SUMMARY:
The randomized phase III FIND trial integrated serial ctDNA methylation testing with protocol-driven CT imaging after colorectal cancer surgery. ctDNA-guided surveillance detected recurrence a median 3.9 months earlier and doubled curative-intent treatment among patients who relapsed (48.1% versus 23.6%).
- The study moves MRD beyond prognosis toward actionable surveillance, although mature overall-survival, economic, and independent replication data remain essential for adoption.
CITATION:
Mo S, Zhou C, Ma M, et al. Dynamic circulating tumor DNA methylation monitoring guiding postoperative surveillance in nonmetastatic colorectal cancer: a prospective, randomized, phase III FIND trial. J Clin Oncol. Published online July 29, 2026. PMID 42525894. doi:10.1200/JCO-25-03009.
https://pubmed.ncbi.nlm.nih.gov/42525894/
The Blood Test That Tells the CT Scanner When to Wake Up
The randomized FIND trial moves ctDNA beyond prognosis and into a working surveillance algorithm—doubling the proportion of recurrent colorectal cancers treated with curative intent
Patrick Adams recently highlighted a deceptively simple question about postoperative colorectal cancer surveillance: can ctDNA tell us more precisely whom to image and when, so that recurrent disease is found while it is still manageable? His LinkedIn note is not independent evidence, but it captures the operational insight of an unusually interesting new trial rather well.
The phase III FIND trial, published July 29 in the Journal of Clinical Oncology, did not merely add another biomarker measurement to an observational registry. The investigators built a complete, protocol-driven system in which serial ctDNA results changed the timing of CT imaging, subsequent negative results could turn intensified surveillance back down, and the clinical endpoint was not simply “earlier detection.” It was whether patients whose cancer recurred could receive metastasis-directed treatment with curative intent.
That is a much more consequential test of molecular residual disease.
What does ctDNA MRD actually contribute?
After apparently curative surgery, conventional imaging faces an unavoidable problem. CT is quite good at showing where recurrent cancer is, but only after the deposits become large enough to see. A blood-based ctDNA assay may detect the molecular signal of relapse earlier, but it generally cannot tell the clinician where the disease is located, whether it is technically resectable, or whether an apparent molecular recurrence can actually be cured.
In other words, ctDNA is an alarm, not a map. CT remains the map.
The potential contribution of MRD is therefore not that it replaces imaging. It identifies a period in which the prior probability of recurrence has abruptly risen and imaging should be performed sooner or more frequently. Used this way, ctDNA converts surveillance from a fixed calendar—scan everybody at approximately the same intervals—into a conditional workflow:
Molecular signal appears → image now → if imaging is still negative, watch more closely → if the molecular signal disappears repeatedly, return to the ordinary schedule.
That is what impresses about FIND. The intervention was not “order ctDNA.” It was an integrated template joining a sensitive molecular signal, CT localization, repeat-testing rules, treatment eligibility, and an explicit off-ramp from intensified monitoring.
The investigators used a fixed, 10-marker methylation assay based on multiplex quantitative PCR. Blood was collected before surgery, within one month afterward, and quarterly for two years. A positive result required immediate contrast-enhanced CT of the chest, abdomen, and pelvis. If CT did not yet show metastasis, CT was repeated every two months while quarterly ctDNA testing continued. Only after two consecutive negative ctDNA results did imaging return to its standard frequency. If CT found recurrence before ctDNA did, ordinary relapse management proceeded without delay.
This last feature matters. FIND did not allow the biomarker to delay action on radiographic disease, and it did not leave patients trapped in an intensified imaging loop after one positive blood test.
The trial
FIND was a prospective, multicenter, randomized phase III study conducted at six academic centers in China. It enrolled adults with stage I-III colorectal cancer who had undergone an R0 resection and had no radiographically detectable distant metastases. The experimental arm received the ctDNA-directed surveillance algorithm in addition to conventional follow-up; the control arm received standard CT-based surveillance.
The primary endpoint was pragmatic: among patients with radiographically confirmed recurrence, what proportion received curative-intent metastasis-directed therapy?
That endpoint moves the field beyond the familiar finding that postoperative ctDNA positivity predicts recurrence. The unresolved question has been whether learning about molecular recurrence earlier changes anything important. FIND asked whether the information could move patients from palliative management into surgery, ablation, or another treatment intended to eradicate limited metastatic disease.
A total of 728 patients were randomized, although the modified intention-to-treat analysis contained 584 eligible patients—289 in the ctDNA arm and 295 controls—after exclusions for findings such as stage IV disease, a non-CRC diagnosis, or failure to undergo surgery. Median follow-up was 23.3 months.
The main result: twice as many patients reached curative-intent treatment
Recurrence occurred at virtually the same rate in the two groups: 18.0% with ctDNA-guided surveillance and 18.6% with conventional surveillance. That is exactly what one should expect. A surveillance strategy does not change whether microscopic cancer remained after the original operation; it changes the circumstances under which that cancer is discovered.
Among the patients who recurred, however, 48.1% in the ctDNA-guided arm received curative-intent metastasis-directed therapy, compared with 23.6% in the control arm. The relative risk was 2.03, with an absolute difference of 24.4 percentage points and a P value of .008.
Put another way, among patients who experienced recurrence, approximately four would need to be managed under the ctDNA-directed strategy for one additional patient to receive curative-intent treatment. That is a derived calculation rather than a prespecified trial statistic, but it conveys the clinical magnitude.
The median time to clinical recognition of recurrence was 9.5 months in the ctDNA group and 13.4 months in controls—a lead of 3.9 months. The value was not simply that recurrence appeared earlier on a graph. Earlier detection produced a visibly different anatomical presentation.
Among patients with liver metastases, 75% in the ctDNA arm had three or fewer lesions, versus only 28.6% in the control arm. Ninety percent had a largest lesion no greater than 3 cm, versus 57.1%, and 80% had disease confined to one hepatic lobe, versus only 28.6%. These are not abstract molecular endpoints. They are the characteristics that often determine whether a surgeon can plausibly remove the disease.
For recurrences confined to the liver and/or lungs, curative-intent treatment was delivered to 42.3% of patients in the ctDNA arm and 18.2% of controls. An exploratory analysis also found longer post-recurrence progression-free survival in the ctDNA group, with a hazard ratio of 0.51, although this should not be mistaken for mature [longer-in-time] proof of an overall-survival benefit.
ctDNA was considerably more sensitive than CEA
Within the ctDNA-guided arm, longitudinal methylation testing had a reported sensitivity of 84.1% and specificity of 91.4% for recurrence. CEA had a sensitivity of only 35.7%, although its specificity was slightly higher at 94.0%. Serial ctDNA positivity was much more strongly associated with shortened recurrence-free survival than elevated CEA; conversely, sustained ctDNA negativity identified patients with substantially more favorable outcomes.
This comparison helps answer the “what the heck does MRD contribute?” question. It enriches the pool of patients being sent for immediate imaging far more effectively than CEA. It does not eliminate false alarms—the 91.4% specificity makes that clear—but the protocol anticipated them. A positive test without visible disease triggered closer observation, not immediate systemic treatment based on blood alone.
The lead time was also heterogeneous. Among recurrences detected molecularly before imaging, 36% had no more than two months of lead time, 56% no more than four months, and 81% no more than eight months. Thus, FIND was not based on a fanciful assumption that ctDNA always provides a year-long warning. In many patients, the warning was relatively brief—but apparently long enough to matter.
A strong trial, but not yet the final verdict
The outcome is highly favorable, but “more curative-intent treatment” is not synonymous with “more patients cured.” An intervention may discover more apparently operable metastases, expose patients to more surgery, and still fail to improve overall survival. The authors properly invoke the cautionary experience of intensive conventional surveillance: more favorable anatomy at recurrence does not by itself guarantee longer life.
FIND’s 3-year and 5-year overall-survival analyses are not mature. Its patient-reported outcomes—including fear of recurrence, cancer-specific distress, psychological morbidity, and quality of life—are also pending, as is a full health-economic analysis. Those data are especially important when a positive result can initiate repeated CT examinations and months of uncertainty.
Generalizability is another issue. This was one methylation assay, one surveillance protocol, and one national health care setting. A favorable trial of this assay-plus-algorithm does not prove that every tumor-informed mutation assay, every fixed-panel assay, or every commercial MRD product will produce the same result. The trial also randomized 728 patients but excluded a substantial number before the primary modified intention-to-treat analysis. The consistency of the per-protocol and modified intention-to-treat findings is reassuring, but a future confirmatory trial should complete eligibility assessment before randomization wherever possible.
Finally, two authors reported employment, ownership, leadership, or patent interests involving Innovation Biomed and the mqMSP assay. That disclosure does not negate a randomized multicenter result, but it strengthens the case for independent replication.
Why FIND matters for coverage
Medicare and commercial insurance have approached MRD from markedly different directions.
MolDX Medicare policies have permitted coverage of qualifying MRD assays when the test detects recurrence before conventional clinical or radiographic evidence, has adequate validity, and—critically—provides information that will change management. Coverage remains test-specific and contractor-based rather than a blanket national declaration that all ctDNA testing is medically necessary. (CMS)
The FIND result maps unusually well onto that Medicare logic. It shows not merely that ctDNA can precede CT, but that a prespecified response to the molecular result changed the proportion of recurrent patients offered curative-intent treatment.
The contrast with commercial coverage is sharp. UnitedHealthcare’s commercial molecular-oncology policy effective August 1, 2026, continues to list tumor-informed and tumor-naïve MRD assays—including well-known commercial products—as unproven or not medically necessary. Its evidence discussion emphasizes the lack of peer-reviewed evidence that MRD testing improves decisions or outcomes. (UHC Provider)
FIND is therefore the kind of study commercial payers have said they were waiting for. But it will not automatically compel coverage of an unrelated American assay. Payers can still reasonably ask whether the benefit belongs to:
ctDNA surveillance as a general concept;
this particular methylation test;
the FIND imaging algorithm;
the quality of the participating multidisciplinary teams; or
the complete combination of all four.
The strongest commercial evidence package would replicate the clinical algorithm with the assay actually being submitted for coverage. Nonetheless, FIND makes a generic dismissal of MRD as “prognostic only” increasingly difficult to sustain.
Guidelines may move—but probably not immediately
The caution reflected in publicly reported NCCN updates has been that ctDNA is clearly prognostic in colorectal cancer, but evidence has been insufficient to recommend routine use outside clinical trials or to direct treatment solely from a positive result. (OncLive)
FIND directly addresses one part of that concern. It demonstrates prospective clinical utility: the test result initiated a defined action, and the action changed a clinically meaningful endpoint. Yet the trial does not establish that a positive ctDNA result should automatically start chemotherapy, and it does not yet establish improved overall survival.
A guideline committee could reasonably begin by acknowledging ctDNA-triggered imaging as an emerging option at experienced centers, while awaiting replication and survival maturation before calling it standard surveillance. This would be a substantial step forward without pretending that every outstanding question has been answered.
Replicate in CRC, or move rapidly into other cancers?
The answer should be both—but not equally.
CRC deserves a strong independent replication because FIND has identified a particularly coherent use case. Colorectal metastases frequently recur in the liver or lungs, low-volume disease can sometimes be removed or ablated, and CT can localize what the blood test detects only indirectly. A confirmatory study should cross geography and technology: ideally a North American or European trial using another methylation platform and at least one widely used tumor-informed assay, with blinded central assessment of resectability, mature survival, resource use, and patient-reported outcomes.
At the same time, the field should not spend the next five years repeating essentially identical CRC studies before testing the broader principle. Parallel studies should move into selected cancers—such as resected lung or urothelial cancer—where there is a credible intervention that can be initiated when molecular recurrence is found and where low-volume disease may still be controllable.
The key selection criterion is not merely that ctDNA can predict relapse. Almost any sufficiently good MRD assay may do that. The cancer must offer an actionable window between molecular detection and incurable clinical recurrence. Tumors in which earlier knowledge produces only earlier anxiety or earlier administration of the same palliative therapy are much weaker candidates.
Will MRD become a must-have?
Probably—but the standard of care will ultimately be the MRD-directed clinical algorithm, not the laboratory result in isolation.
MRD is likely to become “must-measure” before it becomes universally “must-treat.” In some cancers it may determine adjuvant therapy; in others it may trigger imaging; elsewhere it may identify patients for a randomized intervention. The meaning of a positive result will remain tumor-specific, stage-specific, assay-specific, and time-specific.
FIND advances the field because it assigns ctDNA a concrete job. The assay does not simply tell the patient, ominously, that relapse is likely. It tells the care team when to look, defines how intensively to continue looking, and helps uncover metastases while their anatomy remains favorable for attempted cure.
That is much closer to standard-of-care medicine than another Kaplan-Meier curve showing that ctDNA-positive patients do badly.
The remaining test is the hardest one: whether converting more recurrences into curative-intent procedures ultimately converts more patients into long-term survivors.
Sidebar: Seven Particularly Forceful Features of FIND
1. The endpoint was action, not merely prediction.
The trial was powered around receipt of curative-intent metastasis-directed therapy—not assay accuracy, molecular lead time, or a prognostic hazard ratio.
2. Earlier detection changed treatment from mostly palliative to far more often potentially curative.
Among patients whose cancer returned, nearly one in two in the ctDNA-guided group received surgery, ablation, or another metastasis-directed treatment intended to eradicate all visible disease, compared with only about one in four under standard CT surveillance (48.1% versus 23.6%). This does not prove that twice as many patients were cured—but it doubled their opportunity for attempted cure.
3. The recurrence rate did not change at all.
It was 18.0% versus 18.6%. The intervention (taking a blood sample for a ctDNA test) did not prevent relapse; it changed the stage and anatomy at which relapse became visible.
4. Fewer than four months made a large clinical difference.
The median lead was only 3.9 months. Yet liver metastases were markedly fewer, smaller, and more commonly confined to one lobe, a likely clinical benefit. This could be a useful measure in the future; not just early detection of relapse, but status at the time of detection. This will require moving the CT, as in FIND.
5. ctDNA and CT were complementary rather than competitive.
The blood test supplied timing and risk; CT supplied location, extent, and resectability.
6. The algorithm contained both an accelerator and a brake.
Positive ctDNA accelerated imaging. Two consecutive negative results allowed return to ordinary surveillance, avoiding permanent escalation after a transient signal. [*]
7. The positive result is clinically important but still provisional.
Overall survival, health economics, quality of life, psychological effects, and cross-assay reproducibility remain unresolved. FIND is arguably the strongest bridge yet from MRD prognostication to MRD action—but it has not finished crossing the bridge.
##
[*] AI explains a little further re #6. "Positive ctDNA + negative CT = molecularly suspicious, radiographically occult disease. Continue CT every two months. If CT eventually localizes the recurrence, treat it. Should they occur, however, two subsequent negative ctDNA tests provided an off-ramp from intensified imaging, on the assumption that the original positive result was probably false or not reproducible."
