8 INITIAL STEPSnumber of patients, or patients may have to be accepted at several medical centers in
order to have a sufficient sample size.
Random assignment to treatment is still the recommended procedure, and methods
of performing the assignment have been devised to fit the needs of clinical trials. If
possible, it is recommended that neither the patient nor the professionals who interact
with the patient or evaluate the outcome know whether the patient is getting the new
treatment or the control treatment. When this can be accomplished, the clinical trial is
called a double-blind trial. The blinding of the patient and physician can be achieved
in drug trials by having the hospital pharmacist place the two possible medications
in identical-looking pills that are placed in bottles stacked in random order, and the
pharmacist just hands out the next bottle. This ideal condition is not possible in all
instances. For example, it is difficult if the treatment is surgery, where what was done
is obvious to the patient or surgeon, or if different side effects occur in a drug trial that
the physican can recognize. For further information on this topic, see Mould [ 19981,
Pocock [1983], Wooding [1994], Spriet et al. [1994], or Piantadosi [200.5].
A multicenter study is one method of performing clinical trials when the condition
under study is rare and no single medical center has access to a sufficient number
of patients to reach a valid conclusion in a reasonable time period. This requires
that a common protocol be developed where all the centers agree on what patients
will be admitted, how they will be treated, how the measurements will be made,
and a minimum set of data that they will all collect. Since medical centers have a
tendency to have their own procedures for treatment and data collection, multicenter
studies sometimes require considerable negotiation to get agreement on a single set
of procedures. However, it can solve the problem of inadequate sample size, and if
multiple centers with somewhat different patients and physicians reach a conclusion
on a given treatment, it tends to increase confidence in the results.
When a number of researchers have already performed clinical trials on the same
topic, such as the comparison of two specific treatments for a given medical condi-
tion, a meta-analysis can be considered. The idea behind a meta-analysis is to pool
the results from different studies on the same topic, and by reanalyzing the combined
results be able to reach conclusions with greater certainty. A successful meta-analysis
requires a clear statement of the purpose of the study, rigorous identification of rel-
evant studies with a clear statement of inclusion and exclusion criteria, collection of
relevant data, and finally, an appropriate analysis (see Piantadosi [200.5] and Sutton et
al. [2002]). One way of thinking of meta-analysis is that a successful meta-analysis
may yield results from medical studies that are somewhat similar to what would be
obtained with a multicenter study. One potential problem in meta-analysis is the non-
retrieval of relevant information. Another concern is that some journals have what is
called publication bias since they tend to preferentially accept manuscripts that have
positive (significant) results over those with negative results (no differences in the
treatments). A good analysis of poorly done studies is unlikely to yield good results.
Statistical software for performing meta-analysis can be found by going to Google
and typing “meta-analysis.”