Using Adaptive Research Design to Define the Proper Methodology to Use a Data
Peek for Power: Step by Step Process
Introduction
Research on outcomes of particular treatments, whether hospital based, or
pharmacologic/device studies, at the Phase III and IV level are often conducted to
examine effectiveness of treatments as compared to standard of care. These studies
can be performed to determine therapeutic or costs differences, efficacy over time,
patient compliance, longitudinal treatment changes and many other reasons. Late
phase studies can also be performed to demonstrate that two therapies are clinically
equivalent, but that one therapy costs less or is less traumatic to patients when used
in practice. As a result, statistics used in these studies can include tests for
differences, non-inferiority or equivalence.
1,2
To conduct these studies, statisticians
have an arsenal of approaches to use, and standard significance levels of 0.05 are
seldom applicable in application.
For example, a recent study examining two different types of ventilation
used in an ICU. A study was performed to determine if a less expensive ventilator
performed as well as a more expensive one which required more staff involvement
and maintenance with increased cost. In this study patient outcomes were examined
for equivalence with a conservative p-value of p<0.20 to catch any indication of
patient outcome differences. After the study it was determined that the patient
outcomes were not statistically different (p>0.20) and the less expensive ventilator
was used. In a second example two types of forearm fracture fixation were
examined (internal versus external) in a pediatric Randomized Clinical Trial
(RCT). In this study, the researchers used a difference in treatment design, without
specifying which method was thought to work better (2-tailed hypothesis), with a
very strict p-value of 0.01 required for significance and powered the study
accordingly, only willing to consider one method of fixation superior to the other
if there was over-whelming effectiveness with a p-value less than 0.01. At the end
of the study neither method was found to be convincingly more effective.
Often, when designing or performing this research, the exact effect of
therapy and the variation of the treatments are not known. In these studies,
experimentation is often done in the Real-World Environment (RWE): within
hospitals, with a convenience sample of patients that are available during the time
in which the researcher (often a resident physician) has access to them. These
studies are far outside of the pure clinical trial setting where conditions are well
controlled. Patient behavior and characteristics of RWE studies are far less
controlled than the Randomized Clinical Trial (RCT) at Phases I, II and III (Table
1). As a result, the treatment effects and other statistics of interest that are published
internally and externally while pharmaceutical products are being developed in the
RCT setting, are not directly transferable to the RWE experimental context. The
US. Food and Drug Administration understands that sample size recalculation
during the research may be needed in order to properly complete a study
3
. Often