The need to conduct cardiovascular safety assessments for new type 2 diabetes drugs has raised concerns within FDA and industry about statistical methods, data confidentiality and blinding for CV outcomes studies.

During a recent DIA meeting in Washington, D.C., on diabetes drug development, FDA said sponsor proposals for CV safety programs often fail to adequately control for Type 1 error (a false positive), which could make a drug appear safer than it really is.

Furthermore, agency officials appear to disagree on the level of confidentiality that should be afforded to interim analysis of data from an ongoing outcomes study that is used to support NDA approval, but continues post-approval.

FDA’s December 2008 guidance requires diabetes drug sponsors to exclude an 80% or higher CV risk pre-approval. This can be done through a meta-analysis of Phase II/III studies or a single large safety trial.

If the pre-approval clinical data show that the upper bound of the 95% confidence interval for the estimated risk is between 1.3 and 1.8, a post-marketing trial will generally be required to rule out a 30% or higher increased risk. This can be done through a single, adequately powered trial or by combining the results from a pre-marketing safety trial with a similarly designed post-marketing study.

The new guidelines are forcing companies to re-think their commitment to the diabetes area (Also see "Glycemic Control Not Enough For New Diabetes Drugs, Roche Exec Says" - Pink Sheet, 18 October, 2010.).

Many Paths To Reaching The Goalposts

Sponsors have proposed a variety of designs to meet the 1.8 and 1.3 benchmarks, Hylton Joffe, Diabetes Team I lead medical officer in FDA's Division of Metabolism and Endocrinology Products, said. These include a superiority trial to show CV benefit, a dedicated trial in patients with recent acute coronary syndromes to show CV safety, a traditional Phase II/III program plus a dedicated safety trial, and glycemic efficacy sub-studies within a dedicated safety trial.

Some sponsors have proposed starting a CV safety trial and using interim analysis data to satisfy the 1.8 threshold for approval, but also continuing the study to reach the 1.3 post-approval goal.

However, conducting multiple analyses of a trial at different time-points raises numerous statistical problems, particularly with regard to controlling for Type 1 error, Joffe said.

“This is one common problem we’ve seen where sponsors want to take multiple looks to see if they’ve made 1.8 and the concern here is that we’re relying on that decision – whether you’ve made 1.8 or not – as part of the basis for deciding whether we’re going to approve the drug, so we want to be correct on that,” he said. “If you do multiple looks without controlling Type 1 error, we could erroneously conclude evidence of CV safety when there is none.”

FDA may provide more advice on statistical approaches to control for Type 1 error in a forthcoming guidance on diabetes drug development.

FDA may provide more advice on statistical approaches to control for Type 1 error in a forthcoming guidance on diabetes drug development. That document will serve as the final version of a February 2008 draft guidance on general development issues but also will incorporate the December 2008 final guidelines on CV risk assessments (Also see "FDA Plans To Add Specifics To Statistical Part Of Antidiabetics CV Guidance" - Pink Sheet, 12 July, 2010.).

Protecting Trial Integrity

Proposals for outcomes studies that begin before NDA submission and continue afterward have generated questions about how the ongoing trial’s integrity will be protected during FDA’s review of interim data in an NDA.

Ilan Irony, medical leader of FDA’s Diabetes Team II, said the study’s integrity could be maintained by having the data monitoring committee share unblinded data with the agency. “That requires the applicant only to know whether they met the 1.8 or not for filing, not the specific numbers,” he said.

Details about the interim data would not become public during FDA’s review process, Irony said. “The only thing publicly available is … [that] the company has met the 1.8 goalpost for filing for approval.”

However, Center for Drug Evaluation and Research Deputy Director for Clinical Science Robert Temple quickly challenged this view. “Won’t there be a detailed description in labeling and in your reviews of what they found?” he asked.

Irony suggested the interim results could be described but the sponsors would remain blinded to the data. “I can’t imagine that,” Temple responded. “I think we’ve got to get over the idea that you’re not going to have a blinded outcome for the 1.8 and get used to whatever correction you have to make for the 1.3, if any.”

Mary Jane Geiger, senior medical officer at Lilly, said sponsors must pre-specify who will have access to unblinded, interim data that is included in an NDA. Public release of such data would seriously jeopardize an ongoing trial’s integrity, she warned.

Geiger referenced concerns about bias that followed publication of an unplanned interim analysis of GlaxoSmithKline’s open-label RECORD study of Avandia’s CV safety (Also see "FDA Drug Review Staff Split On Avandia's CV Safety In RECORD Study" - Pink Sheet, 9 July, 2010.). “If we don’t learn from those lessons we’ve failed,” Geiger said.”

By Sue Sutter

Diabetes CV Studies Spur Questions On Statistical Methods, Data Integrity

More from Archive

More from Pink Sheet