“In this new paradigm, AI becomes the modern physician’s stethoscope as a powerful, indispensable tool that applies the most sophisticated developments in computer science in the service of medicine.”
—Calvino et al, Genomic Medicine Laboratory
Publication
Article
Author(s):
Precision oncology has entered a new phase, as data sets mature and a new wave of tools emerges to help clinicians manage cancer care over time. A feature from our upcoming special issue of Evidence-Based Oncology: our annual ASCO recap.
At first, the SERENA-6 trial (NCT04964934) seemed a typical choice for the plenary session of the 2025 American Society of Clinical Oncology (ASCO) Annual Meeting in Chicago, Illinois. It involved a new therapy, camizestrant (AstraZeneca), the first next-generation oral selective estrogen receptor (ER) degrader and complete ER antagonist, being used with a CDK4/6 inhibitor to offer superior results for patients with the most common type of breast cancer.1,2
But in SERENA-6, the drug is only part of the story. The trial also broke ground with repeat testing for circulating tumor DNA (ctDNA), used to capture the emergence of a key mutation well before cancer progression showed up on a scan.2,3 This mutation, ESR1, is most likely to appear in patients with hormone receptor–positive, HER2-negative (HR+/HER2–) breast cancer following use of aromatase inhibitors—a sign that endocrine therapy is no longer working.4
Nicholas C. Turner, MD, PhD | Image: Royal Marsden Hospital
Julie R. Gralow, MD | Image credit: ASCO
As lead author Nicholas C. Turner, MD, PhD, of the Royal Marsden Hospital in London, United Kingdom, explained, patients being treated with an aromatase inhibitor and a CDK4/6 inhibitor were tested with a Guardant assay every 2 to 3 months for the ESR1 mutation; among those who developed one, half were switched from their aromatase inhibitor to camizestrant with amazing results. Median progression-free survival for the camizestrant group was 16.0 months compared with 9.2 months for the aromatase inhibitor group for a 56% improvement in the time to progression or death.2,3 With HR+/HER2– breast cancer accounting for 70% of new diagnoses, the results could impact more than 200,000 patients each year just in the United States.4
SERENA-6 encapsulates what is happening at the cutting edge of cancer care: It’s no longer just about new therapies but also about the maturity of precision medicine to account for meaningful advances with both new and approved drugs. During a press briefing, ASCO Chief Medical Officer and Executive Vice President Julie R. Gralow, MD, noted how the 2025 meeting offered multiple examples of how biomarker testing identifies subsets of patients who might benefit from a particular drug or from adjuvant or perioperative therapy.1,5-7
“We’re now taking common cancers and turning them into rare cancers,” she said.
Nine years after the 21st Century Cures Act empowered the FDA to use data analytics in drug approvals,8 precision oncology has entered a new phase. More than a decade of data collection, through initiatives such as Caris Life Sciences’ Caris Precision Oncology Alliance (POA),9 is allowing the marriage of genomics and artificial intelligence (AI) to bear fruit, with new tools to customize treatment, improve early detection, and understand drug resistance, according to authors from the Genomic Medicine Laboratory in Rome, Italy.10
“As the global population continues to age, chronic degenerative diseases and cancer are increasingly challenging the sustainability of worldwide health care systems. The growing burden of these diseases demands not only innovative treatments but also more efficient and equitable health care strategies. In this context, the integration of genomics and AI holds transformative potential,” the authors write. “By enabling earlier diagnosis, more accurate risk prediction, and the development of tailored therapeutic approaches, these technologies can improve clinical outcomes while optimizing the use of health care resources.”
“In this new paradigm, AI becomes the modern physician’s stethoscope as a powerful, indispensable tool that applies the most sophisticated developments in computer science in the service of medicine.”
—Calvino et al, Genomic Medicine Laboratory
However, the authors warn, this shift will require an overhaul of medical education to train a new corps of clinicians with competency in AI. “In this new paradigm,” they write, “AI becomes the modern physician’s stethoscope as a powerful, indispensable tool that applies the most sophisticated developments in computer science in the service of medicine.”10
The authors envision a future of more equitable treatment, but in the United States, experts say prior authorization, clinical inertia, and geography could widen disparities in the short term. Both the potential and pitfalls of this wave of innovation were on full display at ASCO 2025, as many who spoke with Evidence-Based Oncology (EBO) agreed that testing and technology rivaled new therapies for attention.
Pathologists are in the spotlight as never before; 109 were trained in an international study with results that showed how AI could improve diagnostic agreement among pathologists. This improved accuracy when deciding which patients with HER2-low and HER2-ultralow status should be treated with the antibody-drug conjugate (ADC) trastuzumab deruxtecan (Enhertu; Daiichi Sankyo, AstraZeneca).11
Another AI study involved combined analysis of pathologists and AI (CAPAI). Investigators from University Medical Center Utrecht in the Netherlands showed that using the CAPAI marker, which produces an AI-developed score using pathological data and hematoxylin and eosin slides, surpasses current tools to predict risk in patients with colorectal cancer post surgery.12
And TROPION-Lung02 (NCT04526691), a phase 1b/2 trial involving the use of datopotamab deruxtecan (Dato-DXd; Datroway; Daiichi Sankyo, AstraZeneca) and pembrolizumab with or without chemotherapy, offered promising results in advanced non–small cell lung cancer (NSCLC) while highlighting the potential of AstraZeneca’s Quantitative Continuous Scoring, an AI platform used to measure TROP2 expression in tumor samples. The analysis used normalized membrane ratio to identify which patients might respond best to Dato-DXd, an ADC that targets TROP2.13
Yet with all this good news, uneven testing rates, inadequate use of results, and reimbursement worry test developers and payers alike.14-16 Policy questions will only increase as scientists explore the potential of AI.
Take Foundation Medicine, whose CEO, Dan Malarek, PhD, spoke with excitement about ongoing pilots for the company’s FoundationOne Monitor assay, which is on track to go to market by the end of 2025. “We’re still collecting data, getting feedback, and optimizing things—including nailing down the ordering experience,” Malarek said in an interview during ASCO. “What we’re looking to do is build the right evidence to ensure that we have reimbursement,” which he said can be a barrier.
Next up will be a look at patient experience, which Malarek said is of “supreme importance to us.”
Joshua K. Sabari, MD | Image: NYU Langone
During ASCO, Joshua K. Sabari, MD, a lung cancer specialist for NYU Langone Health in New York, New York, presented quality-of-life data for zongertinib (Boehringer Ingelheim), an investigational therapy gaining notice for its ability to target HER2 in NSCLC while sparing EGFR, thus reducing toxicity. In an interview, Sabari described how the rapid advances in drug development have aided progress in biomarker testing.
“We definitely hit critical mass 10 years ago when I started practice,” Sabari said. “Most people were not doing broad-panel next-generation sequencing [NGS], and we were missing many patients with mutations. Now that we have better therapeutics available, we are testing more—at academic centers, it’s probably 99% to 100%. But I just had a discussion with my colleagues from Latin America, and they’re still only at about 40% to 50% of newly diagnosed patients.”
“For most of us, we’re getting broad-panel NGS,” he continued, “with the reason being we have effective therapies; we can match patients with [treatments] that allow them to live longer and to live better.”
Testing rates in academic centers are not matched everywhere, unfortunately. A 2024 study in JAMA Network Open examined rates of NGS among those diagnosed with prostate and urothelial cancers, and results found that despite ample evidence of actionable genomic alterations in both cancer types, most patients did not receive NGS. “Black race, low socioeconomic status, and Medicaid and Medicare insurance coverage were associated with lower NGS rates,” study authors wrote.14
As EBO went to press, a pair of rulings from the Supreme Court of the United States offered mixed news on access to basic cancer screenings, typically the first test that can lead to a cancer diagnosis. Although the court upheld the right of the US Preventive Services Task Force to determine which preventive screenings should be free under the Affordable Care Act,17 a separate ruling found that states could withhold Medicaid payments from Planned Parenthood, which serves as a provider for many low-and middle-income Americans.18 Data from Planned Parenthood state that more than 75% of their patients earn below 150% of the federal poverty level and 60% are covered by Medicaid or a Title X program for those with low incomes.19
Dan Malarek, PhD | Image: Foundation Medicine
Biomarker testing has a much higher profile at ASCO than it did 10 to 15 years ago, said Malarek, who sees several reasons for this evolution: Not only are there more drugs and drug classes that involve testing, but there is also much more an oncologist must absorb to keep pace. “They have the challenge of sifting through the [information] and trying to determine, ‘What is the best therapy for my patient with this specific mutation?’ So I would say it’s getting more complex,” he said.
Although finding new biomarkers and developing companion diagnostics remains critical, Foundation Medicine has bigger things in mind. In May, the company unveiled FlexOMx Lab, described as a “flexible research lab” for use by biopharmaceutical partners to develop new medicines.20 Malarek said this resource will expand beyond the company’s expertise in genomics; plans include easier processing of plasma samples for ctDNA analysis and more offerings for hematological malignancies—an assay for minimal residual disease (MRD) is due later in 2025. The laboratory will offer whole-exome sequencing, which Malarek said will aid development of neoantigen therapies. Foundation Medicine is also moving beyond oncology, with services to support precision medicine applications in areas such as atopic dermatitis.
The company presented 15 abstracts during ASCO, most involving its FoundationOne CDx or FoundationOne Liquid CDx assays.21 On ASCO’s final day, investigators presented results involving FoundationOne Monitor, which is currently used in research settings only. The assay uses a ctDNA tumor fraction biomarker to track changes over time, allowing investigators to assess both responses to therapy and tumor resistance; the assay will go beyond telling researchers that a therapy is not working to assess why that is so.22
“We realized the FoundationOne Monitor can be used as a prognostic to identify patients who will respond to certain therapies,” Malarek said. The study, the results of which found olaparib with radium 223 was superior to radium 223 alone in treating metastatic castration-resistant prostate cancer, used the Monitor assay to assess homologous recombination repair (HRR) gene mutation status; more ctDNA analyses are expected.23
Beyond reimbursement, factors such as turnaround time and a physician’s comfort level with using biomarkers can affect their utility. Results from a well-known study by ASCO and LUNGevity found that turnaround time can play a significant role in whether a clinician orders a test or whether the results are used. When community oncologists must wait more than 3 weeks to receive results, the study results found they are more likely to move on to a nontargeted therapy.24 The 2022 study results also uncovered a generation gap: Clinicians who had completed training more than 5 years prior were more likely to cite wait times as a reason not to order testing than those with less than 6 years of experience (42% vs 19%).24
These factors all affect which patients receive testing. A 2022 report in Value & Outcomes found that reimbursement is needed to ensure market access but that patient access to testing, in this case NGS, is a distinct step that requires “infrastructure, awareness, and education of health professionals, which appear to be suboptimal in many geographical areas.” 25 Endorsements from clinicians who use the technology are critical.25
EBO asked Malarek, “Do some practices have better testing rates than others, or are gaps between individual physicians? And are they generational?”
“I think it’s a combination, to be honest. But I think it’s more on the individual level, and a lot of this has to do with education,” he said, echoing the findings from Italy.10 “Even in medical school today, there’s not the right emphasis on the molecular space as there should be. So that leaves it up to the biopharmaceutical companies and us as the diagnostic companies to drive that education.”
Habits can be hard to break, Malarek explained. A senior physician who sees 20 patients a day may have entrenched work patterns that existed before testing rose in importance. More work is needed to educate oncologists “not only [about] how to interpret the results but also when is the right time to order a test,” he said. Most of all, physicians need help interpreting the results and selecting the best therapy.
Will AI make all this easier?
“In the long term, yes, absolutely,” Malarek said. “In the short term, we still need to improve just the basic, foundational education of oncologists.” He sees progress in that the largest community oncology networks have set priorities, such as testing every patient with metastatic disease or moving the testing rate from 50% to 90%.
“There are commitments to move the needle, and that’s a huge plus for patients.”
Testing diverse populations does not just improve outcomes for individual patients—it also adds to a rich database that fuels AI and allows scientists to ask new questions. That is what Caris Life Sciences has learned through its POA, a collaborative that was closing in on 100 members when its chair, James Hamrick, MD, MPH, spoke with EBO during ASCO.
James Hamrick, MD, MPH | Image: Caris Life Sciences
The alliance launched in 2015, Hamrick said, as a vehicle to not only improve patient outcomes at the point of care but also to gain insights from the data that would be collected in the testing process. “If you’re analyzing tumor biology at scale, over time, that could become a very powerful data set,” he said.
POA launched in the early days of real-world data, Hamrick said, a year before the 21st Century Cures Act called for the FDA to make this kind of data useful in drug approvals or postmarket monitoring.8 “The vision was that as a by-product of the clinical experience, [Caris] would spin off a lot of things to help science advance. And the company realized that to take advantage of the data, they needed partners.”
Thus, POA was born.
Through the alliance, investigators had a way to propose studies that would use shared, deidentified data for research purposes. Over time, more and more research teams used the data set, resulting in more than 1000 publications, Hamrick said. As the POA grew, the data set grew as well to nearly 500,000 clinical outcomes and matched molecular data. POA members include 45 National Cancer Institute–designated sites, he said.
The Caris data come from institutions across the United States and the globe, thus offering data from diverse areas and populations. “By having a footprint that covers basically all of the United States and some places [outside the US], we’re able to add that layer of what’s actually happening in the real world outside of the prospective trial,” Hamrick said.
Such rich data allow investigators to probe multiple points that can contribute to outcomes, whether it is a molecular mechanism, a social driver, or access to care. “There are many different things that might explain inequity,” he said. The Caris data include information from groups often left out of research, “and that’s been very intentional.”
For POA members, access to data and the analytics support is free. “That’s the game changer,” Hamrick said, because it allows alliance members to launch projects at significant savings. “Most individual sites, even big ones, don’t have the scale to answer a lot of questions. By being a member of the POA, you get access not only to your own data but to HIPAA [Health Insurance Portability and Accountability Act]-deidentified data from across the network,” he said.
The process to use the Caris data set has protocols, which are overseen by an external board. Projects can be pitched from POA member investigators, from within Caris, and from the pharmaceutical industry. Not every idea can be pursued, so the POA has a peer review process to select projects. “It’s baked into the founding vision of the company,” Hamrick said. “It’s the right thing to do—and that’s wonderful—but also, we’re good at it.”
And although there is an obvious scientific benefit to the arrangement, he said, “Generating this kind of science is also good for business…. We provide data to really smart investigators who are really qualified to do rigorous science with us. It’s mutually beneficial.”
With precision oncology reaching a new phase, what are the near-term priorities?
Hamrick’s list was long: demonstrating the value of the whole exome, the whole transcriptome, and the whole genome. Malignant hematology and MRD are priorities for POA as testing companies explore how to safely de-escalate therapy. As seen in SERENA-6, exploring tests that can warn of a coming relapse before it arrives is the wave of the future.
“It’s an exciting time,” he said.
References
Stay ahead of policy, cost, and value—subscribe to AJMC for expert insights at the intersection of clinical care and health economics.