Causal Knowledge as a Prerequisite for Interrogating Bias: Reflections on Hernán et al. 20 Years Later

Area of Science:

• Epidemiology
• Causal Inference
• Biostatistics

Background:

• Hernán et al. (2002) stressed theory-based confounding control, differentiating colliders from confounders.
• This commentary re-examines the seminal paper's impact on bias conceptualization.

Purpose of the Study:

• To highlight the link between nonexchangeability, bias, and directed acyclic graphs (DAGs).
• To underscore causal knowledge as fundamental for identifying and addressing research bias.

Main Methods:

• Revisiting Hernán et al.'s (2002) arguments on confounding and colliders.
• Applying DAGs to conceptualize and differentiate bias sources.
• Analyzing examples from the original paper.

Main Results:

• The paper unified bias conceptualization via DAGs and nonexchangeability.
• DAGs facilitate distinguishing confounding from selection bias.
• Unresolved questions remain regarding collider bias, selection bias, and generalizability.

Conclusions:

• Causal knowledge is a prerequisite for confounding control.
• DAGs provide a framework for understanding and mitigating various biases.
• Further research is needed on the interplay of different bias types and generalizability.