Quantifying Exposome-Health Associations with Bayesian Multiple Index Models

, 9 for deciles Decompose coecients of a linear model into • : an “overall mixture e↵ect” • bp: component weights Very interpretable • Weights bp indicate relative contribution of each component Somewhat restrictive: • Assume linearity (on the quantile scale) • Assume no interactions between any exposures Di↵erent versions • Constrained: Weighted quantile sum regression (Carrico et al., 2015) • Un. [...] (2019) 5 Dioxin1 Dioxin2 Dioxin3 Furan1 Furan2 Furan3 Furan4 PCB074 PCB099 PCB118 PCB126 PCB138 PCB153 PCB169 PCB170 PCB180 PCB187 PCB194 A Tradeo↵ Linear index methods: • Easy to interpret • Somewhat restrictive • Assumes linearity and no interactions BKMR: • Very flexible • Allows non-linearities and higher order interactions • Dicult to interpret when there are many exposures (P is large) 6 Pro. [...] • Distinct indices for pregnancy and postnatal exposures • Subject to Bayesian variable selection Adjusted for covariates: • Cohort, gestational age at birth, maternal BMI, maternal weight gain, maternal education, parity, native to country of birth 10 Exposome Analysis: Indexwise PIPs Organochlorines_Postnatal Metals_Postnatal WaterDBPs_Pregnancy Builtenvironment_Postnatal Meteorological_Postnata. [...] interactions between multipollutant index and individual covariates Extensions: • Constraints on index weights • DLM structure for time-varying exposures (Wilson et al., 2019) This research was supported by NIH grants ES028800, ES028811, ES028688, and ES000002 and USEPA grants RD-835872-01 and RD-839278. [...] Kernel machine and distributed lag models for assessing windows of susceptibility to mixtures of time-varying environmental exposures in children’s health studies.