Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations

Neat paper by Tennant, P. W. G. et al. (2020): Use of directed acyclic graphs (DAGs) to identify confounders in applied health research: review and recommendations in the International Journal of Epidemiology.

Picture

Recommendations from the paper

1. The focal relationship(s) and estimand(s) of interest should be stated in the study aims
2. The DAG(s) for each focal relationship and estimand of interest should be available
3. DAGs should include all relevant variables, including those where direct measurements are unavailable
4. Variables should be visually arranged so that all constituent arcs flow in the same direction
5. Arcs should generally be assumed to exist between any two variables
6. The DAG-implied adjustment set(s) for the estimand(s) of interest should be clearly stated
7. The estimate(s) obtained from using the unmodified DAG-implied adjustment set(s)—or nearest approximation thereof—should be reported
8. Alternative adjustment set(s) should be justified and their estimate(s) reported separately