The U.S. Food and Drug Administration’s Accelerated Approval program provides an important route for the development of new medicines for rare diseases where the traditional approval pathway may be lengthy or infeasible. As two people who have devoted our lives to developing a drug for a rare disease — prion disease — Sonia and I have recently started trying to educate ourselves on what Accelerated Approval is and how it works. This blog post is my first attempt at synthesizing what I’ve learned so far. I’m still learning, so if you see something I’ve gotten wrong, please leave a comment to let me know.
The backstory here dates to the FDA’s reaction, in the late 1980s and early 1990s, to the HIV/AIDS epidemic. FDA approved the first HIV/AIDS drug, zidovudine (AZT), in 1987 (see this timeline), after AZT showed remarkable success in a conventional clinical trial paradigm. 282 people with AIDS had been randomized to receive either AZT or placebo for 24 weeks [Fischl 1987]. They were monitored for survival, opportunistic infections, and other clinical phenotypes such as weight loss. The AZT-treated participants fared better than the placebo participants on all of these dimensions. The survival outcomes were especially dramatic: only 1 person who received AZT died during the trial, compared to 19 placebo-treated people.
In parallel, the participants were also monitored for CD4 cell counts, which proved to increase — transiently but dramatically (about 2-fold) — in the AZT group and not in the placebo group. CD4 count is not a clinical endpoint — you can’t directly see or feel what your CD4 cell count is — but the loss of these immune cells results from viral replication and weakens the immune system, allowing the secondary infections that characterize AIDS, so CD4 count is a measurement very tightly linked to disease stage and outcome.
Traditionally, drug approval has been based on (in addition to safety, of course) a demonstration of clinical benefit: either survival, or some measure that affects people’s quality of life, has to be improved in people getting a drug compared to placebo-treated controls. But the link between CD4 cell counts and AIDS progression was so tight, and the need for new HIV/AIDS drugs so urgent, that an idea took hold. Perhaps, in the future, simply showing that a drug raises CD4 cell counts, compared to a placebo, could be enough to strongly suggest that it will yield a clinical benefit. This could be a basis for granting tentative approval for a drug while confirmatory studies to demonstrate clinical benefit are still underway.
A few years later, the FDA formalized this idea in a “rule” (a decree from a regulatory agency, as opposed to a law) that became part of the Code of Federal Regulations [21 CFR 314 §§500-560]. The rule stated that for:
…[for] certain new drug and antibiotic products that have been studied for their safety and effectiveness in treating serious or life-threatening illnesses and that provide meaningful therapeutic benefit to patients over existing treatments… FDA may grant marketing approval for a new drug product on the basis of adequate and well-controlled clinical trials establishing that the drug product has an effect on a surrogate endpoint that is reasonably likely, based on epidemiologic, therapeutic, pathophysiologic, or other evidence, to predict clinical benefit… Approval under this section will be subject to the requirement that the applicant study the drug further, to verify and describe its clinical benefit, where there is uncertainty as to the relation of the surrogate endpoint to clinical benefit… When required to be conducted, such studies must also be adequate and well-controlled. FDA may withdraw approval… if… A postmarketing clinical study fails to verify clinical benefit
In other words, if a drug for a deadly disease could be shown to have a positive effect on a surrogate endpoint (a biomarker such as CD4 cell count) that was reasonably likely to predict a clinical benefit, then that drug could be eligible for a sort of tentative approval. The drug could be marketed and given to patients even as clinical trials to demonstrate an actual clinical benefit were still underway and incomplete. If those trials ended up showing that the drug delivered a clinical benefit, then it would advance to full approval, and if the trials failed to show a clinical benefit, then the drug would be withdrawn. This was a big change from historic practice where a drug had to affect a clinical outcome — how patients feel or function — in order to be distributed outside of a clinical trial.
The political backdrop for this shift is critically important. Historically, the political pressure faced by FDA has been a classic example of baseline bias, whereby if FDA action (approving a drug) results in any harm (adverse events), then heads will roll, whereas if FDA inaction (failing to approve a drug) results in harm (patient deaths that could have been prevented), there are no consequences. In the 1970s, FDA Commissioner Alexander Schmidt was quoted [Stahl 2005] saying:
In all of FDA’s history, I am unable to find a single instance where a Congressional committee investigated the failure of FDA to approve a new drug. But, the times when hearings have been held to criticize our approval of new drugs have been so frequent that we aren’t able to count them… The message to FDA staff could not be clearer. Whenever a controversy over a new drug is resolved by its approval, the Agency and the individuals involved likely will be investigated. Whenever such a drug is disapproved, no inquiry will be made.
The HIV/AIDS patient activism of the 1980s and 1990s was perhaps the first political force to push hard enough in the opposite direction to make a dent. For the first time, the fallout for FDA if it failed to approve a drug soon enough might be big enough to rival the concern about approving a drug that might turn out to be unsafe or ineffective.
By the time the rule was put forth, on April 15, 1992, it reflected a shift that FDA had already undertaken. FDA had already approved didanosine (DDI), a new nucleoside analog reverse transcriptase inhibitor (NARTI; the same mechanism of action as AZT) in October 1991 on the basis of improved CD4 cell counts. Almost immediately after the new rule was proposed, the FDA granted the first official Accelerated Approval, on June 22, 1992, for zalcitabine, another NARTI for the treatment of HIV/AIDS. The early clinical trial data apparently showed that zalcitabine in combination with AZT raised CD4 cell levels by more than AZT alone. (Those early data are described in this NIH document although I didn’t find anything in Google Scholar — the early data may have only appeared in the New Drug Application that Roche filed with FDA). Because AZT raised CD4 levels and also improved survival, it stood to reason that zalcitabine, which rasied CD4 levels, would also improve survival, thus meriting Accelerated Approval.
The post-approval studies eventually confirmed that zalcitabine did indeed improve survival: people treated with AZT plus zalcitabine had a 32% lower mortality rate over the 2.5-year-long trial than people treated with AZT alone [Darbyshire 1996]. Interestingly, in the meantime, it had been suggested that viral load (PCR quantification of HIV RNA) was a better predictor of clinical outcome than CD4 count [Mellors 1995], and studies of zalcitabine confirmed that the drug’s effect on viral load was actually the better predictor of benefit than was CD4 count [Katzenstein 1996], presaging a shift towards viral load as the “surrogate endpoint” for future Accelerated Approvals in HIV/AIDS drugs.
The FDA’s new rule was subject to a comment period (some of the comments and discussion are pretty interesting) and the final rule was issued issued on December 12, 1992 [Stahl 2005]. In time, Congress eventually codified FDA’s new practice (along with implementing several unrelated reforms) in a bill, the FDA Regulatory Modernization Act of 1997 [Pub. L. 105-115]. That law never uses the term “Accelerated Approval”, instead using the term “Fast Track”. This can be confusing because today, the latter term is understood to refer to FDA’s Fast Track designation, which offers drug companies quicker review and more frequent meetings with FDA for drugs that address unmet medical needs, as distinct from Accelerated Approval, though of course many of the same drugs qualify for both programs.
Laws, in general, are remarkably short on detail: the Modernization Act codifies the phrase “reasonably likely” but never defines what it means. The whole section on Accelerated Approval (then known as Fast Track) [21 USC §356] is only about one page long. FDA often issues “Guidance for Industry” documents explaining in much greater detail how exactly it intends to implement the provisions of laws. One 1998 guidance document on “Providing clinical evidence of effectiveness” does touch on the issue of surrogate endpoints:
A pharmacologic effect that is accepted as a validated surrogate endpoint can support ordinary approval (e.g., blood pressure effects, cholesterol lowering effects) and a pharmacologic effect that is considered reasonably likely to predict clinical benefit can support accelerated approval… When the pharmacologic effect is not considered an acceptable effectiveness endpoint, but the linkage between it and the clinical outcome is strong, not merely on theoretical grounds but based on prior therapeutic experience or well-understood pathophysiology, a single adequate and well-controlled study showing clinical efficacy can sometimes be substantiated by persuasive data from a well-controlled study or studies showing the related pharmacologic effect.
In other words, there are three strata of how good your surrogate endpoint can be:
- Surrogate endpoints that are already considered validated, such as blood pressure or cholesterol, can merit full approval
- Surrogate endpoints that are reasonably likely to predict clinical benefit can merit Accelerated Approval.
- Surrogate endpoints that have strong linkage to a clinical outcome don’t merit approval on their own, but can help a drug to achieve full approval based on only one trial showing a clinical benefit, rather than two.
Thus, a surrogate endpoint can merit either Accelerated or regular approval, depending upon the weight of evidence behind it. Here’s an aside: confusingly, not all clinical endpoints are created equal either. Although I’ve focused on surrogate endpoints in this post, the rule [21 CFR 314 §§500-560] also stipulates that Accelerated Approval can be granted for drugs “on evidence of the drug’s effect on a clinical endpoint other than survival or irreversible morbidity”. In 2009-2014, this clinical endpoint rationale was invoked for Accelerated Approval in orphan diseases several times, almost always for cancer [Kakkis 2016, see table in Supplementary Information]. I have to admit that I am still confused on how this works and why we need this provision. Wasn’t it already always possible for drugs to get full approval based on clinical endpoints other than survival or irreversible morbidity? Otherwise how did the drugs that we all take for everything from headaches to seasonal allergies ever get approved? Indeed, even for fatal diseases, drugs sometimes receive full approval just to treat symptoms, even when no one believes that they modify the underlying disease process — for instance, tetrabenazine in Huntington’s or memantine in Alzheimer’s. So why do some drugs only get Accelerated Approval, rather than full approval, even though they do affect a clinical endpoint? One paper I read gave the example that, in cancer, full approval might be given for a drug that causes complete disappearance of a tumor, while Accelerated Approval might be given for a drug that causes some tumor shrinkage but not full remission [Stahl 2005]. Another argued that, again in cancer, progression-free survival over the short term might only be considered an intermediate clinical endpoint, because if that benefit were to turn out to come at the expense of longer-term survival, then the drug wouldn’t merit approval [PCAST 2012]. That makes sense, but I am still not sure I have a full grasp of the rationale. If anyone reading this post can help me understand when and why Accelerated vs. regular approval might be granted for a clinical endpoint, please leave a comment at the bottom of this post.
Anyway, the above breakdown of validated vs. reasonably likely vs. strong linkage clarifies things somewhat, but the meaning of these terms is still not entirely clear. What exactly qualifies as “reasonably likely”? In practice, a plurality of all Accelerated Approvals granted in the 1990s and 2000s were for HIV/AIDS drugs, where the well-worn endpoints of CD4 counts and, later on, viral load, were avaiilable [Kakkis 2016]. The easiest way to get Accelerated Approval appears to be to re-use an endpoint whose reasonable likeliness has already been established through extensive clinical experience, particularly through clinical trials of earlier drugs for the same or related conditions. For diseases that are both rare and untreatable, this is a pretty tough hurdle to clear. Apparently from 1992 to 2014, only one rare disease drug ever received Accelerated Approval based on a novel biomarker, meaning a biomarker that hadn’t already been established through a previous drug approval. That drug was agalsidase beta for Fabry disease, approved in 2003 based on ameliorating lysosomal storage in kidney biopsies [Kakkis 2016]. Meanwhile, the bar for having a “validated” biomarker meriting full approval is at least as high: of the rare disease drugs received full approval based on a surrogate endpoint in 2009-2014, ten out of ten used a biomarker whose predictiveness of clinical benefit had already been established through clinical trials of earlier drugs [Kakkis 2016].
There’s no doubt that Accelerated Approval was instrumental in ushering in new drugs for HIV/AIDS and several other diseases. Still, as of a few years ago, the bar for Accelerated Approval had proven to be out of reach for many conditions, particularly rare diseases. Perhaps partly in response to this, Congress introduced language expanding Accelerated Approval into the Food and Drug Administration Safety and Innovation Act (FDASIA), which was signed into law in July 2012 [Pub. L. 112-144]. The stuff relevant to Accelerated Approval is under §901:
During the 2 decades following the establishment of the accelerated approval mechanism, advances in medical sciences, including genomics, molecular biology, and bioinformatics, have provided an unprecedented understanding of the underlying biological mechanism and pathogenesis of disease. A new generation of modern, targeted medicines is under development to treat serious and life-threatening diseases, some applying drug development strategies based on biomarkers or pharmacogenomics, predictive toxicology, clinical trial enrichment techniques, and novel clinical trial designs, such as adaptive clinical trials.
As a result of these remarkable scientific and medical advances, the FDA should be encouraged to implement more broadly effective processes for the expedited development and review of innovative new medicines intended to address unmet medical needs for serious or life-threatening diseases or conditions, including those for rare diseases or conditions, using a broad range of surrogate or clinical endpoints and modern scientific tools earlier in the drug development cycle when appropriate. This may result in fewer, smaller, or shorter clinical trials for the intended patient population or targeted subpopulation without compromising or altering the high standards of the FDA for the approval of drugs.
Patients benefit from expedited access to safe and effective innovative therapies to treat unmet medical needs for serious or life-threatening diseases or conditions.
For these reasons, the statutory authority in effect on the day before the date of enactment of this Act governing expedited approval of drugs for serious or life-threatening diseases or conditions should be amended in order to enhance the authority of the FDA to consider appropriate scientific data, methods, and tools, and to expedite development and access to novel treatments for patients with a broad range of serious or life-threatening diseases or conditions.
The bill (which also contained loads of other stuff) passed with overwhelming bipartisan support (96-1 in the Senate!). As before, the law itself doesn’t get too deep into the specifics. What, for instance, might this “broad range” of surrogate endpoints include? But the intent is clear: we should take advantage of all that modern molecular biology has taught us about the etiology of disease in order to approve drugs more quickly for patients who really need them.
Within a couple of years, FDA issued a Guidance for Industry document, “Expedited Programs for Serious Conditions - Drugs and Biologics”, explaining how exactly it interprets Congress’s intent and how it intends to implement the provisions of FDASIA. A few main take-homes are:
- The more severe, rare, and untreatable a disease is, the more flexible FDA will be in accepting surrogate endpoints. To wit: “The new provisions provide additional flexibility concerning the implications of available therapy on eligibility for accelerated approval” and “FDA should take into account, ‘…the severity, rarity, or prevalence of the condition…’”
- A wider variety of types of evidence will be considered when deciding whether a surrogate endpoint is acceptable. Such as “pharmacologic or other evidence developed using biomarkers or other scientific methods or tools, in conjunction with other data”.
- When deciding whether to accept a surrogate endpoint, FDA will consider how strong is our biological understanding of the disease, and how a biomarker proposed as a surrogate endpoint relates to it. It helps if a disease is more thoroughly understood: “…the extent to which the pathophysiology of a disease is understood is an important factor in determining whether an endpoint is reasonably likely to predict clinical benefit”, and it helps if it’s clear how a biomarker fits into a disease pathway, that helps: “Whether there is reliable and consistent epidemiologic evidence supporting the relationship between the endpoint and the intended clinical benefit.”
All of this is a bit more specific than the text of FDASIA, but still leaves a lot of latitude for interpretation. This is deliberate: the document states that it does not specifically enumerate the lines of clinical evidence that will or won’t get a surrogate biomarker accepted, because “such evidence is case-specific and is not readily generalizable.” Biology is complicated, every disease is unique, and trying to write any hard-and-fast set of rules that make sense for every drug and every indication would be impossible. Instead, FDA uses its scientific judgment in each case.
Because neither the letter of the law nor the letter of the Guidance provide an unambiguous formula for what surrogate endpoints do and don’t merit Accelerated Approval, pharma and biotech decision-makers considering whether to invest in a drug have to do some reading of the tea leaves. One obvious set of tea leaves to read is the record of precedent: for what types of surrogate endpoints has FDA awarded Accelerated Approval previously? 102 drugs have received Accelerated Approval since 1992, but FDASIA is so new (and the Guidance even newer) that not many New Drug Applications (NDAs) have been reviewed under the new framework yet, and presumably no NDAs have yet been reviewed for drugs that were developed with the new framework in mind from the beginning. Because precedent is thin, in conversations Sonia and I have had with private sector people, I have heard a wide range of interpretations and speculations as to what FDA might or might not do.
The Guidance document states, for example, that “Evidence of pharmacologic activity alone is not sufficient”, but the term “pharmacologic activity” is never defined. Can increasing or decreasing the level of a disease-associated protein be acceptable? For instance, consider the case of an antisense oligonucleotide targeting a pre-mRNA in order to affect the inclusion of exons in the mature mRNA. If you use a Western blot or ELISA to measure the level of the protein translated from that altered mRNA, is that just “pharmacologic activity”? Or would using qPCR to measure the mRNA be the “pharmacologic activity” whereas the protein level is downstream of that activity? A few months ago, there was no precedent for a drug being granted Accelerated Approval based on changes in a protein’s abundance, and so most people I spoke with assumed it was impossible, but last month FDA granted Accelerated Approval to eteplirsen for Duchenne muscular dystrophy on the basis of its raising dystrophin protein levels in muscle biopsies.
Another important question is whether FDA will consider how difficult it would be to show a clinical benefit, when deciding whether to award Accelerated Approval based on a surrogate biomarker endpoint. FDASIA itself states that FDA “…shall consider how to incorporate novel approaches into the review of surrogate endpoints… especially in instances where the low prevalence of a disease renders the existence or collection of other types of data unlikely or impractical” [Pub. L. 112-144 §506(c)(3)], perhaps implying that the more impossible a clinical endpoint would be, the more flexible FDA should be about accepting a biomarker endpoint instead. The Guidance doesn’t explicitly address this, though. There is one paragraph discussing the situations where Accelerated Approval applies, stating that Accelerated Approval is useful when a biomarker can be assessed rapidly, whereas a clinical endpoint would require very long and/or very large trials, either because the disease is slowly progressive like HIV or because clinical events in the disease are acute but occasional. Is this just included as an FYI about when Accelerated Approval is typically sought? Or does it mean that the impracticality of a clinical endpoint is actually one criterion that FDA will weigh in assessing a surrogate biomarker endpoint? Emil Kakkis, a prominent advocate for expanded use of Accelerated Approval, has argued that in many diseases, such as those where there is a long presymptomatic progression phase, or those where a biomarker can be quantified precisely but all available clinical endpoints are noisy, biomarkers might be the only route to ever develop a drug [Kakkis 2016]. But he’s also been quoted saying that when engaging FDA, “you can’t plead poverty… you have to plead science”. Pharma types who I’ve spoken with have expressed differing opinions of whether or not FDA might take this sort of consideration into account.
A question that the Guidance does not address, but that may matter to FDA, is whether the drug in question has potential for widespread off-label use. The President’s Council of Advisors on Science and Technology writes that “The FDA interprets its statutory responsibility to mean that it may deny approval to a drug with a favorable benefit-risk balance in a narrow population if the drug will pose serious risks to a broad population. Drugs have also been removed from the market because they were being used off-label in a way that raised serious concerns” [PCAST 2012]. So it might be the case that FDA will be more flexible about accepting a surrogate endpoint in a biologically unique rare disease for which a drug has little or no potential for off-label use, whereas it might impose stricter standards for a surrogate endpoint if it thinks that Accelerated Approval in a rare disease might lead to rampant off-label prescriptions for a common disease with shared biology. But this is my speculation — I haven’t managed to find anything written by FDA that states this as a criterion.
Kakkis has pointed out that it is only after the drug has been developed, tested, gone into humans and produced Phase 2 data that FDA will pronounce on whether a particular surrogate endpoint is acceptable for a particular drug in a particular disease. Thus, biotech and pharma companies risk making a tremendous up-front investment only to find out that, even if the drug does prove to have the hoped-for effect on a biomarker, the biomarker itself is considered unacceptable. This concern may lead many companies to be quite conservative, never investing in developing a drug in the first place unless a hard clinical endpoint is plausibly in sight [Kakkis 2015]. You can imagine that this in turn probably means that many of the biomarker-based New Drug Applications that FDA might otherwise be reviewing never get submitted in the first place, thus contributing to the dearth of precedents. Kakkis’ proposed solution is that the FDA should implement a formal biomarker qualification process, whereby before an Investigational New Drug (IND) application is ever submitted, indeed, before any preclinical work even begins, a company can apply to have a biomarker certified as meriting Accelerated Approval [Kakkis 2015]. That way, they’d be able to invest in developing a drug having in hand the knowledge that if they can successfully raise or lower X protein or metabolite by Y percentage points, or resolve Z pathology in the disease-relevant tissue of interest, then Accelerated Approval will follow.
For clarity, the type of biomarker qualification process Kakkis is proposing here is distinct from the Biomarker Qualification Program that does exist as part of FDA’s Drug Development Tools program. As explained in the FAQ, that program is oriented towards biomarkers that will be used for multiple drugs, not just for one specific rare disease program. The associated Guidance document also makes it clear that the bar for qualification as a surrogate endpoint is very high, and that most biomarkers evaluated through this program will be used for things like safety assessments, diagnostics, or enrollment criteria.
So far, then, it appears that FDA has not implemented Kakkis’s idea of a pre-qualification program, although it has created more opportunities for meetings and discussions that can provide non-binding guidance on what FDA will or won’t consider acceptable. For instance, Critical Path Innovation Meetings (Guidance here) allow opportunities for early discussion of biomarkers, along with clinical endpoints, natural history studies, and clinical trial design; the Fast Track designation allows companies to have earlier and more frequent meetings with FDA “to discuss the drug’s development plan and ensure collection of appropriate data needed to support drug approval”; and FDA has hosted symposia on biomarkers, such as last year’s workshop on dystrophin quantification.
If there’s one thing that’s clear from all this, it’s that there is no formula and there are no hard-and-fast rules. FDA scientists will use their scientific judgment in each individual case, based on all available data about the drug and about the biomarker and about the disease. FDASIA is still young, and the precedents on record to date may not tell the full story about what will or won’t be considered acceptable in the future.
Sonia Vallabh and Margaret Orseth contributed research to this post.