Identifiability of causal effects in test-negative design studies

Area of Science:

• Epidemiology
• Biostatistics
• Causal Inference

Background:

• Causal directed acyclic graphs (DAGs) are standard tools for variable selection in causal effect identification.
• Outcome-based sampling, like the test-negative design for vaccine effectiveness, poses unique challenges not handled by traditional DAG criteria.
• The common back-door criterion is insufficient for identifying population average causal effects in outcome-based sampling.

Purpose of the Study:

• To explain graphically why the standard back-door criterion is inadequate for outcome-based sampling studies.
• To introduce alternative graphical rules for causal inference in outcome-based sampling.
• To demonstrate updates in the Dagitty software reflecting these new principles.

Main Methods:

• Utilized intuitive graphical explanations of causal directed acyclic graphs (DAGs).
• Developed and described new graphical rules applicable to outcome-based sampling.
• Illustrated modifications within the Dagitty software.

Main Results:

• Demonstrated the limitations of the back-door criterion in outcome-based sampling for causal effect identification.
• Provided alternative graphical criteria for assessing identifiability of the causal odds ratio in such studies.
• Showcased the integration of these principles into the Dagitty software.

Conclusions:

• The standard back-door criterion is inappropriate for population average causal effects in outcome-based sampling.
• New graphical rules are necessary and have been developed for causal inference in these specific study designs.
• The Dagitty software has been updated to support these advanced causal inference methods.