Identification and estimation of causal effects with confounders missing not at random

Area of Science:

• Statistics
• Epidemiology
• Biostatistics

Background:

• Causal inference from observational studies is hindered by missing confounder data.
• Missing confounders not at random (MNAR) often prevents reliable identification of causal effects.
• Existing methods struggle when confounder data is MNAR.

Purpose of the Study:

• To develop a method for identifying causal effects under treatment-independent missingness assumption for confounders.
• To propose estimators for average causal effect (ACE) when confounders are MNAR.
• To evaluate the performance of proposed estimators in simulations and real-world data.

Main Methods:

• Proposed a weighted estimating equation approach for parameter estimation.
• Introduced three ACE estimators: regression-based, propensity score weighting, and doubly robust.
• Utilized a treatment-independent missingness assumption.

Main Results:

• Established identification of causal effects under the specified missingness assumption.
• Simulation studies demonstrated the performance of the proposed estimators.
• A real data analysis illustrated the practical application of the method.

Conclusions:

• The proposed weighted estimating equation approach successfully identifies causal effects with MNAR confounders.
• The developed estimators provide reliable estimates for ACE in challenging observational data.
• This method enhances causal inference capabilities in the presence of missing data.