Doubly Robust Control Outcome Calibration Approach Estimation of Conditional Effects with Uncontrolled Confounding

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Drawing causal conclusions from nonrandomized studies requires assumptions about unmeasured confounding, which are often untestable.
• Existing control outcome calibration (COCA) methods rely on correctly modeling negative control outcomes.

Purpose of the Study:

• To propose a doubly robust COCA estimator for average causal effects.
• To relax the strict modeling requirements of existing COCA methods.
• To allow for effect modification by covariate-exposure interactions.

Main Methods:

• Developed a doubly robust COCA estimator using correctly specified exposure and focal outcome models.
• This approach protects against bias from misspecified negative control outcome models.
• Incorporated covariate-exposure interaction terms to enable effect modification analysis.

Main Results:

• The doubly robust COCA estimator provides unbiased point estimates and inferences.
• Simulation studies confirmed the method's ability to obtain unbiased estimates.
• Empirical evaluation on volunteering and mental health data demonstrated practical utility.

Conclusions:

• The proposed doubly robust COCA method offers a practical and implementable approach for causal inference.
• It enables unbiased estimation of average causal effects in the presence of uncontrolled confounding.
• This advancement improves the reliability of causal conclusions from observational studies.