Simulation-based estimators of analytically intractable causal effects.

Area of Science:

• Statistics
• Causal Inference
• Computational Statistics

Background:

• Estimating causal effects is crucial but often analytically intractable.
• Existing methods like g-formula with Monte Carlo integration are computationally intensive.
• Nested Monte Carlo integration poses significant computational challenges for inference.

Purpose of the Study:

• To develop a widely-applicable method for accelerating Monte Carlo integration in g-computation.
• To reduce the computational burden of existing g-computation algorithms.
• Introduce accelerated g-computation (AGC) for efficient causal effect estimation.

Main Methods:

• Developed accelerated g-computation (AGC) algorithms.
• AGC accelerates the Monte Carlo integration step in g-computation.
• Method is analogous to multiple imputation but adjusts variance estimation differently.

Main Results:

• Significantly reduced computational burden for g-computation algorithms.
• Demonstrated applicability in mediation analysis with beta regression.
• Showcased utility in longitudinal trials with nonignorable missingness using BART models.

Conclusions:

• Accelerated g-computation (AGC) provides a computationally efficient approach.
• The method is broadly applicable across various causal inference problems.
• AGC enhances the feasibility of complex causal effect estimation.