Within clinical trials, exploratory or post-hoc subgroup analyses are widely recognized as only “hypothesis generating” due to their high potential for bias and/or misleading interpretation. This is the main reason why Sponsors cannot make efficacy claims or seek regulatory approval based on evidence of efficacy in a certain subgroup unless that benefit is consistent with the broader trial population and unless the trial is positive overall for the intention-to-treat (ITT) population. This begs the question, “Is it acceptable to use an exploratory subgroup analysis to restrict an indicated population when the data suggest less benefit in a particular subgroup?”
That is exactly what FDA asked the Oncologic Drugs Advisory Committee (ODAC) to consider in the case of the PROpel data, based on their conclusion that the combination of olaparib plus abiraterone has a favorable benefit/risk only in the subgroup of patients with advanced prostate cancer who test positive for a mutation in the BReast CAncer (BRCA) gene, which regulates homologous recombination repair of DNA. However, one might argue that the exploratory/post-hoc analysis on which FDA based their conclusion remains, by its very nature, fraught with potential for bias and/or misleading interpretation and is thus only hypothesis generating.
“Is it acceptable to use an exploratory subgroup analysis to restrict an indicated population when the data suggest less benefit in a particular subgroup?”
In the era of precision medicine, we expect that treatment choices are driven by biomarkers that can predict clinical benefit. In the case of poly ADP-ribose polymerase (PARP) inhibitors, like olaparib, BRCA mutations or deficiencies in homologous recombination repair (HRR) can predict clinical benefit. But there may be clinical situations where biomarker testing is limited or where patients without BRCA mutations might benefit from treatment with a PARP inhibitor. Indeed, the science suggests that patients with metastatic castration-resistant prostate cancer (mCRPC) may benefit from the combination of a PARP inhibitor with an antiandrogen, like abiraterone, regardless of BRCA mutation status, based on the synergistic activity of these 2 drug classes. In addition, the majority of patients with mCRPC (especially in disadvantaged communities) do not have definitive biomarker testing for BRCA mutations, usually due to cost and/or lack of available tumor tissue. That is the context for the PROpel study investigating the combination of the PARP inhibitor olaparib (Lynparza) plus abiraterone (Zytiga) as first-line treatment of mCRPC.
The PROpel trial was designed to assess the activity of this combination in the broad, unselected, ITT population, and data on BRCA mutation status by ctDNA and tissue tests were collected for the purpose of exploratory subgroup analysis. The trial met its primary endpoint in the ITT population, demonstrating a statistically significant 40% improvement in radiologic progression-free survival (rPFS). Therefore, AstraZeneca was seeking a broad indication that includes BRCA mutant, BRCA wild-type, and BRCA unknown patients. The Sponsor also presented evidence that patients without BRCA mutations or with unknown BRCA status benefited from the combination of olaparib plus abiraterone. However, on April 28, 2023, the ODAC voted 11 to 1 (with 1 abstention) to limit use of the combination to men whose tumors tested positive for a BRCA mutation, which represents only about 10% of patients with mCRPC. This was based on post-hoc subgroup analyses that created the perception of a less favorable benefit/risk in the BRCA wildtype or unknown patients.
On April 28, 2023, the ODAC voted 11 to 1 (with 1 abstention) to limit use of the combination to men whose tumors tested positive for a BRCA mutation
Dr. Chana Weinstock articulated the FDA’s position on this issue at the April 28 ODAC meeting. She said that the Agency discourages using subgroup analysis to try to argue for efficacy in a specific group, particularly in a failed trial (although PROpel was a positive study). However, she highlighted historical precedent for limiting indications based on post-hoc subgroup analysis suggesting that certain subgroups might have compromised safety or a potential overall survival detriment. Finally, she cited the FDA guidance that states that if a trial only shows benefit in a selected subgroup, the indication may be limited to a narrower population, especially if that same signal is observed in other comparable trials. (Figure 1)
Jorge Nieva, section head of solid tumors at the University of Southern California, objected to restricting the indication to only those patients with known BRCA mutations, saying “I worry that the approach used in this application can justify removing any subgroup from any application where that subgroup has an OS curve that crosses one. FDA seems to be looking at these OS curves in a vacuum and is ignoring the corroborating evidence that some non-BRCA patients could benefit significantly.”
“I worry that the approach used in this application can justify removing any subgroup from any application where that subgroup has an OS curve that crosses one.”
It is common for Sponsors to find themselves in this situation at ODAC where the data are somewhat ambiguous and the arguments/counter arguments are highly statistical in nature. This is especially true for subgroup analyses. The key communication goal when addressing an advisory committee is to make your position as easy to understand as possible by breaking down your argument into digestible bites. If your messages are too complex, statistical or philosophical, the committee may not fully appreciate your position. When this occurs at ODAC, the committee typically defers to the FDA’s position.
Aaron Csicseri, PharmD
Senior Scientific Director
Dr. Csicseri joined the ProEd team in November 2017 as a scientific director, responsible for scientific leadership, content development, strategic input, and effective moderation of team meetings. Aaron received his PharmD at the University of Buffalo, where he studied the clinical curriculum. He has 10+ years of experience as a medical director/clinical strategist in the accredited medical education field (CME), as well as in the non-accredited PromoEd sphere. Over the past 5 years, he has been guiding Sponsor teams in their preparations for FDA and EMA regulatory meetings in a wide variety of therapeutic areas. Aaron is based in Grand Island, NY, just outside of Buffalo, New York.
At the heart of the FDA’s concerns with the PI3Ki class is an apparent excess of deaths in clinical trials of PI3Kis, indicating a potential detriment to long-term survival in patients who were treated with PI3Kis. However, assessing OS in patients with indolent blood cancers, who have very long lifespans, can be exceedingly difficult—and the results are not always interpretable.
In this second installment of this blog series, we will focus on the OS data for PI3Kis presented by the FDA, the complexities of evaluating OS in blood cancers with long lifespans, and the broader regulatory implications of an increased focus on OS as a safety endpoint in patients with indolent diseases.
Regulatory pressure to withdraw PI3Kis continues…
The Class 1 phosphatidylinositol-3-kinases (PI3Ks) are master regulators of cell proliferation, growth, survival, motility, and immune cell function that are highly dysregulated in blood cancers. Four PI3Kis (idelalisib, duvelisib, copanlisib, and umbralisib) have been approved by the FDA for B-cell malignancies such as chronic lymphocytic leukemia (CLL), and indolent non-Hodgkin’s lymphomas (iNHL), including marginal zone lymphoma (MZL) or follicular lymphoma (FL). Please see Part 1 for additional background on PI3Kis.
As of writing this post, all accelerated approvals for PI3Kis in iNHL have been withdrawn, mostly due to a failure to complete a confirmatory trial. These withdrawals also likely reflect the ODAC decision in favor of a requirement for randomized data to support future approvals for PI3Kis (rather than single-arm trial data, which are often used to support accelerated approvals in oncology).
Currently, there are 2 PI3Kis (idelalisib and duvelisib) with full approval in later-line treatment of CLL that remain on the market.
Since the April ODAC, however, the FDA has increased pressure on sponsors to withdraw even fully approved indications for PI3Kis on the basis of a potential detriment to overall survival. The most recent development came on June 30 with a warning issued by the FDA that Secura Bio’s PI3Ki, duvelisib, may increase the risk of death when used as a third-line treatment for patients with CLL. Duvelisib received full approval for CLL in 2018 on the basis of the phase 3 DUO trial. The FDA warning stated that the agency plans to hold a future public meeting to discuss the updated overall survival data of duvelisib, and whether duvelisib should continue to stay on the market – hinting that another potential ODAC meeting may be on the horizon.
Since the April ODAC, however, the FDA has increased pressure on sponsors to withdraw even fully approved indications for PI3Kis on the basis of a potential detriment to overall survival.
“The ultimate safety endpoint”: Do PI3Kis lead to a detriment in OS?
The FDA views OS as “the ultimate safety endpoint” because it encompasses deaths directly caused by adverse events related to the drug, and deaths that might be related to long-term side effects of the drug, which can be difficult to identify. Importantly, the FDA specifies that OS “does not require [the] same statistical considerations when used as a primary safety endpoint.” The FDA’s viewpoint is straightforward: The ultimate goal with any drug therapy should be to prolong life without inflicting long-term harm. From this viewpoint, OS has a dual purpose of providing information about the effectiveness of a drug, while also informing safety.
The FDA presented data at the April ODAC showing that in six phase 3 randomized trials for the approved PI3Kis, most had OS hazard ratios (HR) >1 (Figure 1) . An HR >1, based on Kaplan-Meier estimates, suggests that more patients died on the experimental arm (in this case the PI3Ki) than on the comparator arm. In 3 trials with idelalisib (the first approved PI3Ki drug), there was a much higher risk of death in the idelalisib arm, with a nearly 5× greater risk of dying in the 313-0124 trial in iNHL, and a more than 3× greater risk of dying in the 312-0123 trial in CLL.
The FDA views OS as “the ultimate safety endpoint” because it encompasses deaths directly caused by adverse events related to the drug, and deaths that might be related to long-term side effects of the drug, which can be difficult to identify.1
The OS results in trials with other PI3Kis across various patient populations and treatment regimens (DUO, CHRONOS-3, and UNITY CLL) did not show a survival imbalance as extreme as that observed in the idelalisib trials. Although most of these trials exhibited OS HRs >1, the 95% confidence interval was very wide and included 1. The CHRONOS-3 trial of copanlisib in iNHL was an exception, with an HR of 0.87 (95% CI: 0.57, 1.35) (Figure 1).
The FDA interprets these data to indicate an overall pattern of potential detriment to survival in patients treated with PI3Ki as a class, possibly linked to an increased risk of infection or immune-mediated adverse events, or unidentified long-term effects of PI3Kis that may negatively impact the efficacy of subsequent therapies.3
Imbalances in OS are not unheard of in blood cancer clinical trials…
The FDA repeatedly describes a pattern of detriment in survival with PI3Kis as “unprecedented” in oncology, but are unfavorable imbalances in OS in blood cancer trials really unprecedented? A quick review of the literature disproves this notion.
The Bruton’s tyrosine kinase inhibitor (BTKi) ibrutinib was approved for previously untreated CLL in January 2019 on the basis of the iLLUMINATE trial. At the final analysis from iLLUMINATE, there was an OS HR of 1.08 in favor of the comparator arm. Similarly, venetoclax, a B-cell lymphoma 2-targeted inhibitor (BCL2i), was approved for first-line CLL treatment in May 2019 on the basis of the CLL14 trial with an initial OS HR of 1.24 at the time of approval, although with 4 years of follow-up the OS HR improved to 0.85, albeit with results that were still not significant. Importantly, both ibrutinib and venetoclax are approved in multiple indications and are considered successful and paradigm-shifting drugs, despite never having shown a significant improvement in OS in blood cancers (nor have they definitively ruled out potential detriment).
These examples cast doubt on a narrative that PI3Kis are somehow uniquely detrimental to survival in indolent blood cancers. Rather, the problem with evaluating OS in blood cancer trials may be a reflection of the disease itself, rather than any particular class of drugs.
These examples cast doubt on a narrative that PI3Kis are somehow uniquely detrimental to survival in indolent blood cancers. Rather, the problem with evaluating OS in blood cancer trials may be a reflection of the disease itself, rather than any particular class of drugs.
The complexities of evaluating OS in blood cancers with long lifespans
There are several considerations that make OS data difficult to interpret in blood cancers with very long life expectancies. Most importantly, because there are few deaths, it is not feasible to appropriately power these trials to assess OS with statistical confidence. As a result, OS analyses tend to have wide confidence intervals that are highly unstable over time. For these reasons, OS is not used as a primary endpoint in indolent blood cancers. Instead, studies are typically powered for the primary endpoint of progression free-survival (PFS), which provides more clearly interpretable results. OS is often included as a secondary or exploratory endpoint, but is not usually considered conclusive.
The interpretation of OS is further challenged by trial design realities in indolent diseases; for example, patients may go on to receive several different lines of therapy before their death and trials often include a crossover arm (wherein patients on the control arm may choose to receive the experimental treatment after disease progression). Consequently, analyses of OS may be confounded and rendered uninterpretable. Finally, identifying the cause of death is fraught with difficulty when a large number of deaths may occur years after patients have discontinued study drug.
It is also worth mentioning that COVID-19 deaths present a unique challenge to interpreting OS in indolent blood cancers (and, to some extent, all clinical trials affected by the pandemic) not only because of the timing of COVID-19 infection waves and deaths, but also because of shifts in standard of care that can no longer be controlled for by randomization. Given these circumstances, it is possible, or even likely, that a pattern of OS detriment seen with the newer PI3Kis in recent clinical trials is a chance finding, rather than a real reflection of risk.
How should sponsors use OS to support benefit-risk assessments?
If it is not possible to adequately power a trial for OS or to interpret the results, how exactly should an “assessment” of OS be performed to support benefit-risk evaluation of PI3Kis? When asked this question directly by a panelist, the FDA fell back on its stance that the burden is on sponsors to prove that PI3Kis are not producing a detriment to survival – a difficult situation for drug developers. However, the FDA provided few clues as to what it is looking for (see slide 75 of the FDA presentation).
Sponsors developing new PI3Kis should be prepared to incorporate detailed analyses of the causes of death on each arm into study protocols, with the goal of providing assurance that the experimental arm is not causing excess deaths related to immune-mediated toxicity or an increased risk of infection due to PI3K inhibition. This type of analysis would require a much more in-depth investigation than is typically done, given the problems discussed above with tracking deaths that occur long after a patient has left the study. Sponsors should also attempt to design future studies so that the OS data are better able to withstand an increased level of scrutiny. For example, trial designs that include an option to cross over to the experimental arm after progression may not be advisable.
Ultimately, even following the FDA’s suggested changes to clinical trial design, there is no guarantee that OS can be evaluated with sufficient statistical clarity to definitively show that PI3Kis are not causing a detriment to survival.
“Going forward and looking more widely at the CLL therapeutic landscape, safety must never be compromised. But drug innovation should continue to be encouraged. We believe the FDA can leverage the power of electronic medical records and real-world data to further assess the safety of PI3K inhibitors. Delays associated with the wait for overall survival data have already started to dampen research efforts in CLL and have slowed patient access to potentially life-saving therapies. CLL is a chronic cancer, and patients are often exposed to multiple therapies over the span of their disease. Survival data will come too slowly for many patients and will never be ‘statistically pure.’ CLL Society and many others have pushed for crossover in clinical trials to ensure equipoise, which while further confounding the survival data, saves lives. In the opinion of CLL Society, clinical trial design focused on overall survival endpoints will ultimately delay or deny the best possible care to patients with CLL.”
Coming Next: Part 3 – How should dose optimization be conducted in drug classes with known immune-mediated toxicity risks?
The FDA has raised concerns that for the approved PI3Kis, dose modifications or treatment interruptions due to AEs are too common in clinical trials, indicating that the dose may be too high. The dosing regimen for all approved PI3Ki was selected based on standard, phase 1, single-arm dose escalation studies in small cohorts of patients. However, in the FDA’s view, this standard approach may lead sponsors to default to selecting the maximum tolerated dose, rather than the optimal (likely lower) dose that might achieve similar efficacy with less toxicity.
InIn Part 3 of this series, ProEd will review the panel discussion of how best to optimize drug dosing for PI3Kis, the broader FDA regulatory initiatives (ie, “Project Optimus”) surrounding dose optimization, and how these considerations may impact clinical trial design and dose-finding studies for targeted oncology drugs that modulate the immune system.
Angela W. Corona, PhD Scientific Director, ProEd RegulatoryAngela is a Scientific Director for ProEd Regulatory. She is responsible for helping sponsors navigate complex regulatory communications such as FDA advisory committee meetings. She develops clinical and regulatory strategy along with high-quality scientific and medical content across a wide range of therapeutic and drug development areas. Angela received her PhD in Neuroscience from The Ohio State University and completed her postdoctoral training at Case Western Reserve University in Cleveland, Ohio.
Moreno C, et al. First-line treatment of chronic lymphocytic leukemia with ibrutinib plus obinutuzumab versus chlorambucil plus obinutuzumab: final analysis of the randomized, phase 3 iLLUMINATE trial. Haematologica. 2022 Jan 13 [Online ahead of print]. https://doi.org/10.3324/haematol.2021.279012.
Al-Sawaf O, et al. Minimal residual disease dynamics after venetoclax-obinutuzumab treatment: extended off-treatment follow-up from the randomized CLL14 study. J Clin Oncol. 2021;39(26):4049-4060. https://pubmed.ncbi.nlm.nih.gov/34709929/
The FDA’s Oncology Division has recently taken a hard stance on PI3K inhibitors, a novel class of drugs that inhibit various isoforms of phosphatidylinositol 3-kinase (PI3K) and are approved for treating blood cancers. Recent actions by the FDA, including withdrawing some approved indications, indicate that they will likely be seeking a higher bar for new approvals of PI3K inhibitors for certain blood cancers. So, why is this promising class of drugs suddenly in the hot seat? At the heart of the issue is an excess of deaths in clinical trials of PI3K inhibitors and the FDA’s concern about their safety profile.
The panel voted resoundingly (16 yes votes; 1 abstention) that future approvals of PI3K inhibitors should be supported by randomized data, which would allow for an assessment of overall survival (OS)—an action that makes accelerated approvals less likely and more difficult to obtain in cancers with a long life expectancy.
We’ll take you inside this evolving saga. This first installment describes the drug class and what happened at a recent ODAC meeting, with top-line takeaways for drug developers in this field. Subsequent installments will analyze the implications of the panel discussion and the broader regulatory implications of that ODAC meeting and the FDA’s recent actions.
“Not a Typical ODAC Meeting”
The FDA used the April 21 ODAC meeting to zero-in on safety concerns with the class as a whole and to reevaluate the appropriate regulatory approach for future approvals. Nicole Gormley, MD, Director the Division of Hematologic Malignancies, opened up the FDA presentation by first explaining that “this committee meeting is not a typical ODAC where we would discuss the risk-benefit profile of a specific product. But instead, we will discuss the class of PI3K inhibitors as a whole, the unique toxicities that they present, and the best development approach for future drugs in this class.”
This is actually quite unusual—typically, FDA does not ask ODAC panels to provide forward-looking advice for hypothetical future new drug approvals.
Some panelists questioned if it was appropriate to make determinations for an entire drug class on the basis of collective data rather than the individual merits of each drug. Indeed, the one panel member who abstained from voting, Anthony Sung, MD, from Duke, cited this fact as the reason for his abstaining, “I still feel uncomfortable labelling an entire class and requiring that future drugs in that class be supported by randomized data. If the phase 1 data [are] not concerning, then I don’t know if randomized studies should be needed in that case.” While Dr. Gormley assured the panel that flexibility would be applied in specific cases, saying “nobody has a ‘crystal ball’,” the nature of the ODAC and recent agency actions suggests that the FDA plans to apply this advice broadly.
I still feel uncomfortable labelling an entire class and requiring that future drugs in that class be supported by randomized data.
Approved PI3K Inhibitors: Promising efficacy, paired with lingering safety concerns
The Class 1 PI3Ks are master regulators of cell proliferation, growth, survival, motility, and immune cell function. They have long been recognized as one of the most frequently aberrantly activated pathways in cancer, making this a “holy grail” drug target. Certain PI3K isoforms are implicated more strongly in different cancers. PI3Kδ and γ, for example, are preferentially expressed in B lymphocytes and are often constitutively activated in B-cell malignancies.1 Consequently small-molecule PI3K inhibitors with varying levels of selectivity for several PI3K isoforms have been developed, mainly for B-cell malignancies like chronic lymphocytic leukemia (CLL), and indolent non-Hodgkin’s lymphomas (iNHL), including marginal zone lymphoma (MZL) or follicular lymphoma (FL).
Table 1 summarizes the FDA-approved PI3K inhibitors that were the subject of the April 21 ODAC, and the history of FDA approvals for each. As of writing this article, most accelerated approval indications in Table 1, and several new marketing applications for PI3K inhibitors in blood cancer, have been withdrawn (including applications for accelerated approval in MZL/FL for MEI Pharma’s zandelisib (ME-401) and Incyte’s parsaclisib, and most recently, TG Therapeutics’ umbralisib).
Table 1: Summary of FDA-approved PI3K inhibitors and current status of approvals and new marketing applications for additional indications.
CLL=chronic lymphocytic leukemia; FL=follicular lymphoma; HER2–, human epidermal growth factor receptor-2–negative; HR+, hormone receptor-positive; iNHL=indolent non-Hodgkin’s lymphoma; MZL=marginal zone lymphoma; PI3K=phosphatidylinositol 3-kinase; PIK3CA, phosphatidylinositol-4,5-bisphosphate 3-kinase, catalytic subunit alpha; R/R= relapsed/refractory; SLL, small lymphocytic leukemia. *Because of the FDA’s focus on PI3K inhibitors in blood cancer, the PI3Kα inhibitor, alpelisib, approved for advanced breast cancer in 2019, was not included in recent ODAC discussions and is not discussed in this post.
These approvals validated the promising clinical activity of PI3K inhibitors in blood cancers based on improvements in progression-free survival (PFS) and overall response rates (ORR); nevertheless, the class has been associated with safety issues. Because PI3K regulates the immune response, inhibition of this pathway leads to immune-related adverse events such as transaminitis, pneumonitis, diarrhea/colitis, and rash that are sometimes severe.2 These agents are also associated with an increased risk of infection. This means that adverse events and infectious complications need to be carefully managed by the treating physician.2
The core of FDA’s concern was a pattern suggesting a potential detriment in overall survival (OS) in patients treated with PI3K inhibitors. The FDA presented data at the ODAC showing that in 6, randomized, phase 3 trials for the approved PI3K inhibitors, most demonstrated an OS hazard ratio (HR) >1. A HR >1 means that there were more patients who died on the experimental PI3K inhibitor arm than on the comparator arm. The FDA hypothesized that this pattern could be caused by an increased risk of infection or immune-mediated adverse events or by unidentified long-term effects of PI3K inhibitors, which may negatively affect the efficacy of subsequent therapies.4
The core of FDA’s concern was a pattern suggesting a potential detriment in overall survival (OS) in patients treated with PI3K inhibitors.
Not all PI3K inhibitors have identical safety profiles. To minimize risks, drug companies have developed next-generation PI3K inhibitors, such as umbralisib and duvelisib, with greater selectivity for specific PI3K isoforms. These developments have led to hope in the scientific community that next-generation PI3K inhibitors are “finally coming of age.”5 However, infections and immune-mediated adverse events are still important recognized risks related to the class.
Topline Takeaway: Randomized trial data with an evaluation of OS to support new approvals of PI3K inhibitors
The primary outcome of the April 21 ODAC was the panel’s unanimous vote that randomized trials should be necessary to support future approvals of PI3K inhibitors. Sponsors should anticipate that the ODAC vote will likely translate into regulatory policy. While this does not necessarily affect the regulatory threshold for full approvals (where randomized data has always been required), this decision would make accelerated approvals based on single-arm trials less likely for PI3K inhibitors. Sponsors seeking new approvals for PI3K inhibitors should plan to base their New Drug Applications on randomized clinical trial data and should design clinical trials to include an evaluation of OS. Sponsors should be aware, however, that FDA appears to be looking for an OS hazard ratio of <1.
Sponsors seeking new approvals for PI3K inhibitors should plan to base their New Drug Applications on randomized clinical trial data and should design clinical trials to include an evaluation of OS. Sponsors should be aware, however, that FDA appears to be looking for an OS hazard ratio of <1.
Coming Next: Part 2 — How Should Overall Survival Data Be Evaluated in Blood Cancers?
The rationale for requiring randomized trials specifically for PI3Kis was to allow for an assessment of OS as a safety endpoint. However, assessing OS in patients with indolent blood cancers, who have very long lifespans, can be exceedingly difficult—and the results are not always interpretable. Many panelists questioned the feasibility of assessing OS and how this information should be used in regulatory decision making moving forward.
In Part 2 of this series, ProEd will review in detail the OS data for PI3K inhibitors presented at the April 21 ODAC and the broader regulatory implications of an increased focus on OS as a safety endpoint in patients with indolent blood cancers.
Angela W. Corona, PhD Scientific Director, ProEd Regulatory
Angela is a Scientific Director for ProEd Regulatory. She is responsible for helping sponsors navigate complex regulatory communications such as FDA advisory committee meetings. She develops clinical and regulatory strategy along with high-quality scientific and medical content across a wide range of therapeutic and drug development areas. Angela received her PhD in Neuroscience from The Ohio State University and completed her postdoctoral training at Case Western Reserve University in Cleveland, Ohio.
Phillips TJ, Michot JM, Ribrag V. Can next-generation PI3K inhibitors unlock the full potential of the class in patients with B-cell lymphoma? Clin Lymphoma Myeloma Leuk. 2021;21(1):8-20.e3.