The European Committee for Medicinal Products for Human Use (CHMP) released a document entitled "Reflection paper on methodological issues in confirmatory clinical trials with flexible design and analysis plan" (CHMP/EWP/2459/02) on March 23, 2006. The paper is available on the EMEA website (Download).

The draft document was released for consultation with the deadline of September 30, 2006.

It is explained in this section that the paper focuses on adaptive designs in confirmatory trials (late-stage Phase II or Phase III trials). It will be helpful to provide guidance or recommendations for the use of adaptive strategies in an exploratory setting (dose-finding Phase II studies). In addition, the role of data monitoring committees in adaptive clinical trials needs to be described with emphasis on sponsor's involvement in data monitoring committees, especially in adaptive trials with complex decision rules.

It is noted in this section that the use of data-driven adaptation can lead to operational bias (for example, intentional or unintentional dissemination of the interim results) thus the trial's sponsor is responsible for protecting the integrity of the trial. This issue is not specific to adaptive trials and the risk of introducing operational bias in clinical trials with adaptive elements is comparable to the risk of introducing operational bias in widely used group sequential studies. Dissemination of the interim results can be prevented by careful planning and implementation of operating procedures used in classic group sequential trials.
One can theoretically consider statistical methods for studying the magnitude of operational bias (for example, tests aimed at detecting trends in the patient characteristics). However, it is important to carefully examine operating characteristics of these methods (a routine application of population drift tests may lead to spurious results). Secondly, it will be difficult to establish a cause-and-effect relationship and prove that the observed population drift is directly related to adaptive decisions made in the trial.
Finally, the section describes operational bias due to implicit or explicit dissemination of the interim results and does not discuss other types of operational bias, for example, bias introduced by the knowledge that patients enrolled later will receive a more effective treatment.

The paper needs to provide a more detailed description of acceptable sample size reassessment strategies, for example, discuss the use of blinded and unblended sample size reassessment methods and comment on the use of other adaptive options, for example, an adaptive selection of the patient population.

It is stated in this section that "...potential differences in the patient population may well produce different estimates of the treatment effect and this may question the combination of results of stages where the placebo arm has been stopped after an interim analysis. Consequently, all attempts should be taken to maintain the blind and restrict knowledge that recruitment to the placebo arm has been prematurely stopped." This refers to explicit operational bias as well as the bias introduced by the knowledge that patients enrolled later in the study will receive a more effective treatment (known as accrual bias) and thus these statements actually apply to a very broad class of adaptive designs. It is not clear to what extend it will be prudent to withhold information on design changes during the study from the study protocol to minimize the magnitude of the accrual bias.

This section appears to be too short and high-level and for an important topic like this one. A more detailed discussion of issues that need to be considered in the context of seamless Phase II/Phase III trials needs to be provided, including conditions under which seamless Phase II/Phase III trials can serve as pivotal trials and data collected in a seamless Phase II/Phase III trial can be combined to perform a single analysis.