Introduction

Project Overview

Our goal here is to examine the social and health determinants of racial disparities in sepsis mortality. We will use Cox Proportional Hazards modeling strategies to examine each variable for its confounding and/or mediation effect on the age- and sex-adjusted Black c. White risk ratio for all-cause and cause-specific mortality.

Dataset

We utilize the 1999-2005 National Health Interview Surveys (NHIS) linked at the individual-level to outcomes in the National Death Index (NDI), years 1999-2011. The outcome of interest for mediation analyses is septicemia deaths. Septicemia death is defined using the National Vital Statistics ICD-10 coding schema (A40-A41). These codes can occur in any position on the death record. The competing outcome of interest is ‘other cause’ mortality, deinfed as any death without an ICD-10 code for septicemia, influenza or pneumonia among the listed caused odeath on the death record.

Mediation

Mediation is assessed using the difference method. First, the age- and sex- adjusted parameter estimate in Cox PH model for Black c. White is taken as the total effect. Then a candidate covariate is entered into this base model. The resultant Black c. White parameter estimate is now taken as the direct effect, conceptualized as the resultant direct effect of race on risk of mortality after adjustment. The subtraction of the two gives the indirect effect, conceptualized as the amount of the total race effect on mortality that is mediated/confounded through the added covariate. Finally, by dividing the indirect effect by the total effect, we get the proportion of total effect mediated.

This process is done separately for each covariate. This is also done for a model of septicemia death and separately for a model of other causes of death. This allows a resultant qualitative comparison of mediators in a cause-specific death to a more general category.

Proportion of Age- and Sex-Adjusted Black c. White Risk of Death Mediated by Covariate

Indirect Effect of Covariate on Age- and Sex-Adjusted Black c. White Risk of Death

I noticed above that for every covariate, the proporiton mediated was higher for other causes of death than for septicemia causes of death. I pondered whether this was due to the fact the the parameter estimate for age-sex-adjusted Black/White for other causes of death was absolutely smaller than that for septicemia causes of death (0.28 vs 0.65, equating to HR of 1.3 and 1.9, respectively). So I thought I would explore same graph but just looking at indirect effects, an absolute measure, rather than a relative measure.