Using Causal Diagrams for Biomedical Research.

Area of Science:

• Biomedical Research
• Epidemiology
• Biostatistics

Background:

• Causal diagrams are essential for modeling causal relationships in biomedical research.
• They provide a scientific basis for multivariable model selection in observational studies and clinical trials.
• This methodology identifies potential biases and classifies variables like confounders, colliders, and mediators.

Purpose of the Study:

• To enhance researchers' understanding of variable selection in statistical analyses.
• To improve communication between researchers and statisticians regarding study design.
• To illustrate the impact of causal diagrams on accurate effect estimation.

Main Methods:

• Introduction of causal diagrams, specifically directed acyclic graphs (DAGs).
• Illustration of variable selection's impact on effect estimates using numeric examples.
• Analysis of the Framingham Heart Study dataset to demonstrate calculations and miscalculations.

Main Results:

• Incorrect variable selection in multivariable modeling can lead to serious miscalculation of effect estimates.
• Causal diagrams provide a framework for identifying appropriate variables based on causal ordering.
• The Framingham Heart Study analysis highlighted potential miscalculations due to flawed variable selection.

Conclusions:

• Causal diagrams are critical tools for clinical researchers to understand complex causal relations.
• Accurate variable selection, guided by causal diagrams, is necessary for valid multivariable statistical models.
• This methodology enhances the scientific rigor of study designs and statistical analyses.