After successful surgery, up to half of patients with early-stage lung cancer still experience relapse. A new study shows that ultrasensitive measurement of circulating tumour DNA in blood can identify which patients are already at risk – potentially changing who receives additional treatment and who may safely avoid it.
For many people diagnosed with lung cancer, surgery is meant to be the moment of relief – the point at which the tumour is gone and the worst is over. But for many patients, that relief never truly arrives. Even after apparently successful surgery, relapse remains a real and invisible threat – turning recovery into a prolonged period of uncertainty: a waiting game in which neither patients nor clinicians can see what is coming.
“For patients, that uncertainty is often the hardest part,” says James Black, a researcher at the Johns Hopkins Bloomberg School of Public Health, Baltimore, United States. “You are told the tumour is gone – but there is no reliable way to know what happens next.”
Now, a large international study shows that microscopic remnants of cancer can be detected in the blood at unprecedented sensitivity – down to just a few parts per million, equivalent to a handful of tumour DNA fragments hidden among millions of normal ones. For some patients, this enables the risk of relapse to be identified months – and sometimes years – before scans or symptoms provide any warning.
“Current approaches are basically flying blind,” says Black, first author of the study, describing a limitation that defines much of post-surgical lung cancer care. “With CT scanning, you are seeing it once it is macroscopically visible – once it is big enough.”
A molecular signal where scans fall short
The findings are based on nearly 3,000 blood samples from more than 400 patients with non–small cell lung cancer, analysed as part of the long-running TRACERx programme.
Rather than asking only whether tumour DNA can be detected at all, the researchers show that how much is present acts as a molecular measure of residual disease, fundamentally altering how a patient’s risk is assessed.
“Patients with very low ctDNA [circulating tumour DNA] levels are not ‘negative’ in the classical sense,” says Black. “They constitute a distinct intermediate-risk group – and that is precisely the group to which we were previously blind.”
The implication is a more precise treatment pathway: some patients may be spared unnecessary chemotherapy and its long-term side-effects, while others can be identified early – at a point at which intervention may still prevent relapse rather than merely respond to it.
“In the end,” says Black, “it is about matching treatment and follow-up to the true biological risk – not just stage and statistics.”
Why relapse risk remains invisible after surgery
Relapse after surgery remains one of the greatest unresolved challenges in treating people with early-stage non–small cell lung cancer. Even when tumours are removed with clear resection margins, many patients later develop recurrence – often without any early clinical warning signs. As a result, decisions about adjuvant treatment and follow-up are still largely based on disease stage, histology and a limited set of risk factors that offer only rough guidance.
“For decades, we have accepted that two patients with the same stage can have very different outcomes,” says James Black. “The problem is that we still lack a way to measure the true biological risk after surgery.”
In practice, clinicians are still forced to rely on indirect proxies rather than a direct readout of whether disease is actually still present. Existing biomarkers offer limited help: they are often neither sensitive nor specific – even when cancer is already known to exist.
Liquid biopsies, which measure ctDNA in the blood, have long been proposed as a way to close that gap. Cancer cells that remain in the body shed small fragments of DNA into the bloodstream – leaving a molecular trace that scans cannot see. In practice, however, translating that idea into a reliable clinical tool has proved difficult, particularly in early-stage disease.
“Early-stage lung cancer is where liquid biopsy is hardest,” says Black. “ctDNA levels are extremely low – which means that the absence of a signal is not the same as the absence of disease.”
Because of these technical limits, earlier studies largely reduced ctDNA to a yes-or-no readout: either cancer DNA was detected or it was not.
“There has been a tendency to treat ctDNA as either present or absent,” Black says. “But we wanted to know whether that simplification was hiding something biologically important. If you measure too early, you may capture DNA from the tumour that has just been removed,” says Black. “If you measure too late, you may miss the opportunity to act on the signal.”
A dataset built to capture disease in motion
The TRACERx programme offered a rare opportunity to address these challenges head-on, following patients systematically from diagnosis through surgery, adjuvant treatment and long-term follow-up with repeated tissue and blood sampling – enabling researchers not only to ask whether ctDNA was present but to watch disease biology unfold over time.
“This level of longitudinal detail is rare,” says Black.
Importantly, patients in this cohort did not receive neoadjuvant treatment. The data therefore reflect a surgical-first pathway with adjuvant chemotherapy as standard of care, enabling ctDNA dynamics to be examined without the confounding effects of preoperative systemic therapy.
At the heart of the study was a simple but unresolved question: what happens at the very lowest levels of ctDNA – a range the researchers suspected was hiding clinically important information with direct consequences for patient stratification and treatment planning.
“We strongly suspected that we were missing something important at the lower end of the scale,” says Black.
Pushing liquid biopsies to the edge of detectability
To detect the faintest traces of cancer in the blood, the researchers inverted the conventional logic of liquid biopsies. Instead of broadly screening for generic cancer signals, they pursued a fully personalised, tumour-informed strategy in which each patient’s own tumour serves as the reference blueprint – pushing ctDNA detection close to its theoretical limit.
The study draws on patients from the longitudinal TRACERx programme. Tissue samples from each original tumour were subjected to whole-genome sequencing, generating a detailed map of clonal and subclonal mutations. From this, the researchers typically selected around 1,800 unique mutations per patient – creating a highly specific genetic fingerprint that could be tracked in that patient’s blood.
“On the tumour-informed side, you can take a tumour out, sequence it; you know what is in there, and you know what you are looking for,” Black explains. “The idea of looking for thousands of mutations and searching for those in the blood in a completely optimised way has never really been done before,” he says – and it is precisely this scale that makes the faintest signals visible.
Tracking thousands rather than hundreds of mutations sharply increases the probability of capturing rare tumour DNA fragments.
“If you track more mutations – more personalised mutations within each tumour – your signal just goes up and up,” says Black. “And in lung cancer in particular, there are lots of mutations, so there is a huge opportunity to exploit that.”
Tracking thousands of tumour mutations – one patient at a time
These personalised mutations were tracked longitudinally in blood samples collected before surgery, after surgery, during adjuvant treatment and when relapse was suspected. In total, the researchers analysed 2,994 plasma samples from 431 patients – making this the largest study to date of ultrasensitive ctDNA monitoring in early-stage non–small cell lung cancer.
The key advance was analytical sensitivity: the limit of detection was about 3 parts per million – meaning just a few tumour DNA fragments among millions of normal ones. In the blood, tumour DNA is an extreme minority.
“There has been a lot of debate about whether sensitivity at that level really matters,” he adds. “But what our study shows is that in the window between around 80 parts per million and 3, there are many clinically relevant relapses you would otherwise miss.”
Timing proved just as critical as sensitivity: blood samples collected too soon after surgery can contain DNA released by the removed tumour, producing misleading signals that do not reflect persistent disease.
“We definitely saw false positives two days afterwards,” says Black. “So we started the window from around 10 days after surgery, to be conservative.”
On this basis, the researchers defined a postoperative landmark window from 10 to 120 days after surgery – late enough to avoid DNA released by surgery itself but early enough to detect residual disease while intervention may still matter.
“The method is not just about sensitivity,” Black emphasises. “It is about biological timing. If you measure at the wrong time, even the best test ends up asking the wrong question.”
Why relapse risk is not a yes-or-no question
Even before surgery, ctDNA could be detected in the vast majority of patients – but crucially, many of these signals lay far below the detection limits of standard assays.
“The key point was that we did not just ask whether ctDNA was there or not,” says James Black. “We asked how much was there – and that completely changed what we could see.”
After surgery, patients were followed within a predefined postoperative landmark window. Within this window, ctDNA status emerged as a powerful predictor of outcome. In the evaluable cohort, ctDNA was detected in 29% of patients and predicted relapse with 93% specificity and 62% sensitivity.
Patients with no detectable ctDNA had a markedly lower risk of relapse, whereas those with high ctDNA levels faced a very high risk of recurrence and death. The most revealing insight, however, was between these two extremes.
“We identified a group of patients with low but detectable ctDNA,” says Black. “They are not negative – but they are also not the group that is destined to do badly.”
The patients in between – and why they matter most
This intermediate-risk group showed outcomes clearly between the other two groups – both in relapse-free survival and overall survival – even after adjustment for classical clinical variables such as stage, age, sex and smoking history.
Risk did not behave as a threshold phenomenon but as a continuous gradient – clinically more like a dimmer switch than an on–off button.
“With every 10-fold rise in ctDNA, risk increases,” Black explains, revealing a clear dose–response relationship with both relapse and death.
By integrating ctDNA measurements before and after surgery, the researchers identified three biologically meaningful trajectories: patients who remained ctDNA-negative throughout; patients who were ctDNA-positive before surgery but cleared ctDNA afterwards; and patients with persistent ctDNA despite tumour resection.
“The middle group – the ones who clear ctDNA after surgery – are particularly interesting,” says Black. “They do not do as well as the completely negative group, but they do far better than patients with a persistent signal.”
When blood becomes a real-time readout of treatment response
ctDNA dynamics during adjuvant chemotherapy added a further layer of insight. Patients with detectable ctDNA after surgery who cleared ctDNA during chemotherapy had a markedly better prognosis than those for whom the signal persisted.
“That clearance is telling you something very real,” says Black. “It is almost like a biological receipt – confirming that the treatment is reaching disease that would otherwise remain invisible.”
In an exploratory analysis, ctDNA clearance after adjuvant therapy was seen only in patients who completed standard chemotherapy, whereas those receiving less intensive treatment neither cleared ctDNA nor avoided relapse.
Long-term follow-up revealed that ctDNA often rose well before relapse became clinically apparent. Across the cohort, the lead time between ctDNA detection and documented relapse ranged from 0 to 1,732 days, with a median of 158 days.
“We are watching relapse develop in the blood long before the scan changes or symptoms appear,” says Black. “This opens a window that does not exist with conventional imaging – a chance to act while the disease is still molecular.”
Finally, ctDNA kinetics carried information not only about whether disease returned but how it returned. Slowly rising ctDNA levels were more often associated with local or intrathoracic recurrences, whereas rapidly increasing signals pointed towards aggressive, systemic relapse.
“ctDNA is not just telling us whether the cancer comes back,” Black concludes. “It is telling us something about the nature of that return.”
From population averages to individual biological risk
Rather than relying on uniform follow-up schedules and broad recommendations for adjuvant treatment, ctDNA enables a more individualised strategy – one guided by biological risk rather than population averages.
“Stage has been a blunt instrument,” says Black. “Biology gives you a scalpel. For a long time, we have treated patients as if their risk were the same simply because they share a stage. What this allows us to do is differentiate far more precisely.”
This precision opens the door to both de-escalation and escalation of care. At the low-risk end, patients who remain ctDNA-negative after surgery have a very low probability of relapse and could potentially avoid unnecessary treatment and intensive surveillance.
“If we can identify low-risk patients with high confidence, we have an obligation not to overtreat them,” says Black. “That is both an ethical and a clinical issue.”
At the opposite end of the spectrum lies a new opportunity: acting earlier in patients with persistent or rising ctDNA – even when absolute levels are extremely low.
“What is exciting is that you can start responding to the molecular behaviour of the disease before it manifests clinically,” says Black. “That gives you a time window you simply do not have otherwise.”
Rethinking post-surgical treatment as a dynamic process
Changes in ctDNA over time – particularly during adjuvant therapy – provide clinically actionable information.
“Clearance of ctDNA during adjuvant therapy is a favourable prognostic sign compared with persistence,” Black notes. “This suggests it could be valuable for dynamic monitoring.”
This dynamic perspective challenges the traditional view of post-surgical treatment as a fixed protocol and points toward a future in which therapy intensity and duration could be adapted in real time based on molecular response – a promise that will need to be tested in prospective trials.
“The next step is to test whether acting on ctDNA signals actually changes outcomes,” says Black. “That means new trial designs in which molecular data are integrated directly into treatment decisions.”
“I do not think you necessarily need a phase-three trial to start using this,” Black says. “You could start treating patients on the basis of the evidence we already have – it just needs someone with the gumption to do it.”
Limits, logistics – and the coming immunotherapy era
The study also highlights the biological limits of blood-based monitoring. Some relapses shed very little ctDNA into the circulation – particularly intrathoracic-only or brain-only recurrences – underscoring that even ultrasensitive assays are not a perfect crystal ball.
Implementation will therefore depend on infrastructure and clinical integration. Tumour-informed ctDNA testing requires access to tumour tissue, sequencing capacity and standardised workflows.
“The technology is ready,” says Black. “The challenge is getting it into everyday clinical use in a way that is robust, timely and scalable.”
Finally, the researchers note that the cohort analysed in this study predates the widespread use of neoadjuvant and adjuvant immunotherapy. This raises new questions about how ctDNA dynamics interact with modern treatment paradigms – and whether molecular monitoring could help to guide treatment intensity and duration in an immunotherapy era.
“We see ctDNA as a shared language between biology and the clinic,” says Black. “That language is only just beginning to be spoken – but it is already changing what it means to watch, wait and act after surgery.”
