A meeting hosted by Friends of Cancer Research put a spotlight on the rising potential of circulating tumor DNA and AI-enabled tumor assessment in cancer drug development. An FDA program designed to validate biomarkers for regulatory use may be ill-suited to quickly advance these tools and others like them, however.
Background on biomarker development
- Biomarkers are a vital tool for disease diagnosis, patient risk assessment and drug development. The FDA describes biomarkers as defined, measurable characteristics of the body that indicate biological and pathogenic processes, or the response to an intervention such as a medical product. While clinical assessments can capture how a patient feels, functions or survives, biomarkers provide researchers with a valuable window into the body’s inner workings. These tests can be molecular, histologic, radiographic or physiologic.
- Biomarkers have a variety of purposes. They can diagnose, such as how sweat chloride is used to confirm cystic fibrosis; determine risk, as BRCA gene mutations do for assessing the likelihood of breast cancer; or monitor disease status, as prostate-specific antigen levels indicate tumor burden in people with prostate cancer. They’re also used to watch for treatment side effects and to measure drug response.
- Measuring drug response is a particularly important use case, allowing biomarkers to be used as clinical trial objectives, including as “surrogate endpoints,” which are meant to be predictive of clinical benefit. One example in oncology is objective response rate: Researchers assess whether a treatment shrinks the size of a target tumor under the assumption that reductions will correlate in some fashion with longer survival. The FDA maintains a list of surrogate endpoints that have supported traditional or accelerated drug approvals.
- The FDA validates biomarkers in several ways. They can be confirmed through the agency’s review and approval of a new drug or biological product, or through a qualification process the agency established in 2007. This process was formalized in 2016 through the 21st Century Cures Act, which established a three-stage pathway for groups to submit prospective biomarkers for FDA verification within a specific context of use. Once qualified this way, a biomarker can be incorporated into any drug development program to support regulatory approval. The FDA may also more generally evaluate scientific consensus around a biomarker measure and its clinical utility.
- The Biomarker Qualification Program is meant to address a market failure of sorts. “One of the problems with biomarkers is there’s really no one in charge of developing them,” explained JANET WOODCOCK, a former top FDA official, in a 2016 video on an FDA webpage about biomarker development. The BQP was designed to offer a more structured and transparent alternative to sponsors validating biomarkers via individual drug submissions. As detailed in a 2020 final guidance, developers first submit a letter of intent, which the FDA reviews for up to three months and decides whether to accept. If the letter is accepted, developers are tasked with assembling a qualification plan, which the FDA aims to review within six months after submission. An accepted qualification plan is then expanded into a full qualification package, which the FDA aims to assess within 10 months after filing.
- The FDA has qualified only 11 biomarkers through the BQP, and most were qualified prior to 21st Century Cures’ December 2016 enactment, according to a database maintained by the agency. Five were qualified to assess safety, one is validated as a surrogate endpoint and the remaining five are either prognostic, diagnostic or meant to measure or monitor disease. The FDA validated three biomarkers in November and December 2025, but prior to that the most recent qualification was in 2018.
- Sluggish utilization of the program has spurred criticism that it may not be as effective as intended. An analysis by Friends of Cancer Research published Oct. 28, 2025, in Therapeutic Innovation & Regulatory Science characterized the program as slow-moving. Median times for FDA review of letters of intent and qualification plans were more than double the agency’s respective three- and six-month goals, according to their analysis. (The law does not hold the agency to any statutory review timelines.) Sponsor development of qualification plans is also slow, taking a median of more than two-and-a-half years among the programs that had analyzable timeline data. “These trends demonstrate the program may not be well-suited for advancing novel response biomarkers,” the Friends researchers wrote. More resources might help, the paper concluded, noting that there are no dedicated funds tied to the BQP through user fees, for instance. [Read AgencyIQ’s analysis of Friends’ research here.]
Biomarkers as endpoints: Two emerging technologies
- A meeting hosted by Friends on Feb. 5, 2026, highlighted two biomarkers – circulating tumor DNA and AI-enabled tumor assessments – that could aid cancer drug development. Briefly, circulating tumor DNA, or ctNDA, describes the genetic fragments shed from cancerous cells into the bloodstream. Researchers can use ctDNA as a biomarker to detect or predict the presence and stage of a tumor, providing a less-invasive option than biopsies. As the amount of ctDNA detected can indicate a patient’s tumor burden, measuring it can be used to predict drug efficacy. AI-enabled tumor assessments refer to the use of artificial intelligence-based algorithms to assess and measure images produced through radiology.
- Panels convened by Friends discussed the potential of each technology, as well as the hurdles standing in the way of their broader use. Both promise to offer researchers and clinicians greater visibility into the cancers they study and treat, while addressing some of the limitations associated with assessing tumor biopsies by hand. In doing so, they might also be incorporated into clinical study as endpoints capable of supporting regulatory decision-making. (Alongside the meeting, Friends also published a new white paper on AI-enabled tumor assessments.)
- Friends has led a group, the ctDNA Monitor Treatment Response, that’s working to establish ctDNA as a trial endpoint. NÉVINE ZARIFFA, of the NMD Group, kicked off the meeting’s ctDNA discussion with a presentation of new data from the ctMonitoR project showing that decreased ctDNA levels are associated with improved overall survival at the individual level. Trial-level associations, which link the effect of treatment on ctDNA response with the effect of treatment on “true” endpoints such as survival, were weaker, however. The project plans to look next at earlier cancer settings for which a blood-based measurement such as ctDNA may be more functional than measuring changes in tumor size through imaging, Zariffa said. And there’s still work to be done defining how and when to collect ctDNA based on the type of tumor and to conduct trials that prospectively measure ctDNA’s association with outcomes, she added. [Read prior AgencyIQ analysis of the ctMonitoR project’s work here.]
- The FDA does not consider ctDNA a validated endpoint for pivotal trials designed to secure drug approval. “We are not ready to say that we’re going to accept it as an endpoint to support approvals,” said PAZ VALLANKI, a medical oncologist and supervisory associate director at the FDA’s Division of Oncology 2. “We are encouraged by the data that we’re seeing. We want to develop this data. But I can’t say that right now if you have a ctDNA endpoint – it would not likely lead to an approval based on ctDNA alone.” She expressed hope, however, that the oncology field could eventually reach such a point. “I think that the ctMonitoR analyses have been really encouraging so far and … are laying groundwork,” she said at another point in the discussion.
- Drugmakers are already putting ctDNA to work in other ways, though. “A lot of pharma are already doing this,” said JON BADEN, Bristol Myers Squibb’s head of precision medicine, citing as examples analyses to select drug doses, modify trial designs or make go/no-go decisions on a specific asset. “CtDNA is part of the story,’ he added. “It’s not the only decision point, but it’s one of the decisional points.”
- The Friends meeting also discussed the potential for AI-based tools to help researchers extract information from tumor scans. Currently, clinicians and radiologists assess imaging of solid tumors using RECIST, or Response Evaluation Criteria in Solid Tumors. This decades-old metric provides a standardized basis for defining whether a tumor has responded to treatment by measuring unidimensional changes in a select sample of cancer lesions. Typically, a 30% or greater reduction in the sum of longest diameters across sampled lesions is classified a “partial response,” for instance. In the trial setting, RECIST evaluations are typically conducted via blinded, independent central review, which often features two radiologists evaluating tumor scans and a third adjudicator to bridge any discordant findings. The aggregate results can contribute to response-based endpoints, which have been used regularly to approve new cancer drugs.
- RECIST has limitations that AI could help address. “We now know that subjectivity in lesion selection limits the consistency of capturing the tumor dynamics,” said LARRY SCHWARTZ, a radiologist at Memorial Sloan Kettering Cancer Center. “We are, with simple unidimensional measurement, not able to capture or understand – really, distinguish – treatment-related effects like pseudo-progression,” he added. “And we just capture a small number of target lesions with RECIST because it’s actually too time-consuming, and certainly artificial intelligence can help with that.” AI can enhance RECIST evaluation, such as by supplementing or speeding along the standard central review process.
- AI tools could also be used to assess tumors in novel ways, such as by measuring changes in tumor consistency, volume or growth. “Can AI help us detect something so nuanced about the disease that maybe a radiologist can’t see with their human eyes?” asked LAUREN BRADY, a director of translational science at Genmab. “I think there’s an important shift to make from thinking of radiology primarily as a diagnostic tool, a monitoring tool, and thinking about it as a source of biomarkers,” said NATHANIEL BRAMAN, the head of AI at biomarker developer Picture Health. “The truth is, radiology scans are dense with tumor biology, phenotypic information, and we’re just barely beginning to scratch the surface in terms of how these scans can inform precision medicine.” Panelists noted, however, that using AI tools in this way will require extensive work to validate their outputs, ensure reproducibility and evaluate how well they work across different settings.
- Friends’ white paper outlines steps biomarker developers should take to validate AI-enabled tumor assessments. The paper recommends that sponsors define the specific unmet need as well as the proposed context of use before embarking on collecting the analytical and clinical data needed to support development of a tool output as a biomarker. The group is currently leading a project, ai.RECIST, that seeks to evaluate how consistent and reliable AI-driven tumor evaluations are compared with human readers. “I think this project is the perfect place to start,” said Braman. “The fact is, we have a long way to go in terms of getting regulatory agencies on board with this idea and clinical trialists on board with this idea. Starting with RECIST – something that everyone understands, it’s verifiable – is exactly the right step one.”
While new cancer biomarkers hold promise, the FDA’s qualification process has arguably underperformed
- The BQP offers one pathway for validating biomarkers such as ctDNA and AI-enabled tumor assessment. However, speakers at the Friends meeting pointed to challenges that help explain the program’s underutilization and could similarly hamper qualification of tools like these through the BQP.
- Some of BQP’s shortcomings may have been baked in from the start. JOHN STONE, a principal at BGR Group who helped develop the 21st Century Cures legislation while working on the House Energy and Commerce Committee, recalled how industry’s enthusiasm waned after a proprietary channel with constraints on review timelines was stripped from the program. “Once the proprietary framework fell out, I think the excitement went away a little bit,” said Stone. “We ultimately landed on this public framework that, as far as I understand, has not been utilized as wholeheartedly or efficiently as some would have hoped.”
- Long review times have diminished the program’s attractiveness, several panelists said. PathAI, which in December 2025 secured qualification of a biomarker for assessing liver biopsies of people with metabolic-dysfunction associated steatohepatitis, began the process more than five years prior, for example. The BQP’s lack of dedicated resources “materialized in long gaps in time between the various stages of the qualification program,” said MICHAEL MONTALTO, a vice president of precision medicine at Amgen who was previously chief scientific officer at PathAI. Development was further slowed by “not really understanding what the agency would accept for validation,” Montalto said.
- Biomarkers can sometimes fall in a “no-man’s-land” between FDA review divisions, Montalto added. TALA FAKHOURI, who was associate director of data science and AI within the FDA’s Center for Drug Evaluation and Research until she departed last year, described an example of this in more detail: “What ends up happening is a submission comes in with AI that is challenging,” said Fakhouri, now a vice president consulting at the contractor Parexel. “It will go to [Center for Devices and Radiological Health]. CDRH will say, ‘This is not CDRH. Send it back to CDER.’ This takes a lot of time, partly because you don’t have that multi-disciplinary expertise that is really required when you’re looking at these innovative AI type of tools.”
- BQP’s focus on validating biomarkers within a specific context of use may be limiting. Fakhouri differentiated between analytical validation, which describes whether a test can accurately measure its target parameter, and clinical validation, which confirms whether that parameter is predictive of a health condition or outcome. She proposed a system by which sponsors could get analytical validation of their tool, which could then be taken up by other groups that pursue clinical validation for specific use cases. “We seem to always conflate analytic validation of tools with the clinical validation of tools, instead of getting qualification on the method itself, so that then it can be used in different contexts,” she said. “I think that would open the door for a lot of sponsors to integrate these tools in their drug development programs.”
- Panelists also highlighted a mismatch between the program’s requirements and the sponsors that most often develop new biomarkers. “We keep forgetting about a lot of the developers of these tools who sit outside of traditional drug development players,” said Fakhouri. “They’re not pharma; they’re not biotech; they are academia and small tech companies that – they don’t want to make information about their tool public because it is proprietary,” she added. “That’s how they will benefit from developing these tools, and there’s no pathway for them right now.” Montalto agreed, noting that emerging platform companies aren’t incentivized to participate “because the process takes so long and they don’t really have assurances that there’s going to be a market even after five years.”
- LOWELL ZETA, the FDA’s deputy commissioner for strategic initiatives, told the panel the agency is looking at what it can do to increase uptake in the program. “The program has the potential for just such profound benefits,” said Zeta, while acknowledging that larger companies often simply choose to integrate biomarker development into their drug programs. Zeta also noted “gaps” in understanding around the program’s data quality standards.
Could PDUFA negotiations offer an opportunity for BQP reform?
- Speakers appeared to agree that the BQP could be reformed. “I think we can all agree that the program needs to be revamped,” said Fakhouri. “My real hope is that, in time, this is going to be a well-used, really strong part of our drug development programs,” added Montalto, who mentioned hearing interest from other attendees to “relook at the BQP and try to come up with some substantive, concrete ways to make those improvements.”
- Biomarker development could find its way into discussions around the renewal of the Prescription Drug User Fee Act, which needs to be reauthorized by September 2027 for a new five-year cycle. “Negotiations are happening now, … so let’s lock in resources for biomarker qualification and getting the FDA and stakeholders to actually talk to each other so we’re prioritizing the biomarkers that will make the biggest difference,” said Rep. DIANA DEGETTE, D-Colo., who is the ranking member of the Energy and Commerce’s Subcommittee on Health.
- The broader Drug Development Tools program, of which BQP is a part, has already come up in PDUFA negotiations. At a Dec. 5 meeting with patient and consumer groups, the FDA acknowledged that “the program faces delays when subject matter experts from various divisions are unavailable due to staff shortages,” according to minutes posted on the agency website. Stakeholders attending the meeting expressed “strong support” for BQP and “stated the need to resource it with PDUFA resources.” The topic has also come up in FDA meetings with industry, according to published minutes. Industry representatives on Jan. 6, 2026, presented proposed language around DDT and biomarker qualification for the “commitment letter” that results from PDUFA negotiations, which the FDA agreed to during the following Jan. 8 meeting.
- PDUFA negotiations are likely to wrap up over the coming months and produce a draft commitment letter, which could direct agency resources toward speeding biomarker review within existing legal authorities. The agency will solicit comments on this draft letter before producing a finalized version about a year from now. Lawmakers will then debate a legislative package ahead of the September 2027 expiry of the current PDUFA framework. If there’s sufficient lawmaker interest and consensus, the new PDUFA package could be used to usher in legal changes to the BQP or the broader DDT program.
- For AI-enabled tools, though, industry is looking for more guidance from the FDA. “Everyone is piloting but not scaling,” Fakhouri said, referring to AI tool developers. “And a part of that is because they don’t know if the tool will be acceptable by the regulator. A lot of them are worried about what the inspectors will say about the application of their tool.” The FDA and European Medicines Agency recently published a set of agreed-upon guiding principles, which Fakhouri called a good “first step,” but indicated that industry is looking for more “consistency and predictability” across FDA review divisions, too. Policy, Montalto said, needs to keep up with a “tsunami of innovation that’s happening” in the development of both drugs and biomarkers. [Read AgencyIQ’s analysis of the FDA-EMA principles on AI here.]
- Other biomarker pathways could be used to advance technology such as ctDNA and AI-enabled tools. Sponsors could choose to incorporate data supporting their use within an individual drug development program, although assembling sufficient evidence is likely to be difficult. The FDA may also work with the scientific community more generally to advance ctDNA and AI tools as potential biomarkers. For example, the agency worked with the University of Miami’s Sylvester Cancer Center and an international consortium known as i2TEAMM to evaluate minimal residual disease as an accelerated approval endpoint in multiple myeloma. Research from both groups and the FDA was discussed publicly at an April 2024 advisory committee meeting and later supported the agency’s development of draft guidance on the topic. [Read AgencyIQ’s analysis of the FDA’s recommendations here.] The FDA previously used a similar approach to establish pathological complete response as an early endpoint for breast cancer trials. Doing so took significant time and resources in both cases, however.
Featuring previous research by Kari Oakes.
Key documents and dates
- Friends of Cancer Research meeting: Modernizing Oncology Endpoints: Pathways for Evidence and Policy, Feb. 5, 2026
- Friends of Cancer Research white paper: Leveraging AI-Enabled Tumor Assessment Tools on Radiological Images to Evaluate Treatment Effect and Support Clinical Trial Endpoints in Solid Tumors, February 2026