The Big “Data Manipulation” Story

One of the big end-of-summer news stories at FDA is the agency’s August 6 statement that Novartis submitted “manipulated” data to support approval of its gene therapy product, Zolgensma.  According to FDA, Novartis knew about the manipulation before FDA approved the product, and yet the company didn’t disclose the manipulation to the agency until June 28 — more than a month after approval. 

A group of Senators (including Presidential hopefuls…) has said the company’s “greed” cannot be condoned and that FDA should hold the company accountable for its “malfeasance.”  They have also asked FDA why it withdrew a proposed regulation requiring companies to report data falsification.

I take you through a brief tour of the publicly available information, after the jump.  In my next post, I’ll explain the abandoned data falsification reporting proposal.

The usual caveats apply.  I don’t have any inside knowledge.  Without a doubt, there is more information that hasn’t been disclosed publicly by FDA or Novartis.  Still, here’s what the publicly available documents seem to suggest.

What is not at issue

FDA approved Zolgensma (onasemnogene abeparvovec-xioi) on May 24.  Zolgensma is a gene therapy product intended to treat very young children (under two years of age) with spinal muscular atrophy with specific mutations in the SMN1 gene. 

No one is suggesting that anything inappropriate happened in the clinical trials supporting approval.  This is really important, because Zolgensma seems to make a huge difference.  Children born with these mutations usually develop flaccid paralysis within six months of birth, and they are generally dead of respiratory failure by their second birthdays.  When the data were cut off for analysis in the phase 3 trial, nearly every child was living without permanent ventilator support, and nearly half could sit independently for at least 30 seconds. Ordinarily none would be expected to.  And again, there’s no suggestion of malfeasance in the clinical trials; presumably the children received the product pursuant to the protocols, and the data were recorded, analyzed, and reported as planned and required. 

No one is suggesting the product isn’t safe or effective.  Indeed, quite the opposite. According to FDA, the phase 3 trial provided “compelling” evidence of effectiveness, and it showed the product has an overall positive benefit-risk profile.  This is, presumably, why FDA “remains confident” the product should remain on the market and why Novartis has stated that it stands behind the safety and effectiveness of the product. 

But there is a data manipulation issue, so let’s back up . . .

What did Novartis submit in the Zolgensma application?

FDA has posted many of the documents it generated during review of the application.  Novartis included five clinical studies:  (1) a completed phase 1 trial, (2) a long-term follow-up study of some patients from that trial, (3) an ongoing phase 1 trial in other patients, (4) an ongoing U.S. phase 3 trial, and (4) an ongoing global phase 3 trial.  It also included pharmacology and toxicology studies in animals.  The completed phase 1 trial and the U.S. phase 3 trial are the key.

The completed phase 1 trial was an open-label single-dose trial looking at both safety and effectiveness.  There were 15 children; 3 received a low dose, and 12 received a high dose.  For effectiveness, the company looked at the time from birth until death or until the child needed at least 16 hours of ventilator support per day for 14 days in a row.  Two years after the dose was administered, every child in the high-dose cohort was alive without requiring permanent ventilation.  Nine could sit independently for at least 30 seconds, and 2 could stand and walk without assistance.

The U.S. phase 3 trial was also an open-label single-dose trial, ultimately in 21 children with a clinical diagnosis of spinal muscular atrophy before six months of age.  The trial was ongoing when Novartis submitted its application, but data collection ended in March 2019, so these data were a key part of the application.  There were several endpoints relating to effectiveness, such as surviving to 14 months (without permanent ventilator support) and sitting independently for at least 30 seconds by 18 months of age.  At the time of data cut off, one child had died, and one had withdrawn from the study, but the remaining 19 were alive and not using permanent ventilation.  By that point, 13 of the 21 children were 14 months old, and 1 was 18 months old.  Ten could sit independently for at least 30 seconds.  Ordinarily none would be expected to do this.

No one is questioning these results.

If these data weren’t touched, which data were manipulated?

The suspect data relate to the effect of a manufacturing change after the phase 1 trial.

Changing the manufacturing process during a premarket R&D program is not unusual.  But when a company does this, it has to compare the two versions of its product, meaning before and after the change.  In plain English, if the versions are comparable, then the results of testing one version apply equally to the other version.  If the versions aren’t comparable, these data aren’t directly relevant any more.  The data are still important, however, and they still go to FDA.  But they don’t apply directly to the version of the drug proposed for approval.

A comparability exercise usually involves a structural comparison of the two versions (which can be complicated with biologics) and analytical tests.  Sometimes there’s more.  Novartis did several things, one of which was the “SOP-285” test.  Groups of newborn mice were injected with various doses of the two versions of the product.  The FDA scientist reviewing this part of the Novartis application concluded that the results from SOP-285 supported comparability of the two versions.  And the data manipulation — whatever it was — affected SOP-285.  In other words, the results from this mouse test comparing the two versions may not be reliable.  It seems Novartis also submitted the results of a second assay, which (according to the same FDA reviewer) showed that the potency of the two versions was similar. 

What exactly happened during this test?

It seems that someone reported to Novartis’s Chief Quality Officer in March that the data in SOP-285 “may have been mismanaged or even potentially manipulated.”  There don’t appear to be any details in the public documents, which is not surprising if (company and FDA) investigations are ongoing.  But manipulation isn’t always nefarious; sometimes it reflects sloppiness, for instance, or poor training.  I’ll say more about this in my next post.

But I have no idea what happened. We do know, however, that the data from this test were mismanaged or manipulated in some fashion that is not acceptable.

What is the scientific/regulatory significance of this?

Let’s assume that the results of the mouse assay are not reliable. Does this mean that the two versions of the product were not comparable?  No, not necessarily.  And remember that Novartis compared them other ways as well.

Let’s suppose, though, that the two versions were not comparable. What does that mean for Zolgensma?  It means that data from studies using the first version don’t relate directly to the version that FDA approved.  As far as I can tell, this applies to (1) the phase 1 trial and (2) the animal pharmacology study.  <The animal pharmacology study involved 500 mice and showed a dose-dependent improvement in survival.>  If the two versions of the product weren’t comparable, the results from these two studies may still “support” approval, but they don’t directly prove anything about the version that FDA approved.

The rest of the application is unaffected. That leaves, for instance, toxicology studies (involving more than 700 mice) and the phase 3 trial, which showed the product is safe and effective in the children for whom it is labeled. 

What, then, is the big deal?

In one respect — perhaps the only respect that matters, if you are a parent whose infant has just been diagnosed with spinal muscular atrophy — the data manipulation is not a big deal.  The Director of the FDA Office that approved Zolgensma has said that the application still meets the standard of approval, based on the phase 3 trial, with the phase 1 data treated simply as “supportive” evidence. 

What about the data manipulation itself; isn’t that a big deal? Maybe, maybe not. I wouldn’t call mismanagement the same thing as manipulation, and falsification can be a whole other kettle of fish. There is more to the story. I’m disinclined to speculate, let alone assume bad faith. Couldn’t the facts change? Couldn’t it turn out to be a big deal? Sure. Maybe it will turn out the falsification was intentional, and maybe it will turn out there is a systemic problem at the company, and maybe other applications and products are implicated. Systemic data integrity problems happen. Usually, they don’t.

But if these are all the facts, what is the concern?  It’s the timing.  Novartis submitted the application in October 2018.  FDA was supposed to issue its decision in May.  Novartis learned about the data manipulation issue in March, but it didn’t say anything.  FDA approved the application on May 24.  Novartis notified FDA of the data manipulation at the end of June (106 days after it learned of the issue), when it submitted the interim conclusions of its investigation.  Why didnt it say anything while the investigation was ongoing? No idea. But I generally subscribe to Hanlon’s razor when thinking about why people do the things they do.

Would telling FDA before approval have mattered?  Yes. And no. According to the Office Director, if Novartis had told the agency before May 24, FDA would have delayed approval to consider the significance of the data issue.  But, also according to the Office Director, FDA would still have approved the product.  In other words, in the alternative universe (in which Novartis reported sooner), the only thing that changes is when the drug gets approved. Still safe and effective, just not available for patients at the end of May 2019.

** Obviously, I am not saying that Novartis did the right thing here. Or that getting the product in the hands of patients quickly was more important than notifying FDA of the data issue. But anyone considering this situation needs to understand that as of right now, the primary responsible FDA official is saying that knowing earlier wouldn’t have changed anything but the approval date.

Leave a Reply