Causal Methods for Observational Research: A Primer.

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Observational studies aim to estimate causal effects but often struggle with confounder adjustment.
• Standard regression models have limitations in handling interactions and time-varying confounding.
• Stepwise confounder selection based on p-values can introduce bias.

Purpose of the Study:

• To highlight the limitations of traditional regression for causal effect estimation.
• To introduce and illustrate advanced causal inference methods.
• To emphasize the benefits of targeted maximum likelihood estimation (TMLE).

Main Methods:

• Inverse-probability-of-treatment-weighting (IPTW)
• Parametric g-formula
• Targeted Maximum Likelihood Estimation (TMLE)
• TMLE combines IPTW and g-formula for double-robustness.

Main Results:

• Causal methods overcome limitations of standard regression, such as inability to handle interactions or time-varying confounders.
• TMLE offers a robust approach to confounder adjustment.
• These methods provide more reliable estimates of causal effects in observational data.

Conclusions:

• Advanced causal inference methods are crucial for accurate effect estimation in observational studies.
• TMLE is a powerful tool for addressing complex confounding scenarios.
• Adoption of causal methods can improve the quality of research in clinical journals.