A recent workshop summed up the state of play for circulating tumor DNA. Although evidence is building of its utility as a biomarker in many oncology settings, its prognostic value has yet to be proven, limiting any current utility as an endpoint for regulatory use for cancer therapeutics.

Oncology endpoints: The growing pains of an evolving field

• FDA approval of a drug or biologic relies on whether there is “substantial evidence” of safety and effectiveness derived from “adequate and well-controlled clinical investigations” according to the Federal Food, Drug, and Cosmetic (FD&C) Act. However, the FDA Modernization Act of 1997 gave FDA the ability to exercise its scientific judgment in deciding exactly what it means to fulfill these requirements. In general, FDA determines the “effectiveness” of a product based on its ability to demonstrate clinical benefit: to prolong life, improve function, and/or improve the way a patient feels. The assessment of this benefit typically relies upon the use of specific endpoints to measure this clinical benefit, generally within the context of a clinical trial. [  For more information, read FDA’s 2019 draft guidance on substantial evidence here.]
• Overall survival (OS) is generally considered to be the gold standard for verifying clinical benefit of cancer products, since it is precise, easy to measure, and nonbiased. OS is an outcome-related endpoint that is defined as the time from trial randomization until death from any cause in the intent-to-treat population. It is also considered the most reliable cancer endpoint, and when studies can be conducted to adequately assess survival, it is usually the FDA-preferred endpoint. The two biggest challenges for calculating overall survival include long follow-up periods and the potential for confounding variables to distort results. It was to address this first challenge of long follow-up periods that the FDA first started accepting tumor-based endpoints such as objective response rate (ORR) and progression-free survival (PFS) for anticancer products. As for the second challenge, potential confounders present at the time of treatment initiation can be controlled for using randomization, but it may be more difficult for sponsors to detect and adjust for factors that arise after the investigational product’s administration (e.g., subsequent superior/inferior supportive care, lifestyle changes, and/or use of additional therapies).
• That’s why, according to FDA’s Oncology Center of Excellence, the determination of an appropriate endpoint in oncology is based on the specific disease and is highly dependent upon numerous factors. These include effect size, effect duration, depth of response, available therapy, disease setting, location of disease, the clinical consequences of delaying or preventing disease progression or delaying administration of more toxic therapies, and the risk-benefit relationship. Also, given the aggressive and serious nature of many cancers, it should come as no surprise that many anticancer products qualify for accelerated approval. In fact, the FDA estimates that the accelerated approval program is responsible for access to life-saving anti-cancer therapies a median of 3.4 years before they would have been available otherwise.
• In contrast to traditional approval, accelerated approval products can demonstrate efficacy using outcome-related endpoints, intermediate endpoints, and surrogate endpoints. Like intermediate endpoints, surrogate endpoints measure an earlier (non-irreversible) effect of a product, but these measurements are deemed reasonably likely to predict clinical benefit, whereas intermediate endpoints for traditional product approval must be known predictors of clinical benefit. While the use of non-outcome-related endpoints helps get these potentially lifesaving products to patients faster, sponsors are still expected to conduct confirmatory studies to verify the product’s clinical benefit following its approval.
• In oncology,  the most commonly used surrogate endpoint in support of accelerated approval products is objective response rate (ORR). ORR is a tumor-based surrogate endpoint that measures the ratio of patients who either see a reduction in tumor size or whose tumor is undetectable, compared with all patients in a given study arm. In addition to its use in accelerated approvals, the ORR endpoint has also been used as a surrogate endpoint for traditional approval, the FDA’s 2018 guidance on oncology endpoints notes. In certain cases, ORR may measure direct clinical benefit depending on “the specific disease, context of use, magnitude of the effect, the number of CRs, the durability of response, the disease setting, the location of the tumors, available therapy, and the risk-benefit relationship,” writes the FDA in the guidance document.
• Progression-free survival (PFS) is another common tumor-based surrogate. PFS is calculated as the amount of time from trial randomization to objective tumor progression or death, whichever occurs first. To evaluate tumor progression, clinicians commonly use the response evaluation criteria in solid tumors (RECIST) guideline to determine a patient’s level of response to treatment based on the percentage increase or decrease in lesion size. Like ORR, FDA considers PFS to be an appropriate surrogate endpoint for accelerated approval, traditional approval, and for measuring direct clinical benefits in certain circumstances. But unlike ORR, PFS is considered by FDA to be a more appropriate measure of stable disease and is thought to be correlated to overall survival. One important note regarding PFS is that it only captures the initial disease-related event following treatment initiation; it does not measure the time between the beginning of tumor progression and death.
• However, the validity of these popular surrogate endpoints has been called into question in recent years. Just three months ago, OCE leadership published a paper based on the analysis of data from approved anticancer products that have measured Objective Response Rate (ORR), Progression-Free Survival (PFS), and Overall Survival (OS). The major finding: Some products approved using tumor-based endpoints (ORR or PFS) did not show improvements in overall survival, and some products that did improve overall survival did not demonstrate improvements in PFS and/or ORR. What’s more, some products that used PFS or ORR for approval have demonstrated a potential detriment to overall survival. The authors proposed several potential reasons for divergences in tumor-based endpoints and overall survival including magnitude of effect, trial design,  inadequate attention to dose optimization, and differential effects on various subpopulations within the overall intent-to-treat population. [ Read AgencyIQ’s analysis of that paper, and the larger issues, here.]
• But if not ORR or PFS, then what? The article noted that unlike conventional cytotoxic drugs where the relationship between early endpoints and OS has been more consistently observed, the unique mechanisms of action for oncology products, like immune checkpoint inhibitors (which may alter tumor growth kinetics rather than solely acting via direct cytotoxicity), could result in improvements in overall survival where PFS or ORR are not necessarily observed. But until there are intermediate endpoints for these products that do strongly correlate with overall survival, ORR and PFS will likely continue to be used, if only because there do not seem to be readily available surrogate endpoints that are superior and able to be used in their stead.

Could ctDNA be the solution?

• In 2019, the advocacy organization, Friends of Cancer Research (FOCR) organized a multi-stakeholder initiative comprised of industry, government, and academia to find out. The FOCR project ctDNA for Monitoring Treatment Response (the ctMoniTR Project) aims to “establish an aligned strategy for developing the necessary data to support the use of circulating tumor deoxyribonucleic acid (ctDNA) as an early endpoint for treatment response for regulatory decision-making and leading a multi-stakeholder group to generate this data.” Prior to this effort, the lack of harmonized methods used for ctDNA collection and analysis made it challenging to draw firm conclusions.
• For context, circulating tumor deoxyribonucleic acid (ctDNA) is fragmented genetic material  of cancer cells that are found in the bloodstream. As dying cancer cells degrade, their cell contents are released  into the patient’s circulation, affording the possibility of detecting fragments of DNA that are unique to the cancer. The technology, less invasive than conventional biopsies, can be  used as a biomarker to detect and predict the presence and stage of some cancers. ctDNA can also be used to predict treatment effectiveness, since it indicates the amount of tumor DNA found in the blood. Finally, since the technique can provide information about tumor genetics and mutations, ctDNA can also be used to help selected directed therapies.
• In the last few years, the group has evaluated the use of ctDNA to track long-term outcomes. The project was rolled out in a two-phase design, with the first step focused on using previously collected data on patients with lung cancer from clinical trial and observational cohort studies. This first step was intentionally narrow in focus so the group could assess the feasibility of harmonizing ctDNA, align on a methodology to combine clinical data from multiple sources, and take a first pass at characterizing the associations between ctDNA values and tumor response.
• At a  virtual meeting held by FOCR in 2020, ctMoniTR team members presented and discussed initial findings from Step 1 of the project. Notably, the results demonstrated that a reduction in ctDNA was associated with improved clinical outcomes for 200 patients with advanced NSCLC (aNSCLC) receiving immunotherapy. At the meeting, FDA’s Chief of Medical Oncology, Julia Beaver, expressed her belief that ctDNA had potential to be used as a novel endpoint – provided it is backed by randomized clinical trial data. “It’s also important in the metastatic setting to differentiate ctDNA from objective response (OR) rate or show how it could be complementary to [it]. Although response rate is used to support drug approval and it’s relevant for clinical decision making, it’s not always been shown to track specifically well with long-term outcomes. So, if ctDNA could improve upon this or add to OR response rate in some way, it could prove very useful in regulatory decision making,” she said. Although the workgroup made note of multiple challenges, such as inconsistent timing of ctDNA sampling and the different amount of clinical patient details provided across various studies, the association between ctDNA and treatment response was still observed after accounting for clinical covariates.
• Next, the group embarked on Step 2 of the project to “determine the extent to which changes in circulating tumor DNA (ctDNA) are an early indicator of response to therapy on a larger scale.” This second phase was subdivided into three stages, all of which were anticipated to be complete by mid-2023. As a whole, Step 2 seeks to more rigorously evaluate the potential application of ctDNA as an intermediate endpoint and to determine whether it is a reliable prognostic indicator. In other words, researchers are finding out during Step 2 whether ctDNA can be used to determine patient response to a particular treatment at an earlier timepoint than currently available methods. In total, the data from the second phase is expected to be sourced from 22 clinical trials with over 3,000 trial participants diagnosed with various types of cancer, spanning 16 types of treatments.
• In parallel, FOCR developed a white paper that provided an overview of the potential regulatory applications for ctDNA in oncology. The paper, published in November 2021, primarily focused on the potential for ctDNA to serve as an early endpoint. Not only did FOCR hypothesize that ctDNA could detect and measure changes earlier than other, commonly used endpoints in oncology; the paper also stated that ctDNA would entail more frequent and less burdensome assessments, as compared with direct sampling of tumor tissue for many cancers and could be used in conjunction with imaging to monitor early response to therapy. Another potential application described in the paper was the use of ctDNA for patient selection by detecting targetable molecular alterations in cancer DNA, “allowing for the development and evaluation of targeted therapeutic approaches that bestow the most benefit to the patient.” Lastly, it described how ctDNA could be used to detect molecular residual disease (MRD), which could then serve as a proxy to determine which patients are at the highest risk of cancer recurrence. [ Read AgencyIQ’s analysis of the white paper and the meeting at which it was discussed here.]
• In May 2022, FDA responded to industry interest with a  draft guidance on the use of ctDNA as a biomarker in investigational new drug (IND) and marketing applications for products intended to treat solid tumor malignancies in the early-stage setting. In the guidance, FDA described four scenarios ctDNA could be used for: selection of the patient population, patient enrichment using molecular residual disease (MRD), as a measure of response in early trials, or to support pathologic complete response information following neoadjuvant therapy and as an early endpoint for clinical trials. However, AgencyIQ notes that the guidance made it clear that FDA was not ready to recognize ctDNA as a validated surrogate endpoint. It also cited many of the challenges previously mentioned in the FOCR publications, which the organizations involved in the ctMoniTR project have been working to address. Two months later, FOCR hosted a meeting to report on the progress of the ctMoniTR project. While FDA speakers at the event seemed open to the possibility of ctDNA’s eventual use, it was clear that in the eyes of regulators, there was still a long way to go before ctDNA would be recognized as a validated early endpoint.
• Two years after the initial findings of the Step 1 research were unveiled, the results were officially published in JCO Precision Oncology in August 2022.

Have we learned anything new about ctDNA?

• Last week, FOCR hosted a meeting to provide stakeholders with the latest updates on the ctMoniTR project. Over the course of the afternoon, Baseline ctDNA Working Group members and top-ranking officials including Patrizia Cavazzoni (CDER), Richard Pazdur (OCE), and Lee Fleisher (CMS) discussed their thoughts on where ctDNA is now and where it’s headed next. Attendees heard that the Step 2 study is being conducted in three phases (or modules as FOCR calls them). The first module focuses on patients with non-small cell lung cancer (NSCLC) undergoing several types of TKI treatments. The second module will focus on patients diagnosed with advanced non-small cell lung cancer who are being treated with anti-PD1 or chemotherapy products. Finally, according to FOCR researcher and presenter Hillary Stires, the third module is a sort of “catch-all” category, which will evaluate the use of ctDNA to measure treatment response in patients with advanced solid tumors undergoing treatment with anti-PD(L)1 or TKI products.
• Just as before, FOCR used the meeting to offer a first glimpse into the findings from the studies underway in Step 2 of the ctMoniTR project. As Stires explained, Step 1 was a promising start to the project but had a relatively small (200 participants) and homogenous patient population: all patients included were diagnosed with advanced NSCLC and were treated with anti-PD(L)1 treatments. In contrast, while only the data from Step 2: Module 1 appears to have been collected and analyzed so far, already the study includes five times more patients than were included in the first study. In total, the Step 2: Module 1 dataset includes more than 1,000 patients enrolled in clinical trials for eight different TKI therapeutics (e.g., anti-EGFR, ALK, RET). While only patients with NSCLC were included in the Module 1 study (like the Step 1 study), Stires pointed out that this new study does not validate the previous one. Instead, this study includes patients with a different treatment modality and uses a different method for defining ctDNA change. Rather than detection (as was used in the Step 1 feasibility study), the Step 2: Module 1 study uses a more precise estimand: percentage change.
• A preview of three high-level analyses from the Step 2: Module 1 study were presented. The first analysis considered whether the magnitude of ctDNA change had any relationship to OS. Similar to the high-level results from Step 1, the data showed that a higher percentage of patients with persistently nondetectable levels of ctDNA over the four years of follow-up after treatment were still living. The researchers then grouped the patients who did eventually display detectable levels of ctDNA within four years following treatment into categories based on how large or small the change in ctDNA levels were from baseline. Here, ctDNA levels did not convey any sort of proportionate response to overall survival and the relationship was not statistically significant. Hence, researchers concluded that the magnitude of change in ctDNA levels in the years following treatment did not provide any predictive insight.
• Next, the researchers grouped the patients more simply into two categories: patients who had a return in detectable levels of ctDNA in the four years after treatment, and patients who did not display detectable levels of ctDNA during the four years of follow-up after treatment. This analysis did result in a statistically significant finding that patients who responded to treatment and did not display detectable levels of ctDNA immediately following treatment and who remained at undetectable levels in the subsequent four years had improved overall survival compared to patients who still displayed detectable levels of ctDNA even immediately after receiving treatment and in the years that followed (regardless of whether ctDNA levels increased or decreased over time).
• Finally, the researchers examined whether detectable levels of ctDNA, when used in combination with radiographic response data, increased their ability to accurately predict overall survival benefit. Here, radiographic response data categorizes imaging following treatment as showing a complete response, partial response, stable disease, or progressive disease. Given the small sample size for the complete response category, the researchers grouped complete and partial response groups together and labeled the category “responders.” All other categories were grouped together as non-responders. While the distribution of patients divided by ctDNA levels was similar to the distribution of patients using the RECIST responder vs. nonresponder categorizations, combining these data together did not improve the ability of the model to make accurate inferences about a patient’s overall survival.

FOCR’s latest initiative: Introducing Project Baseline

• During the FOCR meeting, we also heard about a new initiative led by FOCR to address one of the gap’s discussed in its roadmap: lack of standardization when it comes to assays used to measure ctDNA levels. The new Baseline ctDNA Project will seek to address the current variation seen in ctDNA levels when measured using different assays. Per Brittany Avin McKelvey, FOCR’s Director of Regulatory Affairs, the lack of standardization in assays to detect ctDNA levels has made it challenging to pool data from different studies. With this in mind, FOCR will work over the next few years to “establish evidence of baseline ctDNA levels by cancer type and stage across assays through a collaborative effort with multiple assay developers.”
• It’s not necessarily just the assays themselves that need to be standardized, but when and how they are being used by clinicians. For example, as Avin McKelvey explained, many of the samples in FOCR’s dataset do not have standardized elements explaining at which stage of the trial the assay was used to measure a patient’s ctDNA levels, how much time had passed between the patient’s diagnosis and the ctDNA test, or why the clinician chose to use the test. While in some diseases, like NSCLC, the samples tended to be collected fairly frequently even in early-stage patients, for other types of cancers like bladder or prostate, the tests were almost exclusively conducted in patients with stage IV diagnoses. These facts make it challenging to determine what baseline ctDNA levels look like just after a cancer diagnosis and prior to any treatment.

Has the tide changed for ctDNA?

• Probably not yet. The initial analyses from Step 2 of the CTMoniTR Project seem to prompt more questions than answers. It appears that ctDNA levels offer valuable information regarding patient response to treatment, there’s no clear signal of predictive value in terms of other key endpoints, and ctDNA doesn’t yet have a clear place in the clinical toolkit to assess treatment response and patient prognosis.
• A statistical challenge for FOCR’s dataset for Module 1 of Step 2 is that just four patients saw CR to treatment. This forced researchers to dichotomize the data, grouping CR and partial responses into one category. Additional data is critical before any true determinations can be made regarding the role ctDNA levels might play in augmenting imaging for endpoints that rely on radiographic data. It’s possible that future application of artificial intelligence and machine learning could help in integrating ctDNA data with imaging data, and in uncovering previously undetected associations between ctDNA and prognostic tumor characteristics beyond size.
• FDA and FOCR want you (and by you, we mean your data on ctDNA). While ctDNA may not yet be ready for prime time, that doesn’t mean it never will be. The previous Step 1 study was a small feasibility exercise which yielded positive results, while the Step 2 study has furthered the knowledge base for ctDNA. FOCR researchers, who seek more data, have enabled a secure and anonymous submission system to help sponsors feel comfortable submitting their data. The veil of anonymity can only be pierced by the FDA, which can access data for verification that their records match the data submitted.
• Another frequent topic of discussion was the utility of ctDNA as a potential surrogate endpoint for products seeking accelerated approval. While Cavazzoni expressed her excitement about how far ctDNA has come to date, she also restated many of the same messages we’ve heard from both CBER and CDER leadership before: early engagement between sponsors and FDA is key, confirmatory trials are a must, and FDA isn’t afraid to use its Congressionally-mandated authority to pull products from the market when sponsors don’t hold up their end of the bargain (i.e., completing confirmatory trials).
• An interesting point Cavazzoni repeated several times was that there is a difference in the amount of evidence needed to establish a validated endpoint intended to replace a clinical endpoint versus that needed to support a surrogate endpoint for gaining accelerated approval. She didn’t elaborate on the extent of evidence required for a new surrogate endpoint but did state that while the Agency remains open to the use of new accelerated approval endpoints, they do expect to see the data to support their use. Her advice on ctDNA? “We recommend that trials collect appropriate data on ctDNA before and after treatment as well as a long-term outcome so that we can include these trials into our growing body of evidence about this promising endpoint.”

https://fda.agencyiq.com/article/00000189-6c58-d9aa-a18b-6d7dbb290000