Contextualizing Sensitivity Analysis in Observational Studies: Calculating Bias Factors for Known Covariates

The strength of evidence provided by epidemiological and observational studies is inherently limited by the potential for unmeasured confounding. While methods exist to quantify the potential effect of a specified unmeasured confounder, these methods should be anchored and contextualized within each study. We put forward a method for merging sensitivity to unmeasured confounding analyses with the impacts of the observed covariates. We graphically display what we call the observed bias factors with the tipping point sensitivity analysis. We illustrate the method under various study designs and provide an application created to simplify the implementation of this methodology.

Census Tract-Level Disparities: Examining Food Swamps and Food Deserts

Examining disparities in resources on the census tract-level is currently a public health priority. The Modified Retail Food Environment Index (mRFEI), provided by the CDC, incorporates two food environment metrics, ‘food deserts’, areas with no access to healthy foods, and ‘food swamps’, areas in which the quantity of unhealthy food options overwhelm healthy ones. We assess the association between the census tract racial make-up and food environment. Multiple logistic regression models are fit, controlling for census-tract level covariates from 2008-2012 ACS estimates, as well as state. Percent black is significantly associated with food swamps, with an absolute increase of 14.4 percent black living in food swamps (p< 0.01). Percent Hispanic is associated with food swamps, with an absolute increase of 9.1 percent Hispanic living in food swamps (p< 0.01), but inversely related to food deserts (absolute difference -6.8, p< 0.01). After adjustment, all associations remain significant. The strong association between the census tract-level racial make-up and food swamps shown here will allow for targeted interventions to census tracts where these disparities exist.