The medical concept of interaction is synonymous
to the terms heterogeneity and synergism. Interaction must be
distinguished from confounding. In a trial with interaction effects
the parallel groups have similar characteristics. However, there
are subsets of patients that have an unusually high or low
response. The above figure gives an example of a study in which
males seem to respond better to the treatment 1 than females. With
confounding things are different. For whatever reason the
randomization has failed, the parallel groups have asymmetric
characteristics. E.g., in a placebo-controlled trial of two
parallel-groups asymmetry of age may be a confounder. The control
group is significantly older than the treatment group, and this can
easily explain the treatment difference as demonstrated in the
previous chapter.
Example of Interaction
A parallel-group study of verapamil versus
metoprolol for the treatment of paroxysmal atrial tachycardias.
The numbers of episodes of paroxysmal atrial tachycardias per
patient are the outcome variable.
|
Verapamil
|
Metoprolol
|
||
|---|---|---|---|
|
Males
|
52
|
28
|
|
|
48
|
35
|
||
|
43
|
34
|
||
|
50
|
32
|
||
|
43
|
34
|
||
|
44
|
27
|
||
|
46
|
31
|
||
|
46
|
27
|
||
|
43
|
29
|
||
|
49
|
25
|
||
|
464
|
302
|
766
|
|
|
Females
|
38
|
43
|
|
|
42
|
34
|
||
|
42
|
33
|
||
|
35
|
42
|
||
|
33
|
41
|
||
|
38
|
37
|
||
|
39
|
37
|
||
|
34
|
40
|
||
|
33
|
36
|
||
|
34
|
35
|
||
|
368
|
378
|
746
|
|
|
832
|
680
|
||
Overall metoprolol seems to perform better.
However, this is only true only for one subgroup (males).
|
Males
|
Females
|
|
|
Meanverapamil (SD)
|
46.4 (3.23866)
|
36.8 (3.489667)
|
|
Meanmetoprolol (SD)
|
30.2
(3.48966) −
|
37.8
(3.489667) −
|
|
Difference means (SE)
|
16.2 (1.50554)
|
−1.0 (1.5606)
|
|
Difference between males and females 17.2
(2.166)
|
||
There is a significant difference between the
males and females, and, thus, a significant interaction between
gender and treat-efficacy. Interaction can also be assessed with
analysis of variance and regression modeling. These two methods are
the methods of choice in case you expect more than a single
interaction in your data. They should be carried out on a
computer.