Balancing Score Adjusted Targeted Minimum Loss-based Estimation.

Area of Science:

• Causal inference
• Statistical modeling
• Biostatistics

Background:

• Estimating causal effects requires addressing confounding bias.
• Propensity score methods are common but rely on correct propensity score estimation.
• Alternative scores that balance covariates may offer robustness.

Purpose of the Study:

• Introduce a new targeted minimum loss-based estimator (TMLE) with the balancing score property.
• Evaluate the performance of this novel TMLE against existing causal inference estimators.
• Demonstrate the utility of balancing scores for robust causal effect estimation.

Main Methods:

• Developed a TMLE incorporating the balancing score property.
• Compared the proposed TMLE with propensity score matching, inverse probability of treatment weighting, and regression-based estimators.
• Utilized simulation studies to assess estimator performance and bias reduction.

Main Results:

• The proposed TMLE demonstrated robustness and efficiency in simulations.
• Estimators utilizing a balancing score showed consistency even with misspecified propensity scores.
• The balancing score property provides a more flexible approach to bias reduction.

Conclusions:

• The balancing score property is a valuable concept for robust causal inference.
• The new TMLE offers a promising, efficient, and doubly robust method for estimating treatment-specific means.
• This work advances methods for reliable causal effect estimation in observational studies.