Asymmetric stratification. An outline for an efficient method for controlling confounding in cohort studies

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Confounding is a critical issue in cohort studies, impacting the validity of results.
• Traditional methods like cross-stratification and multivariate modeling have limitations in controlling numerous confounders.
• Propensity score methods offer efficiency but can lack intuitive interpretation.

Purpose of the Study:

• To propose asymmetric stratification as an efficient method for controlling multiple confounders in cohort studies.
• To provide an intuitive framework similar to cross-stratification while enhancing confounder control.
• To demonstrate the implementation of asymmetric stratification using classification and regression trees (CART).

Main Methods:

• Asymmetric stratification defines strata using categories of a subset of confounders, avoiding complex multivariate models.
• The method leverages classification and regression trees (CART) for practical implementation.
• Comparison with standard propensity score analysis through simulations and a real-world example.

Main Results:

• Asymmetric stratification provides an efficient way to control multiple confounders.
• The proposed method retains the intuitive appeal of cross-stratification.
• Computer simulations and an example suggest it is a simpler alternative to propensity score analysis.

Conclusions:

• Asymmetric stratification is a promising method for addressing confounding in cohort studies.
• It offers a balance between efficiency and interpretability.
• Further recommendations are provided for method enhancement.