Measured and accounted-for confounding in pharmacoepidemiologic studies: Some thoughts for practitioners.

Confounding must be considered at both the design and analysis stages of a study. DAGs and data adaptive approaches can help.