Causal machine learning methods and use of cross-fitting in settings with high-dimensional confounding

TL;DR

This study compares advanced causal inference methods, AIPW and TMLE, in high-dimensional confounding settings, highlighting the benefits of cross-fitting and comprehensive ensemble libraries for accurate effect estimation.

Contribution

It provides empirical guidance on implementing AIPW and TMLE with cross-fitting and Super Learner libraries in high-dimensional confounding scenarios.

Findings

TMLE shows more stability than AIPW.

Cross-fitting improves variance estimation and coverage.

Full Super Learner libraries reduce bias and variance.

Abstract

Observational epidemiological studies commonly seek to estimate the causal effect of an exposure on an outcome. Adjustment for potential confounding bias in modern studies is challenging due to the presence of high-dimensional confounding, which occurs when there are many confounders relative to sample size or complex relationships between continuous confounders and exposure and outcome. Doubly robust methods such as Augmented Inverse Probability Weighting (AIPW) and Targeted Maximum Likelihood Estimation (TMLE) have the potential to address these challenges, using data-adaptive approaches and cross-fitting, but despite recent advances limited evaluation and guidance are available on their implementation in realistic settings where high-dimensional confounding is present. Motivated by an early-life cohort study, we conducted an extensive simulation study to compare the relative performance of AIPW and TMLE using data-adaptive approaches for estimating the average causal effect (ACE). We evaluated the benefits of using cross-fitting with a varying number of folds, as well as the impact of using a reduced versus full (larger, more diverse) library in the Super Learner ensemble learning approach used for implementation. We found that AIPW and TMLE performed similarly in most cases for estimating the ACE, but TMLE was more stable. Cross-fitting improved the performance of both methods, but was more important for variance estimation and coverage than for point estimates, with the number of folds a less important consideration. Using a full Super Learner library was important to reduce bias and variance in complex scenarios typical of modern health research studies.