univariate associations between covariates and place of death Variable Number dying in hospitals (%) p Value

Cancer site Leukemia and lymphoma 116 Lung and bronchial 103 Upper gastrointestinal 54 Female cancers 76 Other cancers 155 Severity of illness Local 108 Regional 88 Distant 308 Coexisting diabetes or hypertension No 458 Yes 46 Survivala NA Age at deatha NA Race White 376 Hispanic 64 Black 51 Other 13 Gender Male 289 Female 215 Marital status Married 352 Not married 152 Median household income <US$32,310 152 US$32,310US$37,972 114 US$37,973US$50,940 114 >US$50,940 124 Medicare and/or Medicaid at diagnosis Yes 210 No 294 Short-term hospital beds/1,000 NA habitants per countya Hospices per county of residencea NA Home health agencies per NA county of residence Type of county of residence Urban 408 Rural 96
a

(79) (57) (48) (48) (57) (53) (55) (61) (57) (67)

<0.001

=0.06

=0.13 =0.02 =0.59

(59) (61) (55) (45) (59) (57) (59) (55) (65) (60) (52) (56) (58) (58)

=0.39

=0.54

diagnosed with cancer. Out of 146 residents of rural areas, 96 (66%) died in hospitals, compared to only 408 (57%) of those who lived in urban counties (p=0.04). On the multivariate analysis, severity of illness was excluded from the final analysis. The decision was based in the statistically significant correlation (p<0.01) observed with survival as well as a previous report of collinearity [16] between the same variables. Other variables significantly correlated (p<0.01) were the number of hospices per county, the number of home health agencies per county, and the type of county. Because of this, the variable classifying counties as rural or urban remained for the final analysis. Variables included in the final model and their coding values are presented in Table 3. The result of the logistic analysis (Table 4) showed the type of cancer was significantly associated with death in the hospital after controlling for coexistence of diabetes or hypertension, survival time, age at death, income, and type of county of residency. Patients with hematological cancers were significantly more likely to die in hospitals (p<0.001). The odds of dying in hospitals for those with leukemia or lymphoma were 2.88 [95% CI 1.79, 4.63] times more than for patients diagnosed with other cancers after controlling for the covariates mentioned. On the other hand, patients diagnosed with breast and gynecological cancers were significantly less likely to die in hospitals (p=0.03) OR 0.64 [95% CI 0.420.966]. Household income, a patient factor, was marginally associated (p=0.06) with dying in a hospital.

=0.26

Discussion
=0.02

=0.86 =0.31 =0.07 =0.07

(57) (66)

=0.04

Continuous variable

1,000 county inhabitants than those living in Harris County. Wheeler, Galveston, Tom Green, Lamar, Lubbok, Wharton, and Nueces counties had more short-term beds/1,000 county inhabitants than Harris County. Most of the cases (83%) in our sample were living in an urban area when

The findings of this study confirm that clinical factors are important determinants of the place of death of cancer patients. Patients diagnosed with hematological cancers were at higher risk of dying in hospitals, while patients diagnosed with breast and gynecological cancers were more likely to die elsewhere. Other researchers have reported the association of cancer site with the place of death. Bruera et al. [8] found that patients diagnosed with hematologic cancer at M. D. Anderson Cancer Center and who were residents of the Houston metropolitan area were at higher risk of dying in hospitals (OR 4.4; 95% CI 2.8, 6.8). Gallo et al. [16] showed that in Connecticut, patients whose cause of death was colorectal cancer were less likely (RR=0.74, 95% CI 0.650.90) to die at home than those with other cancers. Practice style in treating hematologic cancers may be the reason these patients tend to die in hospitals. The increased survival of patients diagnosed with breast or early invasive cervical cancers may be the cause of these deaths occurring elsewhere, as patients may have more time to arrange for nursing home or hospice care. In this study, household income was marginally (p= 0.06) associated with the place of death in multivariate

76

Table 3 Variables included in the final multivariate model of factors associated with the place of death of cancer patients

Variable

Type

Coding

p Value in bivariate analysis <0.001

Cancer site

Categorical

Coexistence of diabetes or hypertension Survival time (months) Age at death (years) U.S. Census median household income per zip code of residency Type of county

Categorical Continuous Continuous Categorical

NA Not applicable a Reference category

Categorical

Hematologic Lung and bronchial Upper gastrointestinal Female cancers Othersa 1=Yes 0=Noa NA NA <US$32,310=1 US$32,310US$37,972=2 US$37,973US$50,940=3 >US$50,940=4a 1=Urban 0=Rurala

0.13 0.02 0.59 0.02

0.04

analysis. Our findings showed that patients living in lower income areas were more likely to die in hospitals when controlling for all other factors. Gallo et al. [16] reported similar findings when studying factors associated with home death of cancer patients in Connecticut. The authors reported that patients living in higher income areas tended to die at home. Patients residing in lower income areas may have less family support or less access to health care resources than those living in higher income areas. In contrast with results reported by others [9, 16, 24] who have described the underutilization of hospice service by AfricanAmericans, our study of race/ethnicity was not associated with place of death; this could be attributed to the small percentage of patients (10.4%) in this category or maybe because AfricanAmericans who receive treatment at M. D. Anderson Cancer Center are different than those receiving it elsewhere. The main limitations of this study are the bias due to the use of administrative data from a single institution and the
Table 4 Multivariate model of clinical, patients, and local system factors associated with place of death of cancer patients Variable name

lack of information about the place of death in almost two thirds of the cases. The use of data from a single institution makes it difficult to generalize results to all cancer deaths because patients referred to M. D. Anderson Cancer Center may be different than those attending other clinics and hospitals; that is, patients with certain cancer types or diagnosed with more advanced disease tend to be referred to this hospital. Patients whose place of death was not known are maybe of lower income, are residents of rural areas, or maybe they did not die in hospitals. Interestingly, this study found determinants potentially associated with place of death, as it is the case of comorbidities. Although they are not systematically recorded by registries, comorbidities may be an important determinant of the place of death. Cancer registries should record accurate information about the place of death and comorbidities in order to help researchers to understand factors influencing place of death.

Beta

Odds ratio [95% CI]

p Value 0.001* 0.001* 0.71 0.19 0.03* 0.52 0.11 0.27 0.06 0.12 0.47 0.26 0.37

Model: 2=55.62, df=11, p<0.001 Hosmer and Lemeshow test 11.00, df=8, p=0.20 *Significant p<0.05

Cancer site Hematologic Lung and bronchial Upper gastrointestinal Female cancer Survival time Coexistence of diabetes or hypertension Age at death Income <US$32,310 US$32,310US$37,972 US$37,973US$50,940 Urban residency

1.060 0.074 0.302 0.442 0.002 0.438 0.006 0.322 0.153 0.217 0.188

2.88 1.07 0.73 0.64 1.00 1.55 0.99 1.38 1.16 0.80 1.20

[1.79, [0.72, [0.46, [0.42, [0.99, [0.90, [0.98, [0.91, [0.76, [0.54, [0.79,

4.63] 1.59] 1.66] 0.96] 1.00] 2.66] 1.00] 2.08] 1.77] 1.18] 1.82]

77

In the long term, research about factors influencing the place of death may help to identify potential beneficiaries of palliative care services in hospitals as well as to inform policy decisions about health care financing and delivery of these services. The identification of potential beneficiaries

of palliative care in hospitals is an initial step in reducing disparities in the provision of health care services.
Acknowledgements The authors wish to thank Charissa Higginbotham and Lore Feldman for their editorial contributions which enhanced the quality of this manuscript.

References
1. Addington-Hall J, Altmann D, McCarthy M (1998) Which terminally ill cancer patients receive hospice inpatient care? Soc Sci Med 46:1011 1016 2. Anderson RAE (2003) The effect of revised populations on mortality statistics for the United States, 2000. Natl Vital Stat Rep 51:15 3. Anonymous (2000) Population, housing units, area, and density (geographies ranked by total population), vol 2004. US Census Bureau 4. Anonymous (2003) Directory of hospices and home health agencies in Texas. Texas Cancer Data Center, Houston, TX 5. Bast R, Kufe D, Pollock R, Weichselbaum R, Holland J, Frei E, Gansler T (2000) Cancer medicine, 5th edn. B.C. Decker, Inc., London 6. Bauduer F, Capdupuy C, Renoux M (2000) Characteristics of deaths in a Department of Oncohaematology within a general hospital. A study of 81 cases. Support Care Cancer 8:302306 7. Brooks S, Chen T, Ghosh A, Mullins C, Gardner J, Baquet C (2000) Cervical cancer outcomes analysis: impact of age, race, and comorbid illness on hospitalization for invasive carcinoma of the cervix. Gynecol Oncol 79:107 115 8. Bruera E, Russell N, Sweeney C, Fisch M, Palmer L (2002) Place of death and its predictors for local patients registered at a comprehensive cancer center. J Clin Oncol 20:21272133 9. Bruera E, Sweeney N, Russell JW, Palmer L (2003) Place of death of Houston area residents with cancer over a two-year period. J Pain Symptom Manage 26:637643 10. Costantini M, Camoirano E, Madeddu L, Bruzzi P, Verganelli E, Henriquet F (1993) Palliative home care and place of death among cancer patients: a population-based study. Palliat Med 7:323331 11. Dawson N (1991) Need satisfaction in terminal care settings. Soc Sci Med 32:8387 12. De Conno F, Caraceni A, Groff L, Brunelli C, Donati I, Tamburini M, Ventafridda V (1996) Effect of home care on the place of death of advanced cancer patients. Eur J Cancer 32A:11421147 13. Emanuel L, Von Gunten C, Ferris F (2000) Gaps in end-of-life care. Arch Fam Med 9:11761180 14. Foley K, Gelband H (2001) Improving palliative care for cancer. Institute of Medicine and National Research Council, Washington, DC 15. Fox E, Landrum-McNiff K, Zhong Z, Dawson N, Wu A, Lynn J (1999) Evaluation of prognostic criteria for determining hospice eligibility in-patients with advanced lung, heart, or liver disease. JAMA 282:16381645 16. Gallo W, Baker M, Bradley E (2001) Factors associated with home versus institutional death among cancer patients in Connecticut. J Am Geriatr Soc 771777 17. Gallup Organization Survey (1996) Knowledge and attitudes related to hospice care. Commissioned by the National Hospice Organization 18. Ghafoor A, Jemal A, Ward E, Cokkinides V, Smith R, Thun M (2003) Trends in breast cancer by race and ethnicity. CA Cancer J Clin 53:342 355 19. Goodlin SJ, Winzelberg GS, Teno JM, Whedon M, Lynn J (1998) Death in the hospital. Arch Intern Med 158:1570 1572 20. Greer D, Mor V, Morris J, Sherwood S, Kidder D, Birnbaum H (1986) An alternative in terminal care: results of the national hospice study. J Chronic Dis 39:926 21. HCUPnet (2002) Healthcare cost and utilization project. Agency for Healthcare Research and Quality, Rockville, MD 22. Izquierdo-Porrera A, Trelis-Navarro J, Gomez-Batiste X (2001) Predicting place of death of elderly cancer patients followed by a palliative care unit. J Pain Symptom Manage 21:481490 23. Kochanek KD, Smith BL (2004) Deaths: preliminary data for 2002. Natl Vital Stat Rep 52:147 24. Lackan NA, Ostir GV, Freeman JL, Mahnken JD, Goodwin JS (2004) Decreasing variation in the use of hospice among older adults with breast, colorectal, lung, and prostate cancer. Med Care 42:116122 25. Omara AN, Arenella C (2001) Minority representation, prevalence of symptoms, and utilization of services in a large metropolitan hospice. J Pain Symptom Manage 12:290297 26. Petrisek A, Mor V (1999) Hospice in nursing homes: a facility-level analysis of the distribution of hospice beneficiaries. Gerontologist 39:279290 27. Pritchard R, Fisher E, Teno J, Sharp S, Reding D, Knaus W, Wennberg J, Lynn J (1998) Influence of patient preferences and local health system characteristics on the place of death. J Am Geriatr Soc 46:12421250 28. Tang ST, McCorkle R (2001) Determinants of place of death for terminal cancer patients. Cancer Investig 19:165180 29. The Dallas Morning News L (2001) Texas almanac 2002/2003. Texas A&M University Press Consortium, Dallas, TX 30. Weitzen S, Teno JM, Fennell M, Mor V (2003) Factors associated with site of death: a national study of where people die. Med Care 41:323335 31. White K, Cochran C, Patel U (2002) Hospital provision of end-of-life services who, what, and where? Med Care 40:1725