Dear List,
My supervisor is facing a statistical predicament outlined in the
question below. I hope some list members will recognize his scenario
and the trend it may represent. We would warmly receive any help or
expertise.
-Brett Larive
QUESTION:
1. Frequently we confront clinical investigators who wish to
perform randomized clinical trials using a clinical endpoint as
the primary outcome, but due to funding or logistical constraints
can employ limited sample sizes which are sufficient to detect
only very large effects. The detectable effect size is typically
regarded as not totally outside the realm of possibility, but
larger than is expected and much larger than the minimum clinically
imortant effect. In an attempt to lend respectability to
the trial the investigators tend to use the term "Pilot Study"
to refer to it, as a way of acknowleging the limited power. However,
from our statistical persecptive such studies are not true
pilot studies because there are no concrete plans to follow them
with definative Phase 3 trials with adequate power.
The clinical investigators' logic typically is that if their
study shows "interesting trends", then there may be a reasonable
chance of convincing funding agencies to support a definitive
trial in the future.
There is an extensive statistical literature on Phase II clincial
trials conducted for the explicit purpose of screening interventions
to determine which interventions should be evaluated in Phase III
trials. There is also a literature dealing with preliminary
trials for screening interventions based on surrogate endpoints,
with the idea that interventions shown to effect the surrogate
endpoints would then be investigated in full scale trials. There is
also a literature on the conduct of true pilot studies, defined as
preliminary trials to evaluate logistical issues regarding the conduct
of a planned full-scale phase III trial. But the scenario we are
concerned with does not appear to fall into any of these 3 categories
because there is no clear linkage between the conduct of the
proposed trial with a future adquately powered phase 3 trial.
To our knowlege, the statistical literature refers to such trials
as underpowered Phase 3 studies and strongly discourages
their conduct. But given the high frequency of such "low-powered"
Phase 3 trials in spite of the best efforts of statisticians, we
would like to investigate what statistical approaches may have
been developed that may be appropriate to guide their design and
analysis. Clearly, bayesian approaches might be considered. We
would be interested in the perspecitve of others regarding this
issue.
