(JOSE LUIS PELAEZ, JON BOYES, GETTY IMAGES ILLUSTRATION: PAUL A. BELCI)
|
To ensure commonality of understanding, establishing what is meant by "adaptive trial" is warranted. A working definition
of an adaptive trial is as follows: An adaptive trial is one that uses information obtained within the study to modify, or
potentially modify, the structure or other design feature of the study.
From the perspective of statistical analysis, an adaptive design can affect alpha (require some adjustment for multiplicity)
or can preserve alpha without adjustment for multiplicity, where alpha is the probability of a type I error—which most commonly
is the probability of declaring that a drug is effective when it is not effective. In early phase development studies where
alpha has no important role, adjustment to alpha is not essential because it is not relevant.
|
It is noted that when a study includes a provision to make a modification (e.g., an interim analysis with a reassessment of
the sample size), the study is considered to be an adaptive trial whether or not a change was made. Also, when a study does
not include a provision to make a modification but a modification is made through the amendment process, the study is considered
to be an adaptive trial (e.g., if it is determined that the sample size will not support the study's achievement of its goals
and a decision is made to modify the sample size). When information within the study is used to make a modification or potentially
to make a modification, then the study is an adaptive study.
Apples and orangesBy contrast, in a traditional clinical trial a protocol is developed and executed with no modifications—ideally. Amendments
are tolerated, but when there are too many a stigma is attached to the study. Introducing flexibility in a clinical study
is contrary to the well-planned, well-executed nature of the traditional study.
One driver of the traditional plan-your-work and work-your-plan approach to research is to protect alpha—the probability of
a type I error, which, in the context of a placebo controlled superiority study is the probability of concluding that a drug
works when it doesn't. The rigid nature of the traditional trial protects the statistical analysis from inflated alpha for
efficacy evaluation.
Another driver of the traditional approach to planning the study and executing it without modification is the protection of
the study against sources of bias. Knowledge of results of the study can influence the accruing results, thereby damaging
the study's integrity. A traditional study often mitigates this kind of bias by using an independent Data Monitoring Committee
for reviews of unblinded data or accumulating results.
This source of bias is an issue for an adaptive study because the protocol includes provisions to alter the study structure
or other design feature. The risk of this bias should be evaluated in the context of the overall goal of the study and its
role in the drug development plan; if this bias is not important in context, then the risk is low. If the study is intended
to be used as a source of pivotal evidence of efficacy, this bias would be important.
Regulatory considerations
Although adaptive trials are not new, there is a new enthusiasm for the flexible methodology, and the recent popularity is
in part related to encouragement from the regulatory arena for creative pathways to shorten timelines to bring new therapies
to market. Regulatory agencies have communicated that, under certain conditions, flexibility in a trial can be acceptable.1,2,3
Table 1. Synopsis of the EMEA views on adaptive design based on a document issued in 2007.
|
EMEA issued a document, "Reflection Paper on Methodological Issues in Confirmatory Clinical Trials Planned with an Adaptive
Design,"4 (See Table 1) and FDA has as a published PDUFA goal the release of a draft guidance shortly. It should be emphasized that
encouragement from regulatory agencies to make use of adaptive trials is accompanied by caution against inappropriate use.
Regulatory risk depends on the regulatory role of the study. If the study is not used as a basis of approval, regulatory risk
is minimal. There must be adequate basis for dose selection, but demonstrating statistically significant differences between
the selected dose and a de-selected dose is not required.
If the alpha associated with the analysis of differences between doses is inflated, it is of little or no concern to regulatory
agencies if the evaluation of efficacy does not rely on that study for key support. Confirmatory studies need to be uncompromised.
For basis of approval, the issue still goes to weight of evidence. If remarkable efficacy is found, some compromise can be
tolerated. If the study conclusion is less robust, compromise becomes more of an issue.
Scientific risks
The regulatory risks are truly scientific risks as well. There are, however, other scientific risks that would not be of particular
concern to regulatory authorities. If the study does not have an important regulatory role, then the regulatory agency would
not be concerned with the compromise, hence the sponsor bears the scientific risk.
In an adaptive study, presumably in early development, when data review occurs as data is accrued, it is tempting for sponsors
to make decisions based on small amounts of data. While having a small amount of data is better than having no data, there
is still risk of over-interpreting the trends (favorable or unfavorable) and making wrong decisions. Including statistical
assessment as a routine part of the clinical and medical evaluation can help to mitigate this scientific risk.
When a later stage study is planned as adaptive, scientific compromises are introduced when the personnel involved in the
ongoing study receive information about accumulating results about the study.
In any study with compromise, whether traditional or adaptive, the evidence of efficacy must be sufficiently substantial to
overcome obstacles of potential bias or inflated alpha.
Discussion
Table 2. A description of the types of adaptive design modifications and adjustments to alpha needed.
|
Benefits of an adaptive design are clear: potentially shorter timelines, fewer subjects exposed to de-selected doses or treatments,
fewer subjects in a study, and less cost. The adaptive design permits a study to adjust to the accumulating body of knowledge
and data pool to guide continued clinical research. [The most common types of adaptive studies include modifications of the
type listed in Table 2.]
An adaptive trial that includes more than one study under the same protocol permits a small (or pilot) study to be done followed
immediately by another study without the delays and resource investment of study start-up. Pilot studies are useful in situations
where data are not available or are too limited to support the required decision making.
It should be noted that an adaptive design is not a remedy for poor planning. In the absence of good data to plan a study,
an adaptive study can include doing a pilot study within the larger study. A scenario could be that an initial dose is selected
for evaluation, data are collected, analyzed, evaluated, and the next step in the program is made on the basis of real data
as opposed to relying solely on animal data and guesswork.
