I've entitled this slide The Horse Is Out of the Barn, and it should be evident to this audience why that is the case. Even though most drug approvals so far in prostate cancer have been based on overall survival, or CI-based endpoints, PSMA PET is increasingly used in clinical practice for various reasons. The availability of PSMA PET presents issues with designing a pivotal trial with only CI in 2026. Here are a few potential ones. First, baseline PSMA PET results may inform patient selection and willingness to enter trials. Second, the use of off-study PSMA PET during a trial may lead to treatment changes, or study withdrawal, which may lead to censoring, and difficulty interpreting a CI-based primary endpoint. Finally, non-uniform use of off-study PSMA PET may further complicate the study interpretation by introducing heterogeneity. For all of these reasons, it seems difficult or impossible for prostate cancer trials to completely ignore PSMA PET in 2026.
On the other hand, there are also issues with designing a trial on prostate cancer with only PSMA PET imaging and no conventional imaging. Again, here are a few potential issues. First, there are differing potential definitions of PSMA PET-based endpoints, all of which are relatively less validated versus CI-based endpoints. Second, the currently proposed PSMA PET endpoints have uncertain clinical meaningfulness. And third, there's potential for modulation of PSMA expression with certain therapy classes. So therefore, it seems difficult or impossible to completely exclude CI-based endpoints either.
A related question you might be asking is, what should the primary endpoint be? And I'm not going to answer that in this talk, but suffice to say that both CI-based and PSMA PET-based endpoints have challenges currently that should be considered during clinical trial development. There are some considerations that apply to any radiographic time-to-event endpoint. We frequently give some variation of the comment shown here to sponsors. Use of whatever the time-to-event radiographic endpoint is as a primary endpoint in a marketing application will depend on the totality of evidence of a favorable benefit-risk assessment, including, but not limited to, the magnitude of the observed effect size, consistent study results in subgroup analysis, sensitivity analyses, and clinically meaningful secondary endpoints, no apparent OS detriment, and the safety and tolerability profile. Ultimately, whether any radiographic primary endpoint can support an approval is a review issue at the time an application is reviewed. But still, careful planning and discussion with regulatory agencies at the design stage can increase the chances of success.
Next, I'm going to go briefly through a history of regulatory approvals that were supported by CI-based endpoints. So on this slide, I'm showing selected approvals based on radiographic progression-free survival, which was introduced by PCWG2 in 2008, as shown below the timeline. Approvals based on this endpoint have occurred in both metastatic castration-resistant, and metastatic castration-sensitive disease, and became more frequent after published data became available, showing an association between rPFS and OS. I included a subset of the numerous relevant publications below the timeline. While the initial approvals were strongly supported by other endpoints, such as OS, some more recent approvals have been based primarily on rPFS, with non-significant OS endpoints that nevertheless still suggested no apparent OS detriment.
Next, I've listed approvals based on metastasis-free survival, which occurred in earlier disease settings of non-metastatic, castration-resistant prostate cancer, and high-risk biochemical recurrence. The use of MFS for approvals was supported by several factors. First, there were two oncologic drug advisory committee meetings, or ODACs. One in 2011, and one in 2012. Second, as the FDA discussed in a 2018 publication, and a 2021 guidance, the initial development of metastases was considered by multiple stakeholders to be a clinically meaningful event. Third, published data have shown an association between MFS and OS, with a few examples included below the timeline. Finally, for completeness, I've listed the single accelerated approval in prostate cancer, which was rucaparib, and was based on objective response rate, and duration of response by conventional imaging.
For the last and longest portion of the talk, I'm going to consider the impact of PSMA PET on a hypothetical clinical trial using something called the estimands framework to organize this portion. And for those of you who aren't familiar, the estimands framework is described in the ICH E9(R1) guidance, and is very useful to facilitate multidisciplinary and multi-stakeholder discussions about the design and interpretation of clinical trials. I'm going to go through some ways that PSMA PET might impact each of the key elements of the estimands framework, which are depicted as pillars in the figure here. So it's treatments, population, endpoint, intercurrent events, and population level summary.
The first element is the treatment of interest. As this audience well knows, PSMA PET images physiologic PSMA expression rather than strictly anatomy, and an important question is whether PSMA expression can be altered by certain systemic therapies. If so, could this complicate determination of response or progression? Certainly a well-recognized example is treatment-induced flare phenomenon early in treatment, but we should also consider the potential for development of PSMA-negative disease, particularly after treatment with PSMA-targeted drugs. And by contrast, conventional cross-sectional imaging reflects anatomy more so, and may be relatively unaffected by a treatment type.
The second element is the population. The clinical context of a patient population factors into the eventual benefit-risk evaluation in our review of a marketing application. New patient populations have been defined by PSMA PET, and a better understanding of the natural history of disease in these populations will be very helpful to understand the benefit-risk considerations. For example, we could ask whether single PSMA PET-positive bone metastases or lymph nodes might have a more indolent biology in CI-positive lesions, as well as whether they need to be treated in the same way.
The third element is the endpoint. Multiple potential definitions of PSMA PET-based endpoints exist, none of which is fully validated. This talk is not going to endorse any particular endpoint definition, but I've listed a couple of relevant discussion points here. So first, sponsors may consider including progression by CI as an event, even in a PSMA PET-based endpoint, to capture potential PSMA-negative disease progression that would still be clinically significant. Additionally, any proposals for use of specific quantitative metrics as opposed to qualitative visual assessment should be justified.
The next point has to do with timing of assessments. If PSMA PET imaging is incorporated into the primary endpoint, it should be obtained at baseline, and at set intervals, and not initiated solely based on PSA rise. This is for at least a couple reasons. One is that asymmetric timing of imaging between arms will complicate the statistical interpretation. And furthermore, radiographic progression without PSA rise is not uncommon, and has certainly been documented. The third point about PSMA PET-based endpoints is that they're not well-suited to support accelerated approval at this time, both because the criteria for objective response remain relatively less validated than CI-based objective response, and also because benchmark data are limited to serve as a sort of reference point for an accelerated approval. Finally, on this slide, most of the bullet points focused on PSMA PET-based endpoints, but this last bullet applies to CI-based endpoints, regardless of whether they're primary or secondary. And namely, sponsors should consider blinding the readers of CI scans to PSMA PET-based results to help preserve the independent value of the CI-based endpoint.
Let's see. The fourth element of the estimands framework is intercurrent events, which, again, for those of you who are not familiar, are defined as events that occur after the treatment initiation, that either preclude the measurement of the primary outcome measure, or affect its measurement or interpretation in some way.
I've listed here a few examples of potentially important intercurrent events related to the PSMA PET. First, when does worsening of a PSMA PET image necessitate a change in systemic treatment? Second, how to handle receipt of metastasis directed therapy on study. For example, in the case of oligoprogression. And I'll elaborate a little bit more on this one. What should the specific criteria be for allowing MDT? Should they be the same or different from calling progression from systemic therapy? And should there be a limit to how much MDT is allowed on a trial? Third point, how to handle PSA anxiety that could impact the imaging schedule. I think one could suggest that to preserve the statistical interpretability of the primary endpoint, it may be helpful to try to limit moving up the scheduled scans purely due to PSA rise, without any clinical correlate.
For intercurrent events that are expected to be important to the interpretation of the primary endpoint, sponsors should pre-specify objective criteria for how to define these events, and deal with them. Should also try to limit asymmetry between the arms due to either the events, or resulting missing data, and they should pre-specify supplementary analysis to explore the robustness of the primary endpoint, and the impact of these anticipated intercurrent events.
The fifth and final element of the estimands framework is the summary measure, which, for time-to-event endpoints is frequently a hazard ratio comparing two or more treatments. But really key questions related to this talk are, first, what is the meaningfulness of a PSMA PET-based summary measure? Second, what magnitude of improvement is sufficient to conclude benefit? And third is the PSMA PET-based endpoint associated with established endpoints, such as CI-based endpoints and OS. Going back to the beginning part of the talk, I want to remind everybody that the interpretation of a PSMA PET-based summary measure may be influenced by the clinical context, and the totality of the data, including CI-based endpoints.
One of my last slides is meant to emphasize the importance of multi-stakeholder engagement to help address the key questions I mentioned on the previous slide. I acknowledge that many people in this room contributed to the development of CI-based endpoints, and their acceptability for use in regulatory approvals, and it could be that similar processes may be helpful for PSMA PET-based endpoints. For the first two questions I proposed on the previous slide, these might be addressed by ongoing multi-stakeholder discussions, or future discussions, including, but not limited to public workshops, and potentially one or more advisory committee meetings, like... Happen with MFS. And then the third question may be addressed by emerging data and analysis of patient level and trial level association of PSMA PET-based endpoints with established endpoints such as CI-based endpoints or OS.
So in summary, regardless of the choice of primary endpoint, sponsors should carefully consider how the availability of PSMA PET might affect all aspects of trial design and interpretation. Sponsors should meet with FDA and other regulatory agencies to discuss trial designs intended to support future marketing applications. The inclusion of CI-based endpoints remains important in pivotal trials in prostate cancer, and ideally, emerging data and ongoing multi-stakeholder discussions will clarify the appropriate definitions and clinical meaningfulness of PSMA PET-based endpoints. Finally, I want to acknowledge the core group of people from FDA who helped me prepare this talk, the organizers for inviting me to speak, and the audience for your attention. Thank you.