What's the Weight? Estimating Controlled Outcome Differences in Complex Surveys for Health Disparities Research

Area of Science:

• Health Disparities
• Biostatistics
• Genetics

Background:

• Estimating racial disparities in health outcomes, like telomere length, is crucial.
• The average controlled difference (ACD) quantifies racial effects on health.
• Complex survey data, such as NHANES, presents unique challenges for estimating ACD due to dependent survey weights.

Purpose of the Study:

• To propose novel identification formulas for estimating the ACD in health outcomes between racial groups in complex surveys.
• To adjust for social determinants and confounding factors when assessing the controlled effect of race on telomere length.
• To provide a robust method for analyzing racial disparities in health outcomes using observational data.

Main Methods:

• Developed propensity score methods tailored for complex survey designs where weights depend on the group variable.
• Proposed identification formulas to address covariate imbalance and ensure generalizability in ACD estimation.
• Validated the proposed methods through extensive simulations, comparing them to traditional analytic approaches.

Main Results:

• The proposed methods demonstrated superior performance over traditional approaches, showing reduced bias, lower mean squared error, and improved coverage.
• Analysis of NHANES data revealed that racial differences in telomere length between Black and White individuals attenuated after controlling for socioeconomic factors.
• The study highlights the importance of appropriate propensity score and survey weighting techniques for accurate estimation of racial health disparities.

Conclusions:

• The developed methods offer a statistically sound approach for estimating racial disparities in health outcomes from complex survey data.
• Accounting for socioeconomic confounding significantly reduces observed racial differences in telomere length.
• The R package `svycdiff` is available to implement these advanced statistical techniques for reproducible research.