FDA To Consider Critical Path Trial Design Proposals On Case-By-Case Basis

FDA will generally take a case-by-case approach to considering and accepting development programs using clinical trial designs discussed in its Critical Path opportunities list.

The ¹ list, released March 16, includes a section on opportunities for streamlining clinical trials. The section mentions many areas of trial design and analysis that FDA has addressed in the past, including adaptive, enrichment and non-inferiority designs.

Sponsors have expressed interest in using some of these designs and analytic methods; however, they have said that they are generally hesitant to do so for fear of an unfavorable review by FDA (² (Also see "The Next Phase: FDA Looks Beyond Three-Stage Drug Development Model" - Pink Sheet, 13 Feb, 2006.), p. 10).

Key factors that FDA will consider in reviewing new designs will be whether they are acceptable "on the face" and have statistical rigor. One avenue for sponsors to get feedback on trial designs is the agency's special protocol assessment program.

Sponsors have expressed particular interest in using adaptive trials that allow for the adjustment of trial parameters based on interim analyses.

Such designs could make trials more efficient - both in terms of time and cost - by allowing certain arms to be dropped during the trial, modification of randomization or bridging of Phase II and III studies, the list states.

Adaptive designs could also be used to identify and validate - "learn and confirm" - biomarkers as surrogate endpoints within a pivotal trial. Establishing a process for validating biomarkers is a key goal of the Critical Path initiative (³ (Also see "Biomarker Qualification Guidance To Require Fit For Use Validation" - Pink Sheet, 27 Mar, 2006.), p. 25).

One "learn and confirm" trial design FDA feels would likely be acceptable is modeled on a design for evaluating gene expression signature for sensitive patients proposed by the National Cancer Institute's Boris Freidlin and Richard Simon in the Nov. 1, 2005 issue of Clinical Cancer Research.

The so-called Adaptive Signature Design would compare an active compound with a control arm while also identifying and evaluating patient subsets with higher response rates.

Under the design, the Phase III trial would be divided up into two stages: the first stage would accrue a portion of the overall study population and would be used to identify subgroups of high-responders, while the second stage of the trial would randomize the remaining patients to look at the benefit in the identified subgroups.

The final analysis would compare the treatment effect of the active and control arms in the overall trial population (both stages), and in the subsets of potentially higher responders randomized in the second stage. The trial would be considered positive if either of the two analyses were significant.

To maintain statistical rigor, the ? of the overall trial would be divided for the two analyses. The authors recommend that 80% of the trial's ? be devoted to the primary analysis, with the remaining 20% devoted to the subgroup analysis.

"Because the size of the treatment effect in the identified subset may be much greater than in the overall study population, analysis of the subset in patients accrued during the second stage of the trial at a stringent significance level may still provide substantial power," Freidlin and Simon note.

Such a trial design could be particularly useful for targeted therapies, where certain subsets of patients are often found to benefit more than others. Simultaneously analyzing subgroups and the overall population would allow sponsors to seek approval for the limited population even if the overall test fails.

Another way to evaluate efficacy in specific subpopulations is to enrich a trial with patients with certain characteristics that put them at greater likelihood of an outcome or make them more likely to respond to a compound. As opposed to "learn and confirm" trials, enrichment trial sponsors need to be reasonably confident the subpopulation's characteristics are predictive of higher response in advance of conducting the trial.

The agency believes that biomarkers may be even more useful as a tool for trial population enrichment than as validated surrogate markers for efficacy endpoints.

According to FDA's opportunities list, enrichment designs raise a number of key issues including: "How will data on the marker status of potential trial enrollees be used in trial design?"; "How much data are needed on the un-selected population?"; and "What types of retrospective subset analyses are valid (i.e. what can be reliably learned from subgroup analyses that were not prespecified in the original trial design)?"

FDA said it is willing to consider allowing subgroups as the primary efficacy evaluation population, so long as safety data is available in the broader treatment population (⁴ , p. 9).

The agency has generally said that non-prespecified subset analyses are not reliable and should be viewed as hypothesis-generating; however, FDA has said that with the move to personalized medicine, such analyses may become more valuable.

The clinical trials section of the opportunities list also seeks to develop new methods for establishing confidence intervals for non-inferiority trials. FDA has said it is willing to consider less conservative approaches to establishing non-inferiority intervals, primarily through Bayesian statistical methods (⁵ (Also see "FDA Considering “Less Conservative” Approaches To Non-Inferiority Trials" - Pink Sheet, 27 Jun, 2005.), p. 10).

The opportunities list says that new methods for establishing non-inferiority intervals would be of particular use for therapeutic areas such as oncology that often require a single trial for approval and have a paucity of prior trial data to estimate the treatment effect of the comparator.

Manufacturers have said that FDA's approach to non-inferiority trial design for oncologics is too "rigid" (⁶ (Also see "FDA Approach To Non-Inferiority Trials For Oncologics Too Rigid, Bristol Says" - Pink Sheet, 15 Aug, 2005.), p. 12).

The list suggests "it might be possible to use biomarker data to circumvent some of these difficulties" for these therapeutic areas.

The opportunities list also discusses the potential for the use of virtual control groups for rare diseases or in cases where use of placebo is infeasible or unethical.

"Databases, models, and/or imaging collections could be used by multiple sponsors across different product types as historical controls to reduce the necessary size of control groups in clinical trials," the list states. "Trusted third parties could be used to hold data or images and create an open source library."

FDA and the Pharmaceutical Research & Manufacturers of America are planning to conduct a workshop in November on the use of historical controls.