Triangulating Instrumental Variable, confounder adjustment and difference-in-difference methods for comparative...

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Observational studies are crucial for comparative effectiveness research.
• Randomization in clinical trials balances confounders, simplifying analysis.
• Unmeasured confounding in observational data complicates treatment effect comparisons.

Purpose of the Study:

• To address challenges in estimating causal effects from observational data.
• To develop robust causal inference methods resilient to assumption violations.
• To introduce a framework for assessing the consistency of treatment effect estimates.

Main Methods:

• Utilized Instrumental Variable (IV) and Prior Event Rate Ratio (PERR) methods.
• Employed multivariable regression and propensity score matching for confounder adjustment.
• Proposed a novel prior outcome augmented IV method robust to assumption violations.
• Developed a heterogeneity statistic to assess statistical dissimilarity of estimates.

Main Results:

• The proposed method estimates treatment effects without bias, even when other methods' assumptions are violated.
• The application study demonstrated the utility of triangulation for assessing estimation consistency.
• The prior outcome augmented IV method showed robustness to key assumption violations.

Conclusions:

• Triangulating results from diverse estimation methods is vital for high-quality observational evidence.
• The proposed framework and heterogeneity statistic aid in identifying potential biases.
• This approach enhances the reliability of causal inference in comparative effectiveness studies.