Analysis of multiple-variable missing-not-at-random survey data for child lead surveillance using NHANES

Area of Science:

• Public Health Surveillance
• Biostatistics
• Environmental Health

Background:

• Multi-topic surveys are crucial for public health surveillance but often suffer from high proportions of missing data.
• The National Health and Examination Survey (NHANES) is vital for tracking elevated blood lead levels (EBLLs) in US children, yet up to 35% of respondents have missing key predictor variables.

Purpose of the Study:

• To develop and apply a statistical model to address missing-not-at-random variables in complex survey data for improved public health surveillance.
• To estimate the prevalence of elevated blood lead levels in young children using a novel Bayesian approach applied to the American Community Survey (ACS).

Main Methods:

• Formulated a t-distributed Heckman selection model within a Bayesian framework to handle multiple missing-not-at-random variables in complex survey designs.
• Utilized Gibbs and grid sampling to estimate posterior distributions of parameters.
• Applied the developed model coefficients to ACS data to calculate prevalence estimates for EBLLs.

Main Results:

• Quantified significant disparities in EBLL prevalence based on race/ethnicity, age of housing, and poverty.
• Presented three- to five-fold differences in predicted EBLL prevalence across various US geographies.
• Enabled multivariate analyses of EBLLs, including the critical variable of housing age, which was previously unavailable.

Conclusions:

• The developed Bayesian model enhances the utility of existing survey data for public health surveillance, particularly for tracking elevated blood lead levels.
• The findings highlight significant socioeconomic and environmental factors influencing EBLLs in children.
• This methodology offers a valuable expansion for NHANES and similar health surveillance programs facing challenges with missing data.