Regulatory Approval in ALS; When Is a Single Study Enough?

On March 30, 2022, the US Food and Drug Administration (FDA) Division of Neurology convened an Advisory Committee to recommend whether AMX0035 data "established a conclusion that Sodium Phenylbutyrate/Tauraursodiol is effective in the treatment" of patients with amyotrophic lateral sclerosis (ALS), based on the single randomized placebo controlled trial and the blinded open label follow-up study (CENTAUR trial). The FDA stated their previous view that a single study could justify approval of an agent if the target disease was serious, fatal, and inadequately treated, and the findings of the study were robust. However, they stated their view that, despite the study meeting its primary endpoint, the study was not robust. They based their view on a number of perceived inadequacies; here, we will address them and argue strongly that virtually all of the criticisms are undeserved. The FDA argued that the results of the CENTAUR study were not robust. The Centaur study was a 24 week randomized, placebo controlled trial of 137 people with ALS. The primary outcome measure was the rate of decline in the Amyotrophic Lateral Sclerosis Functional Rating Scale-Revised (ALSFRS-R), a 48-point scale that measures function in several domains and declines with disease progression. The CENTAUR study and its associated open label follow on component met its primary end point, showing a 25% reduction in the rate of progression of the ALSFRS-R and a reduction in the hazard risk of death of approximately 0.60, on top of the standard of care.1, 2 Although this is certainly not a cure for ALS, the open comment section of the meeting clearly demonstrated the robustness of this level of efficacy for people living with ALS. The FDA also judged lack of robustness based on secondary end points not reaching statistical significance, and contended that the overall level of evidence was less than that demonstrated in the single pivotal study of edaravone in ALS.3 However, the CENTAUR study was not powered on secondary end points, yet showed a strong trend toward protection of pulmonary function and a statistically significant impact on upper extremity strength. Coupled with the clear survival effect, this more than adequately compares to the edaravone study, which slowed loss of function on the ALSFRS-R by approximately 33%. In addition, it showed a trend with respect to pulmonary function, did not study strength, and showed no survival signal. There was an effect on the Amyotrophic Lateral Sclerosis Assessment Questionnaire (ALSAQ-40), a 40-question assessment of health status that closely tracks to ALSFRS-R. In a disease where deaths are likely during the conduct of a placebo-controlled trial, it is important to account for the impact of death on a functional end point. Dr. David Schoenfeld, a leading ALS biostatistician, introduced the joint rank test, which was used in the phase III trial of dexpramipexole in ALS.4 Both Dr. James Berry, an ALS neurologist and trialist who spoke during the open comments, and Dr. Schoenfeld recognize that the joint rank test is less sensitive than other methods when deaths are expected to be rare, such as in the CENTAUR study.5 This point was confirmed by a study using multiple simulations in 2018,6 in which the joint rank performed systematically worse than other shared methods when deaths were sparse or if a therapeutic effect was seen primarily in function rather than survival. This information was presented multiple times to the FDA; their persistence in insisting that this test be used is puzzling, as it goes against scientific consensus. However, the joint rank test was performed as suggested by the FDA and showed the same overall findings as in the primary analysis. It was accurately noted at the advisory meeting that the slope of decline of ALSFRS-R was greater in CENTAUR than in previous studies; this was presented as a surprising finding casting doubt on the validity of the placebo group. However, rather than being a surprising finding, this was an intended result of the inclusion criteria for this study, selected using multiple simulations to provide a cohort in which an efficacy signal could be perceived in a short study with the intended sample size. Rather than a problem, this was an example of innovative trial design. The FDA casted doubt on a dramatic survival impact by suggesting that the multiple time points at which mortality was assessed somehow reduced the confidence in the finding that survival was extended by approximately 5 to 6 months. This was a puzzling criticism, as the impact of survival was clear and statistically significant at all time points investigated; just looking at the survival curves makes it obvious that survival was greater on those participants initially randomized to AMX0035. The FDA statistician raised questions about whether the assumption of linearity was appropriate for the analyses used and showed a puzzling figure in which the slopes of decline using both linear and quadratic models looked essentially identical but which he claimed showed a failure of linearity. However, in response to a question from the advisory panel, the FDA statistician confirmed that sensitivity tests used by both the sponsor and the FDA did not show that the linearity assumption was violated. His questions about linearity nevertheless left the mistaken impression that lack of linearity was an issue. There was an initial error at the beginning of the trial, in which the initial 18 participants were randomized to active treatment. After discovery, the following 9 participants were randomized to placebo, and the randomization schedule was as intended. The sponsor and those managing the study were NOT aware of this, so they had no opportunity to make decisions regarding it. This error was discussed publicly and is made clear in the New England Journal of Medicine paper in which the primary results were published. An analysis evaluating the primary outcome with the first 27 participants excluded showed no difference in the estimate of effect of AMX0035. This was certainly an unfortunate event, but neither participants nor the study management team were unblinded. We do not believe that this invalidates the study results in any way. Without any evidence to support their concern, the FDA raised the question of whether there was participant unblinding based on the bitter taste of AMX0035. However, the matching placebo was designed to have an equivalent taste, and a survey of participants at the end of the placebo controlled period clearly demonstrated that study completers could not accurately guess whether they had been on placebo or active treatment. This is despite the observation that gastrointestinal (GI) adverse effects were seen in participants on AMX0035; the differences between placebo and active treatments on GI adverse events were overall not significantly different. Nevertheless, the FDA cast doubt on whether the study was unblinded. When asked at the Advisory meeting, the FDA confirmed that they had no data to suggest unblinding. The issue of the adequacy of the blinding process is relevant to every clinical trial, but there is no reason to be particularly suspicious of blinding in CENTAUR. Placebo and treatment groups were well matched, but small differences were noted in the percent of participants with bulbar onset of disease, and in the use of edaravone. More participants had bulbar onset in the active treatment group, and more were on edaravone in the placebo group. The differences were small, but, to the extent that there might have been an impact of these differences, both should act to reduce any efficacy signal. Nevertheless, the FDA persisted in casting doubt on the results. Neurofilament levels in either plasma or cerebrospinal fluid (CSF) are thought to be representative of axon or neuron loss. In ALS, levels are increased in patients compared to healthy controls.7 However, there is at present no evidence that changes in neurofilament levels predict the efficacy of an agent. The only data relevant to this question in fact demonstrate that changes in neurofilament are not predictive of efficacy measures in ALS. Nevertheless, in response to a question from the advisory board, the FDA asserted that the lack of a difference in neurofilament levels between placebo and active treatment somehow cast doubt on efficacy signals. We conclude that the FDA concerns about the design and conduct of this study are misplaced. The essential question, therefore, is whether a single, small study can support approval of AMX0035. The FDA is empowered by policy to say yes. Patients and their families definitely say yes. For people living with ALS, waiting 2 to 3 years for results of a second study is essentially a death sentence. The disappointment and frustration was palpable at a patient webinar held on Thursday March 31, 2022, for a different clinical trial. The nearly uniform opinion expressed during the open comment section of the advisory meeting strongly supported approval. On July 15, 2021, the FDA asked Amylyx to immediately file a new drug application (NDA) so they could review the entirety of data on a single study of AMX0035 for people with ALS. They had previously requested the company do a second study which Amylyx had begun. This study will continue whether AMX0035 is approved in the United States or not. Waiting for this study to complete is not an option for the current generation of patients with ALS. There was no debate about the safety of AMX0035 – a drug made up of a supplement (Turso) that has been available for decades, and phenylbutrate, an FDA approved agent. Thus, the risk/benefit ratio for AMX0035 is clear, and favors approval. The ALS community has met many times with the FDA over past years and shared that they are willing to take some risks given the seriousness of their disorder. The evidence of efficacy of AMX0035 is strong; it slows disease progression by 25%. In the OLE extension, a survival benefit between 5 and 6 months was also found. We note that survival was ascertained for 136 of 137 participants; a remarkable achievement made possible by innovative data collection. Throughout the open label extension, no participant or investigator knew the original treatment assignment during the placebo-controlled period. The benefit of timely approval prior to completion of the ongoing phase III trial is that patients will have access to a potentially beneficial agent 3 years earlier than if approval is delayed. The risk is that patients may be taking a drug potentially found to be ineffective. Given the acknowledged safety profile for AMX0035, we believe the course is clear. The FDA has acknowledged flexibility in their approval process; and we urge them to use it in the case of AMX0035. M.E.C. received compensation consultant for Avexis, Sunovian, Takeda, Biogen, Anelixis, Aclipse, Cytokinetics, Disarm, ALS Pharma, RRD, Immunity Pharma, Helixsmith, Denali, Wave, Orion, Quralis, Transposon, Faze, Regeneron, Lilly, Ab Sciences, and Arrowhead. She has institutional received research support grants from institutional grants from the ALS Association, ALSFAC, National Institutes of Health (NIH), UCB, Prilenia, Clene, BIohaven, Seelos, Revalsio, and Implicit. M.E.C. was the co-principal investigator of the AMX0035 clinical trial. She has not received any personal consulting from Amylyx. MEC is a current member of the FDA advisory committee but recused herself for this review. J.M.S. received compensation as a consultant for Amylyx, Apic Biosciences, Neurosense, Cytokinetics, Denali, GSK, Mitsubishi Tanabe Pharma America, Orphazyme, Orthogonal, Pinteon, RRD, Swanbio, Helixsmith, Novartis, Sanofi, and EMD Serono and serves as Neuromuscular Section Editor for UpToDate. He has received research support from NIH, ALS Association, Amylyx, Biogen, Biotie Therapies (now Acorda Therapeutics), Cytokinetics Incorporated, Mitsubishi Tanabe Pharma America, Alexion, Medicinova, Ionis, Alector, and Orphazyme. J.M.S. spoke at the recent FDA advisory committee meeting on AMX0035.