|Home | About | Journals | Submit | Contact Us | Français|
This study examined whether the risk of premature mortality associated with living in socioeconomically deprived neighborhoods varies according to health status of individuals, accounting for individual-level socioeconomic and health-risk factors.
566,402 community-dwelling adults (50–71 years of age) in six US states and two metropolitan areas participated in the ongoing prospective NIH-AARP Diet and Health Study, which began in 1995. This analysis included 565,679 eligible subjects. Data were obtained on dietary, lifestyle, self-rated health status, and medical history on baseline mailed questionnaires. Participants were linked to 2000 census data for an index of census-tract socioeconomic deprivation. The main outcome was all-cause mortality ascertained through 2006.
In adjusted survival analyses of persons in good-to-excellent health at baseline, risk of mortality increased with increasing levels of census-tract socioeconomic deprivation. In comparison, neighborhood socioeconomic mortality disparities among persons in fair-to-poor health were not statistically significant after adjustment for demographic characteristics, educational achievement, lifestyle factors and medical conditions of participants.
Neighborhood socioeconomic inequalities lead to large disparities in risk of premature mortality among healthy US adults, but not for persons who were already in poor health.
Research dating back to at least the 1920s has shown that the United States population has experienced persistent and widening socioeconomic disparities in premature mortality over time.(1–5) However, it has been unclear if socioeconomic inequalities impact the longevity of persons in good and poor health equally. Socioeconomic status (SES) and health status are interrelated(6–8) and both are strong independent predictors of mortality.(9) Low SES is associated with greater risk of ill-health and premature death,(1–5, 8, 10–13) partly due to disproportionally high prevalence of unhealthful lifestyles(10, 14, 15) and physical and mental health conditions.(13, 16) Correspondingly, there is a higher risk of premature mortality in poor neighborhoods relative to more affluent areas.(16, 17) Although the association between neighborhood poverty and mortality is independent of individual-level SES,(17, 18) aggregation of low SES populations in poor areas may contribute to variations in health outcomes across neighborhoods. Conversely, economic hardships resulting from ill-health may lead persons in poor physical or mental health to move to poor neighborhoods.(19) This interrelatedness may create spurious associations between neighborhood poverty and mortality.
Although previous studies have found that the risk of premature death associated with poor health status varies according to individual’s SES,(20, 21) no published studies have examined whether the relative risks for premature mortality associated with living in neighborhoods with higher levels of socioeconomic deprivation vary by health status of individuals. Clarifying these relationships will inform social and public health policies and programs aimed at mitigating the health consequences of neighborhood poverty.(22, 23)
Using data from a large prospective study, we examined whether the risk of premature mortality associated with neighborhood socioeconomic context: 1) differs according to health status at baseline; and 2) remains after adjustment for person-level risk factors for mortality including SES, lifestyle factors, and chronic medical illnesses.
This report used data from the ongoing prospective National Institutes of Health (NIH)-AARP (formerly known as the American Association of Retired Persons) Diet and Health Study. The details of the NIH-AARP study, which began in 1995, have been described previously.(24) The primary goal of the NIH-AARP study is to examine the effect of diet and lifestyle factors on cancer incidence as well as mortality. The original cohort was comprised of 567,169 AARP members 50–71 years of age at baseline who resided in six US states (California, Florida, Louisiana, New Jersey, North Carolina, and Pennsylvania) and two metropolitan areas (Atlanta, Georgia, and Detroit, Michigan). This analysis excluded participants who died or moved out of a study area before the start date of the study or withdrew from the study at any point (n=768), whose date of death was the same as the date of recruitment into the study (n=12), or had incomplete or missing geographical information or census measures (n=710). The National Cancer Institute Special Studies Institutional Review Board approved the NIH-AARP study.
A questionnaire mailed to participants at baseline (1995–1996) asked about demographic characteristics, including age, sex, race and ethnicity, and marital status; level of educational achievement; current body weight and height; and lifestyle factors, including frequency of vigorous physical activity that lasted at least 20 minutes, alcohol use, smoking status and frequency, and medical history. Several components of participants’ diet over a prior 12-month period were assessed at baseline using a 124-item food frequency questionnaire (FFQ).(25) Participants were also asked about their health status with the question: “Would you say your health is:” Possible responses were: “poor,” “fair,” “good,” “very good,” or “excellent.” The NIH-AARP study also collected data from separate questionnaires mailed in 1996–1997 and 2004–2006 that included questions about history of hypercholesterolemia or hypertension (n= 334,497), and the occupation of participants’ head of household during childhood (n=317,933), respectively.
The main predictor in our analyses was neighborhood socioeconomic deprivation. The study collected information on the residential address of each participant at baseline, which was transformed (geocoded) into geographical coordinates and linked to information collected on the 2000 US census about the socioeconomic context and racial and ethnic composition of persons within census tracts. Socioeconomic measures obtained at census tract level have been found to be adequate for assessing the health effects of neighborhood socioeconomic context.(26)
We used the census data to generate an empirical index of neighborhood socioeconomic deprivation as has been described previously.(27) In brief, we considered for inclusion in the deprivation index 19 variables representing domains of racial and ethnic composition, education, income and poverty, occupation and employment, and housing and residential stability. Variable selection was performed using principal component analysis stratified on state-of-residence in order to incorporate both the unique and common contributors to deprivation across all study states.(27) Only one variable was retained in the factor analysis model if two or more within the same domain were highly correlated with each other (correlation coefficient ≥0.90). Variables selected for the index loaded at 0.25 or higher in all states and/or the 95% confidence interval (CI) of the loadings in all states included the median of the lower CI limit of loadings across all sites, which, in our analysis was 0.23 (see Table 1). The final index was computed with the following 10 variables using data on all the census tracts from the 8 “states”: percent of persons in the census tract who had less than high school education, or were unemployed, non-Hispanic blacks (NHB), or in managerial jobs [separately for men and women]; as well as percent of households below 1999 federal poverty levels, on public assistance, or with annual income of <$30,000, no car, or headed by females with dependent children.(see Table 1). The NHB variable did not meet the strict criteria but was retained due to a priori interest. The internal consistency or intercorrelation between the 10 indicators was high (Cronbach’s reliability coefficient = 0.93). In our analyses, the index was categorized into quintiles based on the distribution of the study’s census tracts: the fifth quintile corresponds to the most deprived census tracts.
During follow-up from 1995 through December 31, 2006, vital status was determined through annual linkages of the cohort to the United States Social Security Administration Death Master File. This approach made it possible to obtain complete information on all-cause mortality on the cohort.
Survival models using Weibull distribution and gamma frailties were used to estimate the hazard ratio (HR) and 95% confidence intervals (CI) for death from all causes in relation to census-tract socioeconomic deprivation (neighborhood SES). Cox proportional hazard models (without prescribed frailties) yielded almost identical results. Frailty models accounting for clustering of participants within census tract provide an estimate of variability in mortality between areas of residence for the cohort. There was small, but statistically significant, heterogeneity in the risk of mortality between census tracts (frailty parameter (theta): 0.02 (95% confidence interval [CI] 0.01–0.02), likelihood ratio test p-value <0.001). Time to death was assessed from the date when the baseline questionnaire was received at the study center to the most recent date of vital status ascertainment (December 31, 2006).
We examined two-way interactions among the following variables: self-rated health status (SRH), BMI, physical activity, dietary patterns, education, and neighborhood SES. The statistical significance of interactions was assessed using the Wald test with a Bonferroni correction (to account for multiple comparisons) and the likelihood ratio test. Because we found significant interaction effect between SRH and neighborhood SES, estimates of neighborhood SES on mortality were stratified on SRH. In our analyses, we combined the “fair” and “poor” response categories of the SRH variable because previous studies have found similar health trajectories for these two groups.(28) Our analyses also showed similar neighborhood SES mortality gradients and the poor health group comprised a relatively small proportion of the study population.
To assess the extent to which various factors incrementally accounted for the association between neighborhood SES and mortality, we constructed a base model (model 1) that adjusted for age (continuous), sex, race/ethnicity (non-Hispanic whites [whites], non-Hispanic blacks, Hispanics, others, or unknown), marital status (married or living as marriedvs. others), and state of residence at the time of recruitment into the study. Next, we sequentially added educational achievement (lessthan high school, high school, vocational/technical, some college, college graduate, or unknown) (model 2), followed by lifestyle factors (model 3), and history of these chronic medical conditions at baseline: end-stage renal disease [ESRD], diabetes, heart disease, stroke, emphysema, and cancer (model 4). The lifestyle factors were smoking, Mediterranean diet scores (low, medium or high)(29) and logarithmically transformed total daily caloric intake, body mass index (BMI) (<25, 25–30, >30, or missing), and physical activity (never, rarely, 1 time per month to 2 times a week, 3 or more times a week, or missing). The traditional Mediterranean diet score was computed from reported intake of vegetables, legumes, fruits and nuts, fish and seafood, cereals, meat and meat products, dairy products, ratio of monounsaturates to saturates, and alcoholic beverages.(29) Higher Mediterranean diet scores are associated with lower risk of death.(30) Cigarette smoking was categorized as never smoked, quit but previously smoked <=20 cigarettes/day, quit but previously smoked >20 cigarettes/day, current smoking <=20 cigarettes/day, current smoking >20 cigarettes/day, unknown. BMI-diet and smoking-physical activity interaction terms were included in appropriate models.
Among eligible respondents, we had complete data on 90% of respondents. For our primary analyses, we used multiple imputations with chained equations approach for missing values of smoking status (5.3%), education (5.1%), body mass index (4.2%), race/ethnicity (3.0%), SRH (1.6%), marital status (1.5%), and physical activity (1.1%). Although the use of imputed values did not change our findings in comparison with using missing value indicators or performing complete case analysis, the multiple imputation approach provided more stable results and a consistent analytic sample size across various models. All analyses were performed using STATA Release 11.1 (StataCorp LP, 2009, College Station, TX).
Of the 565,679 eligible subjects at baseline, the mean age was 62 years, 60% were male, 9% non-whites, 9% reported a history of cancer, <1% reported ESRD. Approximately 16% of participants reported their health as excellent, 35% very good, 35% good, 12% fair, and 2% poor. Among those who did not report any of the medical conditions considered in this study, only 6% reported their health as fair and <1% poor (data not shown).
There were a total of 18,592 census tracts across the 6 states and 2 metropolitan areas that contained at least one study participant. The component loadings of the variables were consistent across the census tracts except for percent of non-Hispanic blacks (see Table 1).
Compared to persons residing in the least deprived census tracts, a higher percentage of persons in more deprived areas reported diabetes, stroke, hypertension, emphysema, and fair-to-poor health (p-value <0.001) (data not shown). Table 2 shows the social and demographic characteristics of the study population according to the census-tract socioeconomic deprivation index stratified by SRH at baseline. Compared to census tracts in the first quintile (least deprived), a higher proportion of persons in the more deprived areas were black, not married, or had less than 12 years of education.
Among persons in excellent health, a higher percentage of those in the fifth (most socioeconomically deprived) census-tract quintile were obese (BMI >30kg/m2), had lower levels of physical activity, lower Mediterranean diet scores, and higher caloric intake compared to the first quintile (Table 3). The prevalence of cancer was lower in the most deprived neighborhoods. Among persons in poor health, there was a lower reported prevalence of cancer, heart disease, ESRD and hypertension in the most deprived census tracts.
The maximum follow-up time of the cohort was 11.2 years for a total of 5,643,859 person-years. The estimated overall mortality rate among the cohort adjusted for variables in the base model was 14.1 per 1000 person-years: the rate was 6.5 among those in excellent health, 9.1 for very good health, 15.0 for good health, and 36.8 for those in fair-to-poor health (see also Table 3).
Among persons in good-to-excellent health, the adjusted mortality rate was highest for those residing in the most deprived census tracts and the gap continued to widen over the study period (see Figure 1). In contrast, the pattern of differences stratified by the deprivation index was less consistent among those in fair-to-poor health at baseline. The mortality rates for the cohort according to census-tract socioeconomic deprivation alone (without stratification on SRH) was most similar to estimates for subjects reporting good health (data not shown), and the cumulative mortality curves were between the curves for two extremes defined by health status (see Figure 1).
Table 4 shows the results of Weibull frailty models analyses. The comparison group in all survival analyses was the first deprivation quintile (or least deprived census-tracts). In analyses controlling for variables in the base model (model 1) in good-to-excellent health, the hazard ratio of mortality increased with increasing levels of census-tract socioeconomic deprivation (see Table 4). For instance, among persons in excellent health, the adjusted HR was 1.31 (CI: 1.22–1.41) for those residing in the third quintile and 1.68 (CI: 1.49–1.86) for those in the fifth socioeconomic deprivation quintile. However, the strength of the association between census-tract socioeconomic deprivation and risk of premature mortality decreased with worsening health status. The HRs for persons who reported good health were: 1.15 (CI: 1.12–1.19) for the third quintile and 1.35 (1.29–1.41) for fifth quintile. In contrast, among persons in fair-to-poor health, the HRs for each deprivation quintile were smaller than for those who were in good-to-excellent health (p-value for trend <0.001) and did not exhibit a dose-response pattern (third quintile 1.12 [CI: 1.07–1.16]; fifth quintile: 1.08 [CI: 1.03–1.13]) as shown in Table 4.
Further analyses assessed the impact of adjusting for other risk factors for death. Among persons in excellent health, the association between neighborhood socioeconomic deprivation and mortality was slightly attenuated, but remained stable to further adjustment for education, lifestyle factors, and medical illnesses (5th quintile: HR=1.49, CI: 1.33–1.66). The estimates were similarly stable for participants in good or very good health. However, among persons in fair-to-poor health, the observed relatively small neighborhood SES mortality gradients were no longer statistically significant after further adjustment for the lifestyle factors alone (data not shown).
Previous studies suggest that SRH may not measure physical health in a similar way across strata of socioeconomic groups.(31, 32) Therefore, we performed several sensitivity analyses stratified on age groups, sex, and smoking history (Appendix 3a), as well as restricted to persons who: 1) did not report any of the selected medical conditions at baseline (n=327,195); 2) responded to the questionnaire without the help of a proxy or did not report ESRD (n=565,591); and 3) had at least 2 years of follow-up on the study. We also performed analyses controlling for self-reported history of hypertension or hypercholesterolemia, and paternal occupation on the subgroup of participants who responded to both of the two subsequent questionnaires with these items as described above (n=216,989) (Appendix 3b). These analyses yielded findings similar to our main results. Among persons in fair-to-poor health who did not report medical illnesses, the effect sizes from the base model were similar in magnitude to model 2 shown in Table 4 and were no longer statistically significant after further adjustment for education.
This study used a large prospective cohort to examine whether the risk of premature mortality associated with living in socioeconomically deprived neighborhoods differed according to individuals’ health status. We found that neighborhood socioeconomic mortality disparities were less striking among persons in fair-to-poor health than for those in good-to-excellent health. Among healthy adults, those residing in socioeconomically deprived neighborhoods died at a higher rate than persons in relatively non-deprived areas, even after accounting for individual-level SES, lifestyle and medical history. In contrast, among persons in fair-to-poor health at baseline, there were relatively small neighborhood SES mortality disparities, which were no longer significant after controlling for lifestyle factors.
To our knowledge, there are no previous studies for direct comparison. However, two previous studies found a stronger association between individual-level SES and mortality in persons in good-to-excellent health than for those in poor health.(20, 21) Additionally, consistent with our findings, Waitzman and Smith found area-level SES mortality disparities for 25–54 year olds but not for 55–74 year olds.(33) These findings suggest that socioeconomic advantage does not confer an advantage for persons in poor health. Alternatively, people who were in poor health and resided in relatively poor neighborhoods were as resilient to the effects of ill-health as their counterparts in more affluent areas. Additionally, self reported health (SRH) encompasses a broad range of medical conditions and may change over the life course of individuals.(8, 12, 28) We found a higher prevalence of heart disease and cancer, the two leading causes of death in the US,(34) among those in poor health residing in more affluent neighborhoods. It is plausible that the relative proportion of persons with mental illness [which is more prevalent in poor areas(35–37)] or physical health conditions, or differences in perceptions of health according to neighborhood SES may have contributed to our findings. However, the consistency and magnitude of our results, including the sensitivity analyses, show that our findings could not be attributed to systematic neighborhood SES differences in reporting of SRH or medical conditions as has been suggested previously.(21)
Studies have shown that the mortality risk associated with residing in socioeconomically deprived areas is due in part to poverty within neighborhoods, and the extent of relative inequality between neighborhoods.(38–40) Thus, differences in access to health care resources are likely to contribute to neighborhood SES mortality disparities.(35) Although we did not have data on access to health care, existing literature shows that the relationship between access to health care resources and neighborhood SES varies according to the population studied.(41, 42) Our study population was relatively homogenous: non-whites comprised a relatively small proportion (9%) of our cohort and participants were predominantly older, upper-to-middle class Americans in urban centers. This suggests that the AARP populations may have had similar access to health care services irrespective of neighborhood SES, but this needs further study.
We also sought to determine the impact of adjusting for other risk factors for death on neighborhood SES mortality disparities. Consistent with previous studies, we found a strong association between neighborhood SES conditions and lifestyles and prevalence of chronic medical conditions. Among persons in poor health, the relatively small neighborhood SES mortality disparities were explained by differences in lifestyle. However, the large neighborhood SES mortality gaps among persons in good–to–excellent health were slightly attenuated, but persisted even after adjustment for those factors. This further supports the hypothesis that neighborhood socioeconomic deprivation confers additional mortality risks beyond an individual’s SES and lifestyle factors,(10, 17, 18, 43, 44) and the higher mortality risks are not due solely to the clustering of persons with higher prevalence of unhealthful lifestyles in poor neighborhoods.(7, 15, 16) Our study further extends this evidence and shows that neighborhood SES mortality disparities documented in long-term studies may reflect inequalities among people who were healthy at baseline. Our results suggest that valid analyses on neighborhood SES disparities require a careful assessment of participants’ health status.
There are some other limitations of this study. We used baseline lifestyle and dietary measures obtained in late adulthood, which may not reflect health behaviors throughout the life course. Socioeconomically deprived environments increase the risk of both exposure and vulnerability to adverse health risk factors such as violence, prejudice and segregation, psychological stress, and toxins, pollutants, and other environmental hazards.(19) In addition, the socioeconomic context of the neighborhoods within our study population may have changed during the study period, particularly during the period of economic growth prior to the current global recession. These factors and the potential impact of residential mobility(45) could not be measured in this study.
It is well known that socioeconomic circumstances determine the health status and longevity of individuals.(22, 46) We found that the risk of premature mortality associated with living in socioeconomically deprived neighborhoods varies according to health status of individuals. We found large, long-term neighborhood socioeconomic mortality disparities in healthy individuals, but not in people who were already in poor health. This shows that neighborhood socioeconomic mortality disparities are not due solely to clustering of persons in poor health in poor areas. Our study further shows that poor neighborhood socioeconomic conditions impact longevity of healthy persons beyond personal attributes of demographics, education, or lifestyle of individuals. This represents a major public health challenge and reinforces the need for social policies and programs to mitigate health risks posed by neighborhood socioeconomic deprivation in the US. Future studies are needed for a better understanding of why neighborhood SES mortality disparities vary by health status.
Funding Support: Dr. Doubeni was supported by awards from the National Cancer Institute (5 K01CA127118 & 1R01CA151736-01). Drs. Lian and Schootman were supported in part by an award from the National Cancer Institute (5R01CA137750-02). The NIH-AARP study was conducted as part of an Intramural Research Program of the NIH/National Cancer Institute.
This research was supported [in part] by the Intramural Research Program of the NIH, National Cancer Institute. Cancer incidence data from the Atlanta metropolitan area were collected by the Georgia Center for Cancer Statistics, Department of Epidemiology, Rollins School of Public Health, Emory University. Cancer incidence data from California were collected by the California Department of Health Services, Cancer Surveillance Section. Cancer incidence data from the Detroit metropolitan area were collected by the Michigan Cancer Surveillance Program, Community Health Administration, State of Michigan. The Florida cancer incidence data used in this report were collected by the Florida Cancer Data System (FCDC) under contract with the Florida Department of Health (FDOH). The views expressed herein are solely those of the authors and do not necessarily reflect those of the FCDC or FDOH. Cancer incidence data from Louisiana were collected by the Louisiana Tumor Registry, Louisiana State University Medical Center in New Orleans. Cancer incidence data from New Jersey were collected by the New Jersey State Cancer Registry, Cancer Epidemiology Services, New Jersey State Department of Health and Senior Services. Cancer incidence data from North Carolina were collected by the North Carolina Central Cancer Registry. Cancer incidence data from Pennsylvania were supplied by the Division of Health Statistics and Research, Pennsylvania Department of Health, Harrisburg, Pennsylvania. The Pennsylvania Department of Health specifically disclaims responsibility for any analyses, interpretations or conclusions. Cancer incidence data from Arizona were collected by the Arizona Cancer Registry, Division of Public Health Services, Arizona Department of Health Services. Cancer incidence data from Texas were collected by the Texas Cancer Registry, Cancer Epidemiology and Surveillance Branch, Texas Department of State Health Services. Cancer incidence data from Nevada were collected by the Nevada Central Cancer Registry, Center for Health Data and Research, Bureau of Health Planning and Statistics, State Health Division, State of Nevada Department of Health and Human Services. We are indebted to the participants in the NIH-AARP Diet and Health Study for their outstanding cooperation. We also thank Sigurd Hermansen and Kerry Grace Morrissey from Westat for study outcomes ascertainment and management; and Michael Spriggs and Leslie Carroll at Information Management Services for data support and analysis.