Flatiron Health’s Erin Fidyk, MSN, MBA, shares findings from a real-world study of early breast cancer showing that ctDNA positivity is consistently linked to worse outcomes across subtypes—highlighting its growing prognostic value and potential role in refining treatment decisions.

Erin Fidyk, MSN, MBA: My name is Erin Fidyk. I am a senior clinical director at Flatiron Health. I've been at the company now for 5 years, and most of the time I've been working with our machine learning (ML) and artificial intelligence (AI) team doing a lot of extraction on our large data sets. I'm a nurse practitioner by training and also have an MBA. I've been in the oncology space for my entire professional career.

Can you provide an overview of your study presented at ASCO 2025?

Fidyk: To zoom out for a moment and provide a bit of context, we have 5 million patients in our Flatiron real-world electronic health record (EHR) database. Of those, and at the time of our study, we had close to 740 000 patients in our panoramic breast dataset.

This dataset is fully ML-extracted, meaning that we are using ML models to pull out characteristics for these patients by way of natural language processing and/or large language models. It really covers the full patient journey from diagnosis onwards. For this particular project, we focused on early-stage, so patients with stage 1 to 3 breast cancer, who had at least 1 ctDNA test in this setting, before any evidence of locoregional or metastatic disease.

We were also able to dig into testing trends. For example, we found that most testing is happening for patients with stage 1 and 2 disease, and overwhelmingly in the adjuvant setting rather than the neoadjuvant setting, with 94% happening in adjuvant.

We're also seeing not only an increase in testing prevalence over time, but a decrease in median time to first test from initial diagnosis date. If we look at breast cancer subtypes, most testing is occurring in patients with triple-negative breast cancer (TNBC). What we found was that patients with a positive ctDNA test did, in fact, have worse outcomes with increased risk of recurrence and worse overall survival (OS) compared to those with a negative ctDNA test. We ran these analyses at both the 3-year and 5-year time points for recurrence-free survival (RFS), and we were able to see outcome differences already at the 3-year mark, which was pretty incredible.

Our hazard ratios were also so high that we were concerned we might be capturing patients who had testing done close to or at the time of a ctDNA test or at the time of recurrence. We ran a sensitivity analysis to remove those patients who had less than 4 months between a recurrence event and a ctDNA-positive test. We still saw that strong association, which gave us confidence that the relationship we were seeing was actually true.

Ultimately, we were able to show that ctDNA-positive disease was associated with worse OS and higher recurrence risk across all early breast cancer (EBC) subtypes, which highlights the potential prognostic value of ctDNA testing as well as the growing adoption in this setting.

You found that positive ctDNA was associated with a higher risk of recurrence and worse OS. Did this relationship hold consistently across molecular subtypes such as HR+/HER2–, HER2+, and TNBC?

Fidyk: Yes. For a bit of context on the methodology, we first generated unadjusted Kaplan-Meier curves to examine the association of the ctDNA status with disease recurrence and OS. We then ran an adjusted Cox model controlling for potential confounders, everything from ECOG status to treatment, such as neoadjuvant and adjuvant, to estimate that effect of ctDNA positivity on RFS. We didn't end up doing this for OS just because of the small sample size and limited number of events in the early setting.

We included results in the abstract for the 3-year RFS because we were able to see quite a difference in that 3-year mark already. For example, when we looked at all subtypes combined in patients who had ctDNA testing, only 2% of patients with a negative test had recurrence, whereas 24% of patients with a positive test had recurred at the 3-year mark, which was pretty compelling.

To answer your question, yes, the relationship did hold consistent across each of the 4 molecular subtypes, with triple-negative having the strongest association and the highest risk of recurrence.

Given these findings, how do you envision ctDNA testing being integrated into current EBC clinical pathways—particularly around surveillance or escalation/de-escalation of adjuvant therapy?

Fidyk: This is a very exciting space to watch right now, with a number of opportunities for clinical integration, for example, using ctDNA to refine risk stratification and inform treatment personalization, especially in identifying patients who may benefit from escalation or deescalation therapies.

Anecdotally, we're starting to see trends in our real-world data around decision-making as a result of ctDNA testing in the adjuvant setting, where we're seeing evidence of clinicians initiating treatment due to a positive test after surgery. We're actively looking into this to better understand decision-making and clinical rationale for ordering ctDNA testing. For example, is it symptom driven or are clinicians following surveillance parameters? Does this vary by practice setting or tumor type? Our use of large language models (LLMs) is giving us the opportunity at scale to understand these nuanced, super complex clinical concepts and questions in a way that we really couldn't understand before. We'll be able to see in our data how decision-making is playing out in the real world, and how this can potentially inform guidelines.

Additionally, I would say that we need longer follow-up time and prospective validation in clinical trials in order to confirm the potential prognostic value of ctDNA positivity. In fact, this is something we're actively working on right now. We're successfully powering prospective minimal residual disease (MRD) research by generating novel clinical and molecular evidence through a partnership with one of these MRD vendors.

One other note that's actually quite timely, we saw that Natera Signatera MRD assay received broad Medicare coverage just yesterday. This is signaling the growing recognition around the impact and validation of ctDNA testing, which is only going to help the integration of this type of testing into clinical guidelines.

What future studies are needed to validate ctDNA as a clinical decision-making tool in early breast cancer, and are there interventional trials underway?

Fidyk: Great question. Clinical studies and these interventional trials remain the gold standard for how we know if something is working or not. It's one of the reasons that here at Flatiron, we're not just focused on retrospective analysis of ctDNA use, but are involved in prospective studies, as I mentioned earlier. CtDNA and MRD were really hot topics at ASCO this year. We saw this highlighted at the plenary session—the big session that everybody attends—with the SERENA-6 trial, which was demonstrating improved disease-free survival with treatment change based on ESR1 ctDNA-detected recurrence in the absence of radiologic findings. This positioned ctDNA as a valuable tool for real-time treatment adaptation in breast cancer care and many other cancer types too.

However, despite promising results from both real-world and even prospective research, I think there are still a number of gaps in our understanding of how to use ctDNA, and just questions overall. A few things that I'm thinking about, for example, are looking at clinical utility versus clinical actionability. If you think about ctDNA positivity, it has shown pretty clear prognostic value, but it's still not always clear how best to act on a positive result, especially in early-stage disease. The field is still catching up in terms of evidence-based guidelines for escalation or deescalation of therapy based on ctDNA status.

We also have this concept of false negatives and timing of testing. A negative ctDNA test result doesn't guarantee absence of disease. Timing, assay sensitivity, and tumor shedding dynamics all play a big role here. A single time point may miss low burden disease, especially for the tumor types that are low shedding.

There's also a question around test result interpretation, especially in the absence of radiographic disease. We know that detecting molecular recurrence before radiographic evidence—so any type of imaging study—presents not only opportunities, but a lot of uncertainty. Clinicians are navigating how or whether or not to treat patients based on ctDNA positivity alone.

Anecdotally, I remember speaking with a patient who was also a clinician at the poster presentation last year, where we were looking at colorectal cancer (CRC) and ctDNA testing and she said, "I'm a clinician and I'm also a patient, but I have a positive test and we are terrified. I don't know what to do with this." There was no evidence on the scan that she had disease, but the ctDNA test was positive. So, again, there are a lot of unanswered questions for this.

We also need prospective validation and clinical trial integration beyond what we've already seen so far. There's strong real-world and retrospective evidence linking ctDNA positivity to recurrence. These ongoing prospective trials are needed to determine if acting on ctDNA improves survival outcomes. We have to be considering overtreating or undertreating because of some of these results too.

One last and final point, because this is very relevant to our company and what we're doing, is that there's a lot of heterogeneity in real-world use and data. Real-world uptake of ctDNA testing remains highly variable. It's often influenced by clinician preference, geography, the setting, whether it's a community or an academic center, and payer coverage. This variability raises concerns about equity and standardization in patient care above just personalized care.