Causal inference with missing exposure information: Methods and applications to an obstetric study.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Observational studies often face challenges with confounding and missing data, particularly missing exposure information.
• Accurate causal inference is crucial for understanding treatment effects and outcomes in real-world settings.

Purpose of the Study:

• To describe and compare methods for causal inference with missing exposure data in observational studies.
• To evaluate the performance of these methods using the Consortium on Safe Labor data and simulations.

Main Methods:

• The study considers methods involving three models: treatment assignment, outcome-covariate dependence, and missing data mechanism.
• It explores doubly robust estimators and introduces a novel triply robust estimator consistent if any two of three models are correctly specified.
• Assumptions include missing-at-random, positivity for missing data, no unmeasured confounding, and positivity for causal inference.

Main Results:

• Consistent causal effect estimation generally requires correct specification of at least two of the three models.
• Doubly robust and triply robust estimators offer flexibility in which models require correct specification.
• The triply robust estimator demonstrated consistency when any two of the three models were correctly specified.

Conclusions:

• Addressing missing exposure data is critical for valid causal inference in observational studies.
• The proposed triply robust method provides enhanced flexibility and robustness against model misspecification.
• These methods are applicable to large observational datasets like the Consortium on Safe Labor for reliable causal effect estimation.