The most common source of bias is one resulting
from subjects being selected to the different treat-
ment groups in a way that creates an imbalance in
one or more prognostic variables. This type of bias,
known asselection bias, is usually the result of
unconsciousaction onthe part of theinvestigator or
other people involved in the enrolment of subjects
into the trial, or of a faulty treatment allocation
method. Randomization is designed to take the
treatment assignment decision away from the
enrolling investigator and place it in the hands of
chance. Unfortunately, it is not foolproof. An
investigator who has a personal preference for
one treatment over another for a particular type
of subject may decide to postpone enrolling a sub-
ject until the ‘right’ treatment comes up on the
randomization schedule. Also, there are many
other ways, that are not affected by randomization,
in which the investigator can influence the trial
outcome. A simple talk with a subject reinforcing
the subject’s confidence in the efficacy of treatment
can often have a real or transient effect on the
subject’s response to treatment.
Another potential source of bias is the subject
himself/herself. Often, the mere expectation that
the drug will have a therapeutic effect produces an
effect. This effect is known as theplacebo effect,
and in some cases, it could be considerable.
To counteract these types of bias, CCTs are
generallyblinded. That is, the identity of the treat-
ment is concealed from everybody who can influ-
ence the treatment assignments and any procedure
that could impact the trial outcome. When the
treatments are masked from both the investigator
and the subject, the trial is calleddouble blind.In
drug trials, blinding is accomplished by using pla-
cebo, an inert substance, as a non-active control
and identically looking packaging, for the different
treatments with labels that do not reveal the iden-
tity of the drug.
The use of double-blind randomized clinical
trials has become the gold standard for good clin-
ical research. However, it is not always possible to
mask the treatments. A trial designed to compare
the effectiveness of two surgical procedures, for
example, cannot be blinded. Another example is
a trial comparing an intravenous drug to an oral
drug. In principle, one could blind such a trial by
delivering an inert substance (e.g. saline) intrave-
nously to the oral drug group and an oral placebo to
the intravenous group. However, this procedure
might be controversial because subjects are
exposed to additional risk, albeit small, without
direct potential benefit to them. When the com-
parators have distinct characteristics that would
identify them, blinding can be achieved by using
the so-called ‘double-dummy’ method unless it is
ethically unacceptable. The ‘double-dummy’
method means that all subjects receive identically
looking treatments only one of which is active and
the others are placebos. For example, in a compar-
ison of two oral drugs, one of which is a tablet and
the other a capsule, each subject receives a tablet
and a capsule, one of which contains the treatment
assigned to that subject and the other is placebo.
Sometimes even the ‘double-dummy’ method is
not helpful. The drug might have a characteristic
profile, such as identifiable smell, taste, or a spe-
cific adverse event or other biological effect that
would reveal the identity of the treatment either to
the investigator or to the subject or both no matter
how the drug is packaged or labeled. When blind-
ing is not possible, special efforts must be made to
minimize the possibility of introducing bias by
incorporating appropriate bias prevention methods
in the study design. Once bias is introduced, it is
very difficult and sometimes impossible to adjust
for it at the analysis stage.Stratification
An efficient study design is one that maximizes the
‘signal-to-noise ratio’. Thus, controlling the
‘noise’, or variability, is an important aspect of a
good design. Consider the following example.
A graduate student in Public Health is conduct-
ing a research project on the health-related habits
of the students at her University. As part of
the project, she measured the resting heartbeat of
20 student subjects. The results are listed in
Table 25.3.
The mean and standard deviation are 56.8 and
3.57, respectively. The student further divided the
subjects into two groups: Group A consists of
subjects who do aerobic exercises regularly and320 CH25 STATISTICAL PRINCIPLES AND APPLICATION IN BIOPHARMACEUTICAL RESEARCH