Part IX · Early-stage disease: in situ, locally advanced, residual, and the neoadjuvant platform · Chapter 53
Molecular residual disease
A test that names the patients who will relapse, years before they do, and no treatment yet shown to change what follows.
1 · What a positive assay names, and in which compartment
This chapter takes the assay as given. Liquid biopsy and longitudinal monitoring establishes what circulating tumour DNA detects, how well, and how far ahead of clinical relapse. Dormancy, residual disease, and late recurrence establishes what the detected cells are doing in the years before they declare themselves. The question here is the one a clinic is left with. A patient has finished curative-intent treatment and a scan shows nothing. A test says disease is present.
What that result names is a patient rather than a lesion. There is nothing to measure, nothing to resect and nothing to irradiate. Every endpoint oncology normally reaches for is unavailable, and the only outcome that can be observed is whether relapse arrives.
Two compartments report the state, and they are not the same test.
Plasma. Circulating tumour DNA is the route almost every trial uses. It is easily sampled and can be repeated. Its prognostic strength is established across subtypes. Liquid biopsy and longitudinal monitoring carries that record.
Bone marrow. Disseminated tumour cells are the older observation and the more direct one. A marrow aspirate finds the cells themselves rather than fragments of their DNA. CLEVER is the trial that treated on that basis, and it is the only randomised intervention in this chapter that acted on a marrow result.
No trial in this registry randomises one compartment against the other. The two are usually written about as if they were one assay with two sampling routes, and that is an assumption rather than a finding. A drug that clears plasma has not been shown to clear marrow, and the reverse has not been shown either.
Interplay
Residual disease is the clinical face of dormancy. Dormancy, residual disease, and late recurrence argues that the dormant state is actively maintained, that escape has identifiable triggers, and that early and late recurrence are different biological problems. This chapter inherits that argument and one uncomfortable consequence of it. A test that detects shedding detects cells that have already left dormancy. The population most worth intercepting may be the one the test cannot see.
Finding disease that no scan can see, and having nothing to give for it
How reliably does a blood assay identify the patients who will relapse?
Rationale
Recurrence after curative treatment arrives from disease that was present at surgery and invisible to every imaging modality. A plasma assay that detects it would convert an unobservable state into an observable one.
Experiment
Prospective cohorts followed with serial plasma sampling after curative-intent treatment, using personalised assays built from the resected tumour. Separately, a surveillance programme that sampled at fixed intervals and recorded what it missed as well as what it found.
Finding
Detection is among the strongest prognostic separations reported for any biomarker in this disease, and it arrives well before clinical relapse. The warning interval differs by subtype.
Limitation
A cohort assembled around a test measures the test. None of these designs randomised anything, so they establish that detection predicts relapse and nothing about whether detecting it helps. Patients who relapsed with no prior detection at all are the part of the record most often left out of the summary.
Next question
If the test identifies the right patients, is there anything to give them?
ALIENORc-TRAK TNSURVIVE
Is plasma the only compartment worth sampling?
Rationale
Disseminated tumour cells in marrow are the older observation and the more direct one. A marrow aspirate finds the cells themselves rather than fragments of the DNA they shed, and shedding requires turnover that a dormant cell may not have.
Experiment
Bone marrow aspiration to find disseminated tumour cells directly, in survivors within five years of diagnosis, followed by randomised treatment allocation among those found positive.
Finding
Marrow disease can be found, treated and reduced, and the trial that did so was feasible and tolerable. It was not designed to show a recurrence benefit and it did not claim one.
Limitation
No trial randomises one compartment against the other, so the field treats marrow and plasma as two routes to one state on an assumption rather than a finding. Marrow aspiration is also a procedure rather than a blood draw, which limits how often it can be repeated.
Next question
If the two compartments identify different patients, which of them names the population an intervention should be tested in?
CLEVERALIENOR
2 · Why the surveillance question is not the advanced-disease question
Liquid biopsy and longitudinal monitoring states the distinction and this section develops the part that matters at the prescribing pad. In advanced disease a molecular signal triggers a switch to a drug with a known effect on measurable disease. After curative treatment there is no such drug, because no agent has been shown to eradicate disease detected this way.
So every trial here has had to choose a class on an argument rather than on a precedent. Three arguments have been made, and each is worth naming, because each predicts a different kind of failure.
Checkpoint blockade, on the argument that residual disease is small and immunologically visible.c-TRAK TN tested it in triple-negative disease. Liquid biopsy and longitudinal monitoring carries what came back and what three separate things went wrong.
Dormancy-directed treatment, on the argument that the target is the quiescent cell rather than the dividing one.CLEVER combined autophagy inhibition and mTOR inhibition against marrow disease. Its endpoints were feasibility and safety rather than recurrence, which is the honest design for a first test and also the reason it settles nothing about outcome.
Endocrine escalation, on the argument that the disease is still endocrine dependent and the first course was simply not enough.LEADER is testing it in hormone receptor-positive disease.
The three arguments are not compatible with each other. Checkpoint blockade assumes the cells are proliferating enough to be seen. Dormancy-directed treatment assumes they are not proliferating at all. A field that runs all three at once is a field that does not yet know what it is treating, and saying so is more useful than picking a favourite.
Caution
A trial in this setting can fail for a reason that has nothing to do with its drug. If surveillance begins after adjuvant treatment has finished, the patients it finds are the ones in whom relapse was already underway. If sampling is three-monthly in a subtype whose warning is measured in weeks, the same thing happens. Both faults produce a negative trial of an agent that was never given a population it could help. Liquid biopsy and longitudinal monitoring shows what that looks like in practice.
Finding disease that no scan can see, and having nothing to give for it, continued
Does treating molecularly detected disease change what happens next?
Rationale
The advanced setting has shown twice that acting on a molecular signal before radiological progression can buy time. The same logic applied after curative treatment would intercept recurrence rather than delay it.
Experiment
Three different drug classes offered to patients who became positive on surveillance. Checkpoint blockade in triple-negative disease, autophagy and mTOR inhibition against marrow disease, and endocrine escalation in hormone receptor-positive disease.
Finding
Nothing has yet shown that treatment changes the outcome. The one trial that reported a therapeutic result found that most patients already had metastases on imaging when the assay turned positive, and that the drug cleared the assay in nobody.
Limitation
Each of these trials chose its class on an argument rather than on a precedent, and the arguments are not compatible with each other. Checkpoint blockade assumes the cells are proliferating enough to be seen and dormancy-directed treatment assumes they are not proliferating at all. A negative trial therefore cannot be read as a statement about the setting.
Next question
Were these trials testing the wrong drug, or surveilling at the wrong time?
c-TRAK TNCLEVERLEADERDARE
3 · What the subtype changes about what surveillance could buy
Lead time is not a property of the assay alone. It is a property of how fast the disease moves between the first detectable shedding and the first visible deposit, and that differs by subtype. ALIENOR measured it in one cohort and found the gap longest in hormone receptor-positive HER2-negative disease, intermediate in HER2-positive disease, and shortest in triple-negative disease. Liquid biopsy and longitudinal monitoring carries the figures.
Reading that against the three subtype parts gives three different problems.
Hormone receptor-positive, HER2-negative disease has the most room and the least urgency. The recurrence hazard runs for two decades, which Dormancy, residual disease, and late recurrence establishes and Biology and the treatment spine takes as the premise of its part. A long lead time and a long hazard make this the subtype where surveillance has the most time to act in. It is also the subtype where an effective intervention already exists in the form of more endocrine therapy, and where Adjuvant endocrine therapy shows how little the extra years buy.
Triple-negative disease has the most to gain and the least warning. The recurrence risk is concentrated in the first few years, which Biology and the treatment spine makes the organising fact of its part. That is precisely the disease a surveillance programme would most want to intercept. It is also the disease in which the interval between detection and relapse is shortest, so a quarterly sampling schedule may miss the window entirely.
HER2-positive disease sits between them and is the least studied. No trial in this registry surveils it specifically. The subtype has more active agents than either of the others, which makes the absence of a trial harder to explain than the absence of a result.
The summary is uncomfortable. The subtype with the most usable lead time has the least to gain. The subtype with the most to gain has the least usable lead time.
4 · Reading an interventional trial in this setting before it reports
DARE, SURVIVE and the circulating tumour DNA-positive cohort of LEADER have not reported. Neither have TRAK-ER and PERSEVERE, which are named in Liquid biopsy and longitudinal monitoring and are not in this registry. A reader will meet the first of these results before this chapter is next revised, so the useful thing to record now is what to ask of one.
What was randomised. A trial that randomises treatment among test-positive patients tests the treatment. A trial that randomises the test itself tests the surveillance programme. These answer different questions and only the second one tells a clinic whether to send the assay.
When surveillance started, and how often it sampled. Both are design parameters rather than incidental detail, and Why the surveillance question is not the advanced-disease question gives the reason. A schedule has to be read against the lead time of the subtype enrolled.
What the control group received. Observation is not the only honest comparator. In hormone receptor-positive disease the alternative to escalation is continued endocrine therapy, and a trial that compares escalation against nothing will overstate what escalation adds.
Which endpoint carries the claim. Clearance of the assay is the endpoint most likely to be reported first and the one that establishes least. It shows that the drug reaches the cells it can reach. It does not show that the patient lives longer, and SWOG S0500 is the demonstration in metastatic disease that a strongly prognostic blood marker can support no useful action at all.
Whether the population was defined before or after the fact. A cohort assembled from patients who turned out to be positive is not a programme. A programme enrols everyone and follows them. The first design reports on the test. Only the second reports on the strategy.
Finding disease that no scan can see, and having nothing to give for it, continued
Does running a surveillance programme help the population it is run on?
Rationale
A test is only worth sending if the strategy built around it improves outcomes, and lead time by itself moves the date of diagnosis without moving the date of death.
Experiment
Randomisation of the surveillance strategy itself rather than of the treatment given to those it finds, so that the comparison is between a followed population and a conventionally followed one.
Finding
No result. The designs that ask this question in its randomised form have not reported.
Limitation
This is the question a clinic needs answered and the hardest one to fund, because it requires enrolling everyone to study the few who become positive. A trial that randomises treatment among test-positive patients is a cheaper design that answers a different question.
Next question
If a programme does help, does the benefit come from the earlier treatment or from the earlier imaging that followed the positive result?
SURVIVEDARELEADER
5 · What to do before the answer arrives
In practice
Molecular residual disease testing is not recommended outside a trial, and Liquid biopsy and longitudinal monitoring carries the guideline position that says so. That recommendation is not a hedge. A positive result outside a trial creates a patient who knows that relapse is coming, has nothing to treat, and has been offered no option that changes the outcome.
Three things follow for a clinic.
Ask what will be done with the result before the sample is sent. If the answer is closer imaging or earlier scans, the honest statement is that neither has been shown to help, and that Liquid biopsy and longitudinal monitoring explains why lead time alone is not benefit.
Treat a result already in hand as prognostic rather than predictive. It changes what the clinician expects. It does not yet change what the clinician gives, and the distinction is the whole content of this chapter.
References
All 3 references
Blaye C, Bidard FC, Hue L, et al. Tracking structural variants in early breast cancer using a high-sensitivity ctDNA assay predicts relapse in the post-neoadjuvant setting: the multicenter ALIENOR trial. ESMO Open 2026 11:107766. PMID 42242004
DeMichele A, Clark AS, Shea E, et al. Targeting dormant tumor cells to prevent recurrent breast cancer: a randomized phase 2 trial. Nat Med 2025 31:3464-3474. PMID 40897974
Turner NC, Swift C, Jenkins B, et al. Results of the c-TRAK TN trial: ctDNA mutation tracking to detect molecular residual disease and trigger intervention in moderate- and high-risk early-stage triple-negative breast cancer. Ann Oncol 2023 34:200-211. PMID 36423745