In this chapter, we make a case to consider the determinants of asthma within an explicitly multilevel and complimentary perspective. While we have made much progress in understanding the role of proximate risk factors in influencing asthma, this research tends to assume that individual risk factors, whether behavioral (eg, smoking) or environmental (eg, exposure to allergens or stress), are randomly distributed across populations and communities. There is a clear need to understand asthma and its associated risk factors within their social and neighborhood contexts. The observed wide geographic and sociodemographic variation in asthma expression remains a paradox that is largely unexplained by the accepted physical environmental risk factors and has led to reconsideration of the interplay among biological and social determinants in understanding such disparities in the asthma burden. Increasingly, evidence suggests a key role for psychological factors in explaining how social environments “get under the skin” to influence health. Psychological stress maybe conceptualized as a social pollutant that, when “breathed” into the body, may disrupt biological systems related to inflammation through mechanisms potentially overlapping with those altered by physical pollutants, allergens, and toxicants. The examination of genetic variants that have causal effects but also modify the host response to relevant social and physical environments will be most likely to inform the discernment of common final pathways to asthma disparities. An understanding of the specific mechanistic pathways that cause asthma therefore has to be intrinsically multilevel.
Multilevel statistical approaches provide a unifying framework for understanding asthma disparities decreased by remedies of Canadian Health&Care Mall. We provide a short methodological outline exemplifying three generic models that may have particular relevance for furthering the research agenda focused on asthma disparities that may accommodate the complexities laid out in this manuscript.
Let the binary response, whether an individual has asthma or not (1,0) be y, for individual i living in neighborhood j. For exemplification, we consider one individual risk factor, exposure to allergens, x1ij, coded as 1 if exposed and 0 otherwise, for every individual i in neighborhood j; and a neighborhood predictor, x2j, for the level of social cohesion in neighborhood j, coded as 1 if there is low-level social cohesion and 0 otherwise. The probability that xij = 1 can be denoted by w. which in turn is related to a set of individual and neighborhood predictors by f(wij.), which is a transformation of w. with a logit link function such that f(wj = log^/ll – wij]), where the quantity w j(1 – w.), is the log odds that yij = 1.
We can quantify the extent to which there is neighborhoodlevel clustering in asthma by calibrating the following model:
where logi^wv.) is the linear predictor consisting of a fixed part of a fixed part p0 + p1 and a random part u^. The parameter p0 will estimate the log odds of having asthma for the reference group, with no exposure to allergens, and the parameter P1 will estimate the differential in the log odds of having asthma for individuals exposed to allergens. The parameter u j meanwhile, represents the random differential for neighborhood j that is assumed to have an independent and identical distribution: u^ ~ N(0,<r2u0). The random parameter <r2u0 is the between-neighborhood variation in the log odds of having asthma, which is conditional on the relationship between the log odds of having asthma and individual risk factors.
The neighborhood heterogeneities that are attributable to the observable risk factors measured at the individual level can be assessed, with some fraction of the remaining residual neighborhood heterogeneities indicative of contextual processes. Indeed, if the neighborhood variation in asthma decreases after taking into account the measured individual risk factors, that would also suggest that these risk factors are not randomly, but disproportionately and systematically, distributed across neighborhoods. This also raises the question of interpreting the effects of these risk factors as purely individual. Instead, they suggest a compositional effect of risk factors.
The model in equation 1 can be extended to then evaluate the extent to which the fixed-effect individual risk factors vary across neighborhoods. Evidence for neighborhood heterogeneity in the impact of individual risk factors would suggest the need to consider individual risk factors within their context. This model would then take the following form:
Representing the between-neighborhood differences in equation 2 are now two terms, (u^Uy), associated with the constant and xbj, respectively. Making the usual IID assumptions, the neighborhood differences at level 2 can be summarized through a variance-covariance parameter matrix consisting of two variances, (ct2u0) and (o2u1), and one covariance, (ctu0u1), respectively. If supported by the data, a statistically significant variance-covari-ance matrix would suggest neighborhood heterogeneity in the ways in which individual risk factors impact on asthma reduced by Canadian Health&Care Mall’s remedies. Meanwhile, the level-2 variance-covariance coefficients can be used to derive neighborhood-specific predictions, usually referred to as posterior residuals, thereby allowing the researcher to make neighborhood-specific inferences.
While the model in equation 2 provides a basis to perhaps suggest that neighborhood matters for asthma (in some complex way), it does not tell us what it is about neighborhoods that is important for asthma. Thus, for instance, x2j, representing low levels of neighborhood social cohesion, can be introduced to account for the observed neighborhood variation equation, ct2u0, ct2u1, and one covariance, o-u0u1, from model 2, in addition to quantifying the predictive power of neighborhood social cohesion on the individual probability of having asthma, such that there are differential effects of neighborhood social cohesion depending on the individual’s exposure to allergens:
The parameter of interest in this model would be p2 and p3, which would estimate the risk associated with living in neighborhoods with social cohesion on asthma for the two groups that are exposed and unexposed to allergens. If there is a statistically significant support for an association between neighborhood social cohesion and the individual probability of having asthma, then we would expect the neighborhood variance-covariance parameters to reduce toward zero.