Confounding Bias

Lead Author(s): Jeff Martin, MD

Properties of a True Confounder

Is the exposure (or predictor variable), marked E, associated with disease (or outcome), D? (See diagram)

A confounding factor or more simply a confounder can create or show an apparent effect even when there is truly no effect.

SCHEMATIC OF CONFOUNDING

0325_1confound.JPG

This is a general stick figure schematic for confounding based on a causal diagram.

This diagram emphasizes the essence of confounding, that is, there is another pathway to disease that is present, that is getting in the way of our ability to investigate our main question - a pathway different from the one you are interested in studying.

Examples of Confounding

Properties of a True Confounder - Associaton with Exposure

A confounding variable can be associated with the exposure either:

(1) Confounder causes the exposure

(2) Confounder results from the exposure

(3) Confounder is related to the exposure in a non-causal fashion

Properties of a True Confounder - Association with Disease

A confounding variable can be associated with the disease either:

(1) The confounding variable causes or is a marker of the disease:
(1a) Confounder causes the disease
(1b) Confounder is a marker of the disease
(1c) Confounder causes the disease and the exposure
(2) Confounder is related to the disease in a non-causal fashion

How Do You Adjust for Confounding?

To adjust or control for confounding, we need to stratify the variables following methods for

Factors to Consider in Study Planning

When we plan our studies, which factors should we be considering as potential confounders? WELL UNDERSTOOD AREA: Example Smoking LESS WELL UNDERSTAND AREA: But often, we are working in less well understood areas.

Confounder MUST Be Measured

Confounding can be dealt with in the analysis phase of a study but CANNOT be dealt with if the confounder is not measured.

0330_9last.JPG

References