Everything All at Once: On Choosing an Estimand for Multi-component Environmental Exposures

Area of Science:

• Environmental health
• Epidemiology
• Causal inference

Background:

• Many environmental health research questions involve complex exposure mixtures, not just single exposures.
• Existing methods for quantifying relationships between exposure mixtures and outcomes are limited.
• Non-discrete exposure components in multivariate mixtures pose analytical challenges.

Purpose of the Study:

• To propose a novel approach for quantifying the relationship between a shift in a multivariate exposure mixture and an outcome.
• To develop flexible definitions for exposure mixture shifts, including interactions and varying component amounts.
• To provide a non-parametric estimation method using machine learning, avoiding tenuous parametric assumptions.

Main Methods:

• Define a flexible shift in the exposure mixture, supported by observed data.
• Assess and minimize extrapolation by modifying the exposure shift.
• Employ non-parametric machine learning for estimating the relationship between the exposure mixture shift and the outcome.
• Utilize longitudinal data from the Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS) cohort.

Main Results:

• The proposed method allows for flexible quantification of exposure mixture effects.
• Minimizing extrapolation ensures more reliable and data-driven results.
• Non-parametric estimation using machine learning provides robust analysis in complex settings.
• The approach is demonstrated using longitudinal pesticide exposure data and hypertension risk.

Conclusions:

• The developed approach offers a robust framework for analyzing complex environmental exposure mixtures.
• It provides a valuable tool for environmental epidemiology, particularly in health outcome research.
• The method facilitates understanding the impact of shifts in exposure mixtures on health, using real-world data and reproducible code.