Monday, January 13, 2020

Pre-specifications of the statistical analysis plan - story of Ampligen in Chronic Fatigue Syndrome

In the previous post, 'Pre-specification and SAP', various regulatory guidelines and guidance were cited to demonstrate the importance of the pre-specification of the statistical analysis plan in clinical trials - especially the pivotal clinical trials. In the eye of the FDA reviewers, the most critical issue is not to choose or switch the statistical analysis method after the study unblinding or after the sponsor has the knowledge of the unblinded results. If the cherry-picking is allowed, a negative study might be turned into a positive study. 

One example is a drug Ampligen developed by Hemispherx Biopharma Inc. Hemispherx developed Ampligen for the treatment of chronic fatigue syndrome (Myalgic encephalomyelitis). Prior to the PDUFA date, an advisory committee meeting was organized. The discussion of whether or not Ampligen is efficacious was focused on the statistical analysis using untransformed data (result not statistically significant)or log-transformed data (result statistically significant). FDA accused the sponsor of changing to the statistical analysis using untransformed data after the planned analysis using log-transformed data was not statistically significant. 

To “log transform” or not to was one of the major issues the FDA raised in their background report; the fact that log untransformed data had shown that Ampligen had a significantly positive benefit while log-transformed data indicated the drug (almost did but) didn’t have that effect.

This technical question would dog Hemispherx throughout the hearing; and they would ultimately answer it but one had the feel that it was too late…that the damage had been done. Hemispherx’s statistician, with years of FDA experience under his belt, showed instructions from the FDA stating that log transformation should not be used unless necessary because it could skew the data, and then

The FDA officials were focused on something else, though, when and why Hemispherx decided to log transform or not the data. The big question was whether Hemispherx saw the data before it decided whether or not to transform the data. The statistician appeared to argue that Hemispherx had to check the variance to determine if the transformation was warranted but in the end stated the biostatistician who prepared the data was no longer with the company and they didn’t know. That was a huge blow…

Never mind that Hemispherx had demonstrated that the log transformation data was not warranted or that non transformed data was appropriate …the FDA was mostly interested in whether the biostatistician had ‘followed the rules’. It was a bizarre thing as a patient to watch a drug that could help ill people be held up on procedural issues but there it was.

This is how the FDA approves drugs? More impromptu debate than rigorous analysis the discussion session kind of flowed along from topic to topic with a pro-FDA moderator calling the shots. In fact, an actual debate, with each side arguing pro’s and con’s of each issue would have been much better.

Instead of issues being drawn up, presented on the screens and then discussed in an organized manner with each side given equal time, the conversation lurched from topic to topic with the Hemispherx reps being frustrated spectators too many times.

After lunch, Hemispherx had appeared confident even after the morning pummeling they’d taken from the FDA team. They felt they had answers to the FDA’s concerns but after the meeting, several members of the team felt they simply were not provided the opportunity to produce them. The moderator did call on Hemispherx several times but, for the most part, the FDA personnel to held the floor.

The short early discussion period clearly left many questions hanging at a time when the issues were fresh in the reviewers minds but the later discussion period felt hurried as well.

Given the ad hoc nature of the discussion period drugs, companies must shake in their boots and investors must tremble when they approach these meetings. Then again, this is an FDA that gives companies 220 pages of background materials and a list of questions to be answered two days before the meeting. The FDA team clearly has the upper hand in these hearings and that’s apparently how they want it.
The consequence is that Ampligen was voted down by the advisory committee and NDA was subsequently rejected by FDA. The sponsor gave up the further pursuit of the Ampligen for the treatment of CFS. Instead, the sponsor changed its name to AIM ImmunoTech Inc. and retool the Ampligen for oncology indications.
On the question: "Based on the information included in the briefing materials and presentations, has the applicant provided sufficient efficacy and safety data to support marketing of Ampligen for the treatment of CFS?," the AAC voted 8 no, 5 yes and 1 non-vote.


Norman Williams, UCL said...

Very interesting! So a potentially useful drug was "discarded" because of a procedural glitch. Can the FDA be certain that they have acted in the best interests of patients in this particular case?

Anyway, I'm interested in what follows "and then..." in paragraph four.

EJ said...

Thank you for sharing this story.
Yes, it is essential to provide the pre-specification and SAP for every clinical trial.
As a biostatistician, sometimes it is hard to know the outcome measures, I mean the exact values for outcomes to decide what statistical methods should be used. Do you have any suggestions for this?