Understanding Geographic Disparities in Mortality.

A rich literature shows that early-life conditions shape later-life outcomes, including health and migration events. However, analyses of geographic disparities in mortality outcomes focus almost exclusively on contemporaneously measured geographic place (e.g., state of residence at death), thereby potentially conflating the role of early-life conditions, migration patterns, and effects of destinations. We employ the newly available Mortality Disparities in American Communities data set, which links respondents in the 2008 American Community Survey to official death records, and estimate consequential differences based on the method of aggregation we use: the unweighted mean absolute deviation of the difference in life expectancy at age 50 measured by state of birth versus state of residence is 0.58 years for men and 0.40 years for women. These differences are also spatially clustered, and we show that regional inequality in life expectancy is higher based on life expectancies by state of birth, implying that interstate migration mitigates baseline geographic inequality in mortality outcomes. Finally, we assess how state-specific features of in-migration, out-migration, and nonmigration together shape measures of mortality disparities by state (of residence), further demonstrating the difficulty of clearly interpreting these widely used measures.

Comparisons of individual- and area-level socioeconomic status as proxies for individual-level measures: evidence from the Mortality Disparities in American Communities study.

BACKGROUND: Area-level measures are often used to approximate socioeconomic status (SES) when individual-level data are not available. However, no national studies have examined the validity of these measures in approximating individual-level SES. METHODS: Data came from ~ 3,471,000 participants in the Mortality Disparities in American Communities study, which links data from 2008 American Community Survey to National Death Index (through 2015). We calculated correlations, specificity, sensitivity, and odds ratios to summarize the concordance between individual-, census tract-, and county-level SES indicators (e.g., household income, college degree, unemployment). We estimated the association between each SES measure and mortality to illustrate the implications of misclassification for estimates of the SES-mortality association. RESULTS: Participants with high individual-level SES were more likely than other participants to live in high-SES areas. For example, individuals with high household incomes were more likely to live in census tracts (r = 0.232; odds ratio [OR] = 2.284) or counties (r = 0.157; OR = 1.325) whose median household income was above the US median. Across indicators, mortality was higher among low-SES groups (all p < .0001). Compared to county-level, census tract-level measures more closely approximated individual-level associations with mortality. CONCLUSIONS: Moderate agreement emerged among binary indicators of SES across individual, census tract, and county levels, with increased precision for census tract compared to county measures when approximating individual-level values. When area level measures were used as proxies for individual SES, the SES-mortality associations were systematically underestimated. Studies using area-level SES proxies should use caution when selecting, analyzing, and interpreting associations with health outcomes.

Drug overdose mortality is associated with employment status and occupation in the National Longitudinal Mortality Study.

BACKGROUND: Since 1999, over 702,000 people in the US have died of a drug overdose, and the drug overdose death rate has increased from 6.2 to 21.8 per 100,000. Employment status and occupation may be important social determinants of overdose deaths. OBJECTIVES: Estimate the risk of drug overdose death by employment status and occupation, controlling for other social and demographic factors known to be associated with overdose deaths. METHODS: Proportional hazard models were used to study US adults in the National Longitudinal Mortality Study with baseline measurements taken in the early 2000s and up to 6 years of follow-up (n = 438,739, 53% female, 47% male). Comparisons were made between adults with different employment statuses (employed, unemployed, disabled, etc.) and occupations (sales, construction, service occupations, etc.). Models were adjusted for age, sex, race/ethnicity, education, income and marital status. RESULTS: Adults who were disabled (hazard ratio (HR) = 6.96 (95% CI = 6.81-7.12)), unemployed (HR = 4.20, 95% CI = 4.09-4.32) and retired (HR = 2.94, 95% CI = 2.87-3.00) were at higher risk of overdose death relative to those who were employed. By occupation, those working in service (HR = 2.05, 95% CI = 1.97-2.13); construction and extraction (HR = 1.69, 95% CI = 1.64-1.76); management, business and financial (HR = 1.39, 95% CI = 1.33-1.44); and installation, maintenance and repair (HR = 1.32, 95% CI = 1.25-1.40) occupations displayed higher risk relative to professional occupations. CONCLUSIONS: In a large national cohort followed prospectively for up to 6 years, several employment statuses and occupations are associated with overdose deaths, independent of a range of other factors. Efforts to prevent overdose deaths may benefit from focusing on these high-risk groups.

Association Between Socioeconomic Status and Tumor Grade Among Black Men with Prostate Cancer.

BACKGROUND: Prostate cancer affects black men disproportionately. Black men have an increased incidence of prostate cancer diagnoses at earlier ages and higher grade as indicated by Gleason score, compared to other races. This study investigates the impact of socioeconomic status (SES) on prostate cancer tumor grade among black men. METHODS: Black men with a prostate cancer diagnosis during 1973-2011 were examined using individual-level data from the SEER NLMS database. Logistic regression model estimated the likelihood of receiving a diagnosis of high versus low grade prostate cancer based on self-reported SES status at the time of diagnosis. RESULTS: Men who completed high school only were statistically significantly more likely to have a higher prostate cancer grade than those with a bachelor's degree or higher. However, there was no dose-response effect across educational strata. Retirees were 30% less likely to have higher grade tumors compared to those who were employed. CONCLUSIONS: SES differences among black men did not fully explain the high grade of prostate cancer. Further research is needed on the biology of the disease and to assess access to medical care and prostate health education, discrimination, stress exposures, and social norms that might contribute to the aggressiveness of prostate cancer among black men.

A longitudinal study of smokeless tobacco use and mortality in the United States.

Few studies in the United States have examined longitudinally the mortality risks associated with use of smokeless tobacco (SLT). The sample of our study was composed of participants from the National Longitudinal Mortality Study who completed a single Tobacco Use Supplement to the Current Population Survey between the years 1985 and 2011. Using survival methods, SLT use at the baseline survey was examined as a predictor of all-cause mortality and cause-specific mortalities in models that excluded individuals who had ever smoked cigarettes, cigars or used pipes (final n = 349,282). The participants had median and maximum follow-up times of 8.8 and 26.3 years, respectively. Regression analyses indicated that compared to the never tobacco users, the current SLT users did not have elevated mortality risks from all cancers combined, the digestive system cancers and cerebrovascular disease. However, current SLT users had a higher mortality risk for coronary heart disease (CHD) [hazard ratio (HR) (95% CI) = 1.24 (1.05, 1.46)] relative to never tobacco users. In a separate model, the elevated risk for CHD mortality corresponded to the use of moist snuff [HR (95% CI) = 1.30 (1.03, 1.63)]. The associations with CHD mortality could be attributed to long-term nicotine exposure, other SLT constituents (e.g., metals) or the confounding effects of CHD risk factors not accounted for in our study. The study's findings contribute to the ongoing dialogue on tobacco harm reduction and the US FDA's evaluation of Modified Risk Tobacco Product applications submitted by American SLT manufacturers.

Racial misclassification of American Indians and Alaska Natives by Indian Health Service Contract Health Service Delivery Area.

OBJECTIVES: We evaluated the racial misclassification of American Indians and Alaska Natives (AI/ANs) in cancer incidence and all-cause mortality data by Indian Health Service (IHS) Contract Health Service Delivery Area (CHSDA). METHODS: We evaluated data from 3 sources: IHS-National Vital Statistics System (NVSS), IHS-National Program of Cancer Registries (NPCR)/Surveillance, Epidemiology and End Results (SEER) program, and National Longitudinal Mortality Study (NLMS). We calculated, within each data source, the sensitivity and classification ratios by sex, IHS region, and urban-rural classification by CHSDA county. RESULTS: Sensitivity was significantly greater in CHSDA counties (IHS-NVSS: 83.6%; IHS-NPCR/SEER: 77.6%; NLMS: 68.8%) than non-CHSDA counties (IHS-NVSS: 54.8%; IHS-NPCR/SEER: 39.0%; NLMS: 28.3%). Classification ratios indicated less misclassification in CHSDA counties (IHS-NVSS: 1.20%; IHS-NPCR/SEER: 1.29%; NLMS: 1.18%) than non-CHSDA counties (IHS-NVSS: 1.82%; IHS-NPCR/SEER: 2.56%; NLMS: 1.81%). Race misclassification was less in rural counties and in regions with the greatest concentrations of AI/AN persons (Alaska, Southwest, and Northern Plains). CONCLUSIONS: Limiting presentation and analysis to CHSDA counties helped mitigate the effects of race misclassification of AI/AN persons, although a portion of the population was excluded.

Individual education, area income, and mortality and recurrence of myocardial infarction in a Medicare cohort: the National Longitudinal Mortality Study.

BACKGROUND: The Medicare program provides universal access to hospital care for the elderly; however, mortality disparities may still persist in this population. The association of individual education and area income with survival and recurrence post Myocardial Infarction (MI) was assessed in a national sample. METHODS: Individual level education from the National Longitudinal Mortality Study was linked to Medicare and National Death Index records over the period of 1991-2001 to test the association of individual education and zip code tabulation area median income with survival and recurrence post-MI. Survival was partitioned into 3 periods: in-hospital, discharge to 1 year, and 1 year to 5 years and recurrence was partitioned into two periods: 28 day to 1 year, and 1 year to 5 years. RESULTS: First MIs were found in 8,043 women and 7,929 men. In women and men 66-79 years of age, less than a high school education compared with a college degree or more was associated with 1-5 year mortality in both women (HRR 1.61, 95% confidence interval 1.03-2.50) and men (HRR 1.37, 1.06-1.76). Education was also associated with 1-5 year recurrence in men (HRR 1.68, 1.18-2.41, < High School compared with college degree or more), but not women. Across the spectrum of survival and recurrence periods median zip code level income was inconsistently associated with outcomes. Associations were limited to discharge-1 year survival (RR lowest versus highest quintile 1.31, 95% confidence interval 1.03-1.67) and 28 day-1 year recurrence (RR lowest versus highest quintile 1.72, 95% confidence interval 1.14-2.57) in older men. CONCLUSIONS: Despite the Medicare entitlement program, disparities related to individual socioeconomic status remain. Additional research is needed to elucidate the barriers and mechanisms to eliminating health disparities among the elderly.

Effects of individual-level socioeconomic factors on racial disparities in cancer treatment and survival: findings from the National Longitudinal Mortality Study, 1979-2003.

BACKGROUND: This is the first study to use the linked National Longitudinal Mortality Study and Surveillance, Epidemiology, and End Results (SEER) data to determine the effects of individual-level socioeconomic factors (health insurance, education, income, and poverty status) on racial disparities in receiving treatment and in survival. METHODS: This study included 13,234 cases diagnosed with the 8 most common types of cancer (female breast, colorectal, prostate, lung and bronchus, uterine cervix, ovarian, melanoma, and urinary bladder) at age ≥ 25 years, identified from the National Longitudinal Mortality Study-SEER data during 1973 to 2003. Kaplan-Meier methods and Cox regression models were used for survival analysis. RESULTS: Three-year all-cause observed survival for cases diagnosed with local-stage cancers of the 8 leading tumors combined was ≥ 82% regardless of race/ethnicity. More favorable survival was associated with higher socioeconomic status. Compared with whites, blacks were less likely to receive first-course cancer-directed surgery, perhaps reflecting a less favorable stage distribution at diagnosis. Hazard ratio (HR) for cancer-specific mortality was significantly higher among blacks compared with whites (HR, 1.2; 95% confidence interval [CI], 1.1-1.3) after adjusting for age, sex, and tumor stage, but not after further controlling for socioeconomic factors and treatment (HR, 1.0; 95% CI, 0.9-1.1). HRs for all-cause mortality among patients with breast cancer and for cancer-specific mortality in patients with prostate cancer were significantly higher for blacks compared with whites after adjusting for socioeconomic factors, treatment, and patient and tumor characteristics. CONCLUSIONS: Favorable survival was associated with higher socioeconomic status. Racial disparities in survival persisted after adjusting for individual-level socioeconomic factors and treatment for patients with breast and prostate cancer.

New variable selection methods for zero-inflated count data with applications to the substance abuse field.

Zero-inflated count data are very common in health surveys. This study develops new variable selection methods for the zero-inflated Poisson regression model. Our simulations demonstrate the negative consequences which arise from the ignorance of zero-inflation. Among the competing methods, the one-step SCAD method is recommended because it has the highest specificity, sensitivity, exact fit, and lowest estimation error. The design of the simulations is based on the special features of two large national databases commonly used in the alcoholism and substance abuse field so that our findings can be easily generalized to the real settings. Applications of the methodology are demonstrated by empirical analyses on the data from a well-known alcohol study.

Healthy food retail availability and cardiovascular mortality in the United States: a cohort study.

OBJECTIVES: We investigated the association of healthy food retail presence and cardiovascular mortality, controlling for sociodemographic characteristics. This association could inform efforts to preserve or increase local supermarkets or produce market availability. DESIGN: Cohort study, combining Mortality Disparities in American Communities (individual-level data from 2008 American Community Survey linked to National Death Index records from 2008 to 2015) and retail establishment data. SETTING: Across the continental US area-based sociodemographic and retail characteristics were linked to residential location by ZIP code tabulation area (ZCTA). Sensitivity analyses used census tracts instead, restricted to urbanicity or county-based strata, or accounted for non-independence using frailty models. PARTICIPANTS: 2 753 000 individuals age 25+ living in households with full kitchen facilities, excluding group quarters. PRIMARY AND SECONDARY OUTCOME MEASURES: Cardiovascular mortality (primary) and all-cause mortality (secondary). RESULTS: 82% had healthy food retail (supermarket, produce market) within their ZCTA. Density of such retail was correlated with density of unhealthy food sources (eg, fast food, convenience store). Healthy food retail presence was not associated with reduced cardiovascular (HR: 1.03; 95% CI 1.00 to 1.07) or all-cause mortality (HR: 1.05; 95% CI 1.04 to 1.06) in fully adjusted models (with adjustment for gender, age, marital status, nativity, Black race, Hispanic ethnicity, educational attainment, income, median household income, population density, walkable destination density). The null finding for cardiovascular mortality was consistent across adjustment strategies including minimally adjusted models (individual demographics only), sensitivity analyses related to setting, and across gender or household type strata. However, unhealthy food retail presence was associated with elevated all-cause mortality (HR: 1.15; 95% CI 1.11 to 1.20). CONCLUSIONS: In this study using food establishment locations within administrative areas across the USA, the hypothesised association of healthy food retail availability with reduced cardiovascular mortality was not supported; an association of unhealthy food retail presence with higher mortality was not specific to cardiovascular causes.

Racial disparities in mortality in the adult hispanic population.

OBJECTIVE: We addressed three research questions: (1) Are there racial mortality disparities in the adult Hispanic population that resemble those observed in the non-Hispanic population in the US? (2) Does nativity mediate the race-mortality relationship in the Hispanic population? and (3) What does the Hispanic mortality advantage relative to the non-Hispanic white population look like when Hispanic race is considered? METHODS: We estimated a series of parametric hazard models on eight years of mortality follow-up data and calculated life expectancy estimates using the Mortality Disparities in American Communities database. RESULTS: Hispanic white adults experience lower mortality than their Hispanic black, American Indian and Alaska Native, Some Other Race, and multiple race counterparts. This Hispanic white advantage is found mostly among the US born. The Hispanic advantage relative to the non-Hispanic white population operates for most Hispanic race groups among the foreign born but either disappears or converts to a disadvantage for most of the non-white Hispanic groups among the US born. CONTRIBUTION: Our study extends the literature on the Hispanic Mortality Paradox by revealing that the adult Hispanic population experiences racial mortality disparities that closely resemble those observed in the non-Hispanic population. The Hispanic mortality advantage is mediated not only by nativity but by race. These results indicate that race is a critical factor that should be considered in any study with the goal of understanding the health and mortality profiles of the Hispanic population in the US.

Indicators of Socioeconomic Status for Individuals, Census Tracts, and Counties: How Well Do Measures Align for Demographic Subgroups?

OBJECTIVES: Researchers often approximate individual-level socioeconomic status (SES) from census tract and county data. However, area-level variables do not serve as accurate proxies for individual-level SES, particularly among some demographic subgroups. The present study aimed to analyze the potential bias introduced by this practice. METHODS: Data included (1) individual-level SES from the Mortality Disparities in American Communities study (n ≈ 3,471,000 collected in 2008), and (2) census tract- and county-level SES from the 2006-2010 American Community Survey. Analyses included correlations among SES indicators (eg, median household income, having a high school degree, unemployment) across individual versus census tract and county levels, stratified by sex, age, race/ethnicity, and urbanicity. Finally, generalized estimating equations evaluated demographic differences in whether area-level SES matched or underestimated individual-level SES. RESULTS: Low correlations were observed between individual- and area-level SES (census tract: Spearman's r range = 0.048 for unemployment to 0.232 for median household income; county: r range = 0.028 for unemployment to 0.157 for median household income; all P < .0001). SES indicators were more likely to match for males, older participants, and urban groups. Area-level SES indicators were more likely to underestimate individual-level SES for older participants and rural groups, indicating that individuals who are part of these groups may live in systematically lower-SES communities than their own SES might connote. CONCLUSIONS: In this population-based study of 3.5 million participants, area-level indicators were poor proxies for individual-level SES, particularly for participants living in rural areas.

Impact of socioeconomic status on cancer incidence and stage at diagnosis: selected findings from the surveillance, epidemiology, and end results: National Longitudinal Mortality Study.

BACKGROUND: Population-based cancer registry data from the Surveillance, Epidemiology, and End Results (SEER) Program at the National Cancer Institute (NCI) are mainly based on medical records and administrative information. Individual-level socioeconomic data are not routinely reported by cancer registries in the United States because they are not available in patient hospital records. The U.S. representative National Longitudinal Mortality Study (NLMS) data provide self-reported, detailed demographic and socioeconomic data from the Social and Economic Supplement to the Census Bureau's Current Population Survey (CPS). In 1999, the NCI initiated the SEER-NLMS study, linking the population-based SEER cancer registry data to NLMS data. The SEER-NLMS data provide a new unique research resource that is valuable for health disparity research on cancer burden. We describe the design, methods, and limitations of this data set. We also present findings on cancer-related health disparities according to individual-level socioeconomic status (SES) and demographic characteristics for all cancers combined and for cancers of the lung, breast, prostate, cervix, and melanoma. METHODS: Records of cancer patients diagnosed in 1973-2001 when residing 1 of 11 SEER registries were linked with 26 NLMS cohorts. The total number of SEER matched cancer patients that were also members of an NLMS cohort was 26,844. Of these 26,844 matched patients, 11,464 were included in the incidence analyses and 15,357 in the late-stage diagnosis analyses. Matched patients (used in the incidence analyses) and unmatched patients were compared by age group, sex, race, ethnicity, residence area, year of diagnosis, and cancer anatomic site. Cohort-based age-adjusted cancer incidence rates were computed. The impact of socioeconomic status on cancer incidence and stage of diagnosis was evaluated. RESULTS: Men and women with less than a high school education had elevated lung cancer rate ratios of 3.01 and 2.02, respectively, relative to their college educated counterparts. Those with family annual incomes less than $12,500 had incidence rates that were more than 1.7 times the lung cancer incidence rate of those with incomes $50,000 or higher. Lower income was also associated with a statistically significantly increased risk of distant-stage breast cancer among women and distant-stage prostate cancer among men. CONCLUSIONS: Socioeconomic patterns in incidence varied for specific cancers, while such patterns for stage were generally consistent across cancers, with late-stage diagnoses being associated with lower SES. These findings illustrate the potential for analyzing disparities in cancer outcomes according to a variety of individual-level socioeconomic, demographic, and health care characteristics, as well as by area measures available in the linked database.

Are associations between psychosocial stressors and incident lung cancer attributable to smoking?

PURPOSE: To learn whether reported associations between major psychosocial stressors and lung cancer are independent of smoking history. METHODS: Subjects were at least 25 years old and without lung cancer at enrollment in the United States Census Bureau's National Longitudinal Mortality Survey in 1995-2008. Follow-up via Surveillance Epidemiology and End Results and National Death Index continued until lung cancer diagnosis, death, or December 2011. Involuntary unemployment, widowhood, and divorce, stratified by sex, were tested for association with subsequent lung cancer using proportional hazards regression for competing risks. Smoking status, years smoked, cigarettes per day, and years since quitting were imputed when missing. RESULTS: At enrollment, subjects (n = 100,733, 47.4% male, age 49.1(±15.8) years) included 17.6% current smokers, 23.5% former smokers. Of men and women, respectively, 11.3% and 15.0% were divorced/separated, 2.9% and 11.8% were widowed, and 2.9% and 2.3% were involuntarily unemployed. Ultimately, 667 subjects developed lung cancer; another 10,071 died without lung cancer. Adjusted for age, education, and ancestry, lung cancer was associated with unemployment, widowhood, and divorce/separation in men but not women. Further adjusted for years smoked, cigarettes per day, and years since quitting, none of these associations was significant in either sex. CONCLUSIONS: Once smoking is accounted for, psychosocial stressors in adulthood do not independently promote lung cancer. Given their increased smoking behavior, persons experiencing stressors should be referred to effective alternatives to smoking and to support for smoking cessation.

Race/ethnicity, socioeconomic status, and ALS mortality in the United States.

OBJECTIVE: To determine whether race/ethnicity and socioeconomic status are associated with amyotrophic lateral sclerosis (ALS) mortality in the United States. METHODS: The National Longitudinal Mortality Study (NLMS), a United States-representative, multistage sample, collected race/ethnicity and socioeconomic data prospectively. Mortality information was obtained by matching NLMS records to the National Death Index (1979-2011). More than 2 million persons (n = 1,145,368 women, n = 1,011,172 men) were included, with 33,024,881 person-years of follow-up (1,299 ALS deaths , response rate 96%). Race/ethnicity was by self-report in 4 categories. Hazard ratios (HRs) for ALS mortality were calculated for race/ethnicity and socioeconomic status separately and in mutually adjusted models. RESULTS: Minority vs white race/ethnicity predicted lower ALS mortality in models adjusted for socioeconomic status, type of health insurance, and birthplace (non-Hispanic black, HR 0.61, 95% confidence interval [CI] 0.48-0.78; Hispanic, HR 0.64, 95% CI 0.46-0.88; other races, non-Hispanic, HR 0.52, 95% CI 0.31-0.86). Higher educational attainment compared with < high school was in general associated with higher rate of ALS (high school, HR 1.23, 95% CI 1.07-1.42; some college, HR 1.24, 95% CI 1.04-1.48; college, HR 1.10, 95% CI 0.90-1.36; postgraduate, HR 1.31, 95% CI 1.06-1.62). Income, household poverty, and home ownership were not associated with ALS after adjustment for race/ethnicity. Rates did not differ by sex. CONCLUSION: Higher rate of ALS among whites vs non-Hispanic blacks, Hispanics, and non-Hispanic other races was not accounted for by multiple measures of socioeconomic status, birthplace, or type of health insurance. Higher rate of ALS among whites likely reflects actual higher risk of ALS rather than ascertainment bias or effects of socioeconomic status on ALS risk.

A further study of life expectancy by socioeconomic factors in the National Longitudinal Mortality Study.

OBJECTIVES: The objective of this article is to provide estimates of life expectancy for White, Black, and Hispanic populations by socioeconomic factors. Effects of educational, income, employment, and marital status on life expectancy are presented and interpreted. DESIGN: The National Longitudinal Mortality Study, consisting of a number of Current Population Surveys (CPS) linked to mortality information obtained from the National Death Index, provides data to construct life tables for various socioeconomic and demographic groups. Probabilities of death are estimated using a person-year approach to accommodate the aging of the population over 11 years of follow up. RESULTS: Across various ethnicity-race-sex groups, longer life expectancy was observed for individuals with higher levels of education and income, and for those who were married and employed. The differences in life expectancy between levels of the socioeconomic characteristics tended to be larger for men than for women. Also, differences were found to be larger for the non-Hispanic Black population compared to the non-Hispanic White population. Hispanic White men exhibited patterns similar to those of non-Hispanic White and Black men. CONCLUSIONS: For selected ethnicity-race-sex groups, the impact of socioeconomic variables on life expectancy is dramatic. The shorter life expectancy observed among the poor, the less educated, the unmarried, and those not in the labor force, highlights the impact of socioeconomic disadvantage on survival. Further, the substantial 14-year differential favoring the employed over those not in the labor force may be partially explained by unemployment due to poor health. Another reason may be that employed individuals have greater access to health care than do those not in the labor force.

Individual and neighborhood socioeconomic status and healthcare resources in relation to black-white breast cancer survival disparities.

Background. Breast cancer survival has improved significantly in the US in the past 10-15 years. However, disparities exist in breast cancer survival between black and white women. Purpose. To investigate the effect of county healthcare resources and SES as well as individual SES status on breast cancer survival disparities between black and white women. Methods. Data from 1,796 breast cancer cases were obtained from the Surveillance Epidemiology and End Results and the National Longitudinal Mortality Study dataset. Cox Proportional Hazards models were constructed accounting for clustering within counties. Three sequential Cox models were fit for each outcome including demographic variables; demographic and clinical variables; and finally demographic, clinical, and county-level variables. Results. In unadjusted analysis, black women had a 53% higher likelihood of dying of breast cancer and 32% higher likelihood of dying of any cause (P < 0.05) compared with white women. Adjusting for demographic variables explained away the effect of race on breast cancer survival (HR, 1.40; 95% CI, 0.99-1.97), but not on all-cause mortality. The racial difference in all-cause survival disappeared only after adjusting for county-level variables (HR, 1.27; CI, 0.95-1.71). Conclusions. Improving equitable access to healthcare for all women in the US may help eliminate survival disparities between racial and socioeconomic groups.