Ecological study of socio-economic indicators and prevalence of asthma in schoolchildren in urban Brazil

The prevalence of asthma is increasingly high in affluent countries[1]. This has been attributed by the hygiene hypothesis to lower exposure to infection resulting from improved living standards. Challenging this hypothesis, there is growing evidence that in many affluent countries the prevalence is higher among those in low socio-economic status (SES) [2–11]. These socio-economic differentials in asthma support a role of environmental factors in the development of asthma. The fact that these factors might be susceptible to change raises the possibility of identifying effective interventions to prevent asthma. In addition, if the disease burden and risk factors for asthma vary in different groups of the society, specific control measures can be targeted to these groups.

Brazil has a very unequal income distribution[12], with consequent pockets of poverty. There is some evidence that in Brazil the prevalence of asthma is higher in non-white children and in children with low maternal schooling[13, 14]. In this ecological study we examined the relationship between estimates of wheezing prevalence based on surveys conducted in children in different Brazilian cities and explored whether these estimates were related to common health and socioeconomic indicators measured at the population level in the same cities.

We searched the literature in the medical databases (PubMed and Scielo) and in the annals of scientific meetings, retrieving population-based surveys of asthma that were conducted in Brazil using the methodology defined by the International Study of Asthma and Allergies in Childhood (ISAAC). We used the following inclusion criteria: (1) surveys conducted among children aged 6–7 years or 13–14 years in Brazil between 1994 and 2003, (2) use of the official Brazilian translation of the ISAAC questionnaire [15], and (3) the study sample was chosen by probabilistic sampling and was representative of the city.

We performed separate analyses for the age groups 6–7 years and 13–14 years. First, we conducted bivariable analysis of the relationship between asthma prevalence rates and health and socio-economic indicators by visual inspection and using the linear regression between proportion of asthma and each study variable one at a time, weighed by the inverse of the variance of each survey. For each age group, we selected the 4 variables with the lowest P value for multivariable analysis.

Multivariable analysis was performed using linear regression models weighed by the inverse of the variance. Fifteen linear regression models were built with all combinations of the selected four variables for each age group (all-possible-regressions selection procedure) [17]. The model with the smallest mean square error (MSE) and the highest value of adjusted R square was selected. Calculations were done with and without log transformations of the prevalence rates (outcomes); given that results were very similar (same variables were selected and similar P values and patterns as in the Figure 1) we only present the models without log transformation for ease of interpretation. Results from linear regression were: the regression coefficient, which can be interpreted as the prevalence points related to the change in 1 unit of the study variable; and the corresponding P value.

Surveys conducted in 20 cities in Brazil met the study criteria providing 13 prevalence estimates for the age group 6 to 7 year and 19 estimates for the age group 13 to 14 years [16, 18–26]. When the survey results were published more than once, information from the first published report was included, as in all cases the rates were similar. We excluded reports from 5 surveys because of the following reasons: different age group (2 to 10 years) [27]; sampling procedure was not described [28]; presented only combined results for more than one study site [29]; study population was not a city but a selected area of the city and area-specific socio-economic data comparable to the other study sites were not available [30]; change of the question on wheezing in the questionnaire [31], as it has been argued that results might not be comparable with those from other surveys [32].

At visual inspection (Figure 1), variables that seemed to have a clear pattern in relation to asthma prevalence were those with statistically significant coefficients in the linear regression or that remained in the selected models. Other variables studied had an erratic pattern at visual inspection (data not shown).

For children aged 6–7 years, the best multivariable model (adjusted R = 0.69) included the variables mortality for external causes rate, percent of houses with poor sanitation, and infant mortality at the year of birth. Only mortality for external causes showed a significant result (P = 0.007), the coefficient (β = 0.15; 95% C.I.: 0.05; 0.24) being interpreted as a rise in asthma prevalence of 0.15 percent points for each 1/10,000 increase in mortality for external causes; that is, an increase of 33.3/10,000 in mortality for external causes would increase asthma prevalence by 5 percent points.

For children aged 13–14 years the best multivariable model (adjusted R = 0.68) included two variables: illiteracy rate (β = -1.06; 95% C.I.: -1.50; -0.62) and infant mortality rate at the survey year (β = 0.47; 95% C.I.: 0.27; 0.68). An increase in the illiteracy rate of 1 percent point would decrease the prevalence of asthma by 1.06 percent points; that is, an increase in illiterate rate of 4.72 percent points would result in an decrease in asthma prevalence by 5 percent points. Conversely, An increase by 1 per 1,000 in infant mortality in the survey year would increase the prevalence of asthma by 0.47 percent points, and an increase by 10.64 in infant mortality would result in an increase by 5% percent points in asthma prevalence. No evidence of colinearity was found as standard errors of the estimators only changed slightly in different models.

In this ecological analysis, there is evidence that health and socioeconomic indicators were associated with asthma prevalence at ecological level, prevalence increasing with worsening of socio-economic conditions: poorer sanitation and higher infant mortality at birth (for 6–7 years) and at survey year, GINI index, external mortality; but decreased with higher illiteracy rates.

The ecological nature of the study, calls for a cautious interpretation of results, since associations with asthma found at the city level may not reflect associations at the individual level. The number of surveys is relatively small and random variation may also partly explain some of our findings. We used statistical procedures and visual inspection to guide interpretation.

Most indicators were not statistically associated with asthma prevalence. Mortality for external causes, infant mortality at the survey year (mortality 1), poor sanitation and infant mortality at the year of birth (mortality 2) remained in the best model, suggesting a higher prevalence in the cities with the worse living conditions. In contrast, higher literacy rates were associated with lower asthma prevalence. Two of the variables studied are indicators of socioeconomic inequality rather then poverty itself: mortality rate for external causes and GINI index, strongly associated in the bivariable analysis. Unless inequality in these cities is uniquely indicator of poverty at individual level, which seems unlikely, our findings suggest that asthma is also associated with population-based measures of socioeconomic disparity.

How consistent are our results with other investigations of the role of socioeconomic factors and prevalence of asthma? There is some evidence that gross national product (GNP) per capita may be associated with prevalence of wheezing in the last 12 months [33]. While high income is generally considered a positive factor for health, in more developed societies relative poverty and income seem to be more strongly related to health than absolute measures of poverty and income, specially in the field of non-communicable diseases [34]. Indeed, the evidence of association between GNP and wheezing prevalence was found to be only moderate and among children aged 13–14 years [33]. SES indicators measured at group level and not only at individual level can be associated with asthma morbidity [2, 3, 35, 36]. SES indicators measured at an individual level are primarily related to family resources and living standard, for instances; whereas indicators at a collective level are related to, for example, community resources, social barriers and outdoor environmental exposures. Since SES indicators at individual level and at group level are likely to reflect different biological mechanisms, the use of multilevel models would be indicated to investigate the role of socioeconomic factors in the causation of asthma and their causal pathways. A further complication is that asthma is likely to be a syndrome [37], the final presentation of different aetiologies and pathways, and the relationship with SES indicators may result from the balance between effects on different types of asthma.

Finally asthma detection is very sensitive to severity; and it is possible that what we are interpreting as higher prevalence of asthma is in fact higher prevalence of severe or uncontrolled asthma. It is therefore possible that higher estimates of prevalence in the poor reflect differential access to high quality asthma care, which reduces severity. Also, the proportion of asthma cases reported by individuals of different SES is likely to differ (e.g. according to the level of education, access to health care).

In conclusion, we found evidence of the association between asthma prevalence in urban centres of Brazil, a middle income country with high social disparity, with poor socio-economic indicators and inequality. To explore this better, further studies should investigate the relation between prevalence rates of different types of asthma and asthma severity, and social conditions at both the area and the individual levels. In addition, a meta-analysis based on individual and group data from the present surveys could be performed in collaboration with different research centres, as long as individual data on SES and asthma risk factors, such as schooling or income, are available.