DOI:10.1093/jncimonographs/lgu020
© The Author 2014. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: [email protected].
Racial and Ethnic Disparities in Cancer Survival by Neighborhood Socioeconomic Status in Surveillance, Epidemiology, and End Results (SEER) Registries Jonathan K. Kish, Mandi Yu, Antoinette Percy-Laurry, Sean F. Altekruse Correspondence to: Sean Altekruse, Surveillance Research Program, Division of Cancer Control and Population Sciences, National Cancer Institute, 9609 Medical Center Drive Room 4E536, Rockville, MD 20850 (e-mail: [email protected])
Reducing cancer disparities is a major public health objective. Disparities often are discussed in terms of either race and ethnicity or socioeconomic status (SES), without examining interactions between these variables.
Methods
Surveillance, Epidemiology, and End Results (SEER)-18 data, excluding Alaska Native and Louisiana registries, from 2002 to 2008, were used to estimate five-year, cause-specific survival by race/ethnicity and census tract SES. Differences in survival between groups were used to assess absolute disparities. Hazard ratios were examined as a measure of relative disparity. Interactions between race/ethnicity and neighborhood SES were evaluated using proportional hazard models.
Results
Survival increased with higher SES for all racial/ethnic groups and generally was higher among non-Hispanic white and Asian/Pacific Islander (API) than non-Hispanic black and Hispanic cases. Absolute disparity in breast cancer survival among non-Hispanic black vs non-Hispanic white cases was slightly larger in low-SES areas than in highSES areas (7.1% and 6.8%, respectively). In contrast, after adjusting for stage, age, and treatment, risk of mortality among non-Hispanic black cases compared with non-Hispanic white cases was 21% higher in low-SES areas and 64% higher in high-SES areas. Similarly, patterns of absolute and relative disparity compared with non-Hispanic whites differed by SES for Hispanic breast cancer, non-Hispanic black colorectal cancer, and prostate cancer cases. Statistically significant interactions existed between race/ethnicity and SES for colorectal and female breast cancers.
Discussion
In health disparities research, both relative and absolute measures provide context. A better understanding of the interactions between race/ethnicity and SES may be useful in directing screening and treatment resources toward at-risk populations.
J Natl Cancer Inst Monogr 2014;49:236–243
The National Cancer Institute (NCI) defines cancer disparities as adverse differences in cancer outcomes between population groups (1). Each year, summary cancer survival statistics are updated for the major racial and ethnic groups (2,3). Research has focused largely on racial and ethnic inequalities. Less information is available on the role of socioeconomic status (SES) in cancer outcomes such as survival (4–6). Racial/ethnic disparities may be explained by SES (7,8), access to affordable health care coverage (9), or biological differences (10). If racial disparities are driven by inequalities (11) such as adverse social experiences and economic deprivation (12,13), it is possible that race/ethnicity and SES have an interactive effect on survival (8,14,15). Minorities and the poor often experience worse health outcomes than non-Hispanic whites and more affluent individuals, respectively (16). Because cancer registries do not routinely collect data on SES, only a few studies of cancer survival in the United States have investigated interactions between income and education (17,18). To improve the value of its data, NCI’s Surveillance, Epidemiology, and End Results (SEER) Program recently developed a variable based on a census-tract level SES index (19). 236
Although most research on health disparities has focused on relative measures, such as hazard and rate ratios, absolute measures provide an intuitive metric, giving context to relative disparity measures (20). In this study, we used the census-tract level SES index to estimate absolute and relative survival disparities in lung, liver, kidney, colorectal, breast, and prostate cancer burdens. As an absolute measure of disparity, we examined the difference in five-year, cause-specific survival. As a relative measure of disparity, we present hazard ratio measures that are adjusted for race/ethnicity, SES, age, stage, and treatment.
Methods Data Sources and Study Population Disparities in five-year, cause-specific survival were estimated by race/ethnicity and SES. The study cohort included microscopically confirmed malignant cancers of the lung/bronchus (hereafter “lung”), liver/intrahepatic bile duct (hereafter “liver”), kidney/renal pelvis (hereafter “kidney”), colon and rectum (hereafter “colorectal”), prostate, and female breast (hereafter “breast”) diagnosed from January 1, 2002, through December 31, 2008, and followed through December 31, 2009. Journal of the National Cancer Institute Monographs, No. 49, 2014
Downloaded from http://jncimono.oxfordjournals.org/ at New York Medical College on March 26, 2015
Introduction
Sociodemographics Race/ethnicity was classified as non-Hispanic white, non-Hispanic black, non-Hispanic Asian or Pacific Islander (API), AmericanIndian or Alaska Native (AI/AN), and Hispanic. Age at diagnosis was categorized as greater than 45 years of age, in 10-year intervals from ages 45–85, and age 85 years or older. To improve the quality of area-based SES estimates, our analysis is based on a new census-tract level linkage created by SEER and an index variable of SES status in 16 registries, after excluding the Alaska Native and Louisiana registries, where census tract SES data were not available. Cases were mapped to their year 2000 census tract of residence based on their address at time of diagnosis. Validated indices of socioeconomic position were developed using factor analysis from census tract-level estimates from the 2000 census, and from five-year estimates from the 2005–2009 American Community Survey (ACS) (24). Both indices were extracted from the same seven SES measures identified by Yost et al. (25). to represent three components of SES (occupation, education, and income). The variables included: proportion with a blue-collar job, proportion older than age 16 in the workforce without a job, education index, median household income, proportion below 200% of the poverty level, median rent, and median house value. We assigned an index value to each census tract and categorized them into quintiles. The first quintile (Q1, lowest neighborhood SES) corresponded to the 20th percentile or less of SES, and the fifth quintile (Q5, highest neighborhood SES) corresponded to the 80th percentile or higher. For cases diagnosed in 2002–2003, the 2000 census index was used; for cases diagnosed in 2004–2008, the 2005–2009 ACS index was used. (This research resource is accessible by submitting a suitable concept to the NCI SEER Program, care if the corresponding author or M. Yu.) Cases with unknown race/ethnicity or addresses that could not be mapped to a census tract were excluded from the analysis. Cancer Staging and Treatment Cancer survival is strongly associated with stage at diagnosis and treatment. Staging was based on SEER Summary Stage 2000 (1998+). Summary staging, although not of clinical utility, was used as a measure of population-based cancer control efforts. Stage was
categorized as localized, regional, distant, or unknown/unstaged. For prostate cancer, localized and regional stages were combined according to American Joint Committee on Cancer staging procedures (26). First-course cancer-directed surgery and radiation therapy were determined from the SEER variables “cancer-directed surgery” and “radiation sequence with surgery.” Statistical Analysis Cancer survival was studied using both absolute and relative disparity measures (27). To assess disparities on the absolute scale, a survival rate difference was calculated based on Kaplan–Meier estimates of five-year, cause-specific survival by race/ethnicity and SES using SEER*Stat software (28). Because suitable life tables were not available for the SES subgroups, cause-specific death classification was used (29), with other causes of death censored. Pairwise comparisons were considered statistically significant when the Z-test statistic had a P value of less than .05. Sensitivity analyses were used to compare age-standardized and unadjusted survival. Relative disparities were assessed by calculating hazard ratios for various race/ethnicity and SES comparisons, as well as their interactions, for each cancer site. Stratified Cox proportional hazard models were developed with SAS v.9.3 software (30). Hazard ratios reflected relative risk of death in the 5 years after diagnosis adjusted for confounding factors. Each site-specific model was stratified by age at diagnosis and tumor stage after inspection of the [log, −log] plot of the survival function revealed signs of non-proportionality, as recommended by Breslow (31). Models also were adjusted for surgery and radiation treatment. Cases with missing or unknown treatment status were coded as having no reported treatment. Wald χ2 tests were used to assess the statistical significance of the interaction between SES and race/ethnicity. Comparisons of survival are presented by SES and racial/ethnic group. For pairwise comparisons, confidence intervals (CIs) that did not include 1.0 were considered statistically significant.
Results Table 1 presents the demographic and clinical distribution of cases by cancer site. Total numbers of cases ranged from a low of 32 571 for liver cancer to a high of 357 078 for prostate cancer. Excluding prostate cancer, the site with the highest percentage in men was liver cancer (71.7%), and the site with the lowest percentage in men was colorectal cancer (50.9%). Excluding breast cancer, the site with the highest percentage in women was colorectal cancer (49.1%) and the lowest percentage in women was liver cancer (28.3%). Most lung (64%), colorectal (58%), and prostate cancer (64%) cases were diagnosed among individuals aged 65 years and older; and most liver (55.6%), kidney (55.5%), and breast (62.2%) cancer cases occurred in individuals aged less than 65 years. Stage at diagnosis varied by site: the majority of breast and kidney cancers were localized (60.3% and 62.3%, respectively), whereas 56.4% of lung cancer cases were diagnosed at distant stages. Overall, most cancer cases occurred in non-Hispanic whites, followed by non-Hispanic blacks. With the exception of prostate cancer cases (at 2.8%), less than 1% of all cancer cases were excluded due to unknown race. A disproportionately high percentage of liver
Journal of the National Cancer Institute Monographs, No. 49, 2014 237
Downloaded from http://jncimono.oxfordjournals.org/ at New York Medical College on March 26, 2015
Screening and treatment protocols and cancer survival rates differed for these cancer sites during the study period. For lung cancer, there were no population-based screening recommendations, the cancers typically were diagnosed in late stages, and survival rates were low (21). Similarly, although five-year relative survival rates for liver cancer tripled from less than 5% in the 1970s to 15% in 2003, this cancer often was diagnosed at regional or distant stages (2). In contrast, more than 60% of kidney cancer cases were diagnosed at a localized stage, and survival increased from 50% to 75% during 1970–2003 (2). For colorectal cancer, there were well-established screening protocols, decreasing trends in incidence, and increasing trends in survival. The five-year survival estimates for breast cancer were above 85%, and nearly 72.4% of women were screened yearly after age 40 per recommended guidelines (22). Finally, for prostate cancer, the five-year survival rate was above 90% and, despite changes in screening recommendations, approximately 45% of men over age 75 were screened in 2008 (23).
Table 1. Frequency of cases by cancer site and proportion by demographic characteristics* Cancer site Lung
Liver
Kidney 62 371
CRC
Prostate Breast†
213 274 357 078
313 482
61.4 38.7
50.9 49.1
100 —-
— 100
10.8 18.4 26.3 23.2 16.6 4.7
5.8 14.3 21.2 24.3 24.0 10.4
0.7 9.6 31.2 35.7 19.1 3.8
13.6 24.0 24.6 19.0 14.1 4.7
62.3 16.8 17.1 3.7
39.4 36.2 19.7 4.7
92.8‡ —‡ 4.1 3.1
60.3 32.7 5.1 1.9
82.0
85.8
38.7
92.8
3.3
10.4
2.9
47.7
71.5 10.5 4.5 0.8 12.1 0.6
71.6 11.3 7.0 0.5 8.9 0.7
70.5 13.6 4.4 0.3 8.3 2.8
72.7 10.1 6.8 0.5 9.4 0.6
14.7 16.6 17.6 17.9 18.0 15.3
14.9 16.8 17.6 18.1 17.9 14.6
12.5 15.0 17.0 18.7 22.6 14.2
12.2 15.1 17.3 19.5 22.7 13.3
* AI/AN = American Indian/Alaska Native; API = Asian/Pacific Islander; CRC = colorectal cancer; HISP = Hispanic; NHB = non-Hispanic blacks; NHW = non-Hispanic whites; Q = quintile; SES = socioeconomic status. † Breast cancer restricted to women. ‡ Localized, regional combined.
cancer cases occurred among API and Hispanic individuals (17.6% and 17.4%, respectively). The frequency of lung and liver cancer cases decreased from Q1 (low SES) census tracts to Q5 (high SES) census tracts. This pattern was reversed for kidney, colorectal, prostate, and breast cancers. SES data were missing for between 11.7% (liver) and 16.7% (lung) of cases for each site. These cases were excluded. Five-year, cause-specific survival is presented for all six cancer sites by race/ethnicity and SES in Table 2 and Figure 2. Overall survival was lowest for lung and liver cancers (18.3% and 19.3%, respectively), intermediate for colorectal and kidney cancers (64.7% and 72.6%, respectively), and highest for breast and prostate cancers (87.9% and 93.7%, respectively). Survival increased from lowto high-SES groups, with the smallest difference between Q1 and Q5 occurring for prostate cancer (−4.8%) and the largest difference between Q1 and Q5 occurring for breast (−9.8%) and liver cancers (−10.4%). Among racial/ethnic groups, APIs experienced the best five-year, cause-specific survival for all SES strata for each cancer 238
Discussion In this cancer survival study, five-year, cause-specific survival increased with higher SES; and non-Hispanic white and API cases often experienced better cancer survival than non-Hispanic black, Hispanic, and AI/AN cases. These absolute disparities are Journal of the National Cancer Institute Monographs, No. 49, 2014
Downloaded from http://jncimono.oxfordjournals.org/ at New York Medical College on March 26, 2015
N 266 439 32 571 Gender, % Men 53.8 71.7 Women 46.2 28.3 Age at diagnosis, % 0–44 2.3 5.9 45–54 10.3 21.5 55–64 23.2 28.2 65–74 31.4 22.4 75–84 26.0 17.0 85+ 6.9 5.0 Stage, % Localized 15.1 40.3 Regional 22.0 26.3 Distant 56.4 18.5 Unstaged 6.5 14.8 Cancer-directed surgery, % Yes 20.9 24.0 Radiation therapy, % Yes 7.6 1.3 Race/ethnicity,% NHW 78.2 51.8 NHB 10.9 11.8 API 5.1 17.6 AI/AN 0.4 1.1 HISP 5.2 17.4 Unknown 0.2 0.3 SES, % Q1 (low) 17.4 19.2 Q2 17.6 18.3 Q3 17.4 18.1 Q4 16.5 17.8 Q5 (high) 14.3 15.0 Missing 16.7 11.7
site except kidney, for which non-Hispanic blacks had the best survival. Non-Hispanic blacks had the lowest survival rates for liver (15.1%), colorectal (58.2%), and breast (78.7%) cancers. Figure 1 depicts race/ethnicity- and SES-stratified survival with 95% CIs. Sensitivity analyses, with age-standardized survival, showed that variation from these results was small, at between 1% and 3% (data not shown) . Absolute and relative cancer survival disparities by SES within racial/ethnic groups are presented in the last two columns of Table 2. For lung cancer, the absolute difference in survival between Q1 and Q5 within racial/ethnic groups ranged from −15.4 (American Indian/Alaska Native [AI/AN]) to −5.3 (non-Hispanic blacks). The relative disparity comparison, based on stratified Cox proportional hazards models, shows a 24% increase in mortality risk for Q1 compared with Q5 for non-Hispanic whites and a 15% increase for non-Hispanic blacks. Except for lung and prostate cancer among AI/AN cases and kidney cancer among all minority groups, significant SES absolute disparities within racial/ethnic groups aligned with significant relative disparities. Absolute and relative disparities between racial/ethnic groups by SES are presented in Table 3. Compared with non-Hispanic whites, the absolute disparity in breast cancer survival among nonHispanic blacks was larger in Q1 than Q5, with −7.1% and −6.8% lower five-year survival, respectively. In contrast, the relative disparity in breast cancer survival between non-Hispanic black cases and non-Hispanic white cases, based on adjusted hazard ratios, was larger in Q5 than in Q1 (64% higher risk compared with 31% higher risk, respectively). Similar patterns were seen among nonHispanic black and non-Hispanic white colorectal and prostate cancer cases. For lung, liver, and kidney cancers, only three of 18 absolute disparity comparisons (17%) in Table 3 were statistically significant. In contrast, 50% of absolute disparity comparisons for breast, colorectal, and prostate cancers were statistically significant. When Hispanic and non-Hispanic white cases were compared, breast, colorectal, and prostate cancer absolute disparities were greater than 1 for Hispanics in Q1 (better survival than non-Hispanic whites) and less than 1 in Q5 (worse survival than non-Hispanic whites). Hispanic cases experienced an increased risk of death compared with non-Hispanic whites in the highest SES category for breast cancer (hazard ratio of 1.21, 95% CI: 1.08 to 1.36) and prostate cancer (hazard ratio of 1.20, 95% CI: 1.01 to 1.42). Compared with non-Hispanic white cases, the risk of death for APIs was lower in both Q1 and Q5 for breast and prostate cancers and for colorectal cancer in Q5. For colorectal and breast cancers, two of the three leading cancers with robust screening and treatment programs, there was a significant interaction between race/ethnicity and SES in the model (see Table 3). Supplementary Table 1 provides standard errors for Q1 and Q5 survival estimates and CIs for the hazard ratios in Table 3.
Table 2. Five-year, cause-specific survival by race/ethnicity and census tract socioeconomic status (SES), absolute (rate difference) and relative (hazard ratio) disparity, low (Q1) vs high (Q5) SES strata* Five-year, cause-specific survival, % Race/ethnicity
Q2
Q3
Q4
Q5 high
AD rate difference Q1–Q5
RD† HR Q1 vs Q5
18.3 18.6 15.7 19.4 11.8 17.2
15.0 15.3 14.2 15.2 11.3 14.8
17.0 17.2 16.0 17.5 11.0 15.2
11.3 18.2 18.6 19.0 11.3 18.2
19.5 19.6 16.3 20.8 9.0 19.7
22.3 22.4 19.5 22.0 26.7 22.3
−7.3 −7.1‡ −5.3‡ −6.8‡ −15.4‡ −7.5‡
1.24‡ 1.15‡ 1.20‡ 1.19 1.22‡
19.3 18.2 15.1 24.4 16.1 19.3
14.6 13.7 11.5 17.0 14.3 16.6
16.2 14.8 14.2 18.8 14.1 18.1
19.3 18.1 19.8 22.5 16.9 19.9
22.5 21.4 17.6 25.7 19.4 23.0
25.0 20.9 29.0 33.8 28.6 25.8
−10.4 −7.2‡ −17.5‡ −16.8‡ −14.3 −9.2‡
1.32‡ 1.29‡ 1.32‡ 1.28 1.36‡
72.6 72.5 74.1 72.2 68.5 72.6
69.6 68.5 71.6 65.9 69.3 70.9
71.0 70.1 73.1 75.2 75.5 72.2
72.3 72.0 76.5 72.4 62.0 72.9
72.7 72.7 78.3 69.1 66.1 72.5
76.6 76.6 77.0 75.3 51.9 78.1
−7.0 −8.1‡ −5.4‡ −9.4‡ 17.4 −7.2‡
1.23‡ 1.00 1.26 0.88 1.13
64.7 65.1 58.2 69.3 58.9 63.9
60.5 61.4 56.4 60.9 50.7 62.8
62.4 62.7 57.0 67.8 59.5 62.4
65.0 64.9 60.8 68.6 60.7 66.2
66.0 66.1 60.1 69.3 74.2 63.1
68.5 68.2 63.7 74.0 57.9 66.5
−8.0 −6.8‡ −7.3‡ −13.1‡ −7.2 −3.7‡
1.28‡ 1.17‡ 1.36‡ 1.26 1.18‡
87.9 88.8 78.7 91.0 83.6 86.5
81.6 82.6 75.5 85.5 80.9 84.5
85.9 86.5 78.1 89.6 78.5 85.7
87.9 88.1 81.1 89.8 85.6 88.2
89.4 89.6 82.2 92.3 92.6 88.2
91.4 91.6 84.8 93.1 87.0 89.5
−9.8 −9.0‡ −9.3‡ −7.6‡ −6.1 −5.0‡
1.75‡ 1.40‡ 1.34‡ 1.62 1.44‡
93.7 93.9 92.5 94.3 89.6 93.0
90.9 91.1 90.4 91.6 81.1 91.6
92.4 92.2 92.8 93.4 93.9 92.5
93.5 93.6 93.5 92.9 90.7 93.2
94.3 94.2 94.6 94.9 92.0 94.4
95.7 95.7 95.2 95.9 97.4 94.8
−4.8 −4.6‡ −4.8‡ −4.3‡ −16.3‡ −3.2‡
1.56‡ 1.31‡ 1.28 1.80 1.13
* AD = absolute disparity; AI/AN = American Indian/Alaska Native; API = Asian/Pacific Islander; CRC = colorectal cancer; HISP = Hispanic; HR = hazard ratio; NHB = non-Hispanic blacks; NHW = non-Hispanic whites; Q = quintile; RD = relative disparity; SES = socioeconomic status. † Stratified by stage and age with adjustment for sex, cancer-directed surgery, radiation therapy. ‡ P
© The Author 2014. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: [email protected].
Racial and Ethnic Disparities in Cancer Survival by Neighborhood Socioeconomic Status in Surveillance, Epidemiology, and End Results (SEER) Registries Jonathan K. Kish, Mandi Yu, Antoinette Percy-Laurry, Sean F. Altekruse Correspondence to: Sean Altekruse, Surveillance Research Program, Division of Cancer Control and Population Sciences, National Cancer Institute, 9609 Medical Center Drive Room 4E536, Rockville, MD 20850 (e-mail: [email protected])
Reducing cancer disparities is a major public health objective. Disparities often are discussed in terms of either race and ethnicity or socioeconomic status (SES), without examining interactions between these variables.
Methods
Surveillance, Epidemiology, and End Results (SEER)-18 data, excluding Alaska Native and Louisiana registries, from 2002 to 2008, were used to estimate five-year, cause-specific survival by race/ethnicity and census tract SES. Differences in survival between groups were used to assess absolute disparities. Hazard ratios were examined as a measure of relative disparity. Interactions between race/ethnicity and neighborhood SES were evaluated using proportional hazard models.
Results
Survival increased with higher SES for all racial/ethnic groups and generally was higher among non-Hispanic white and Asian/Pacific Islander (API) than non-Hispanic black and Hispanic cases. Absolute disparity in breast cancer survival among non-Hispanic black vs non-Hispanic white cases was slightly larger in low-SES areas than in highSES areas (7.1% and 6.8%, respectively). In contrast, after adjusting for stage, age, and treatment, risk of mortality among non-Hispanic black cases compared with non-Hispanic white cases was 21% higher in low-SES areas and 64% higher in high-SES areas. Similarly, patterns of absolute and relative disparity compared with non-Hispanic whites differed by SES for Hispanic breast cancer, non-Hispanic black colorectal cancer, and prostate cancer cases. Statistically significant interactions existed between race/ethnicity and SES for colorectal and female breast cancers.
Discussion
In health disparities research, both relative and absolute measures provide context. A better understanding of the interactions between race/ethnicity and SES may be useful in directing screening and treatment resources toward at-risk populations.
J Natl Cancer Inst Monogr 2014;49:236–243
The National Cancer Institute (NCI) defines cancer disparities as adverse differences in cancer outcomes between population groups (1). Each year, summary cancer survival statistics are updated for the major racial and ethnic groups (2,3). Research has focused largely on racial and ethnic inequalities. Less information is available on the role of socioeconomic status (SES) in cancer outcomes such as survival (4–6). Racial/ethnic disparities may be explained by SES (7,8), access to affordable health care coverage (9), or biological differences (10). If racial disparities are driven by inequalities (11) such as adverse social experiences and economic deprivation (12,13), it is possible that race/ethnicity and SES have an interactive effect on survival (8,14,15). Minorities and the poor often experience worse health outcomes than non-Hispanic whites and more affluent individuals, respectively (16). Because cancer registries do not routinely collect data on SES, only a few studies of cancer survival in the United States have investigated interactions between income and education (17,18). To improve the value of its data, NCI’s Surveillance, Epidemiology, and End Results (SEER) Program recently developed a variable based on a census-tract level SES index (19). 236
Although most research on health disparities has focused on relative measures, such as hazard and rate ratios, absolute measures provide an intuitive metric, giving context to relative disparity measures (20). In this study, we used the census-tract level SES index to estimate absolute and relative survival disparities in lung, liver, kidney, colorectal, breast, and prostate cancer burdens. As an absolute measure of disparity, we examined the difference in five-year, cause-specific survival. As a relative measure of disparity, we present hazard ratio measures that are adjusted for race/ethnicity, SES, age, stage, and treatment.
Methods Data Sources and Study Population Disparities in five-year, cause-specific survival were estimated by race/ethnicity and SES. The study cohort included microscopically confirmed malignant cancers of the lung/bronchus (hereafter “lung”), liver/intrahepatic bile duct (hereafter “liver”), kidney/renal pelvis (hereafter “kidney”), colon and rectum (hereafter “colorectal”), prostate, and female breast (hereafter “breast”) diagnosed from January 1, 2002, through December 31, 2008, and followed through December 31, 2009. Journal of the National Cancer Institute Monographs, No. 49, 2014
Downloaded from http://jncimono.oxfordjournals.org/ at New York Medical College on March 26, 2015
Introduction
Sociodemographics Race/ethnicity was classified as non-Hispanic white, non-Hispanic black, non-Hispanic Asian or Pacific Islander (API), AmericanIndian or Alaska Native (AI/AN), and Hispanic. Age at diagnosis was categorized as greater than 45 years of age, in 10-year intervals from ages 45–85, and age 85 years or older. To improve the quality of area-based SES estimates, our analysis is based on a new census-tract level linkage created by SEER and an index variable of SES status in 16 registries, after excluding the Alaska Native and Louisiana registries, where census tract SES data were not available. Cases were mapped to their year 2000 census tract of residence based on their address at time of diagnosis. Validated indices of socioeconomic position were developed using factor analysis from census tract-level estimates from the 2000 census, and from five-year estimates from the 2005–2009 American Community Survey (ACS) (24). Both indices were extracted from the same seven SES measures identified by Yost et al. (25). to represent three components of SES (occupation, education, and income). The variables included: proportion with a blue-collar job, proportion older than age 16 in the workforce without a job, education index, median household income, proportion below 200% of the poverty level, median rent, and median house value. We assigned an index value to each census tract and categorized them into quintiles. The first quintile (Q1, lowest neighborhood SES) corresponded to the 20th percentile or less of SES, and the fifth quintile (Q5, highest neighborhood SES) corresponded to the 80th percentile or higher. For cases diagnosed in 2002–2003, the 2000 census index was used; for cases diagnosed in 2004–2008, the 2005–2009 ACS index was used. (This research resource is accessible by submitting a suitable concept to the NCI SEER Program, care if the corresponding author or M. Yu.) Cases with unknown race/ethnicity or addresses that could not be mapped to a census tract were excluded from the analysis. Cancer Staging and Treatment Cancer survival is strongly associated with stage at diagnosis and treatment. Staging was based on SEER Summary Stage 2000 (1998+). Summary staging, although not of clinical utility, was used as a measure of population-based cancer control efforts. Stage was
categorized as localized, regional, distant, or unknown/unstaged. For prostate cancer, localized and regional stages were combined according to American Joint Committee on Cancer staging procedures (26). First-course cancer-directed surgery and radiation therapy were determined from the SEER variables “cancer-directed surgery” and “radiation sequence with surgery.” Statistical Analysis Cancer survival was studied using both absolute and relative disparity measures (27). To assess disparities on the absolute scale, a survival rate difference was calculated based on Kaplan–Meier estimates of five-year, cause-specific survival by race/ethnicity and SES using SEER*Stat software (28). Because suitable life tables were not available for the SES subgroups, cause-specific death classification was used (29), with other causes of death censored. Pairwise comparisons were considered statistically significant when the Z-test statistic had a P value of less than .05. Sensitivity analyses were used to compare age-standardized and unadjusted survival. Relative disparities were assessed by calculating hazard ratios for various race/ethnicity and SES comparisons, as well as their interactions, for each cancer site. Stratified Cox proportional hazard models were developed with SAS v.9.3 software (30). Hazard ratios reflected relative risk of death in the 5 years after diagnosis adjusted for confounding factors. Each site-specific model was stratified by age at diagnosis and tumor stage after inspection of the [log, −log] plot of the survival function revealed signs of non-proportionality, as recommended by Breslow (31). Models also were adjusted for surgery and radiation treatment. Cases with missing or unknown treatment status were coded as having no reported treatment. Wald χ2 tests were used to assess the statistical significance of the interaction between SES and race/ethnicity. Comparisons of survival are presented by SES and racial/ethnic group. For pairwise comparisons, confidence intervals (CIs) that did not include 1.0 were considered statistically significant.
Results Table 1 presents the demographic and clinical distribution of cases by cancer site. Total numbers of cases ranged from a low of 32 571 for liver cancer to a high of 357 078 for prostate cancer. Excluding prostate cancer, the site with the highest percentage in men was liver cancer (71.7%), and the site with the lowest percentage in men was colorectal cancer (50.9%). Excluding breast cancer, the site with the highest percentage in women was colorectal cancer (49.1%) and the lowest percentage in women was liver cancer (28.3%). Most lung (64%), colorectal (58%), and prostate cancer (64%) cases were diagnosed among individuals aged 65 years and older; and most liver (55.6%), kidney (55.5%), and breast (62.2%) cancer cases occurred in individuals aged less than 65 years. Stage at diagnosis varied by site: the majority of breast and kidney cancers were localized (60.3% and 62.3%, respectively), whereas 56.4% of lung cancer cases were diagnosed at distant stages. Overall, most cancer cases occurred in non-Hispanic whites, followed by non-Hispanic blacks. With the exception of prostate cancer cases (at 2.8%), less than 1% of all cancer cases were excluded due to unknown race. A disproportionately high percentage of liver
Journal of the National Cancer Institute Monographs, No. 49, 2014 237
Downloaded from http://jncimono.oxfordjournals.org/ at New York Medical College on March 26, 2015
Screening and treatment protocols and cancer survival rates differed for these cancer sites during the study period. For lung cancer, there were no population-based screening recommendations, the cancers typically were diagnosed in late stages, and survival rates were low (21). Similarly, although five-year relative survival rates for liver cancer tripled from less than 5% in the 1970s to 15% in 2003, this cancer often was diagnosed at regional or distant stages (2). In contrast, more than 60% of kidney cancer cases were diagnosed at a localized stage, and survival increased from 50% to 75% during 1970–2003 (2). For colorectal cancer, there were well-established screening protocols, decreasing trends in incidence, and increasing trends in survival. The five-year survival estimates for breast cancer were above 85%, and nearly 72.4% of women were screened yearly after age 40 per recommended guidelines (22). Finally, for prostate cancer, the five-year survival rate was above 90% and, despite changes in screening recommendations, approximately 45% of men over age 75 were screened in 2008 (23).
Table 1. Frequency of cases by cancer site and proportion by demographic characteristics* Cancer site Lung
Liver
Kidney 62 371
CRC
Prostate Breast†
213 274 357 078
313 482
61.4 38.7
50.9 49.1
100 —-
— 100
10.8 18.4 26.3 23.2 16.6 4.7
5.8 14.3 21.2 24.3 24.0 10.4
0.7 9.6 31.2 35.7 19.1 3.8
13.6 24.0 24.6 19.0 14.1 4.7
62.3 16.8 17.1 3.7
39.4 36.2 19.7 4.7
92.8‡ —‡ 4.1 3.1
60.3 32.7 5.1 1.9
82.0
85.8
38.7
92.8
3.3
10.4
2.9
47.7
71.5 10.5 4.5 0.8 12.1 0.6
71.6 11.3 7.0 0.5 8.9 0.7
70.5 13.6 4.4 0.3 8.3 2.8
72.7 10.1 6.8 0.5 9.4 0.6
14.7 16.6 17.6 17.9 18.0 15.3
14.9 16.8 17.6 18.1 17.9 14.6
12.5 15.0 17.0 18.7 22.6 14.2
12.2 15.1 17.3 19.5 22.7 13.3
* AI/AN = American Indian/Alaska Native; API = Asian/Pacific Islander; CRC = colorectal cancer; HISP = Hispanic; NHB = non-Hispanic blacks; NHW = non-Hispanic whites; Q = quintile; SES = socioeconomic status. † Breast cancer restricted to women. ‡ Localized, regional combined.
cancer cases occurred among API and Hispanic individuals (17.6% and 17.4%, respectively). The frequency of lung and liver cancer cases decreased from Q1 (low SES) census tracts to Q5 (high SES) census tracts. This pattern was reversed for kidney, colorectal, prostate, and breast cancers. SES data were missing for between 11.7% (liver) and 16.7% (lung) of cases for each site. These cases were excluded. Five-year, cause-specific survival is presented for all six cancer sites by race/ethnicity and SES in Table 2 and Figure 2. Overall survival was lowest for lung and liver cancers (18.3% and 19.3%, respectively), intermediate for colorectal and kidney cancers (64.7% and 72.6%, respectively), and highest for breast and prostate cancers (87.9% and 93.7%, respectively). Survival increased from lowto high-SES groups, with the smallest difference between Q1 and Q5 occurring for prostate cancer (−4.8%) and the largest difference between Q1 and Q5 occurring for breast (−9.8%) and liver cancers (−10.4%). Among racial/ethnic groups, APIs experienced the best five-year, cause-specific survival for all SES strata for each cancer 238
Discussion In this cancer survival study, five-year, cause-specific survival increased with higher SES; and non-Hispanic white and API cases often experienced better cancer survival than non-Hispanic black, Hispanic, and AI/AN cases. These absolute disparities are Journal of the National Cancer Institute Monographs, No. 49, 2014
Downloaded from http://jncimono.oxfordjournals.org/ at New York Medical College on March 26, 2015
N 266 439 32 571 Gender, % Men 53.8 71.7 Women 46.2 28.3 Age at diagnosis, % 0–44 2.3 5.9 45–54 10.3 21.5 55–64 23.2 28.2 65–74 31.4 22.4 75–84 26.0 17.0 85+ 6.9 5.0 Stage, % Localized 15.1 40.3 Regional 22.0 26.3 Distant 56.4 18.5 Unstaged 6.5 14.8 Cancer-directed surgery, % Yes 20.9 24.0 Radiation therapy, % Yes 7.6 1.3 Race/ethnicity,% NHW 78.2 51.8 NHB 10.9 11.8 API 5.1 17.6 AI/AN 0.4 1.1 HISP 5.2 17.4 Unknown 0.2 0.3 SES, % Q1 (low) 17.4 19.2 Q2 17.6 18.3 Q3 17.4 18.1 Q4 16.5 17.8 Q5 (high) 14.3 15.0 Missing 16.7 11.7
site except kidney, for which non-Hispanic blacks had the best survival. Non-Hispanic blacks had the lowest survival rates for liver (15.1%), colorectal (58.2%), and breast (78.7%) cancers. Figure 1 depicts race/ethnicity- and SES-stratified survival with 95% CIs. Sensitivity analyses, with age-standardized survival, showed that variation from these results was small, at between 1% and 3% (data not shown) . Absolute and relative cancer survival disparities by SES within racial/ethnic groups are presented in the last two columns of Table 2. For lung cancer, the absolute difference in survival between Q1 and Q5 within racial/ethnic groups ranged from −15.4 (American Indian/Alaska Native [AI/AN]) to −5.3 (non-Hispanic blacks). The relative disparity comparison, based on stratified Cox proportional hazards models, shows a 24% increase in mortality risk for Q1 compared with Q5 for non-Hispanic whites and a 15% increase for non-Hispanic blacks. Except for lung and prostate cancer among AI/AN cases and kidney cancer among all minority groups, significant SES absolute disparities within racial/ethnic groups aligned with significant relative disparities. Absolute and relative disparities between racial/ethnic groups by SES are presented in Table 3. Compared with non-Hispanic whites, the absolute disparity in breast cancer survival among nonHispanic blacks was larger in Q1 than Q5, with −7.1% and −6.8% lower five-year survival, respectively. In contrast, the relative disparity in breast cancer survival between non-Hispanic black cases and non-Hispanic white cases, based on adjusted hazard ratios, was larger in Q5 than in Q1 (64% higher risk compared with 31% higher risk, respectively). Similar patterns were seen among nonHispanic black and non-Hispanic white colorectal and prostate cancer cases. For lung, liver, and kidney cancers, only three of 18 absolute disparity comparisons (17%) in Table 3 were statistically significant. In contrast, 50% of absolute disparity comparisons for breast, colorectal, and prostate cancers were statistically significant. When Hispanic and non-Hispanic white cases were compared, breast, colorectal, and prostate cancer absolute disparities were greater than 1 for Hispanics in Q1 (better survival than non-Hispanic whites) and less than 1 in Q5 (worse survival than non-Hispanic whites). Hispanic cases experienced an increased risk of death compared with non-Hispanic whites in the highest SES category for breast cancer (hazard ratio of 1.21, 95% CI: 1.08 to 1.36) and prostate cancer (hazard ratio of 1.20, 95% CI: 1.01 to 1.42). Compared with non-Hispanic white cases, the risk of death for APIs was lower in both Q1 and Q5 for breast and prostate cancers and for colorectal cancer in Q5. For colorectal and breast cancers, two of the three leading cancers with robust screening and treatment programs, there was a significant interaction between race/ethnicity and SES in the model (see Table 3). Supplementary Table 1 provides standard errors for Q1 and Q5 survival estimates and CIs for the hazard ratios in Table 3.
Table 2. Five-year, cause-specific survival by race/ethnicity and census tract socioeconomic status (SES), absolute (rate difference) and relative (hazard ratio) disparity, low (Q1) vs high (Q5) SES strata* Five-year, cause-specific survival, % Race/ethnicity
Q2
Q3
Q4
Q5 high
AD rate difference Q1–Q5
RD† HR Q1 vs Q5
18.3 18.6 15.7 19.4 11.8 17.2
15.0 15.3 14.2 15.2 11.3 14.8
17.0 17.2 16.0 17.5 11.0 15.2
11.3 18.2 18.6 19.0 11.3 18.2
19.5 19.6 16.3 20.8 9.0 19.7
22.3 22.4 19.5 22.0 26.7 22.3
−7.3 −7.1‡ −5.3‡ −6.8‡ −15.4‡ −7.5‡
1.24‡ 1.15‡ 1.20‡ 1.19 1.22‡
19.3 18.2 15.1 24.4 16.1 19.3
14.6 13.7 11.5 17.0 14.3 16.6
16.2 14.8 14.2 18.8 14.1 18.1
19.3 18.1 19.8 22.5 16.9 19.9
22.5 21.4 17.6 25.7 19.4 23.0
25.0 20.9 29.0 33.8 28.6 25.8
−10.4 −7.2‡ −17.5‡ −16.8‡ −14.3 −9.2‡
1.32‡ 1.29‡ 1.32‡ 1.28 1.36‡
72.6 72.5 74.1 72.2 68.5 72.6
69.6 68.5 71.6 65.9 69.3 70.9
71.0 70.1 73.1 75.2 75.5 72.2
72.3 72.0 76.5 72.4 62.0 72.9
72.7 72.7 78.3 69.1 66.1 72.5
76.6 76.6 77.0 75.3 51.9 78.1
−7.0 −8.1‡ −5.4‡ −9.4‡ 17.4 −7.2‡
1.23‡ 1.00 1.26 0.88 1.13
64.7 65.1 58.2 69.3 58.9 63.9
60.5 61.4 56.4 60.9 50.7 62.8
62.4 62.7 57.0 67.8 59.5 62.4
65.0 64.9 60.8 68.6 60.7 66.2
66.0 66.1 60.1 69.3 74.2 63.1
68.5 68.2 63.7 74.0 57.9 66.5
−8.0 −6.8‡ −7.3‡ −13.1‡ −7.2 −3.7‡
1.28‡ 1.17‡ 1.36‡ 1.26 1.18‡
87.9 88.8 78.7 91.0 83.6 86.5
81.6 82.6 75.5 85.5 80.9 84.5
85.9 86.5 78.1 89.6 78.5 85.7
87.9 88.1 81.1 89.8 85.6 88.2
89.4 89.6 82.2 92.3 92.6 88.2
91.4 91.6 84.8 93.1 87.0 89.5
−9.8 −9.0‡ −9.3‡ −7.6‡ −6.1 −5.0‡
1.75‡ 1.40‡ 1.34‡ 1.62 1.44‡
93.7 93.9 92.5 94.3 89.6 93.0
90.9 91.1 90.4 91.6 81.1 91.6
92.4 92.2 92.8 93.4 93.9 92.5
93.5 93.6 93.5 92.9 90.7 93.2
94.3 94.2 94.6 94.9 92.0 94.4
95.7 95.7 95.2 95.9 97.4 94.8
−4.8 −4.6‡ −4.8‡ −4.3‡ −16.3‡ −3.2‡
1.56‡ 1.31‡ 1.28 1.80 1.13
* AD = absolute disparity; AI/AN = American Indian/Alaska Native; API = Asian/Pacific Islander; CRC = colorectal cancer; HISP = Hispanic; HR = hazard ratio; NHB = non-Hispanic blacks; NHW = non-Hispanic whites; Q = quintile; RD = relative disparity; SES = socioeconomic status. † Stratified by stage and age with adjustment for sex, cancer-directed surgery, radiation therapy. ‡ P
