Skip to content

The Cancer Letter — Using ctDNA as an endpoint in your cancer clinical trial? FDA wants to hear from you

The Cancer Letter — Using ctDNA as an endpoint in your cancer clinical trial? FDA wants to hear from you

Circulating tumor DNA has the potential to not only change the way medical oncologists assess and treat cancer patients, but also how cancer drugs are reviewed, oncology experts at FDA and pharmaceutical companies said in response to research findings published July 11 by a collaboration led by Friends of Cancer Research.

To date, the Friends ctMoniTR initiative has provided robust evidence, based on retrospective studies of aggregated data from clinical trials, that patients with advanced non-small cell lung cancer who have reduced ctDNA levels on treatment with PD-1/L1 drugs or tyrosine kinase inhibitors show improved overall survival and progression-free survival (The Cancer LetterJuly 7, 2023).

Moving forward, data to support the use of ctDNA need to be collected in prospective randomized controlled clinical trials to formally evaluate its utility as a regulatory endpoint, said Patrizia Cavazzoni, director of the FDA Center for Drug Evaluation and Research.

“It’s very important that the developers come and talk to [FDA Oncology Center of Excellence Director] Rick Pazdur and his team, because it’s very important to have the right guideposts when these endpoints are included in clinical trials,” Cavazzoni said at the July 11 Friends meeting in Washington, DC. “For instance, we would want them to be included in clinical trials and then reflect or are consistent with the clinical trials, in which the novel endpoint could be used potentially in the future for regulatory decision-making.”

In addition to evaluating long-term survival outcomes, the Friends collaboration is focused on two fundamental factors that determine whether ctDNA can be used  as an early endpoint in clinical trials:

  • Understanding tumor biology, e.g., heterogeneity in the shedding of ctDNA by different tumor types, and
  • Assessing assay technology, i.e., determining whether the different commercially available assays can reliably detect ctDNA biomarkers across cancer types.

“There’s work to be done, but it’s very exciting to see that once again, the cancer community is at the leading edge and is really charting the way for other areas,” Cavazzoni said. “Let’s face it, what we’re seeing here, for instance, in more utilization of accelerated approval in areas that had been dormant for decades where there was no hope and no treatments for decades, such as neurodegenerative diseases, we’re beginning to see those therapeutic areas beginning the journey that the cancer community began well over 20 years ago.”

Following are two key takeaways from the collaboration’s ctDNA Baseline Project and the Step 2 Module 1 phase of the ctDNA for Monitoring Treatment Response Project:

  1. Baseline ctDNA was similarly detected across most late-stage cancer samples—NSCLC, breast, bladder, prostate, as well as head and neck squamous cell carcinoma—and across assays, and
  2. In aNSCLC patients, ctDNA molecular response to targeted therapy was associated with OS and PFS, while radiographic response at first RECIST was not.

A webcast of the meeting and slide decks are available on the Friends website. For a more detailed accounting of the research phases of the ctMoniTR project and its initial conclusions, see The Cancer Letter’s July 7 article.

These findings, if validated in clinical trials across multiple cancer types and drug classes, could mean that ctDNA has a role to play in predicting survival outcomes, becoming a novel endpoint that could be used in FDA’s accelerated approval framework, experts say.

“The fact that first RECIST alone did not associate with long-term outcomes, but the three ctDNA categories that we looked at did associate with improved outcomes, was very surprising—an interesting takeaway,” said Minakshi Guha, LBx strategy lead and associate director of precision medicine and companion diagnostics at Takeda Pharmaceuticals. “We might start considering looking at ctDNA as a complementary biomarker to predict long-term responses.

“We don’t need to wait years for OS and PFS readouts to determine whether our drug is working or not. We can look much more earlier on and in scenarios of accelerated drug improvement, it perhaps makes sense to look at ctDNA as a predictor of response.”

Even among patients who are categorized as having stable disease at first RECIST, the data show that molecular response identified patients with improved associations with OS and PFS.

“The thing we’ve known for so many years is that the dreaded stable disease doesn’t tell the whole story. It comes to life here; right? Stable disease adds nothing. And in fact, ctDNA can clarify a stable disease,” said Geoff Oxnard, a thoracic oncologist at Boston Medical Center and senior vice president and head of clinical development at Foundation Medicine. “So, I think this model system allows us to now apply to other more mysterious therapies to really help amplify our drug development and figure out where are these drugs working and not working.

“In this model system, we’ve learned that ctDNA shows dramatic clearance. If you want to develop a drug that’s like one of these drugs, you better be seeing ctDNA responses; right?” Oxnard said.

“But it’s also not a setting of unpredictability. Like we know how it’s going to play out for these patients; right?

“This is very different than the first experience on immunotherapy where the therapy is unpredictable. In this setting, where we know what the outcome should look like—ctDNA trumps response.”

Baseline ctDNA Project

To evaluate change in ctDNA levels in response to treatment and its relationship with survival outcomes, it would first be necessary to establish the baseline ctDNA level and demonstrate that it is feasible to reliably detect ctDNA across cancer types.

In the metastatic setting, current baseline ctDNA data are variable across different cancer types, and there is a dearth of data on baseline ctDNA shed rates in early-stage cancers. Also, the data are disparate across ctDNA technologies and assays, making it difficult to conduct pooled analyses across studies.

A slide showing ctDNA assay characteristics and assay types used in the Friends ctDNA Baseline Project.

In response to these challenges, the Friends collaboration conducted an analysis of baseline ctDNA levels using assays from eight ctDNA test developers. The results, which were based on 2,327 early-stage NSCLC cancer samples and 87,209 late-stage samples across five different cancer types, are encouraging.

“Looking at the proportion of samples with detected ctDNA, we do see that within each assay, the baseline ctDNA is similarly detected across these late-stage cancer types for the majority of samples,” said Brittany Avin McKelvey, director of regulatory affairs at Friends. “And then, when we look at the median VAF [Variant Allele Frequency], we see again that they’re relatively similar as well across these five cancer types, which really truly highlights the potential opportunity to use ctDNA and all of these cancer types.”

A slide showing late-stage median VAF by assay in the Friends ctDNA Baseline Project.

Readouts from the project data show that in late-stage non-small cell lung cancer, there is a higher proportion of detected ctDNA as well as ctDNA levels compared to early stage samples.

Similar detection of ctDNA was also observed across all five late-stage cancer types.

“There was heterogeneity looking at assay characteristics, which includes the biomarker features that were assessed, the intended use of the assays as well as analytical performance, as well as the heterogeneity and the available clinical pathological data, which could lead to challenges interpreting aggregated data,” McKelvey said. “Therefore, additional data as well as common data standards are needed to help make more robust comparisons and support future data harmonization efforts.”

A slide showing frequency of ctDNA detection in late-stage cancers in the Friends ctDNA Baseline Project.

Assessing baseline ctDNA levels is important, because emerging evidence is elucidating its role as a prognostic factor. Also, the trackability of a patient may depend on whether there’s a signal present at baseline, said Anand Pathak, a medical officer with FDA’s Division of Molecular Genetics and Pathology.

“In terms of the technical and biological aspects of ctDNA and assays, there are many technical issues,” Pathak said. “There could be tumor-informed assays, tumor-naive assays. There could be ddPCR assays that only target a few genes. And then there’s the issue of how do you subtract the germline; right? Like, is the germline being addressed appropriately?

“Besides the tumor stage, I think timing of the collection of ctDNA, with respect to the intervention or whatever is going on before baseline sampling, is really essential to know,” Pathak said. “ctDNA holds a lot of promise in advancing patient care. And I really believe in that. And I believe that it might give you an early signal for disease.”


Over the past five years, the ctMoniTR collaboration designed a series of retrospective studies to systematically assess whether changes in ctDNA reflect response to treatment.

A slide providing an overview of Module 1 in the Friends ctMoniTR Project.

In Module 1, which focused on aNSCLC patients treated with TKI, the researchers set out to answer three questions:

  1. How are early changes in ctDNA associated with long-term clinical outcomes?
  2. Do changes in ctDNA complement first RECIST or best overall response (BOR)?
  3. If changes in ctDNA are combined with first RECIST, can improvements in associations with outcomes be observed?
A slide showing molecular response based on detection in the Friends ctMoniTR Project.

Of particular interest is the question of whether patients who had non-detected ctDNA on treatment—molecular responders—are correlating with radiographic responders as defined by first RECIST.

“Once we had our ctDNA categories, we decided to compare molecular and radiographic response,” said Hillary Stires, director of regulatory and research partnerships at Friends. “So, within that first 10 weeks after index, we selected that time point specifically because it aligned with when our ctDNA values were collected as well, or best overall response, which could have happened at any point during the study.

“Because ctDNA on treatment time point was within 10 weeks, we wanted to have a similar timing with the first RECIST, also BOR timing was variable and sometimes took many months, which meant that indexing would be challenging.”

To answer research question #3, the researchers performed four analyses when combining ctDNA and radiographic response.

“The first focused on radiographic response alone and associations with outcomes. So, here we’re not looking at changes in ctDNA at all,” Stires said.

A slide showing radiographic response + outcomes without ctDNA in the Friends ctMoniTR Project.

“The next analysis combined radiographic response with changes in ctDNA to look at outcomes.

A slide showing radiographic response + ctDNA + outcomes in the Friends ctMoniTR Project.

“And the third analysis was an opportunity for us to compare models to determine whether we see improved associations with outcomes.

A slide comparing the full model (ctDNA x 1st RECIST) to the reduced model (ctDNA only) in the Friends ctMoniTR Project.

“Finally, we looked at molecular response within specific radiographic categories to see if ctDNA is associated with outcomes within these categories, specifically looking at partial responders in our stable disease categories.”

A slide showing OS and PFS Kaplan-Meier curves for partial response and stable disease by ctDNA response in the Friends ctMoniTR Project.

The results were based on retrospective data harmonized across eight clinical trials.

“Overall, we noticed that incorporating radiographic response at first RECIST into molecular response categories does not improve the degree of association with survival for those with non-detected ctDNA on treatment,” Stires said. “For our likelihood ratio test, we walked away from this analysis, seeing that radiographic response at first RECIST does not provide additional value in characterizing the association with overall survival beyond the contribution of molecular response.”

What does this mean for drug development?