Invited commentary: Mediation analysis and cancer screening, identifying areas to target to improve cancer disparities.

Abstract

Surgical health services research has traditionally focused on reporting “associations” between specific risk factors and healthcare outcomes. Fewer studies have sought to address the underlying “how” or “in what way” the risk factors/exposures affect the outcome in question. Mediation analysis is a statistical approach that attempts to define underlying mechanisms through which an exposure influences an outcome. Mediation analysis differs from classic multivariable regression analysis in that it aims to understand the factors involved in the pathways between the exposure and the outcome. In particular, mediation models seek to identify and explain the mechanism or process that underlies an observed relationship between an independent variable and a dependent variable via the inclusion of a third hypothetical variable, known as a mediator variable.1 In essence, the effect of the exposure (X) on the outcome (Y) is decomposed into the indirect effect (X → M → Y) of the exposure (X) through which other factors act as mediators (M) versus the direct effect (X → Y) that denotes the impact of the exposure not explained by these mediators (Figure 1).2 Mediation analysis can help deconstruct how various factors may be driving outcomes of interest and help identify both indirect and direct factors to target for intervention. In turn, this analytic tool may help delineate the various contributions of different mediators associated with social determinants and health outcome inequities. Determining which variables may function as mediators can be difficult to discern, however, as there are multiple ways that factors in the causal pathway may interact with each other. Unmeasured confounders may cause an opposite directional effect or, in certain cases, entirely nullify the total effect; even true opposite direct and indirect effects may coexist.3 In turn, variable selection for mediation analysis should be informed relative to the possible directionality of associations based on prior knowledge, experience, and existing data in the literature. In addition, sensitivity analyses may be helpful to examine the strength of unmeasured confounders and interaction effects, as well as estimate whether these factors might alter the conclusions drawn from the analysis.2 The current study examined the role of structural factors that underlie racial/ethnic and economic disparities related to cancer screening using mediation analysis.4 Specifically, we utilized mixed-effect beta distribution models to assess direct and indirect effects of race/ethnicity and economic disparities on county-level cancer screening rates. Mediation analysis was conducted using generalized structural equation models, which is a validated methodologic approach.5 Mediators were assessed as continuous variables and standardized (mean centered) before inclusion into the final model. A Gaussian distribution with an identity link for all mediators was used and the random intercept was assumed to have a normal distribution, which was appropriate for our dataset yet needs to be tailored based on the specific data being analyzed. In the current study, we noted an association between racial/ethnic and economic privilege on USPSTF-recommended cancer screening. Perhaps more importantly, mediation analysis revealed that the observed disparities between county-level privilege and cancer screening were explained by mediators such as poverty status, lack of health insurance or employment, urban-rural location and access to primary care physicians that accounted for 64% (95% confidence interval [CI]: 61%–67%), 85% (95% CI: 80%–89%), and 74% (95% CI: 71%–77%) of the effect on breast, colorectal and cervical cancer screening, respectively. The combination of more rigorous statistical approaches such as mediation analysis with the use of publicly available datasets such as the Center for Disease Control and Prevention (CDC)'s PLACES dataset and the American Community Survey can help better elucidate area-level social determinants of health that mediate disparities in cancer prevention strategies. In turn, rather than simply identifying and describing disparities, these analyses may hopefully inform which specific disparities need to be addressed to improve health equity. Over the last decades, disparities in surgical outcomes have been well described. A new era of surgical health services research should now focus on more “mechanistic” analyses that identify which specific direct and indirect factors should be targeted by healthcare institutions, payers, policymakers, and professional organizations to improve access to quality healthcare and enhance health equity through state and national policies. NA.