On estimating average effects for multiple treatment groups.

Area of Science:

• Epidemiology
• Biostatistics
• Observational Data Analysis

Background:

• Estimating average exposure effects from observational data is challenging, especially when treatment assignment is not ignorable.
• Existing methods often rely on assumptions of ignorable treatment assignment, which may not hold in real-world scenarios.

Purpose of the Study:

• To propose and evaluate a novel method for estimating average exposure effects from observational data with multiple exposure groups.
• To address situations where the assignment rule depends on the response variable, violating the ignorable treatment assignment assumption.

Main Methods:

• Fitting an approximation of the marginal sample distribution of the response variable in each exposure group.
• Utilizing the population expectation (PE) of the outcome variable for effect estimation.
• Comparing the PE approach with instrumental variable and propensity score-based methods.

Main Results:

• The proposed PE method is robust under model misspecifications.
• Demonstrated the method's application using data on soy consumption and estrogen metabolites in Asian American women.
• The PE approach provides a valid alternative when ignorable treatment assignment cannot be assumed.

Conclusions:

• The population expectation (PE) method offers a flexible and robust approach to estimating average exposure effects in observational studies.
• This method is particularly valuable in complex epidemiological research where treatment assignment mechanisms are non-ignorable.
• The study highlights the importance of considering non-ignorable assignment rules in causal inference from observational data.