| |
RCT investigators
should be highly focused on the hypothesis they wish to test.
Because RCTs are expensive to design and conduct, and because
pragmatic RCTs (see M. Max,
Chapter 1, The Design of Clinical Trials for Treatment of Pain)
can influence how TMD will be managed for many patients, trials
evaluating clinical interventions should focus on an important
endpoint, best reflected as a single primary outcome variable
(or two) that can be measured accurately and reliably.
It is not too much
to say that the design and conduct of an RCT begins and ends with
considerations of what the primary outcome measure should be -first
in conceptual terms and then pragmatically. Pragmatically, the
primary measure must be operationalized, that is, defined in terms
of numerical values that can be statistically analyzed, whether
they are binary (pain improved or not), categorical (pain gone,
reduced, same or worse) or continuous (pain level 4.1 out of a
0 to 10 scale) in nature.
Experience gained with
the use of VAS scales in TMD (see http://rdc-tmdinternational.org
for examples of these scales) allows estimates to be made regarding
the change in pain that is expected or hypothesized to result
after a planned, controlled clinical treatment intervention. The
psychometric properties--i.e. the reliability and validity--of
VAS for assessing human pain have been extensively studied (Seymour
et al., 1985; Price
et al., 1994).
Deciding on using the
outcome measure of pain report and how much change in pain report
from baseline to a later date is expected to be observed with
that measure are essential first steps in the design of the actual
RCT because it is the statistical properties of the outcome measure
and the predicted effect size of the clinical treatment intervention
that will have the strongest effect in determining the sample
size of the study.
The types of outcome
measures used in TMD clinical trials have not been uniform and
have varied from study to study, from number of muscle sites painful
to palpation, amount of increase in mandibular opening, composite
jaw function indices, to VAS scales as discussed (Davies
and Gray, 1997; Lundh
et al., 1992; Anderson
et al., 1985). Less than half of published TMD splint RCT's
have used self-reported pain scales such as the VAS or CPI, although
the use of these scales is growing. Overall quality of life measures
have been rarely used, and there is a strong impetus with other
conditions to include these as one of the outcome measures.
Finally, if more than
one outcome measure is used, additional hypotheses are inevitably
being tested-at least one for each outcome measure. Under conditions
of more than one primary outcome measure, the statistical power
of measures to detect significant differences has to be corrected
for the increase in the number of hypotheses being tested with
each measure; the very practical implication is that more--sometimes
many more--subjects are needed to test RCTs using several primary
outcome measures.
To avoid this sticky
but important theoretical issue underlying the power of statistical
tests, it is customary to identify the main outcome of interest,
such as pain as the primary outcome measure, and then relegate
to secondary analyses the testing of hypotheses associated with
other outcome measures, such as the amount of time a splint was
worn or a numerical rating of patient satisfaction with the splint.
|