Effect modification happens when a particular variable has separate, different exposure effects depending on another variable.
One way of looking at this is by thinking of your test pool as having subgroups, or strata; Effect modification happens when an exposure or treatment has a different effect and leads to a different outcome among different subgroups or strata. A very simple example of effect modification would be when a particular cancer drug has different effects in men than in women.
Note that in effect modification, the dependent variable changes—even while the independent variable is kept constant—because of change in a third variable.
Why Effect Modification is Important
Effect Modification is very important in clinical studies because it allows us to:
- Define high-risk subgroups,
- Take preventive, protective actions to ensure the health of subgroups.
- Be more precise in estimating effect,
- Conduct meaningful comparisons among different studies that might have different proportions of effect-modifying groups.
- It can also be crucial to medical research as it gives a starting point for developing a causal hypothesis for the disorder or the disease.
How to Study Effect Modification
Locating effect modifications is often a trial-and-error process. The first step is usually collecting information on potential effect modifiers. Then, if you are designing an experiment or study, you will want to modify the study to test these potential effect modifiers, by incorporating comparisons between different subgroups.
If the experiment or research project is already performed, you can still study effect modifiers while analyzing the data. A first step would be stratifying the data by possible effect modifiers, and then running stratum-specific estimates. If you locate a difference, you will need to run a test to find out if it is statistically significant given the size of your data set.
Corraini et. al. Effect modification, interaction and mediation: an overview of theoretical insights for clinical investigators. Clinical Epidemiology. 2017; 9: 331–338. Published online 2017 Jun 8. doi: 10.2147/CLEP.S129728. Retrieved from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5476432/ on August 5, 2018.
Pennsylvania State University Eberly College of Science. Stat 507 Lesson 3: Measurement Exposure Frequency; Association between Exposure and Disease; Precision and Accuracy. 3.5 – Bias, Confounding and Effect Modification. Course Notes. Retrieved from https://onlinecourses.science.psu.edu/stat507/node/34/ on August 5, 2018.