In the industrialised world, no matter how health is measured—as life expectancy, rate of disease or perceived quality of life—those with more socioeconomic resources do better.1 2 In the United States and other developed countries, higher socioeconomic status (SES) is associated with a lower prevalence of high blood pressure,3 4 cardiovascular risk factors,5 depression,6 heart disease,7^–11 disability12 and with longer life expectancy.13 With a few exceptions (eg, melanoma, breast cancer) in which the opposite associations occur and some variation among subgroups,14 higher SES generally protects physical and mental health in the industrialised countries.

Given the strength of these associations, one might expect that more resources would also be linked to better health in the developing world. One early review of the literature from several developing countries including Chile, Brazil and Colombia showed that increased SES was associated with decreased blood pressure.15 A review published in 1998, however, of 13 studies reporting associations between SES, measured by multiple indicators, and blood pressure revealed mixed results.4 Recent studies have also shown mixed results, which may be due to methodological differences, heterogeneity of samples or differences in the degree of economic development.16

One study from Nigeria using a one-city sample of sedentary adults aged 30–60 years found results similar to those from the developed world.17 In a reversal of these findings and of the Whitehall study of British civil servants, however, a study of civil servants in Nigeria found that those at higher occupational levels had higher blood pressure when compared with those at lower occupational levels.18 Similarly, a convenience sample of rural Indian adults from the highest SES group showed almost double the prevalence of hypertension as those from the lowest SES group.19 A younger, adolescent sample in India also showed a small but significantly positive relationship between SES (self-reported family social class) and systolic blood pressure (SBP).20 In a sample in Jamaica, blood pressure and the prevalence of hypertension were highest in the wealthiest women compared with poorer women.21

Comparisons across existing studies are difficult as a result of the use of small non-representative samples, a variety of design flaws, the use of different measures of health outcomes and the measurement of SES. Although household income has been shown to correspond to general measures of health22 and obesity,23 it is problematical because of the complex nature of families and work.24 The measurement of household income in the developing world is particularly difficult, given that household members may contribute sporadically to income and household size may vary. Educational attainment has been widely used as a measure of SES in industrialised and developing countries and is related to many health outcomes;25^–28 educational attainment may reflect a household’s ability to cope with external shocks.24 29 Current occupation is another common measure of SES linking economic factors to health outcomes but may be difficult to use in the context of extreme poverty.22 30 Measurements of housing and assets are commonly used in the developing world as a proxy measure for household income because these variables have been shown to provide good estimations of the economic concept of consumption.29 31 An alternative to measures of specific aspects of socioeconomic resources is an assessment of individuals’ perceptions of where they stand on the socioeconomic hierarchy, which captures multiple aspects of socioeconomic position.32 These ladders have been used in studies in different countries and have shown significant associations with health-related outcomes.33^–36

The aim of the study reported here was to examine the nature of the SES gradient among low-income residents of a middle-income country by determining the association of SBP with measures of different aspects of SES. We hypothesised that at least some of the SES measures would be positively related to blood pressure, given the existing data from developing countries, but did not predict which specific ones would be. The population under study is unique in that it is a large sample of women living in impoverished, rural areas of Mexico. The sample has a relatively limited range of SES and has an annual income per capita (US$2500) much lower than the national average per capita income of approximately US$10 000. Even within this relatively poor segment of the population, however, there is some variation in the key components of SES and the associations may provide insights into the health effects of changing socioeconomic conditions that accompany economic development.

METHODS

Data source

The survey reported here was conducted in 2003 among 9362 women, over 18 years old, from 364 communities across Mexico. This was part of a National Social Welfare Survey, which was designed to be representative of the poorest (income <20th percentile), rural (defined as towns with <2500 inhabitants), communities in seven of Mexico’s 31 states.37 Within states, households were selected in two stages: first by identifying low-income communities in each state and then by choosing low-income households within those communities.38 Households identified in advance from census records were surveyed by a trained field worker, most of whom were nurses, on all days of the week except Sundays from September to December 2003. In most cases (>95% households), one adult woman in each household was interviewed. If more than one adult woman was present, however, up to two were assessed in each home.

Measures

Blood pressure

SBP was measured using a mercurial sphygmomanometer. Three blood pressure measures were obtained from each respondent: one upon arrival in the home, the second 25 minutes later and the third 25 minutes thereafter. For the purposes of the analysis described here, the mean values of the three measures for SBP were used as dependent measures.

Anthropometry

Height and weight were measured using standard techniques.39 40 Body mass index (BMI) was calculated as weight in kilograms divided by the square of height in meters. Overweight was defined as BMI ≥25–<30 and obesity as BMI ⩾30.41

Education

Respondents were asked to report the highest year of education they had attained and responses were categorised as two indicator variables (completion of some elementary school; completion of some secondary school), with reference to the baseline group (0, no education).

Occupation

Survey participants reported their current occupation in response to an open-ended question. Women were asked whether they were working as domestic servants, in a family business or whether they owned a business. As a result of the small number of women working outside the home, occupation was coded as one indicator variable (0, not working/housewife; 1, working outside the home in any capacity).

Household income

Household income was divided by household size to generate per capita household income, which was then divided into tertiles.

Housing and assets

An additional measure of SES was generated, which was a summary measure of household assets (12 variables, including car, van, refrigerator, blender, television, gas heater, boiler, radio, stereo, video cassette recorder, washing machine and fan) and housing quality (six variables, including quality of roof, wall and floor, number of rooms, presence of indoor bathroom and presence of indoor electricity). The housing and asset measurements were obtained in a baseline survey that was carried out in 1997 in a subset (approximately two-thirds) of these households. Principal components analysis was used to summarise the housing and asset variables into one measure24 42 and the first principal component was retained for use in the analyses.43

Subjective status

Respondents completed the MacArthur Scale of Subjective Social Status. They were shown a graphic of a ladder consisting of 10 rungs and then made two ratings: (1) self-perceived status within their community, with no reference to socioeconomic considerations; (2) self-perceived status within Mexico in terms of educational attainment, income and occupation. Each ladder was also divided into tertiles signifying “low status”, “mid status” and “high status”.

Data analysis and statistical methods

Statistical analyses were conducted using Stata 9.2 for Windows (Stata Corporation, College Station, Texas, USA). Sampling design (cluster sampling) was taken into account for the estimations and survey commands were used for all analyses, which make adjustments to account for sampling weights, clustering and stratification. For SBP, non-plausible values and outliers greater than 3 SD larger or smaller than the mean were identified and excluded (n  =  4 values removed). Women younger than 18 years of age (n  =  65) were not included in the analysis and neither were those older than 65 years (n  =  233) or those missing a value for age (n  =  13). The final sample for analysis was 8944 for analyses using SES, 8917 for analyses using BMI and 8912 for analyses using additional demographic variables (marital status and occupation). There were no systematic differences between participants with missing data and those with complete data.

Descriptive statistics were generated, as were correlations among SES variables. In order to examine the associations between SBP and various measures of SES, linear regressions were performed with SBP as the dependent variable and the following variables added to the statistical models: age, educational attainment, household income and occupation, community ladder, country ladder, BMI and marital status. Given the important contribution of BMI to SBP, interactions between BMI and each measure of SES were probed and plotted using standard techniques.44 45 All analyses were adjusted for clustering at the household level and robust standard errors are reported.

RESULTS

SBP findings were available for 9079 women, and the mean SBP was 122.2 mm Hg (table 1). The SES data reflect the very high level of poverty of the households targeted for recruitment into this study; the median annual household income was approximately US$2500. The large majority of participants had completed no schooling or only some primary school and most women were not working outside the home. Almost 60% of the women were overweight or obese and 11% had current, uncontrolled hypertension.

There was a small but significant correlation of education and per capita household income (r  =  0.09, p<0.001). Educational attainment also showed a significant, moderate correlation with higher scores on the community (r  =  0.15, p<0.001) and country (r  =  0.11, p<0.001) ladders and a strong correlation with the assets and housing measure (r  =  0.22, p<0.001). Per capita income was correlated with the community ladder (r  =  0.08, p<0.001), the country ladder (r  =  0.07, p<0.001), working outside the home (r  =  0.05, p<0.001) and the housing and assets measure (r  =  0.20, p<0.001). Housing and assets were correlated with both the country (r  =  0.13, p<0.001) and the community (r  =  0.11, p<0.001) ladders. Working outside the home was also correlated with the assets and housing measure (r  =  0.04, p<0.001). In bivariate correlations, SBP was significantly positively correlated with age (r  =  0.38, p<0.001), BMI (r  =  0.26, p<0.001), per capita household income (r  =  0.04, p<0.001), the assets and housing measure (r  =  0.08, p<0.001), position on the community ladder (r  =  0.03, p = 0.01) and being married (r  =  0.03, p = 0.006). SBP was negatively correlated with educational achievement (r  =  −0.13, p<0.001).

As shown in model 1 (table 2), higher age-adjusted SBP was associated with higher per capita income, when comparing the highest and lowest tertiles (b  =  +2.26, p<0.001), when comparing the middle and lowest tertile (b  =  +1.08, p = 0.004) and when comparing the highest and middle tertiles (b  =  +1.18, p = 0.001, data not shown). There was a non-significant trend for age-adjusted SBP to be negatively related to education in model 1 (secondary school versus no school: b  =  −0.97, p = 0.08). There was a significant interaction between educational achievement and per capita income (fig 1). At low levels of income, there were no differences in SBP by educational attainment; however, at higher levels of per capita income, those women who had not attended any formal schooling had significantly higher SBP than those who had achieved some secondary school: (interaction term: b  =  −1.14, p = 0.04).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Predicted simple slopes for household income per capita and educational attainment as independent variables and age-adjusted systolic blood pressure as dependent variable in adult Mexican women, plotted according to standard techniques.41 Equation for interaction: Ŷ  =  1.41 E1 + 1.74 E2 + 1.77 INC − 0.89(E1 × INC) − 1.48 (E2 × INC) + 121, where E1 is education dummy for primary school, E2 is education dummy for secondary school, BMI is body mass index and INC is per capita household income (log-transformed).

BMI was entered in model 2 (table 2) to see if the income effects were caused partly by changes in diet afforded by more income. BMI showed a strong association with SBP (b  =  +0.57, p<0.001) and the association of income and SBP was attenuated slightly (highest versus lowest tertile: b  =  +2.02, p<0.001; middle versus lowest tertile: b  =  +1.04, p = 0.004; highest versus middle tertile: b  =  +0.98, p = 0.007, data not shown) once BMI was entered. Unexpectedly with adjustment for BMI, the effect of secondary school education became highly significant, with women with some secondary school education having lower blood pressure than those with no education (b  =  −1.77, p = 0.001); women with secondary education also had lower age and BMI-adjusted SBP than once BMI was entend with only a primary education (b  =  −1.04, p = 0.007, data not shown) once BMI was entered.

In models 3 and 4, subjective evaluations of status in one’s community and in the country as a whole, respectively, were added. The community ladder showed a significant positive relationship with SBP (b  =  +0.23, p = 0.001) but the country ladder was not significantly linked to SBP (b  =  +0.06, p = 0.41). In both models, educational achievement also retained its negative association with SBP; per capita household income retained its positive association with SBP, as did BMI.

Finally, in model 5 we added marital status and occupational status to test whether these variables modified the associations of the SES variables with SBP. Although both marital status and occupation had significant relationships to blood pressure—married women (b  =  −1.09, p = 0.001) and those working outside of the home (b  =  −1.28, p = 0.002) had lower blood pressure—adjustment for these variables did not change any of the other associations.

Given the limitations of using household income as an independent variable in a low-income sample without regular sources of income, we repeated the analyses using a measure of housing and assets in place of income; the subsample for whom these data were available had also been assessed 5 years before the current study. As shown in model 1 (table 3), higher age-adjusted SBP was associated with a higher housing and assets score (b  =  +0.32, p<0.001) and this association was completely attenuated (b  =  +0.08, p = 0.37) with the inclusion of BMI, which was highly significant (b  =  +0.60, p<0.001) (model 2). As in the models presented using per capita household income, the completion of secondary school was negatively associated with SBP (b  =  −1.55, p = 0.02) with the inclusion of BMI and continued to be significant (b  =  −1.66, p = 0.01) with the inclusion of the SES ladders, being married (b  =  −1.57, p<0.001) or working outside the home (b  =  −1.28, p = 0.007) (model 3).

To understand why education alone was not related to blood pressure but became significant when adjusted for BMI, we evaluated whether there was an interaction between education and BMI. As seen in fig 2A, the interaction between educational achievement and BMI, adjusted for age, was significant (interaction term: b  =  −0.26, p = 0.02). At low levels of BMI, there were relatively small differences in blood pressure among women with different degrees of education. At higher levels of BMI, however, the educational differences increased and women with no education or some primary education had higher SBP than women with some secondary education. There was no difference between women who had received some primary education and those who had received no education (b  =  3.05, p = 0.20). In brief, there was a protective effect of secondary education at high levels of BMI, whereas education was not important for blood pressure at low levels.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2 Systolic blood pressure by body mass index and educational attainment (A) and subjective social status (B).

Similarly, there was a significant interaction between the country ladder and BMI, although the pattern was somewhat different (fig 2B). Among women with low BMI, there was a direct gradient; women rating themselves lowest on the country ladder had the lowest SBP, followed by those in the middle. Those rating themselves highest on the ladder had the highest SBP (interaction term: b  =  −0.03, p = 0.05). At high levels of BMI, however, the association reversed, with those lowest on the ladder showing the highest SBP and those highest on the ladder having relatively lower SBP. The associations at the high end of the BMI scale were equivalent to the interaction findings for education.

When the interaction term (BMI × ladder) was included in model 4 (table 2), the main effect for the country ladder also became significant (b  =  0.89, p = 0.04), which was not apparent in the original analyses. The results indicate that the country ladder is associated with SBP in the same positive direction as the community ladder but that the results are not significant without the inclusion of the BMI × ladder interaction. No other interactions between BMI and SES variables (income, occupation or community ladder) were significant.

DISCUSSION

In this sample of low-income rural women in a middle-income country, two aspects of SES (educational attainment and working outside the home) showed an inverse association with SBP, whereas three other aspects (income, a housing and asset index and status within one’s community) showed a positive association. This is the first time that associations between SBP and differing measures of SES have been shown to go in opposite directions within the same population. These findings indicate that in the context of a poor population in a middle-income developing country, having a higher SES is not necessarily as protective of health as it is in the developed world and that different aspects of SES provide different types of protection as well as of risk.

There are several mechanisms by which higher SES may contribute to lower blood pressure in industrialised nations. More advantaged populations in these countries generally exercise more, have access to better diets, have less exposure to chronic stressors and benefit from more social support.2 The question remains as to why, in the context of developing or middle-income countries, some aspects of higher SES would be linked with worse, rather than better, health outcomes. Some have argued that the positive associations between SES and SBP could be caused by dietary factors. Rapid changes have occurred over the past two decades in Mexico, and traditional bean and corn-based diets have been supplanted by industrialised food, even in very rural populations.46 47 Although we did not have data on dietary intake to allow for a test of diet as a mediator of income effects, we tested this pathway indirectly by looking at the role of BMI and our results suggest partial mediation through diet. Increased income among very poor populations may, among other things, be used to purchase “unnecessary” calories. This relationship may disappear or even be reversed with further economic development if those with more income shift consumption to healthier foods more quickly than those in poverty.

Another possible mechanism explaining the positive relationship between SES and SBP could be that those who succeed in periods of transition experience greater psychological stress, which may contribute to an increased risk of hypertension.48 Within tight-knit communities, having more resources than neighbours, family or friends can be a source of social strain. Social connections, which can provide support when needed, can also be a source of demands. Similarly, those having higher status within the community may disproportionately be turned to for help. Another possible mechanism is through the kinds of efforts needed to obtain higher income and status. High effort active coping, termed “John Henryism”, considered to be a measure of striving, was shown to be a partial mediator of the relationship between higher relative SES and higher blood pressure in Nigerian civil servants.49 We do not have a direct measure of striving and cannot evaluate its potential role in the positive associations with SES. The finding that increased social status relative to neighbours and relatives was associated with higher SBP is, however, consistent with this hypothesised pathway.

Another critical question arising from the current findings is why some measures of SES (eg, education) operate differently from other measures (eg, income and social ladder) and why they operate differently at varying levels of BMI. We speculate that secondary education provides women with knowledge and skills that help them to avoid health problems. Although these resources may be less important when women have low BMI, they may come more to the fore when the women are overweight. In contrast, higher income may provide more disposable resources that may be used partly for purchasing calorie-dense foods, contributing to greater BMI and higher blood pressure.

We have shown that at equivalent values of BMI, those women with at least some secondary education have a lower blood pressure than women with less education who have the same BMI; we showed a similar interaction for perceived social status. A recent study showed that multiple measures of SES (education, housing, occupation, income, perceived status) were positively associated with BMI in this same population of low-income Mexican adults.50 The findings reported here add to this previous analysis by suggesting that in spite of this consistent positive association between SES and BMI, the contribution of BMI to blood pressure varies according to SES.

The strength of the study was that the sample was large and representative of the poorest population living in seven states in rural Mexico, which is an under-researched group of critical importance to policy makers. Height and weight were measured by nurses, so the values for BMI were not biased as if they had been self-reported measures. A wide range of measures of SES was collected, including current and past measures, which responds to a recent call for the inclusion of multiple measures of SES.51 In addition, subjective social status was assessed, which has not previously been included in such a large survey.

In spite of the strengths of the study, some limitations are apparent. First, the findings are limited in generalisability to Mexican adults in the poorest segment of the population, because all study participants were from the lowest quintile of the income distribution. Second, the cross-sectional nature of the design means that it is not possible to comment on the causal relations among variables. Third, there are several mechanistic pathways (eg, salt consumption) that remain unexplored. A final limitation is that we did not have complete data for diastolic blood presssure and thus used just SBP in our analyses. We repeated our analyses for the large subset of women who had diastolic blood presssure data available, however, and the results were virtually identical to those using SBP.

The prevalence of high blood pressure is increasing rapidly in low-income and middle-income countries.52 Non-communicable chronic diseases, such as hypertension, contributed to approximately 60% of the 56.5 million total reported deaths in the world and to approximately 46% of the global burden of disease in 2001.53 Understanding the risk factors for these diseases in the context of rural poverty is a critical first step in developing effective interventions for this population.

In contrast to traditional assumptions about the relationships between SES and health, we found that women in low-income rural populations who are at the upper end of the income spectrum within their community are more likely to have higher SBP, as are those who perceive that they have higher status in the community. Relatively more resources and status may thus not always be protective. Given that our findings relate to women living within a curtailed range of SES, future research should examine what patterns emerge across a wider range of incomes; analyses should also be expanded to urban samples where resource availability and constraints might be different. Future work should also explore the pathways at work mediating the associations between SES, SBP and BMI in low-income women, including psychosocial variables such as depression.