Figure 1.

Directed Acyclic Graph (DAG) for the identification of potential confounders. The exposure of interest is haem iron and the outcome of interest is self-reported diagnosed anaemia. The directed acyclic graph makes explicit the causative model we are testing and the relationship of the co-variates. In order to identify confounders, the arrows coming out of haem iron are removed, since these are causative, and any “back door” paths that remain between haem iron and anaemia are identified. The back door paths need not follow the direction of the arrows. Adjusting for a co-variate “closes” the back door path, i.e., removes confounding, assuming that the co-variate is measured without error. A back door path is considered already closed if it passes through a “collider”, i.e., a node with 2 arrows pointing into it. Adjusting for a collider re-opens that path and increases the potential confounding.