A New Causal Decomposition Paradigm towards Health Equity

TL;DR

This paper introduces a novel causal decomposition method for health disparities that accounts for all sources of disparity, automatically identifies covariates, and provides policy-relevant insights, demonstrated on synthetic and real datasets.

Contribution

It proposes a new causal decomposition framework under structural causal models that includes all disparity sources and automatically identifies covariates using maximal ancestral graphs.

Findings

Method accurately decomposes health disparities in synthetic data.

Effective in analyzing disparities in spine disease dataset.

Theoretical guarantees established for the proposed decomposition approach.

Abstract

Causal decomposition has provided a powerful tool to analyze health disparity problems, by assessing the proportion of disparity caused by each mediator. However, most of these methods lack \emph{policy implications}, as they fail to account for all sources of disparities caused by the mediator. Besides, their estimations \emph{pre-specified} some covariates set (\emph{a.k.a}, admissible set) for the strong ignorability condition to hold, which can be problematic as some variables in this set may induce new spurious features. To resolve these issues, under the framework of the structural causal model, we propose to decompose the total effect into adjusted and unadjusted effects, with the former being able to include all types of disparity by adjusting each mediator's distribution from the disadvantaged group to the advantaged ones. Besides, equipped with maximal ancestral graph and context variables, we can automatically identify the admissible set, followed by an efficient algorithm for estimation. Theoretical correctness and the efficacy of our method are demonstrated on a synthetic dataset and a spine disease dataset.