Assessing the impact of unmeasured confounding for binary outcomes using confounding functions

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Causal inference relies on the no unmeasured confounding assumption, often unmet in observational studies.
• Unmeasured confounding can bias effect estimates, necessitating methods to assess its impact.
• Existing methods for bias assessment have limitations, especially for binary outcomes.

Purpose of the Study:

• To extend the confounding function approach for assessing unmeasured confounding in binary outcome settings.
• To illustrate the application of confounding functions for binary outcomes with a practical example.
• To provide guidance on implementing this method for odds or risk ratio estimation.

Main Methods:

• The confounding function approach is utilized to model the potential impact of unmeasured confounding.
• The methodology is adapted for binary outcomes, considering different assumptions about effect modification.
• The study provides Stata and R code for practical implementation.

Main Results:

• The confounding function approach offers a flexible way to assess the total bias from unmeasured confounding and other biases.
• The choice of confounding function implicitly encodes assumptions about effect modification.
• The approach is particularly valuable when alternative methods like instrumental variables are not feasible.

Conclusions:

• The confounding function approach is a valuable tool for evaluating unmeasured confounding, especially for binary outcomes.
• This method allows researchers to quantify the potential impact of unmeasured factors on causal effect estimates.
• The provided code facilitates the application of this technique in epidemiological and biostatistical research.