Causal inference with a quantitative exposure.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Epidemiologic research frequently involves quantitative exposures, yet statistical methods often focus on binary or categorical ones.
• Existing causal inference methods for quantitative exposures have limitations, such as implicit constraints on dose-response curves.

Purpose of the Study:

• To review and propose flexible statistical methods for estimating dose-response curves with quantitative exposures.
• To address limitations of current methods, particularly outcome regression models' implicit constraints.
• To introduce new, robust, and broadly applicable methods for causal inference in epidemiology.

Main Methods:

• Review of existing methods: ordinary regression, inverse propensity weighting, stratification, and augmented inverse propensity weighting.
• Proposal of a flexible modeling strategy to avoid constraining the dose-response curve.
• Introduction of two new methods: weighted regression and stratified regression, combining ordinary regression with propensity-based methods.

Main Results:

• The proposed weighted and stratified regression methods are doubly robust, similar to augmented inverse propensity weighting.
• These new methods are easier to implement and more generally applicable than existing augmented methods.
• Simulation studies and an obstetric example demonstrate the utility and performance of the proposed methods.

Conclusions:

• The developed weighted and stratified regression methods provide flexible and robust approaches for quantitative exposure analysis in epidemiology.
• These methods overcome limitations of traditional outcome regression models and offer practical advantages.
• The findings enhance causal inference capabilities for quantitative exposures in epidemiologic research.