Depicting patient-reported outcome measures within directed acyclic graphs: practice and implications for causal...

Area of Science:

• Causal inference methodology
• Psychometrics and measurement theory
• Observational health research

Background:

• Estimating causal effects on patient-reported outcome measures (PROMs) is complicated by the internal structure of PROMs.
• Directed acyclic graphs (DAGs) offer a visual framework to represent these complex relationships.

Purpose of the Study:

• To demonstrate how to represent PROM internal causal structures using DAGs.
• To explain implications for causal effect estimation in observational data, considering external variables.

Main Methods:

• Utilized DAGs to model PROM indicator-construct relationships under reflective and formative measurement theories.
• Applied models to unidimensional (e.g., PHQ-9) and multidimensional (e.g., EQ-5D) PROMs.

Main Results:

• Unidimensional PROMs under reflective models are straightforward for causal analysis.
• Multidimensional PROMs under formative models require specific attention to indicators and external variables for accurate causal effect estimation.

Conclusions:

• Multidimensional formative PROMs increase causal analysis complexity but are valuable for outcome-wide studies.
• Incorporating PROMs into DAGs supports causal analyses in observational settings, particularly for nuanced exposure-outcome relationships.