LOW-RANK LONGITUDINAL FACTOR REGRESSION WITH APPLICATION TO CHEMICAL MIXTURES

Area of Science:

• Environmental Epidemiology
• Developmental Toxicology
• Statistical Modeling

Background:

• Developmental epidemiology investigates early life exposures and childhood health.
• Analyzing multiple, correlated exposures and their time-varying effects presents statistical challenges.
• Understanding prenatal chemical exposures' long-term health impacts is crucial.

Purpose of the Study:

• To propose a novel statistical model, LowFR, for analyzing complex longitudinal exposure data.
• To assess the association between prenatal bisphenol A (BPA) and phthalate exposures and adolescent glucose metabolism using the ELEMENT study data.

Main Methods:

• Developed a low-rank longitudinal factor regression (LowFR) model.
• Utilized a Bayesian dynamic factor model for handling highly correlated exposures.
• Employed a novel factor regression approach to jointly model exposures and health outcomes.

Main Results:

• The LowFR model demonstrated effectiveness in simulations.
• Analysis of ELEMENT study data revealed associations between specific phthalate metabolites and adolescent glucose metabolism.
• Diethyl and dibutyl phthalate metabolite levels in early pregnancy were linked to altered glucose metabolism.

Conclusions:

• The LowFR model provides a flexible and tractable approach for analyzing complex early-life exposure data.
• Prenatal exposure to diethyl and dibutyl phthalates may influence adolescent glucose metabolism.
• This research highlights the importance of considering early-life environmental exposures in understanding long-term metabolic health.