Standardization as a Tool for Causal Inference in Medical Research

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Traditional standardization methods address single confounders like age in medical research.
• Model-based standardization extends these methods for multiple confounders, yielding population-averaged marginal causal effects.

Purpose of the Study:

• To discuss traditional model-based standardization methods for estimating marginal causal effects.
• To apply these methods to real-world data and evaluate their performance.

Main Methods:

• Review of traditional model-based standardization techniques.
• Application of methods to data from the Tehran Thyroid Study.
• Simulation studies to compare standardization approaches.

Main Results:

• Covariate standardization is generally the preferred method for estimating marginal causal effects.
• Propensity score standardization is suggested when outcome is rare and exposure is frequent, or when exposure mechanisms are well-understood.

Conclusions:

• Model-based standardization offers robust methods for causal effect estimation in complex epidemiological studies.
• The choice between covariate and propensity score standardization depends on study-specific factors like outcome prevalence and exposure frequency.