EPI10-2: CONFOUNDING: AN ILLUSTRATED EXAMPLE |
A prospective study of liver disease amongst bartenders found the following data:
Alcohol | Tobacco | Person-years | Cases | Incidence Rate |
---|---|---|---|---|
No | No | 75 000 | 75 | 0.001 |
No | Yes | 25 000 | 50 | 0.002 |
Yes | No | 25 000 | 50 | 0.002 |
Yes | Yes | 75 000 | 300 | 0.004 |
Ignoring the information on tobacco use:
Alcohol | Incidence Rate | Crude Incidence Rate Ratio |
---|---|---|
Yes | 0.00350 | 2.8 |
No | 0.00125 |
If smoking is more common among those who use alcohol than among those who don't and if smoking is a cause of the disease then the observed results could occur without alcohol being a cause of the disease.
To examine this we stratify the data according to smoking in Tables 3 and 4.
Tobacco | Alcohol | Incidence Rate | Stratum Specific relative Risk (RR) |
---|---|---|---|
Yes | Yes | 0.004 | 2 |
Yes | No | 0.002 |
Tobacco | Alcohol | Incidence Rate | Stratum Specific relative Risk (RR) |
---|---|---|---|
No | Yes | 0.002 | 2 |
No | No | 0.001 |
As you can see, a confounder will change the crude effect to an adjusted effect within each stratum. The stratum based effects will however be identical or very similar and combinable in some way to provide an effect adjusted for the confounder, if the extraneous variable is indeed a confounder..