Racial disparities in the use of hospice services according to geographic residence and socioeconomic status in an elderly cohort with nonsmall cell lung cancer

Abstract

BACKGROUND:

The authors investigated whether there were racial disparities in the receipt of hospice services within geographic residence and socioeconomic status (SES) levels.

METHODS:

In total, 117,894 patients aged ≥66 years with nonsmall cell lung cancer (NSCLC) were identified from the Surveillance, Epidemiology, and End Results-Medicare linked database who had received hospice services in the last 6 months before death and who died between January 1, 1991 and December 31, 2005. Multivariate odds ratios (ORs) with 95% confidence intervals (CIs) using mixed-effects logistic regression models were used to describe relations.

RESULTS:

In urban areas, there were significant disparity trends in age/sex-adjusted rates for blacks and Asians/Pacific Islanders compared with whites (P = .003 and P = .036, respectively). Blacks (OR, 0.79; 95% CI, 0.75-0.82), Asians/Pacific Islanders (OR, 0.42; 95% CI, 0.39-0.46), and Hispanics (OR, 0.81; 95% CI, 0.73-0.91) were less likely to receive hospice services. In rural areas, blacks were 21% less likely to receive hospice services (OR, 0.79; 95% CI, 0.63-0.98). Patients in the poorest socioeconomic status (SES) quartile were 7% less likely to receive hospice services (OR, 0.93; 95% CI, 0.86-1.00). Moreover, within stratified SES quartiles, blacks and Asians/Pacific Islanders had lower receipt of hospice services, and Asians/Pacific Islanders were the least likely to receive hospice services, particularly those in the poorest SES quartile. In general, older patients and women were more likely to receive hospice services.

CONCLUSIONS:

There were substantial disparities in the receipt of hospice services among ethnic minorities within urban and rural areas and within SES levels. The results indicated that efforts are needed to identify barriers, enhance support, and educate patients on the benefits of hospice services. Cancer 2010. © 2010 American Cancer Society.

Lung cancer, which accounts for approximately 34% of hospice patients with cancer,1 is 1 of the most frequent causes of death in the United States for both men and women.2 Patients who are diagnosed with late-stage (III and IV) American Joint Committee on Cancer (AJCC) nonsmall cell lung cancer (NSCLC) have a 5-year survival rate of only 2.8%.2

Hospice care, which is the standard of care for patients with life-threatening illness, is recommended when it is anticipated that a patient will live for ≤6 months, during which time palliation of symptoms becomes the primary focus rather than curative treatment. It has been demonstrated that palliative care improves symptoms, offers psychological relief, and improves quality of life. To enroll in hospice services, the patient must give up curative treatment in exchange for palliative care. The main objective of hospice is to enable the patient to continue to live an alert, pain-free life with dignity with managed side effects and symptoms.3 There is concern that the variation in receipt of hospice services may suggest that individuals who live in geographic areas with low hospice-Medicare expenditures are not getting the appropriate end-of-life care; conversely, regions with high hospice-Medicare expenditures may be overly compensated for the receipt of hospice services.3 In 2008, more than 1 million patients received hospice services, which cost in excess of $11 billion.3

Racial disparities among patients who receive hospice care have been reported previously by several investigators.4-7 Those studies documented that hospice services are used more frequently among whites, individuals with a higher educational level, and individuals who live in urban communities.4, 5, 7 In addition, the receipt of hospice services by African Americans has declined, whereas the rates for whites have increased over the same time period.4, 6 Hispanics6 and Asians/Pacific Islanders8 also reportedly have used hospice services less frequently compared with whites.

The receipt of hospice differs among urban and rural areas and by Surveillance, Epidemiology, and End Results (SEER) registry area.9, 10 In addition, Virnig et al7 reported that hospice services varied significantly among geographic areas in major cities across the United states. Hospice use reportedly was high in urban areas and lower in rural areas, and there was an 11-fold increase in hospice use in metropolitan areas compared with rural areas. Furthermore, in urban areas with populations >1,000,000, all zip codes were served by a hospice agency, whereas only 2.8% of zip codes in urban areas with populations <1,000,000 were not served by a hospice agency.11 In contrast, in rural areas that were adjacent to urban areas, 91% of zip codes were served by a hospice agency compared with only 76% of zip codes in rural areas that were not adjacent to an urban area.11

It is not clear, however, whether there are racial disparities in the receipt of hospice services for those who live in urban or rural areas adjusting for SES or, conversely, for individuals stratified by the highest to lowest SES quartiles adjusting for living in urban or rural areas. Furthermore, it is important to examine the potential interaction between geographic residence (urban/rural) and SES. Specifically, we aimed to assess the extent to which racial disparities are explained by living in urban or rural areas within each level of SES. We hypothesized that there is variation in the receipt of hospice services among specific racial/ethnic groups within urban and rural areas and that there are racial disparities in the receipt of hospice services within urban and rural areas and within different SES levels.

MATERIALS AND METHODS

Data Sources

It has been demonstrated that Medicare data are valuable as a resource for the assessment of disparities in the receipt of hospice service in cancer patients.4, 8, 12 Medicare is a federally funded program that covers hospice care, including a wide range of palliative care and support services for patients aged ≥65 years who are expected to live ≥6 months.3 The SEER-Medicare-linked data files were used to provide information about hospice services, incident cancer cases, and vital status. SEER registries represent 26.2% of the US population. Alaska was not included because of the very small number of cases reported. Atlanta and rural Georgia also were combined, and all 4 registries in California also were combined because there were too few cases reported in some registries.13 Medicare eligibility could be identified for 94% of the individuals aged ≥65 years who appeared in the SEER records.13

Study Population

The study cohort included 117,894 Medicare patients who had died, including 101,990 whites (86.5%), 10,648 blacks (9%), 3694 Asians/Pacific Islanders (3.1%), and 1562 Hispanics (1.3%) aged ≥66 years who were diagnosed with AJCC stage I, II, III, or IV NSCLC and who died between January 1991 and December 2005. Patients who were categorized with other or unknown race (n = 4222), those who did not have both Parts A and B of Medicare and belonged to a Health Maintenance Organization (n = 2795), and those who were aged 65 years at the time of death (n = 5052) were excluded from the analysis. In addition, in SES analyses, patients who were missing information (n = 7) also were excluded.

Study Variables

The outcome variable was whether or not the patients who died had received hospice care after their diagnosis of NSCLC within 6 months of their death. The exposure variable, race/ethnicity, was designated as whites (non-Hispanic whites), blacks (non-Hispanic African Americans and blacks from other countries), Asians/Pacific Islanders, and Hispanics.

The following patient and tumor characteristics were considered for covariate adjustment in multivariate analyses: age (categorized as ages 66-69 years, 70-74 years, 75-79 years, 80-84 years, or ≥85 years), sex, year of diagnosis (from 1991 to 2005), marital status (married, unmarried [single, divorced, widowed], or unknown), SES (quartiles of poverty), disease stage (AJCC stages I-IV or unstaged), and urban/rural area (categorized from SEER geographic areas, in which urban areas are defined as counties in metropolitan areas with ≥250,000 population, and rural areas are defined as nonmetropolitan counties with ≤20,000 population). Comorbidities were ascertained from Medicare claims by identifying 18 diagnoses or related procedures that were recorded between 1 year before and 1 month after the diagnosis of NSCLC. Details on creating a weighted comorbidity score (0, 1, 2, 3, or ≥4) have been reported previously.14

Statistical Analysis

Analyses were conducted using the Stata software package (version 11.0; Stata Corp., College Station, Tex). Descriptive statistics for each ethnic/racial group with the numbers and percentages in each subgroup for all covariates were performed using Pearson chi-square tests or the Fisher exact test, and P<.05 was used as the cutoff value (Table 1). Age/sex-adjusted rates were computed using the SEER 2005 total population as the standard for whites and ethnic minority groups (Table 2). Disparity rates for the receipt of hospice services among ethnic minority groups compared with whites were calculated, and graphs were constructed to test linear trends over time (from 1991 to 2005) for disparities in the receipt of hospice care within urban and rural areas (Figs. 1, 2). Trend tests were carried out using normalized, log-transformed, age/sex-adjusted disparity rates in linear regression models.

Figure 1. Racial disparities are illustrated (with whites as the reference group) in age and sex standardized rates of receiving hospice care for a nonsmall cell lung cancer diagnosis by year of diagnosis from 1991 to 2005 (rates were standardized to the 2005 US population).

Table 1. Distribution of Patient Characteristics Among Decedents With Nonsmall Cell Lung Cancer by Race/Ethnicitya

Characteristic	Percentage of Patients
	White, n=101,990	Black, n=10,648	Asian/PI, n=3694	Hispanic, n=1562
PI indicates Pacific Islanders; SEER, Surveillance, Epidemiology, and End Results Program of the National Cancer Institute. a P values were calculated using Pearson chi-square tests of hypothesis for independence to assess statistical significance in relation between ethnic groups. All P values were <.0001. b Small patient numbers were reported as <1.1%, because the data user agreement required no reporting of <11 patients.
Median age, y (range, 66-89 y)	75	74	76	76
66-69	18.7	24.8	13.9	13.3
70-74	27.5	30.1	26.9	28
75-79	26.5	23.9	28.5	31.1
80-84	18.5	14.7	20.6	18.8
≥85	8.8	6.5	10.2	8.8
Sex
Men	55.1	58.7	63.7	61.6
Women	44.9	41.3	36.3	38.4
Marital status
Married	51.1	35.1	63.6	45.1
Unmarried	45.4	60.4	34.7	52.1
Unknown	3.5	4.5	1.8	2.8
Disease stage
I	15.7	14.2	13.8	13.9
II	3.1	2.2	2.9	2.9
III	28.4	31.7	31.6	28.9
IV	38.3	39.3	41.5	40.8
Unknown	14.5	12.6	10.3	13.4
Comorbid score
0	38.5	34.9	44.1	38.4
1	33	29.7	31	29.1
2	15.5	16.6	13.9	16.5
3	8.9	11.4	7.6	10.8
≥4	4.2	7.4	3.4	5.2
SEER registry area
South	14.9	26.8	1	1.3
North	40.5	42.5	4.6	12.4
West	44	30.6	75.3	85.2
Pacific	0.6	0.1	19.1	1.2
Geographic residence
Urban	90	95.9	>98b	94.2
Rural	10	4.1	<3b	5.8
Year of lung cancer diagnosis
1991	4.8	4.8	0.4	<1.1b
1992	4.8	5	1	<1.1b
1993	4.8	4.5	1.7	1.6
1994	4.8	4.7	3.3	1.8
1995	4.6	4.7	4.1	3.6
1996	4.5	4.3	5.5	3.8
1997	4.2	4.4	5.3	4.4
1998	4.1	4.2	5.2	6.1
1999	3.9	4.3	5.8	4.7
2000	8.4	8.2	8.4	8.5
2001	8.6	8.5	9	9.2
2002	8.3	8.8	9	10.8
2003	12.2	12	14.9	16.3
2004	11.5	11.2	12.9	14.9
2005	10.5	10.4	13.5	13.3
Socioeconomic status
Highest quartile	27.2	5.1	15.8	6.3
Second quartile	26.9	7.7	20.7	11.2
Third quartile	25.8	16.9	25	23.4
Poorest quartile	19.4	69.8	32.5	57.8
Unknown	0.8	0.6	6.1	1.3

Table 2. Age/Sex-Adjusted Standardized Rates for Decedents With Nonsmall Lung Cancer Who Received Hospice Services by Race/Ethnicity, Year of Diagnosis, and Urban or Rural Residence

Year of Diagnosis	Age/Sex- Adjusted Rates
	Whites	Blacks	Asians/PI	Hispanics
PI indicates Pacific Islanders. ^a Rates were standardized to the 2005 population. b Asian/Pacific Islander decedents were too few to support model analysis for rural area.
Urban residence
1991	0.24	0.19	0.05	0.15
1992	0.30	0.26	0.12	0.03
1993	0.34	0.28	0.20	0.18
1994	0.37	0.34	0.18	0.11
1995	0.39	0.34	0.14	0.32
1996	0.42	0.35	0.27	0.38
1997	0.44	0.37	0.26	0.44
1998	0.47	0.42	0.21	0.44
1999	0.50	0.38	0.28	0.35
2000	0.50	0.40	0.21	0.32
2001	0.50	0.44	0.26	0.30
2002	0.52	0.44	0.24	0.47
2003	0.56	0.46	0.31	0.49
2004	0.57	0.47	0.37	0.48
2005	0.54	0.46	0.32	0.50
Rural residenceb
1991	0.18	0.00	—	0.00
1992	0.17	0.00	—	0.00
1993	0.26	0.00	—	0.38
1994	0.33	0.15	—	0.00
1995	0.38	0.55	—	0.38
1996	0.43	0.35	—	0.38
1997	0.42	0.19	—	0.06
1998	0.53	0.39	—	0.63
1999	0.51	0.50	—	0.00
2000	0.49	0.46	—	0.33
2001	0.46	0.48	—	0.00
2002	0.50	0.53	—	0.33
2003	0.50	0.36	—	0.39
2004	0.50	0.45	—	0.50
2005	0.52	0.35	—	0.25

Mixed-effects logistic regression analyses were performed to clarify the relations between ethnic minority groups compared with whites for the receipt of hospice services. This method takes into consideration that observations within the same cluster are correlated. Stratified analyses were carried out within urban and rural areas (Table 3) and within levels of SES (Table 4). The combination of overall fixed effects (covariates) and variations in trends (random effects) were used to estimate these models. In the models that were stratified by urban or rural area, the clustering variable was the 12 SEER areas, for which the random effects were estimated. Likewise, for the stratified SES models, random effects were assessed for the percentage of residents below the poverty level within the census tract (in deciles). The effects of various factors, such as age, sex, urban/rural area, and SES, on the receipt of hospice services stratified by urban/rural area or within SES levels are presented (Tables 3 and 4). The percentages of relative excess risk for ethnic minorities compared with whites living in urban or rural areas and within each level of SES using mixed-effects logistic regression with 95% confidence intervals (CIs) were calculated and are reported.

Table 3. Multivariate Adjusted Odds Ratios and 95% Confidence Intervals of Receiving Hospice Services in Urban and Rural Areas Among 117,894 Decedents Who Had Nonsmall Cell Lung Cancer, 1991-2005

Characteristic	OR (95% CI) of Receiving Hospice Services
	Urban Residence, n=107,149	Rural Residence, n=10,745
OR indicates odds ratio; CI, confidence interval. ^a The mixed-effects logistic model included hospice care (outcome variable), fixed effects as race/ethnicity, age, sex, socioeconomic status, marital status, tumor stage, comorbid score, and year of nonsmall cell lung cancer diagnosis; and random effects within Surveillance, Epidemiology, and End Results areas. The model was stratified according to urban or rural geographic residence. b Statistically significant OR (P < .05).
Race/ethnicity
White	1.00	1.00
Black	0.79(0.75-0.82)b	0.79(0.63-0.98)b
Asians/Pacific Islander	0.42(0.39-0.46)b	0.65(0.11-3.99)
Hispanic	0.81(0.73-0.91)b	1.17(0.75-1.84)
Age, y
66-69	1.00	1.00
70-74	1.09(1.05-1.13)b	1.07(0.96-1.20)
75-79	1.16(1.11-1.20)b	1.17(1.04-1.32)b
80-84	1.27(1.22-1.33)b	1.15(1.01-1.31)b
≥85	1.26(1.20-1.33)b	0.99(0.84-1.17)
Sex
Men	1.00	1.00
Women	1.34(1.30-1.37)b	1.23(1.13-1.34)b
Socioeconomic status
Highest quartile	1.00	1.00
Second quartile	0.94(0.91-0.98)b	1.49(1.02-2.18)b
Third quartile	0.92(0.88-0.95)b	1.35(0.93-1.96)
Poorest quartile	0.79(0.76-0.83)b	1.15(0.79-1.68)
Unknown	0.79(0.95-1.24)	2.17(1.00-4.72)

Table 4. Effect of Various Factors on the Receipt of Hospice Services Among Decedents Who Had Nonsmall Cell Lung Cancer Stratified by Levels of Socioeconomic Status, 1991-2005a

Characteristic	OR (95% CI) of Receiving Hospice Services
	Highest Quartile, n=29,948	Second Quartile, n=29,179	Third Quartile, n=29,353	Poorest Quartile, n=29,289
OR indicates odds ratio; CI, confidence interval. a The mixed-effects logistic model included hospice care (outcome variable); fixed effects as race/ethnicity, age, sex, urban/rural area, marital status, tumor stage, comorbid score, and year of nonsmall cell lung cancer diagnosis; and random effects within the percentage of the census tract below the poverty level (in deciles). The model was stratified according to quartiles of socioeconomic status. b Statistically significant OR (P < .05).
Race
White	1.00	1.00	1.00	1.00
Black	0.70(0.59-0.84)b	0.77(0.66-0.89)b	0.78(0.70-0.86)b	0.87(0.82-0.93)b
Asian/Pacific Islander	0.42(0.35-0.50)b	0.39(0.34-0.46)b	0.40(0.35-0.46)b	0.30(0.26-0.35)b
Hispanic	0.85(0.57-1.28)	0.87(0.64-1.18)	0.72(0.58-0.89)b	0.76(0.66-0.88)b
Age, y
66-69	1.00	1.00	1.00	1.00
70-74	1.09(1.01-1.17)b	1.07(1.00-1.15)	1.09(1.02-1.17)b	1.09(1.02-1.16)b
75-79	1.14(1.06-1.23)b	1.13(1.06-1.22)b	1.15(1.07-1.23)b	1.14(1.06-1.22)b
80-84	1.28(1.18-1.39)b	1.24(1.14-1.34)b	1.26(1.16-1.36)b	1.17(1.08-1.26)b
≥85	1.26(1.15-1.40)b	1.23(1.12-1.36)b	1.17(1.07-1.30)b	1.14(1.03-1.26)b
Sex
Men	1.00	1.00	1.00	1.00
Women	1.32(1.24-1.37)b	1.30(1.23-1.36)b	1.31(1.24-1.38)b	1.34(1.27-1.41)b
Geographic residence
Urban	1.00	1.00	1.00	1.00
Rural	0.68(0.48-0.98)	1.17(1.05-1.30)b	0.93(0.87-1.00)b	0.86(0.81-0.92)b
Year of lung cancer diagnosis
1991-1995	1.00	1.00	1.00	1.00
1996-1999	1.56(1.44-1.68)b	1.68(1.54-1.82)b	1.81(1.66-1.97)b	1.89(1.79-2.13)b
2000-2005	1.96(1.84-2.09)b	2.23(2.08-2.38)b	2.44(2.29-2.61)b	2.38(2.23-2.55)b

A multilevel analysis also was performed to determine whether random effects were shared among SEER areas, with the percentage of residents below the poverty level nested within SEER areas, and whether there was any impact in the receipt of hospice care among ethnic minorities compared with whites. Interaction between geographic residence (urban/rural) and SES for the receipt of hospice services was assessed to determine whether SES modified the effect of living in urban/rural areas. This analysis was performed using logistic regression in which the likelihood-ratio test with a P value cutoff <.05 was used to evaluate statistical significance. Interaction also was assessed with mixed-effects logistic regression to determine whether there were differences in hospice use between urban/rural areas and SES levels. A statistically significant covariance using the 95% CI for the interaction term was used to evaluate this correlation.

RESULTS

Characteristics of the Study Population

Most of the elderly patients resided in urban areas, were men, and had advanced-stage NSCLC (stages III and IV), as indicated in Table 1. A higher proportion of blacks were diagnosed at younger ages (ages 66-74 years). In addition, more blacks and Hispanics were unmarried at the time of diagnosis, had comorbidity scores ≥2, and were clustered in the poorest SES quartile.

Age/Sex-Adjusted Rates for Receiving Hospice Care

Table 2 presents the age/sex-adjusted rates for receiving hospice care. In urban areas, blacks, Asians/Pacific Islanders, and Hispanics appeared to have lower rates of receiving hospice care compared with whites, whereas Asians/Pacific Islanders seemed to have the lowest rates. We noted that hospice care increased for all racial groups over time from 1991 to 2005; however, by 2005, all ethnic minority groups appeared to receive less hospice care compared with whites. Furthermore, Asians/Pacific Islanders were the least likely to receive hospice care among all ethnic minority groups. In recent years, Hispanics (2002 to 2005) appeared to be more likely to receive hospice care than blacks and Asians/Pacific Islanders.

Disparity in the receipt of hospice services in urban areas among the various ethnic minority groups compared with whites is depicted in Figure 1. The receipt of hospice services among blacks from 1991 to 1998 was relatively stable; there was an increase in 1999, then a decrease in 2000, and stabilization at a slightly higher level by 2005 (P = .003). Asians/Pacific Islanders had the lowest receipt of hospice services, and the trend was statistically significant (P = .036); therefore, it appears that the gap in the receipt of hospice services was not narrowing over time. From 1991 to 1994, Hispanics had a noticeable disparity in the receipt of hospice services. The disparity decreased in 1995, increased by 2000, and then decreased in 2001. By 2005, there was only a slight increase in the disparity for Hispanics, and it was not statistically significant (P = .29).

Similarly, in rural areas, blacks and Hispanics in general appeared to receive less hospice care compared with whites (Fig. 2). During 1991 to 1994, there was a disparity in the receipt of hospice services among blacks; however, by 1995, blacks were receiving slightly more hospice services than whites. However, the disparity in hospice services for blacks increased in 1996, decreased again by 1999, but still was apparent in 2005. Nevertheless, the disparity trend was not statistically significant (P = .208). Hispanics had an upward/downward trend in the receipt of hospice services in rural areas. In 1993 and 1998, Hispanics received more hospice care than whites; and, in 2004, there was no disparity in the receipt of hospice care. However in 2005, the disparity rate increased. The overall disparity trend was not stable over time and was not statistically significant for Hispanics (P = .290). Asian/Pacific Islander decedents were too few in number in rural areas to be included in this analysis.

Figure 2. Racial disparities are illustrated (with whites as the reference group) in age and sex standardized rates of receiving hospice care for a nonsmall cell lung cancer diagnosis in rural areas by year of diagnosis from 1991 to 2005 (rates were standardized to the 2005 US population).

Multivariate Adjusted Odds Ratios for Receiving Hospice Care

Table 3 lists the adjusted odds ratios (ORs) for race/ethnicity associated with the receipt of hospice services. The estimates stratified by urban areas were similar to those for both urban and rural areas combined after adjusting for risk factors and confounders. In urban areas, blacks were 21% less likely to receive hospice services (OR, 0.79; 95% CI, 0.75-0.82), Asians were 58% less likely to receive hospice services (OR, 0.42; 95% CI, 0.39-0.46), and Hispanics were 19% less likely to receive hospice services (OR, 0.81; 95% CI, 0.73-0.91). In rural areas, blacks were 21% less likely less likely to receive hospice services (OR, 0.79; 95% CI, 0.63-0.98). Overall, younger patients, men, and individuals in the poorest SES quartile were the least likely to receive hospice services.

We investigated the receipt of hospice care within SEER areas in the Northern states (Iowa, Michigan [Detroit], Connecticut, and New Jersey), the Western states (New Mexico, California, Utah, and Washington [Seattle]), and the Pacific (Hawaii) compared with the Southern states (Kentucky, Louisiana, and Georgia). Because our data indicated that the Southern states had the highest mean and median length of stay in hospice (49.2 days and 19 days, respectively), this region was used as the referent group. SEER areas in the Northern region were 10% less likely to receive hospice services (OR, 0.90; 95% CI, 0.87-0.94) than SEER areas in the Southern region, and SEER areas in the Western region were 18% less likely to receive hospice services (OR, 0.82; 95% CI, 0.79-0.86) than SEER areas in the Southern region.

SES was associated with the receipt of hospice services (Table 4). Patients in the poorest SES quartile were 7% less likely to receive hospice care (OR, 0.93; 95% CI, 0.86-1.00). In further analyses that were stratified by quartiles of SES, black patients and Asian/Pacific Islander patients in all 4 quartiles of SES had lower receipt of hospice services, and Asian/Pacific Islander patients had the lowest receipt of hospice services. However, the gap was greater between Asians/Pacific Islanders and whites. Asians/Pacific Islanders in the poorest SES quartile were 70% less likely to receive hospice services (OR, 0.30; 95% CI, 0.26-0.35), whereas Asians/Pacific Islanders in the highest SES quartile were 58% less likely to receive hospice services (OR, 0.42; 95% CI, 0.35-0.50). The gap between blacks and whites was relatively smaller than that for Asians/Pacific Islanders. Blacks in the highest SES quartile were 30% less likely to receive hospice services (OR, 0.70; 95% CI, 0.59-0.84), whereas blacks in the poorest SES quartile were 13% less likely to receive hospice services (OR, 0.87; 95% CI, 0.82-0.93).

We also estimated the excess risks for ethnic minorities compared with whites for the under use of hospice services. The percentage of excess risk assessed for blacks who lived in urban areas was 109% (OR, 1.09; 95% CI, 0.95-1.25) compared with whites who lived in the same areas. Patients who lived in urban areas at the poorest SES level were 125% more likely to under use hospice care (OR, 1.25; 95% CI, 1.01-1.54) compared with patients who lived in urban areas at the highest SES level.

By using the highest SES level as the background risk quartile, the percentage excess risk assessed for blacks who lived in urban areas at the second SES quartile was 103% (OR, 1.03; 95% CI, 0.87-1.23) compared with whites within the same SES quartile. The percentage excess risk for Asians/Pacific Islanders was spread across SES quartiles. The increased excess risk for under use of hospice services among Asians/Pacific Islanders was 96% in the second SES quartile (OR, 0.96; 95% CI, 0.87-1.07), 92% in the third SES quartile (OR, 0.92; 95% CI, 0.83-1.02), and 87% at the poorest SES level (OR, 0.87; 95% CI, 0.79-0.96) compared with whites in corresponding SES quartiles. It is important to note that, when there is no biologic interaction, the relative excess risk and the attributable proportion are equal to zero, and the synergy index is equal to 1.15

In a multilevel analysis, random effects for the percentage of residents below the poverty level in the census tract division nested within SEER areas had an impact on the receipt of hospice services for ethnic minorities compared with whites. Random effects for the “SEER area” variable and the “percentage of residents below the poverty level in the census tract (in deciles)” variable appeared to be statistically significant judging from the 95% CIs. However, there were small increases in the effect estimates, which were similar to those for the urban model that included only SEER area as the random-effect variable.

Another objective of this study was to determine whether SES interacts with the effect of living in urban/rural areas with regard to the receipt of hospice services among ethnic/racial groups. The fixed-effects estimates in the interaction model did not change appreciably compared with the model that did not include this interaction term. The covariances for the random-effects interaction terms between urban/rural areas and SES were not statistically significant. We also assessed for the possibility of interaction using a logit model to determine whether SES modified the effect of living in urban/rural areas and, consequently, had an impact on the receipt of hospice services among ethnic minorities compared with whites. The P value for the interaction was significant (P<.0001) in the logit model, but there was no considerable change in the effect estimates.

DISCUSSION

In this study, black patients, Asian/Pacific Islander patients, and Hispanic patients with NSCLC, compared with white patients, were less likely to receive hospice services within the last 6 months of life. Hospice services were received more commonly by patients in advanced age groups and by women. In urban areas, disparities in age/sex-adjusted rates for blacks and Asians/Pacific Islanders were stable from 1991 to 2005. Overall, patients in the poorest SES quartile were less likely to receive hospice services. Furthermore, within stratified SES quartiles, blacks and Asians/Pacific Islanders were less likely to receive hospice care, and Asians/Pacific Islanders in the poorest quartile were the least likely to receive these services. SEER areas in the Northern and Western regions of the United States had decreased use of hospice services compared with SEER areas in the Southern region.

Several studies have noted disparities in the receipt of hospice services among ethnic minority groups,4, 6-11 individuals who live in urban areas, and individuals with fewer years of education.5 Kessler et al,16 in a British cross-sectional study, observed that individuals in the lowest social class (unskilled occupations) were less likely to die in a hospice compared with individuals in Social Class I, who were engaged in professional occupations (OR, 0.49; 95% CI, 0.22-1.09); however, the effect estimate was not statistically significant in that study. Higginson et al,17 in another British study, noted that the proportion of cancer decedents in the most deprived group had less home deaths (24%) compared with those in the medium deprived group (27%) and the least deprived group (30%). Our study extended these results and demonstrated that ethnic minority groups in urban and rural areas were less likely to receive hospice services compared with whites. Overall, less receipt of hospice services was noted among patients in the poorest SES quartile. More specifically, blacks and Asians/Pacific Islanders were less likely to receive hospice services regardless of their SES level. However, Asians/Pacific Islanders in the poorest quartile were the least likely to receive hospice services. The percentage of excess risk that contributed to the disparities was 109% and 103% for blacks in urban areas and at the second SES level, respectively; and it was 87% and 96% for Asians/Pacific Islanders at the poorest SES level and the second SES level, respectively. The excess risk was greatest (125%) for individuals who lived in urban areas at the poorest SES level. Our results are in concordance with the results reported by Conner et al,18 who demonstrated that hospice rates were highest in the Southern region of the United States and that the highest rates were in Florida.

The current study was conducted during a period when there was relatively minor change in the treatment of NSCLC. However, over the last 5 years, major changes have been brought about by the development of targeted therapies with much lower toxicities and side effects for the treatment of NSCLC as well as other malignancies. Patients willingly may forgo hospice care to pursue these well tolerated treatment regimens until the time of their death.

Several studies have reported that blacks and other ethnic/minority groups have decreased access to healthcare compared with whites.19-21 Another report indicated that patients with low income who received their care in urban medical centers were less likely to receive adequate support.22 The reason for these disparities may be multifactorial. In patients with lower SES, it is possible that limited resources require family members to continue working, and the physical characteristics of home and neighborhood (such as long distance from a hospice agency) may make it more difficult to deliver end-of-life care at home. Other barriers to receiving hospice care at home among ethnic minorities include having cultural beliefs about death and dying that conflict with the philosophy of hospice care, misconstrued ideas about hospice care, difficulty accepting end-of-life care, guilt feelings about burdening the family with their care, mistrust in the healthcare system, insufficient representation of ethnic groups within hospice agencies, and inadequate, culturally based physician training among ethnic minority groups.22-26 It is believed that provider-level stereotypes and bias, such as prejudice, discrimination, racial profiling, therapeutic decisions based on a patient's skin color and/or ethnicity, or the horror of being part of an unethical experiment, are fears that may play a role in influencing health outcomes for ethnic minority patients.23, 25-27

One of the strengths of this study was the large, retrospective population based on 17 SEER areas, which account for >26% of the US population and allowed us to present results that are easily generalized. We have demonstrated that ethnic minority patients with NSCLC who live in urban areas as well as blacks who live in rural areas, in addition to ethnic minority patients from the highest SES level to the poorest SES level, receive less hospice services compared with whites.

There are several limitations to this study. First, because our population was limited to Medicare beneficiaries aged ≥66 years, the findings may not be generalizable to younger patients. Second, there was no information on patient preferences or refusals despite physicians' recommendations or other related issues that may have played a role in decision-making. For example, it has been demonstrated that blacks who are terminally ill may continue treatment that can be more costly compared with whites who receive hospice services more frequently.28 This may represent cultural beliefs that conflict with the philosophy of hospice care. Certain variables that have been associated with hospice services in the literature were not included in our current analyses because they were unavailable. For example, in 1 study managed-care patients were more likely to receipt hospice services than Medicare fee-for-service patients (OR, 0.88; 95% CI, 0.82-0.93).12 Other factors that can influence the receipt of hospice services, such as having a caregiver, referral source, and payment method4, 6; distance from the home to the hospital29 with a palliative care program; and whether the hospital is a nonprofit institution or a for-profit or public hospital,30 were not available for this analysis.

A concern in this study was the small sample size in rural areas among ethnic minority groups. Table 2 indicates that, in rural areas, the ORs for Asians/Pacific Islanders and Hispanics were 0.65 and 1.17, respectively, but the 95% CIs were >1.0. The sample sizes for these ethnic groups were not sufficiently large; therefore, these estimates may reflect type 2 errors, because there was not enough power to detect statistically significant effect estimates. Hence, larger sample sizes for ethnic minorities in rural areas are needed to resolve the question of the differential association in the receipt of hospice services.

In conclusion, there were substantial disparities with respect to the receipt of hospice services by ethnic minority patients with NSCLC who lived in urban and rural areas and within SES levels. Research efforts are needed to investigate the regional variation among patients with different ethnic/cultural backgrounds and among those within SES levels. Efforts are required to identify barriers and enhance support for patients and their families and to provide education on the benefits of hospice services for all patients at the end of life. The impact of targeted therapies on access to hospice among patients with lung cancer will need to be considered as an important covariate in future research in these populations.

Acknowledgements

We acknowledge the creation of this database from the National Cancer Institute; Centers for Medicare and Medicaid Services; Information Management Services; and Surveillance, Epidemiology, and End Results tumor registries. The analyses, interpretation, and reporting are the sole responsibilities of the authors.

CONFLICT OF INTEREST DISCLOSURES

This study was supported by grant from the Agency for Healthcare Research and Quality (RO1-HS016743).