Abstract
Objectives. We determined whether community-level income inequality was associated with mortality among a cohort of older adults in São Paulo, Brazil.
Methods. We analyzed the Health, Well-Being, and Aging (SABE) survey, a sample of community-dwelling older adults in São Paulo (2000–2007). We used survival analysis to examine the relationship between income inequality and risk for mortality among older individuals living in 49 districts of São Paulo.
Results. Compared with individuals living in the most equal districts (lowest Gini quintile), rates of mortality were higher for those living in the second (adjusted hazard ratio [AHR] = 1.44, 95% confidence interval [CI] = 0.87, 2.41), third (AHR = 1.96, 95% CI = 1.20, 3.20), fourth (AHR = 1.34, 95% CI = 0.81, 2.20), and fifth quintile (AHR = 1.74, 95% CI = 1.10, 2.74). When we imputed missing data and used poststratification weights, the adjusted hazard ratios for quintiles 2 through 5 were 1.72 (95% CI = 1.13, 2.63), 1.41 (95% CI = 0.99, 2.05), 1.13 (95% = 0.75, 1.70) and 1.30 (95% CI = 0.90, 1.89), respectively.
Conclusions. We did not find a dose–response relationship between area-level income inequality and mortality. Our findings could be consistent with either a threshold association of income inequality and mortality or little overall association.
The distribution of incomes in society has been hypothesized to influence a population’s health status.1 Unequal societies tend to have a greater number of people in poverty who lack access to resources (e.g., health care and preventive measures) to achieve good health. Unequal conditions are also more apt to generate invidious social comparisons that lead to frustration and stress.2 A more contentious claim made by a growing number of researchers is that unequal societies are damaging to the health of everybody—the poor as well as the comfortably well-off.1 The putative mechanism for this effect is that income inequality erodes social solidarity. Reduced social cohesion in turn hampers a society’s ability to provide for many kinds of public goods, such as education, health care, and public health infrastructure.3 For example, when the wealthiest members of society begin to purchase education for their children through private means, or purchase their health care through private channels, there is a corresponding clamor to cut taxes on the rich (since they are no longer benefiting from subscribing to the publicly financed system). Falling tax revenues eventually lead to reduced social spending and declining quality of public institutions for the rest of society.
Although the detailed mechanisms through which growing inequality harms society need to be sketched out more fully, considerable evidence has accumulated on the association between income inequality and the health of individuals. Multilevel analyses have demonstrated that there is an excess risk of morbidity and mortality associated with living in a society with high levels of income inequality, even after adjustment for the confounding effects of individual income.4 In other words, there appears to be a contextual influence of income inequality on the health of individuals, over and above their personal socioeconomic circumstances.
Kondo et al.5 conducted a meta-analysis of all multilevel studies linking income distribution to health, which included 9 longitudinal studies and 18 cross-sectional studies. In the pooled analysis of the prospective cohort studies, the authors reported that each 0.05-unit increment in the Gini index (a summary measure of income inequality) was associated with a 7.8% excess risk of all-cause mortality. Nonetheless, data remain sparse from Latin America, where the degree of income inequality is among the highest in the world. Previous studies have looked at the association between income inequality and health in Chile6 and Brazil,7,8 but these have been cross-sectional or ecological. In addition, debate continues concerning what kinds of individuals are most vulnerable to the harmful effects of income inequality. In the US National Longitudinal Mortality Study,9 the association between higher income inequality and increased mortality risk was shown only among working-age individuals; among older individuals (> 65 years), there was no such association.
We address 2 gaps in the literature. We provide a longitudinal test of the association between community-level income inequality and mortality in São Paulo, Brazil, a country with one of the highest degrees of income inequality in the world. We also provide a test of the income inequality hypothesis in a predominantly elderly population.
METHODS
Data for this analysis come from the Health, Well-Being, and Aging (SABE) survey. In 2000, a sample of community-dwelling older adults (≥ 60 years) in the city of São Paulo was recruited to participate in a retrospective cohort study. The SABE is a multicenter study implemented by the Pan-American Health Organization. The study’s overall objective was to investigate the living and health conditions of older adults in 7 cities in Latin America and the Caribbean: Bridgetown, Barbados; Buenos Aires, Argentina; Mexico City, Mexico; Havana, Cuba; Montevideo, Uruguay; São Paulo; and Santiago, Chile. The São Paulo version of the SABE was the only one to continue as a longitudinal study.
The SABE study has been previously described.10 Briefly, the sample consisted of 2143 community-dwelling older adults living in the urban area of São Paulo who were followed from 2000 to 2007. Participants were selected by clustered sampling. An initial sample of 72 census tracts from the Municipality of São Paulo was chosen by the probability-proportional-to-size approach. The first random sample was composed of 1568 older men and women combined with another random sample of 575 elders taken from a pool of older men and women aged 75 years or older. The respondents were nested within 49 districts (average of 43.7 individuals per district). To offset the higher mortality rate in older men, men were oversampled.
Inclusion criteria were participation in the SABE survey in 2000 and availability of information for all variables. Of the sample, complete data were available for 1024 participants, or 47.8% of the original sample. Information on income and education was missing for 421 and 514 participants, respectively. Those with missing data were more likely to be female. We also did not have the date of death for 89 participants who died; these people were less likely to be female but were more likely to be older than 70 years. In addition, 379 participants, or 17.7% of the sample, were lost to follow-up; these people were more likely to be female, to be aged 65 to 70 years, to have lower income, and to have a high school or a form of postsecondary education.
Study Population
São Paulo, Brazil’s largest metropolitan center, has a population of 10.3 million and comprises 31 subprefectures.11 These subprefectures are divided into 96 districts (average population = 107 697), which may contain 1 or more neighborhoods, or bairros. São Paulo is characterized by social, racial, and economic disparities.12 Although the Brazilian Government has legislated income redistribution policies in the form of progressive taxes, Brazil has a Gini coefficient of 0.51, which is higher than the figure for the United States and China.13
Study Variables
Trained health professionals conducted household interviews to collect sociodemographic data, which included gender, age, income (reported in Brazilian reals), and education. Investigators verified the date of death of a participant using the Mortality Information System of the Municipality of São Paulo.
To measure inequality, we used the Gini coefficient, which has a theoretical range of 0 (perfect equality, where every household earns exactly the same income) to 1.0 (perfect inequality, where households earn a diverse range of incomes). Calculation of the Gini coefficient has been described elsewhere.14 Briefly, the Gini coefficient is based on the Lorenz curve, a cumulative frequency curve that compares the distribution of a specific variable with the uniform distribution that represents equality.14 In the current analysis, we calculated the Gini coefficient in each of the 49 districts using 2000 census data. The mean, median, and standard deviation of the Gini across the 49 districts were 0.27, 0.25, and 0.07, respectively. We then used the distribution of the Gini coefficients to categorize the districts into first (≤ 0.20), second (0.20–0.23), third (0.23–0.27), fourth (0.27–0.32), and fifth (> 0.32) quintiles.
Data Analysis
We matched deaths to state and municipal death records by analyzing name, address, gender, and date of birth. We first calculated cumulative incidence and incidence rates (person-years) of deaths and then stratified analyses by district Gini quintile.
We modeled time to death in Cox proportional hazard regressions. The first model identified the crude relationship between district Gini quintile and risk for mortality. We conducted a second regression model that included only the individual characteristics. We then conducted multiple Cox proportional hazard regression, controlling for gender, age, income, and education. We used Kaplan–Meier curves to test the proportional hazard assumption, which resulted in graphs with parallel curves. Because the participants were clustered within districts, the assumption that individuals are independent from each other could not be made. Therefore, we conducted clustered survival analysis, using the SAS PROC PHREG procedure (SAS Institute, Cary, NC) with the robust sandwich estimate option.15 We also tested an interaction term, Gini × income, to determine whether the effect of income inequality differed between high- and low-income participants, but the results (not reported) were insignificant.
Because 835 participants were missing education or income information, we used multiple imputations to replace these missing data, which resulted in a total sample size of 1675 (78.2% of the original sample). We performed multiple imputations of data as described elsewhere.16 Briefly, we imputed missing values within 5 copies of the data set. We then used survival analyses to fit the model of interest to each of the imputed data sets. Next, we averaged estimates to obtain overall estimated associations.16 We then repeated the survival analyses, as described in the previous paragraph, to determine whether consistent results were obtained.
All responses to the SABE survey were weighted to account for the fact that the survey was developed to generate representative data of the city but not at the district level. We developed poststratification weights to adjust for the extent to which participants may not have been representative of the districts in which they lived. Further explanation of how these weights were determined has been provided elsewhere.17 Older age and being male were the characteristics most correlated with mortality. We therefore developed poststratification weights based on the distribution of the age and gender of survey respondents. We then used these weights to adjust the risk for mortality in the SABE study (using the weight procedure in SAS).
We also examined alternative specifications of the Gini coefficient and repeated the analyses described earlier in this section. Because the Gini coefficient was not normally distributed, we used z transformation to standardize the values into z scores and examined the coefficient as a continuous variable. We also created a dichotomous split using the 20th percentile of the Gini coefficient as the threshold value.
RESULTS
Characteristics of the sample with complete-case, multiply imputed, and poststratification weighted data are shown in Table 1. More than half of the sample was female (n = 551; 53.8%) and the average age was 73.4 years (SD = 8.2). We obtained similar descriptive statistics using weighted data, with the exception of gender. Of the weighted sample participants, more than half were male. Features of the 5 district Gini quintiles are presented in Table 2; for quintiles 1 through 5, the mean Gini coefficient was 0.19, 0.22, 0.25, 0.30, and 0.38, respectively.
TABLE 1—
Characteristics of Study Participants: Health, Well-Being, and Aging Study, São Paulo, Brazil, 2000–2007
| Complete-Case Data (n = 1024), No. (%) or Mean ±SD (Median) | Multiply Imputed Data (n = 1675), No. (%) or Mean ±SD (Median) | Poststratification Weighted Data (n = 1574), No. (%) or Mean ±SD (Median) | |
| Gender | |||
| Male | 473 (46.2) | 684 (40.8) | 901 (57.3) |
| Female | 551 (53.8) | 991 (59.2) | 672 (42.7) |
| Education | |||
| Elementary school | 849 (82.9) | 1168 (69.7) | 1183 (75.2) |
| Some high school | 65 (6.4) | 297 (17.7) | 189 (12.0) |
| High school | 21 (2.1) | 92 (5.5) | 70 (4.4) |
| Higher education | 89 (8.7) | 118 (7.0) | 132 (8.4) |
| Age at baseline, y | 73.4 ±8.2 (74.0) | 73.5 ±8.5 (74.0) | 72.2 ±8.6 (72.0) |
| Income, minimum wage unitsa | 6.5 ±17.2 (2.1) | 7.3 ±14.9 (2.3) | 7.7 ±14.3 (2.8) |
One minimum wage unit is equivalent to 152 Brazilian reals per month (1 real = US $0.46).
TABLE 2—
Characteristics of the 5 Gini Quintiles of Participating Districts: Health, Well-Being, and Aging Study, São Paulo, Brazil, 2000–2007
| Quintile 1 (n = 9)a |
Quintile 2 (n = 11)a |
Quintile 3 (n = 9)a |
Quintile 4 (n = 11)a |
Quintile 5 (n = 9)a |
|||||||
| Mean (SD) | Median (Range) | Mean (SD) | Median (Range) | Mean (SD) | Median (Range) | Mean (SD) | Median (Range) | Mean (SD) | Median (Range) | P | |
| Gini coefficient | 0.19 (0.02) | 0.19 (0.15–0.20) | 0.22 (0.01) | 0.22 (0.21–0.23) | 0.25 (0.01) | 0.25 (0.23–0.26) | 0.30 (0.02) | 0.30 (0.27–0.32) | 0.38 (0.03) | 0.29 (0.35–0.42) | < .01 |
| Living in slums, % | 5.1 (8.2) | 1.1 (0–23.0) | 3.8 (3.5) | 2.7 (0–10.8) | 3.3 (5.3) | 0.6 (0–15.8) | 6.8 (8.6) | 2.5 (0–23.2) | 14.1 (8.3) | 18.1 (1.0–24.1) | .01 |
| Living in poverty, % | 18.6 (25.2) | 6.1 (0–65.5) | 19.5 (21.5) | 15.6 (0–69.1) | 9.1 (13.9) | 1.5 (0–39.7) | 11.4 (11.8) | 4.9 (0–41.2) | 18.7 (11.3) | 18.5 (1.0–35.7) | .60 |
| Median income, reals | 487.0 (457.5) | 236.4 (114.5–1179.6) | 256.6 (138.9) | 195.3 (108.9–584.4) | 622.4 (507.0) | 450.2 (144.1–1526.5) | 430.6 (402.3) | 288.1 (143.2–1454.2) | 286.2 (286.2) | 270.4 (152.6–500.0) | .17 |
| Mortality rate, per 100 000 | 267.8 (55.4) | 297.9 (170.8–325.5) | 318.6 (34.3) | 306.0 (286.5–382.8) | 293.0 (57.6) | 290.6 (186.6–365.5) | 313.0 (53.4) | 311.2 (235.5–423.2) | 284.3 (27.0) | 281.2 (251.6–319.4) | .12 |
| ≥ 65 y old, % | 6.7 (4.2) | 6.4 (2.0–13.2) | 7.8 (4.3) | 7.1 (2.3–15.2) | 10.2 (4.2) | 12.1 (3.8–15.2) | 8.3 (3.8) | 7.9 (2.8–13.9) | 5.8 (2.0) | 5.4 (3.5–9.9) | .17 |
The number of districts in the quintile.
In the sample using complete-case data, the average follow-up was 5.15 person-years. Total cumulative incidence of death during follow-up was 31.7% (n = 325). When stratified by district Gini quintile, the cumulative incidence of death in quintiles 1 through 5 was 15.1% (n = 49), 16.6% (n = 54), 23.1% (n = 75), 20.9% (n = 68), and 24.3% (n = 79), respectively. The mortality rate of the total sample was 0.064 deaths per person-year. When stratified by district Gini quintile, the mortality rate for quintiles 1 through 5 was 0.051, 0.060, 0.083, 0.055, and 0.074 person-years, respectively.
Results of the survival analyses are shown in Table 3. Compared with those living in the first Gini quintile, those in the second (adjusted hazard ratio [AHR] = 1.44; 95% confidence interval [CI] = 0.87, 2.41), third (AHR = 1.96; 95% CI = 1.20, 3.20), fourth (AHR = 1.34; 95% = 0.81,2.20), and fifth (AHR = 1.74; 95% CI = 1.10,2.74) quintiles were more likely to die. We did not observe a dose–response relationship between Gini quintile and mortality risk.
TABLE 3—
Mortality Hazard Ratios Within Residential Neighborhood of Participants, by Gini Quintile: Health, Well-Being, and Aging Study, São Paulo, Brazil, 2000–2007
| Complete-Case Data (n = 1024) |
Multiply Imputed Data (n = 1675) |
Multiply Imputed and Poststratification Weighted Data (n = 1675) |
||||||||||
| Variable | Crude HR (95% CI) | P | AHR (95% CI) | P | Crude HR (95% CI) | P | AHR (95% CI) | P | Crude HR (95% CI) | P | AHR (95% CI) | P |
| Gini quintile | ||||||||||||
| First (Ref) | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | ||||||
| Second | 1.43 (0.87, 2.37) | .16 | 1.44 (0.87, 2.41) | .16 | 1.59 (0.98, 2.59) | .06 | 1.53 (0.98, 2.41) | .06 | 1.91 (1.16, 3.14) | .02 | 1.72 (1.13, 2.63) | .01 |
| Third | 2.02 (1.21, 3.38) | < .05 | 1.96 (1.20, 3.20) | < .01 | 1.69 (1.09, 2.62) | .02 | 1.58 (1.06, 2.38) | .03 | 1.62 (1.07, 2.45) | < .05 | 1.41 (0.99, 2.05) | .06 |
| Fourth | 1.32 (0.77, 2.26) | .31 | 1.34 (0.81, 2.20) | .26 | 1.23 (0.74, 2.04) | .43 | 1.16 (0.73, 1.85) | .52 | 1.28 (0.80, 2.06) | .21 | 1.13 (0.75, 1.70) | .57 |
| Fifth | 1.76 (1.11, 2.79) | .02 | 1.74 (1.10, 2.74) | .02 | 1.50 (0.96, 2.36) | .08 | 1.36 (0.91, 2.04) | .13 | 1.58 (1.03, 2.40) | .03 | 1.30 (0.90, 1.89) | .16 |
| Gender | ||||||||||||
| Male (Ref) | 1.00 | 1.00 | 1.00 | |||||||||
| Female | 0.59 (0.45, 0.78) | < .01 | 0.64 (0.52, 0.78) | < .01 | 0.71 (0.57, 0.88) | < .01 | ||||||
| Age, y | ||||||||||||
| < 65 (Ref) | 1.00 | 1.00 | 1.00 | |||||||||
| 65–70 | 1.92 (1.25, 2.93) | < .01 | 1.90 (1.38, 2.64) | < .01 | 1.84 (1.32, 2.56) | < .01 | ||||||
| > 70 | 4.52 (3.29, 6.22) | < .01 | 3.99 (3.17, 5.03) | < .01 | 4.57 (3.59, 5.82) | < .01 | ||||||
| Education | ||||||||||||
| Elementary school (Ref) | 1.00 | 1.00 | 1.00 | |||||||||
| Some high school | 1.39 (0.93, 2.07) | .11 | 1.29 (1.07, 1.55) | < .01 | 1.19 (0.96, 1.47) | .11 | ||||||
| High school | 1.41 (0.72, 2.75) | .34 | 1.21 (0.86, 1.70) | .27 | 1.13 (0.80, 1.60) | .48 | ||||||
| Higher education | 0.85 (0.46, 1.58) | .6 | 0.81 (0.50, 1.31) | .39 | 0.84 (0.49, 1.45) | .53 | ||||||
| Income quartile | ||||||||||||
| First (Ref) | 1.00 | 1.00 | 1.00 | |||||||||
| Second | 0.69 (0.52, 0.91) | .01 | 0.80 (0.67, 0.96) | .02 | 0.80 (0.63, 1.01) | .06 | ||||||
| Third | 0.67 (0.48, 0.93) | .02 | 0.81 (0.67, 0.98) | .03 | 0.82 (0.67, 1.01) | .07 | ||||||
| Fourth | 0.64 (0.48, 0.87) | < .01 | 0.81 (0.67, 0.98) | .03 | 0.81 (0.64, 1.03) | .09 | ||||||
| Proportion of district living in povertya | ||||||||||||
| Low (Ref) | 1.00 | 1.00 | 1.00 | |||||||||
| High | 1.10 (0.77, 1.57) | .61 | 1.04 (0.79, 1.35) | .78 | 1.07 (0.84, 1.37) | .59 | ||||||
Note. AHR = adjusted hazard ratio; CI = confidence interval.
Low and high refer to below and above the median proportion living in poverty.
When we conducted analyses using the sample with imputed data, hazard estimates differed (Table 3). In multiple regression analyses of imputed data, compared with those living in the first Gini quintile, those living in quintiles 2 through 5 had adjusted hazard ratios of mortality of 1.53 (95% CI = 0.98, 2.41), 1.58 (95% CI = 1.06, 2.38), 1.16 (95% = 0.73, 1.85), and 1.36 (95% CI = 0.91, 2.04), respectively. When we applied the poststratification weights to the imputed data, the adjusted hazard ratios of mortality (relative to the lowest Gini quintile) were 1.72 (95% CI = 1.13, 2.63), 1.41 (95% CI = 0.99, 2.05), 1.13 (95% CI = 0.75, 1.70), and 1.30 (95% CI = 0.90, 1.89) in quintiles 2 through 5, respectively (Table 3). In summary, our analyses of 3 sets of data—the complete-case data, the multiply imputed data, and the data set with both multiple imputations and poststratification weights—indicated elevated hazard ratios for mortality in all of the districts beyond the reference group (the areas with Gini values below 0.2). There was no dose–response relationship between Gini values and mortality risk. Our data could be interpreted either as indicating a threshold effect for Gini values of 0.2 or as evidence of little overall association between income inequality and mortality.
When we treated the Gini coefficient as a continuous variable, we found no significant relationship between income inequality and risk for mortality. When we used 0.20 as the cutoff to dichotomize Gini coefficient values, among the complete-case group, those living in districts with higher inequality were at significantly increased risk of mortality (AHR = 1.61; 95% CI = 1.05, 2.46) compared with those in low-inequality districts. We obtained marginally significant hazard ratio estimates for those in the districts with higher inequality among the multiply imputed samples (AHR = 1.39; 95% CI = 0.94, 2.05) and weighted samples (AHR = 1.36; 95% CI = 0.95, 1.95).
DISCUSSION
In this study based on a sample of elderly city dwellers in São Paulo, Brazil, we found mixed evidence for the hypothesis that income inequality is detrimental to health. Compared with participants living in districts in the first Gini quintile (with least inequality), those living in other district quintiles were at greater risk for mortality. However, we failed to demonstrate a dose–response gradient between Gini values and mortality risk. The empirical results from our study appear to support the possibility of a threshold effect of income inequality on risk for mortality above Gini levels of about 0.20. On the other hand, the absence of a monotonic relationship between income inequality and mortality could be interpreted as being consistent with no overall association.
We conducted analyses using complete-case, multiply imputed, and poststratification weighted data. Although there were differences between the 3 sets of analyses (e.g., the estimated hazard ratios for specific Gini categories were or were not statistically significant depending on the data set), the overall conclusions are similar: (1) the point estimates for hazard ratios of mortality were higher in the upper Gini quintiles than in the lowest quintile (which had the least inequality), and (2) there was no dose–response gradient between Gini quintiles and mortality hazard.
A recent study using SABE data found that income inequality was associated with an increased likelihood of reporting poor self-rated health.17 In that study, there was a monotonic, dose–response relationship between district-level Gini coefficient and the odds of reporting poor self-rated health. Another recent investigation identified an inverse association between life expectancy and Gini index among the 27 Brazilian states,8 which is at variance with our findings. A potential explanation is the geographic scale examined in the 2 studies. Rasella et al.8 used the states as the unit of analysis, whereas we used the smaller districts in 1 city of Brazil.
Our use of districts as the area-level unit might therefore explain the lack of a statistically significant association between income inequality and mortality in our study. Previous studies investigating the relationship between area-level income inequality and mortality have tended to use larger area units, such as municipalities, counties, states, or even nations.5 In US studies, for example, the correlation between income inequality and health has been found to be more robust at large spatial units (states, metropolitan areas) than at lower levels of aggregation (counties, census tracts).4 A possible reason for this discrepancy is that at small scales, households tend to become more similar with respect to income because of economic residential segregation (i.e., households in middle-class neighborhoods resemble each other in terms of their income; similarly, households in high-poverty neighborhoods are equally poor). Nonetheless, smaller area units are more proximal to individuals and therefore smaller spatial scales might be more relevant for social cohesion and other health outcomes.
Most studies of income inequality and mortality have not attempted to incorporate lag effects, or to simultaneously control for a series of preceding income inequalities. Studies that have addressed this issue suggest that income inequality may exert its peak influence on population health up to a decade following exposure to income inequality.18 Using 1986–2004 US National Health Interview Survey data with 1986–2006 mortality follow-up data (n = 701 179),19 researchers investigated the lagged effects of national-level income inequality on individual mortality risk by using a discrete-time hazard model where contemporaneous and preceding income inequalities were treated as time-varying person-specific covariates. The study suggested that income inequality did not have an instantaneous effect on individual mortality risk, but began exerting its influence 5 years later. The association peaked at 7 years and then diminished after 12 years. A limitation of the present study is that we did not have information on the exposure of individuals to preceding levels of income inequality. However, given that the average follow-up was 5.16 person-years in our study, we may have underestimated the impact of income inequality on mortality risk.
Although we found evidence of a relationship between greater inequality and mortality among the elderly, previous research is limited. In the US National Longitudinal Mortality Study, among those aged 25 to 64 years, state-level income inequality was associated with a 40% excess in state-level mortality rates (95% CI = 26%, 56%) for men and a 14% excess (95% CI = 3%, 27%) for women.9 However, no such relationship was found for men or women older than 65 years. Similarly, another study in Norway found a significant positive association between area-level income inequality and mortality only among men and women aged 26 to 66 years.20 Researchers in the United States, studying individuals aged 18 to 74 years at baseline, found that those living in high-inequality states were at higher risk of mortality than those living in low-inequality states.21 A possible explanation for the diminution of the association between income inequality and mortality at older ages in the United States is that Social Security and Medicare play a protective role in shoring up the financial security (and health care access) of retirement-age individuals. Both programs (originating in Lyndon Johnson’s War on Poverty) have played a substantial role in reducing the level of poverty among the elderly in the United States. Another explanation may be mortality selection: people living in areas with higher levels of inequality are more likely to die at an earlier age—thus leaving a more robust older population—compared with people living in areas with lower levels of inequality.
Limitations
A number of additional limitations need to be acknowledged. The large loss to follow-up (17.7%) may have introduced bias, especially in the context of a study where one third of the participants died during the follow-up period. The direction of the bias is likely to be conservative—that is, toward underestimating the mortality hazard in the communities with the highest Gini coefficients. The reason is that the communities with higher Gini coefficients had a higher proportion of residents living below the poverty threshold, and the poor were more likely to be lost to follow-up (AHR = 1.44; 95% CI = 1.29, 1.61) and to be at higher risk for mortality (AHR = 1.11; 95% CI = 1.01, 1.22). This type of differential dropout and loss to follow-up may account for the flat dose–response we found between levels of community income inequality and mortality risk. We also lacked information on specific causes of death, so we were unable to conduct a cause-specific analysis of Gini coefficient and mortality risk. A previous ecological analysis of income inequality and mortality in São Paulo suggests that more unequal districts have significantly higher mortality from homicide, ischemic heart disease, HIV/AIDS, and respiratory illnesses.7
Another limitation is that we did not have a history of how long the residents lived in their district; in other words, the length of exposure to the degree of income inequality was not measured. Similarly, there was no indication of whether participants moved in or out of districts, and therefore we could not measure any change of exposure to income inequality during follow-up. Finally, we used a convenience sample of the districts of São Paulo, and therefore estimates were not representative of the entire city. However, as described in the “Data Analysis” section, we used poststratification weighting to make the sample more representative of the districts in which the participants resided.
The average Gini coefficient across the districts was well below the Gini coefficient of São Paulo. One reason for this is that people from similar income backgrounds tend to cluster together within a district.22 In addition, district boundaries may have been created to keep slums or poorer areas together in one district, while keeping wealthier ones in the same district, resulting in greater equality.7 Another possible reason is that wealthier residents might be underreporting their income as a tax-dodging tactic or through fear of violence.
The strengths of this study include a large sample size; the use of longitudinal analysis, which allowed us to establish temporality; the variability of inequality among the São Paulo districts; and the availability of individual data on death as opposed to a self-reported outcome.
Conclusions
To our knowledge, this is one of the first studies in Latin America to address the longitudinal relationship between area-level income inequality and risk for mortality, above and beyond the effects of individual covariates such as age, gender, household income, and education. Future research should include looking at similar small-area units within developed nations such as the United States. In addition, a better understanding of the mechanisms by which income inequality leads to a greater risk for mortality among individuals is warranted. Caution must be used in interpreting the results because the risk of death resulting from increasing district income inequality was not consistent among complete-case data, multiply imputed data, and poststratification weighted data. Nonetheless, our results indicate that redistribution of wealth to create more equitable societies—such as by raising the minimum wage or improving the personal socioeconomic circumstances of the most vulnerable groups in society—represents an important strategy for improving population health.
Acknowledgments
R. Pabayo is a Canadian Institutes of Health Research (CIHR) postdoctoral fellowship recipient. The Health, Well-Being and Ageing study was funded by Fundação de Amparo à Pesquisa do Estado de São Paulo (grants 12/09717-2 and 09/53778-3).
Human Participant Protection
This study received approval from the Human Research Ethics Committee of the School of Public Health, University of São Paulo. We obtained written informed consent from each participant at the time of his or her interview.
References
- 1.Wilkinson RG, Pickett KE. Income inequality and socioeconomic gradients in mortality. Am J Public Health. 2008;98(4):699–704. doi: 10.2105/AJPH.2007.109637. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Adjaye-Gbewonyo K, Kawachi I. Use of the Yitzhaki Index as a test of relative deprivation for health outcomes: a review of recent literature. Soc Sci Med. 2012;75(1):129–137. doi: 10.1016/j.socscimed.2012.03.004. [DOI] [PubMed] [Google Scholar]
- 3.Kawachi I, Kennedy BP. The Health of Nations: Why Inequality Is Harmful to Your Health. New York, NY: New Press; 2003. [Google Scholar]
- 4.Subramanian SV, Kawachi I. Income inequality and health: what have we learned so far? Epidemiol Rev. 2004;26:78–91. doi: 10.1093/epirev/mxh003. [DOI] [PubMed] [Google Scholar]
- 5.Kondo N, Sembajwe G, Kawachi I, van Dam RM, Subramanian SV, Yamagata Z. Income inequality, mortality, and self rated health: meta-analysis of multilevel studies. BMJ. 2009;339 doi: 10.1136/bmj.b4471. b4471. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Subramanian SV, Delgado I, Jadue L, Vega J, Kawachi I. Income inequality and health: multilevel analysis of Chilean communities. J Epidemiol Community Health. 2003;57(11):844–848. doi: 10.1136/jech.57.11.844. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Chiavegatto Filho ADP, Gotlieb SLD, Kawachi I. Cause-specific mortality and income inequality in São Paulo, Brazil. Revista de Saúde Pública. 2012;46(4):712–718. doi: 10.1590/s0034-89102012005000039. [DOI] [PubMed] [Google Scholar]
- 8.Rasella D, Aquino R, Barreto ML. Impact of income inequality on life expectancy in a highly unequal developing country: the case of Brazil. J Epidemiol Community Health. May 1, 2013 doi: 10.1136/jech-2012-201426. Epub ahead of print. [DOI] [PubMed] [Google Scholar]
- 9.Backlund E, Rowe G, Lynch J, Wolfson MC, Kaplan GA, Sorlie PD. Income inequality and mortality: a multilevel prospective study of 521 248 individuals in 50 US states. Int J Epidemiol. 2007;36(3):590–596. doi: 10.1093/ije/dym012. [DOI] [PubMed] [Google Scholar]
- 10.Albala C, Lebrao ML, Leon Diaz EM et al. The Health, Well-Being, and Aging (“SABE”) survey: methodology applied and profile of the study population [in Spanish] Rev Panam Salud Publica. 2005;17(5–6):307–322. doi: 10.1590/s1020-49892005000500003. [DOI] [PubMed] [Google Scholar]
- 11. Subprefectures in São Paulo. Available at: http://www.Capital.sp.gov.br. Accessed June 9, 2012.
- 12.Marques E. São Paulo: São Paulo: segregação, pobreza e desigualdades sociais (segregation, poverty and social inequalities) São Paulo, Brazil: Senac; 2005. pp. 9–15. [Google Scholar]
- 13. Central Intelligence Agency. The World Factbook. 2012. Available at: http://www.cia.gov/library/publications/the-world-factbook. Accessed July 15, 2012.
- 14.Measuring health inequalities. Gini coefficient and concentration index. Epidemiol Bull. 2001;22(1):3–4. [PubMed] [Google Scholar]
- 15. Gharibvand L, Lei L. Analysis of survival data with clustered events. SAS Global Forum, Statistics and Data Analysis. 2009. Available at: http://support.sas.com/resources/papers/proceedings09/237-2009.pdf. Accessed December 21, 2012.
- 16.Sterne JA, White IR, Carlin JB et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ. 2009;338 doi: 10.1136/bmj.b2393. b2393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Kennedy BP, Kawachi I, Prothrow-Stith D, Lochner K, Gupta V. Social capital, income inequality, and firearm violent crime. Soc Sci Med. 1998;47(1):7–17. doi: 10.1016/s0277-9536(98)00097-5. [DOI] [PubMed] [Google Scholar]
- 18.Blakely TA, Kennedy BP, Glass R, Kawachi I. What is the lag time between income inequality and health status? J Epidemiol Community Health. 2000;54(4):318–319. doi: 10.1136/jech.54.4.318. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Zheng H. Do people die from income inequality of a decade ago? Soc Sci Med. 2012;75(1):36–45. doi: 10.1016/j.socscimed.2012.02.042. [DOI] [PubMed] [Google Scholar]
- 20.Dahl E, Ivar Elstad J, Hofoss D, Martin-Mollard M. For whom is income inequality most harmful? A multi-level analysis of income inequality and mortality in Norway. Soc Sci Med. 2006;63(10):2562–2574. doi: 10.1016/j.socscimed.2006.06.002. [DOI] [PubMed] [Google Scholar]
- 21.Lochner K, Pamuk E, Makuc D, Kennedy BP, Kawachi I. State-level income inequality and individual mortality risk: a prospective, multilevel study. Am J Public Health. 2001;91(3):385–391. doi: 10.2105/ajph.91.3.385. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Watson T. Inequality and the measurement of residential segregation by income in American neighborhoods. National Bureau of Economic Research. 2009. Available at: http://myaccess.library.utoronto.ca/login?url=http://www.nber.org/papers/w14908. Accessed January 21, 2013.