Principles and Practice of Pharmaceutical Medicine

Generalizability

A multicenter trial may be viewed as a number of
identical small trials each conducted at a different
center. From this perspective, each center can be
viewed as repeating the study conducted in other
centers. In addition, different centers draw their
subjects from different geographic areas and thus
a multicenter trial is more likely to enroll subjects
who are representative of a cross-section of the
general population. Consistency of the results
among the different centers adds to the level of
confidence that the results could be replicated any-
where. It is possible that the results across centers
are inconsistent. There are two types of inconsis-
tencies.

(a)The magnitude of the effect is different across
centers. When the magnitude of the response to
treatment is different across centers, the rela-
tive effect between the two treatments is
approximately constant, treatment, referred to
by statisticians as ‘center effect’. The existence
of a center effect means that the different cen-
ters contribute differently to the measured
effect of treatment, but this contribution is the
same for both the experimental treatment and
the comparator. This situation is illustrated in
Figure 25.1 below.

Figure 25.1 shows schematically the effects of

two treatments across six centers. The magni-

tude of the treatment differs from center to

center, but the difference between the effect of

treatment A and treatment B is the same. Such a

situation does not present a problem in compar-

ing the treatments. It makes it impossible,

though, to talk about the absolute magnitude of

the treatment effect, as it is not constant. Obser-

ving a center effect is not unusual in clinical

trials. The reasons for this may be many. It could

be the result of a difference inthe type of patients

seen at the different centers, the center proce-

dures and general nursing care, subjects compli-

ance in taking their medication, the equipment

used in the different centers and so on.

(b)Treatment-by-center interaction: This is the

type of inconsistency that may cause an inva-

lidation of the entire study. Here, the relative

response to the different study treatments is

different across centers. There are two situa-

tions that present qualitatively different levels

of difficulties:

(i) Quantitative interaction. We say that the

interaction is quantitative if the relative

effect of the different treatments is in the

same direction across centers, although the

magnitude may be different. Figure 25.2

illustrates this type ofinteraction. This type

of interaction means that the relative effi-

cacy of the treatments is different in differ-

ent centers, but the direction is always the

same. That is, treatment A is more effica-

cious than treatment B at all centers, but the

magnitude of the difference between the

treatment effects is different in different

centers. When this type of interaction

occurs, one can say that one treatment is

more efficacious than the other, but cannot

say by how much because the relative effi-

cacy of the two treatments is not constant.

(ii)Qualitative interaction. The type of inter-

action is the one that could invalidate the

entire study. This occurs when the relative

efficacy of the two treatments is different

across the different centers both in magni-

tude and direction. This is illustrated in

Figure 25.3. Here, treatment A produces a

0

2

4

6

8

10

12

14

16

A B C D E F

Center

Response

TRT A

TRT B

Figure 25.1 Center effect

334 CH25 STATISTICAL PRINCIPLES AND APPLICATION IN BIOPHARMACEUTICAL RESEARCH