Epidemiology has a problem with causality. More than merely the science of the distribution of diseases in populations, epidemiology deals with protective and risk factors for the respective diseases, which is essentially the quest for causality.
The key principles of this theory are positivity, exchangeability, and well-defined interventions.
Positivity essentially means that a reference group is needed for a sensible comparison, whereas exchangeability denotes a setting in which members of the groups being compared are exchangeable in all respects except for assignment to their respective groups and the outcome under study. Exchangeability is thus the counterpart to what is called confounding in classical terminology. The principle of well-defined interventions means that causal effects in observational studies are conceptualized as effects of hypothetical randomized controlled trials (RCTs).
In a model for health inequality, however, the effect of race might be seen as a combination of the effects of genetic background, physical features, cultural context, socioeconomic status, and perception by others.
The latter conditions are theoretically modifiable, although whether this can be achieved by “reasonable hypothetical interventions” might be doubted.
From our usual understanding of interventions and RCTs, “providing cultural competency training might be consonant with RCT thinking; estimating the effect of institutionalized racism is not.”
This reveals another difficulty with a too literal understanding of “reasonable interventions,” which may fail with complex phenomena. Rather, the “hypothetical” aspect alludes to statistical solutions to render 2 groups exchangeable and thus to perform an “in silico intervention.”
By modeling self-perceived race on various dimensions including allergen exposure, access to medication, socioeconomic status, and hardship, a propensity score was calculated. When the inverse of the propensity score was used for weighting subsequent analyses, the features of the racial groups essentially became exchangeable and the effect of race itself was largely removed. In other words, the effect of race was mostly explained by the variables contributing to the propensity score. Instead of a linear association, the underlying factors may actually form a complex causal network resulting in health inequality.
Instead of self-reported race, they determined African ancestry by using a continuous score of global ancestry (GA) derived from 1.37 million genetic variants with the African and European samples of the 1000 Genomes Project as a reference. They found a higher risk of asthma readmission in individuals with an African ancestry of 20% or more versus less than 20%. This association of GA and readmission was mediated about 40% by financial and social hardship, whereas the direct effect was nonsignificant and reflected only 30% of the total effect.
which mediates a part of the effect of hardship on readmission through problems with disease management. Of note, this path does not explain the entire effect of hardship, which suggests that hardship may affect other reasons for readmission, thus possibly including disease severity.
used GA as a statistical construct that integrates high-dimensional genetic information. The resulting score may reflect the continuousness of genetic ancestry better than a multinomial variable such as self-perceived race, although both variables proved to be closely related. Apart from that, such a score illustrates that humans do not fall into genetic categories; they just vary in their ancestry continuously.
point out, genetic ancestry analyses proved to be useful in identifying genetic risk of polygenic diseases such as asthma.
Hence, a GA score inherently covers the risk of asthma attributable to genetic ancestry. By implication, the corresponding effects are included in the direct path (Fig 1 [gray box]), which may stand for an amalgam of unmeasured and possibly yet-unidentified mediators. As the direct path is estimated in the same model, the path through hardship and disease management is independent of all effects subsumed under the direct path including genetic effects.
emphasize that “the absence of a direct GA effect on readmission after accounting for these mediating pathways is consistent with the notion that asthma-related racial disparities are likely driven by factors linked to structural racism and social adversities.” This argumentation is somewhat difficult to follow, as it combines several thoughts. The first part sounds somewhat misleading because a nonsignificant direct path does not provide conclusive evidence for the absence of a direct GA effect on readmission. There is just not enough statistical power to prove or disprove a direct effect. As already stated, the direct effect might stand for “residual mediation,” which may comprise single effects that are significant. Yet, the potential existence of unmeasured mediators does not challenge the mediation through hardship, which is a strong significant effect. What Mersha et al
actually intend to emphasize is that the effect mediated through hardship is independent of other concurring effects.
Among others, they involve “[h]istorical legacies of racism and consequent unequal residential and economic opportunity.”
These constitute an inequitable system that again reinforces discriminatory beliefs, values, and distribution of resources, which meets the definition of structural racism.
Consequently, why Mersha et al
conclude that “structural racism and social adversities” are on the causal path from GA to asthma readmission becomes clear.
Likewise, their analysis is an incomplete model of a rather complex phenomenon, which includes many more aspects than those measured in the current study.
The analysis might have been complemented by, among other things, GA-associated genetic asthma risk loci
and asthma severity scores at baseline to disentangle the effects of genetics and of disease severity directly. Moreover, the study population was selected for White and African American children and ignored other ethnicities.
The ultimate goal of such analyses is that they transform into complex “reasonable real interventions” to remove health inequalities due to any form of discrimination.
Does water kill? A call for less casual causal inferences.
Ann Epidemiol. 2016; 26: 674-680
Is the “well-defined intervention assumption” politically conservative?.
Soc Sci Med. 2016; 166: 254-257
Commentary: race and sex are causes.
Epidemiology. 2014; 25: 488-490
On the causal interpretation of race in regressions adjusting for confounding and mediating variables.
Epidemiology. 2014; 25: 473-484
Rejoinder: how to reduce racial disparities?: Upon what to intervene?.
Epidemiology. 2014; 25: 491-493
Explaining racial disparities in child asthma readmission using a causal inference approach.
JAMA Pediatrics. 2016; 170: 695
Genetic ancestry differences in pediatric asthma readmission mediated by socio-environmental factors.
J Allergy Clin Immunol. 2021; 148: 1210-1218
Meta-analysis of genome-wide association studies of asthma in ethnically diverse North American populations.
Nat Genet. 2011; 43: 887-892
Structural racism and health inequities in the USA: evidence and interventions.
Lancet. 2017; 389: 1453-1463
Epidemiologic analysis of racial/ethnic disparities: some fundamental issues and a cautionary example.
Soc Sci Med. 2008; 66: 1659-1669
Published online: September 18, 2021
Supported by the German Center for Lung Research .
Disclosure of potential conflict of interest. The author declares that he has no relevant conflicts of interest.
© 2021 American Academy of Allergy, Asthma & Immunology