Adjustment for Multiple Confounders

Lead Author(s): Jeff Martin, MD

Stratification of Multiple Confounders with More Than Two Levels

Stratification can also be used when there are multiple potential confounders with more than two levels.

Potential Confound effects

We are looking at the association between Chlamydia pneumonia infection and coronary artery disease (CAD)

Strata for CONFOUNDING Effects

To perform stratification, we would form six strata where we could put each subject only once by crossing smoking by age (see figure below). This would result in the following strata:
  1. < 40 year old smokers,
  2. < 40 year old non-smokers,
  3. 40-60 year old smokers,
  4. 40-60 year old non-smokers,
  5. > 60 year old smokers, and >
  6. 60 year old non-smokers.

Schema for Potential Founding Effects

Schematically (below), it would look like this with six mutually exclusive and exhaustive strata based on the two potential confounders looked at jointly.

Evaluating Joint Confounding

There are a couple of approaches you could take.

1) You could look at each of the other variables one variable at a time to see if it is a confounder.

2) You could look at combinations of potential confounders.

Looking at One Variable at a Time

You could look at each of the other variables one variable at a time to see if it is a confounder.

Consider this case-control study where the crude OR is 2.2.


This is a dangerous practice because it ignores the joint effects of the two potential confounders (the third and fourth variables).

Looking at Combinations of Potential Confounders

If we formed four strata based on the combinations of the two potential confounders, we would see the following.
Each of the four strata shows no effect. This illustrates the point that variables which when evaluated alone show no effect (ie no confounding or no interaction) may show confounding when evaluated jointly.

Approaches for More than One Potential Confounder

Backward versus forward confounder evaluation strategies are relevant both for stratification and multivariate regression modeling. In other words, we are selecting for the best model