EPI10-3: EFFECT MODIFICATION DEFINITION AND ILLUSTRATION |
If one stratifies the data by the levels of the third variable (not the exposure or the outcome) it may turn out that in the different levels or strata of the third variable the effects are different from each other. It follows that these different stratum-specific effects will also be different from the crude effect. This type of third variable because of what it does, is called an Effect Measure Modifying variable or in a more concise way, an effect modifier.
This is because an effect modifier modifies the effect of exposure on the outcome as one goes from stratum to stratum. This changing effect by stratum is not something one wants to control for or get rid of, as it describes reality as it is, and the usual approach is to show the different effects for each stratum.
Data from another similar study showing this time effect measure modification instead of confounding are shown in Table 5:
Alcohol | Tobacco | Person-years | Cases | Incidence Rate |
---|---|---|---|---|
No | No | 75 000 | 75 | 0.0010 |
No | Yes | 25 000 | 37 | 0.0015 |
Yes | No | 25 000 | 37 | 0.0015 |
Yes | Yes | 75 000 | 450 | 0.0060 |
Ignoring the information on tobacco use, we get the crude rate ratio in Table 6:
Alcohol | Incidence Rate | Crude Incidence Rate Ratio |
---|---|---|
Yes | 0.00487 | 4.35 |
No | 0.00112 |
The crude rate ratio of 4.35 in Table 6 becomes 4 and 1.5 in the strata of tobacco use in Tables 7 and 8.
Tobacco | Alcohol | Incidence Rate | Stratum Specific relative Risk (RR) |
---|---|---|---|
Yes | Yes | 0.0060 | 4 |
Yes | No | 0.0015 |
Tobacco | Alcohol | Incidence Rate | Stratum Specific relative Risk (RR) |
---|---|---|---|
No | Yes | 0.0015 | 1.5 |
No | No | 0.0010 |
Here the stratum specific effect measures are quite different and need to be reported as such.