Estimating the effect of latent time-varying count exposures using multiple lists

Area of Science:

• Environmental health
• Biostatistics
• Epidemiology

Background:

• Longitudinal built-environment health studies face challenges with commercial business database accuracy for food environments.
• Conflicting exposure measures from different databases introduce bias in health effect estimates.
• On-site verification of historical data is often infeasible.

Purpose of the Study:

• To propose a novel statistical method for integrating multiple commercial business databases to accurately characterize dynamic food environments.
• To correct for measurement error and bias in health effect estimates arising from discrepancies in exposure data.
• To assess the longitudinal health effects of true food environment exposures, specifically convenience store proximity to schools.

Main Methods:

• A joint statistical model was developed for time-varying health outcomes, observed count exposures, and latent true count exposures.
• The model estimates time-specific source quality and incorporates time-dependent true count exposure using a Poisson integer-valued first-order autoregressive process.
• A Bayesian nonparametric approach was employed to flexibly model location-specific exposures.

Main Results:

• The proposed method effectively resolves discordance between different commercial databases, improving the accuracy of food environment exposure assessment.
• Bias in health effect estimates due to measurement error in single data sources was significantly reduced.
• The study demonstrated reduced bias in longitudinal childhood obesity health effects related to convenience store exposures.

Conclusions:

• Combining multiple commercial business databases using a joint modeling approach is a viable strategy to mitigate measurement error in longitudinal built-environment health studies.
• This method enhances the reliability of exposure data, leading to more accurate estimations of health effects.
• The findings have implications for public health research, particularly in understanding the environmental determinants of childhood obesity.