Liquid Biopsy Explained: What Blood Can Reveal About Cancer

A liquid biopsy looks for cancer in a tube of blood instead of a sliver of tissue. The tube can carry fragments of tumor DNA, rare tumor cells, and tiny membrane packages that cells release, and each of those can be read in the laboratory. The approach already helps choose drugs for some lung, breast, prostate, and colorectal cancers, and one blood test is now FDA-approved for colorectal cancer screening. It also has limits that decide whether a given result means anything at all.

Medical information disclaimer: This article explains biomedical research and is not medical advice. It cannot diagnose cancer, recommend screening, select a treatment, or interpret a personal laboratory result. For urgent symptoms or an emergency, contact local emergency services rather than relying on an article or an AI system.

What a liquid biopsy actually measures

The National Cancer Institute (NCI) describes liquid biopsy tests as tests that look at tumor markers in blood and other body fluids, built on the observation that tumors can shed cells or bits of biological material into the bloodstream (NCI tumor markers fact sheet). So "liquid biopsy" is a family name, much like "imaging." A CT scan and an ultrasound both make pictures, yet nobody asks whether "imaging" is accurate without saying which kind. The same discipline applies here, because a test that genotypes a known lung tumor has almost nothing in common, statistically, with a test that screens healthy adults.

A useful way to picture the difference with a conventional biopsy: a tissue biopsy is a core sample drilled out of a building, while a liquid biopsy is a chemical analysis of the dust drifting out of its windows. Dust from the building tells you something real about what is inside, and you can collect it again next week without drilling. It cannot show you the floor plan, and if the building barely leaks dust, you learn very little.

The main analytes, meaning the things being measured, are summarized below.

Analyte	What it is	What it can show	Main weakness
Circulating tumor DNA (ctDNA)	Fragments of DNA released from tumor cells, mixed into all other cell-free DNA	Mutations, copy-number changes, methylation, fragment-size patterns	Often a tiny share of total DNA; blood-cell clones add noise
Circulating tumor cells (CTCs)	Intact cells that left a tumor and entered blood	Counts, protein markers, whole-cell genetics	Extremely rare; capture methods favor some cell types
Extracellular vesicles	Membrane-bound packages carrying RNA, protein, lipid, or DNA	Possible markers of tumor origin and state	Isolation methods vary; mostly research use
Cell-free RNA, proteins, metabolites	Molecules released by tumors and by normal tissue	Expression signatures, tumor markers	Inflammation and normal tissue produce overlapping signals
Methylation patterns	Chemical tags on DNA that mark gene activity	Tissue of origin, detection signals	Shifts with age and blood-cell composition

Three questions sort almost any claim about a liquid biopsy. What exactly is measured? For which decision? And validated against what reference? A test that answers the first question well but has no answer to the second is a research tool, however impressive its sequencing depth.

Tissue biopsy versus liquid biopsy: where each one wins

Tissue preserves structure. A pathologist looking at a tissue sample sees cell shapes, how the tumor invades its surroundings, which immune cells have moved in, and where in the sample a biomarker sits. None of that survives the trip into the bloodstream. Tissue can also show differences between regions of a tumor, although one needle sample may miss the region that matters.

Blood offers breadth and repeatability. Material from several tumor sites can pool in the same tube, so one draw may reflect more of the disease than one needle path. A second or third draw is far easier than a second or third procedure, which makes blood attractive for watching change over time, and for patients whose tumors are hard to reach.

Question	Tissue biopsy	Liquid biopsy
Does it show tumor architecture?	Yes	No
Can it be repeated easily?	Often not	Usually yes
Does it sample multiple tumor sites at once?	Only the sampled site	Possibly, if each site sheds
Does it work when a tumor sheds little?	Yes, if tissue is obtained	Can miss the tumor
Is it the standard for confirming a first diagnosis?	Yes in most settings	Not in most settings

The two methods are better thought of as complementary than as rivals. Because plasma assays can miss tumors that shed little, tissue remains the fallback when a blood result is negative but suspicion stays high.

A short history: from 1948 to the first FDA-approved blood screening test

The idea is older than most people assume. The timeline below uses only dated, published milestones.

Year	Milestone	Source
1948	Mandel and Métais report nucleic acids in human blood plasma	C R Seances Soc Biol, PMID 18875018
1997	Lo and colleagues detect male fetal DNA in the plasma of 80% of women carrying male fetuses (43 pregnant women tested); the authors suggested this could support noninvasive prenatal diagnosis	Lancet, PMID 9274585
1999	The same group measures a mean half-life of 16.3 minutes (range 4 to 30) for fetal DNA in maternal plasma	Am J Hum Genet, PMID 9915961
2004	Circulating tumor cell counts predict survival in metastatic breast cancer	NEJM, PMID 15317891
2014	Digital PCR detects ctDNA across 640 patients with many cancer types	Sci Transl Med, PMID 24553385
2016	Plasma use of the cobas EGFR Mutation Test v2 appears on FDA's companion diagnostic list (June 1, 2016)	FDA list
2020	Guardant360 CDx (August 7) and FoundationOne Liquid CDx (August 26) are authorized as plasma-based companion diagnostics	FDA list
2022	The DYNAMIC trial uses ctDNA to guide chemotherapy in stage II colon cancer	NEJM, PMID 35657320
2024	FDA approves Guardant's Shield blood test for colorectal cancer screening (decision date July 26, 2024) after the ECLIPSE study	FDA PMA database

Notice the 1997 paper's logic. Lo's team asked whether fetal DNA circulates in maternal blood the way tumor DNA circulates in patients with cancer, so pregnancy and oncology were treated as two versions of the same problem from the start. Sequencing itself has also become far cheaper since then, a shift covered in the human genome revolution.

How tumor DNA gets into blood, and why the signal is faint

Cells die constantly, and when they do, enzymes chop their DNA into short pieces that spill into the surrounding fluid. Healthy cells go through programmed cell death on a schedule (how apoptosis works), and tumor cells die too, including when treatment kills them. Most cell-free DNA (cfDNA) in a healthy person's plasma comes from blood cells. In a cancer patient, a fraction of the cfDNA comes from the tumor, and that fraction is the circulating tumor DNA.

The fraction can be small. Here is a hypothetical piece of arithmetic to make the problem concrete. If tumor DNA makes up 0.1% of all cfDNA, only one fragment in 1,000 carries a tumor-specific change. A sample containing 5,000 copies of the genome would hold about five mutant copies. An assay that needs to see a variant several times before trusting it is operating on the edge of what the tube can supply. It is like finding a handful of misprinted pages in a library whose books have all been run through a shredder.

Why a blood sample is close to a live snapshot

DNA in the bloodstream does not linger. In the fetal-DNA experiments described above, the mean half-life was 16.3 minutes, with individual values from 4 to 30 minutes (Lo et al., 1999). That measurement was made in pregnancy, not cancer, but it explains why researchers treat plasma DNA as a fast-turnover signal. A sample reflects recent cell death rather than the history of the disease, which is useful for monitoring and unhelpful if the tumor is quiet at the moment of the draw.

Which tumors shed enough

The tumor type matters enormously. In a 2014 study of 640 patients, ctDNA was detectable in more than 75% of patients with advanced pancreatic, ovarian, colorectal, bladder, gastroesophageal, breast, melanoma, hepatocellular, and head and neck cancers, but in fewer than 50% with primary brain, renal, prostate, or thyroid cancers (Bettegowda et al., Sci Transl Med 2014). Detection in early, localized disease was lower: 73% for colorectal, 57% for gastroesophageal, 50% for breast adenocarcinoma, and 48% for pancreatic cancer.

Those figures describe digital PCR on known tumor mutations in a research setting, and newer assays differ. What they establish is the pattern: advanced disease sheds more than early disease, and some tumor types shed far less than others. Why, in some cases, is only partly understood. Tumor size, blood supply, rate of cell turnover, and anatomical barriers are all candidates.

Fragment size and methylation as extra clues

Tumor-derived fragments are not just rarer, they are different. A 2018 analysis of 344 plasma samples from 200 patients found that ctDNA was enriched in fragments of 90 to 150 base pairs, shorter than the bulk of cfDNA. Selecting that size range gave a median enrichment of more than twofold in over 95% of cases (Mouliere et al., Sci Transl Med 2018). Models combining fragment length and copy-number features reached an area under the curve (AUC) above 0.99 for advanced cancer, versus below 0.80 without fragmentation features.

AUC is a score of how well a test separates two groups, where 0.5 is a coin flip and 1.0 is perfect separation. That headline needs two cautions. It was a research cohort of people already known to have cancer, and screening healthy people is a harder problem. Methylation, the pattern of chemical tags on DNA that reflects which genes are switched on, works along similar lines, and readers curious about how methylation is read as a signal can see epigenetic clocks explained.

When healthy blood cells imitate a tumor

A subtle problem lurks in every ctDNA readout. As people age, individual blood stem cells pick up mutations and expand into large families of identical cells, a process called clonal hematopoiesis. These clones are not cancer, but their mutated DNA lands in plasma just like tumor DNA does.

A 2019 study in Nature Medicine sequenced cfDNA alongside matched white blood cells from 124 patients with metastatic cancer and 47 cancer-free controls. It attributed most cfDNA variants to clonal hematopoiesis: 53.2% in the patients and 81.6% in the controls (Razavi et al., Nat Med 2019). The authors called the phenomenon pervasive and concluded that sequencing matched white blood cells is important for interpreting variants correctly.

Think of it as a forged signature. The mutation can look exactly like the cancer's handwriting, but the pen was held by a blood cell. Assays that look only for mutations in blood, without a tumor sample or a white-cell control to compare against, can therefore call a healthy-cell mutation a tumor mutation. This is one of the reasons a variant in a report should be read with the clinical context and not alone.

Four jobs a liquid biopsy can do

Finding a treatable mutation

Many targeted cancer drugs work only against tumors carrying a specific genetic change. A plasma test can look for that change when tissue is scarce or hard to obtain. FDA's list of companion diagnostics (tests tied to the safe and effective use of a particular drug) includes plasma-based entries, such as Guardant360 CDx for EGFR changes in non-small-cell lung cancer with osimertinib (August 7, 2020) and FoundationOne Liquid CDx for several indications across lung, breast, and prostate cancer (FDA companion diagnostics list). Each approval is tied to a named drug, cancer type, and biomarker, which is the opposite of a general-purpose cancer detector.

Looking for leftover cancer after surgery

After a tumor is removed, imaging cannot see a few remaining cells. ctDNA can sometimes hint at them. This is called minimal residual disease (MRD) testing.

In a 2016 study of 230 patients with stage II colon cancer, who were followed with 1,046 plasma samples, ctDNA was detected after surgery in 14 of 178 patients (7.9%) who did not receive adjuvant chemotherapy, and 11 of those 14 relapsed. Among 164 ctDNA-negative patients, 16 (9.8%) still relapsed. The hazard ratio for recurrence with a positive result was 18 (Tie et al., Sci Transl Med 2016). Read both halves of that result. A positive test marked high risk, but a negative test was not a guarantee, since roughly one in ten negative patients relapsed anyway.

The DYNAMIC trial then asked whether acting on the result helps. In 455 patients with stage II colon cancer, randomized 2:1 to ctDNA-guided or standard management, chemotherapy use was lower with ctDNA guidance (15% versus 28%), and two-year recurrence-free survival was noninferior (93.5% versus 92.4%), with median follow-up of 37 months (Tie et al., NEJM 2022). The finding is narrow: one cancer, one stage, one question. It shows a path to less treatment for some patients without worse short-term outcomes, not a blanket rule for other cancers.

Tracking response and resistance

Tumors evolve under drug pressure, much like bacteria that survive an antibiotic and then dominate the culture. A liquid biopsy can catch some of that evolution early. In the 2014 study, 24 patients with colorectal cancer who responded to EGFR-blocking therapy and then relapsed were examined, and 23 of them (96%) had developed mutations in genes of the MAPK signaling pathway (Bettegowda et al.).

In lung cancer, the TRACERx study profiled ctDNA from its first 100 participants and found that blinded profiling of postoperative plasma could suggest adjuvant chemotherapy resistance and identify patients likely to relapse (Abbosh et al., Nature 2017). Whether a resistance mutation should change treatment depends on the drug options available, and that decision belongs to the treating team. Immune-based treatments pose a related monitoring question, explored in immunotherapy and how doctors train the immune system to fight cancer.

Screening people with no diagnosis

Screening is the most ambitious use and the hardest one. The ECLIPSE study evaluated a cfDNA blood test for colorectal cancer in 10,258 participants, 7,861 of whom were evaluable. It reported 83.1% sensitivity for colorectal cancer, 13.2% sensitivity for advanced precancerous lesions, and 89.6% specificity for advanced neoplasia (Chung et al., NEJM 2024). FDA's decision on the Shield test covers adults aged 45 or older at average risk (FDA PMA database).

The gap between 83.1% and 13.2% is the story. The test caught most cancers in the study but few of the precancerous growths that a colonoscopy can find and remove. That is a very different value proposition from a procedure that can prevent cancer.

Multi-cancer detection tests try to find many cancer types from one sample. The NCI notes that several are already being marketed, that a critical open question is whether treating the cancers they identify would reduce deaths from them, and that the institute plans a clinical trial to better understand their implications (NCI). Marketed is not the same as shown to help. A small analysis within the PATHFINDER study reported that imaging was used in about 90% of both true-positive and false-positive cases, and that procedures followed in 38.9% of false positives (Cancer Prev Res 2025), a reminder that a false alarm has physical and emotional costs.

Use	Example evidence	What it supports	What is still open
Choosing a drug	FDA-listed plasma companion diagnostics	Genotyping when the assay is validated for that drug	Negative results may need tissue confirmation
Residual disease	Tie 2016; DYNAMIC 2022	Risk stratification; less chemotherapy in one setting	Benefit in other cancers and stages
Response and resistance	Bettegowda 2014; TRACERx 2017	Early view of tumor evolution	Which changes should alter treatment
Screening	ECLIPSE 2024; FDA decision 2024	Detecting most colorectal cancers in the study	Low sensitivity for precancer; multi-cancer benefit unproven

Why a positive result in healthy people is harder than it looks

Screening runs into arithmetic that no laboratory improvement removes. Sensitivity is the share of people with the disease whom the test flags. Specificity is the share of people without the disease whom the test correctly clears. Positive predictive value (PPV) is the share of positive results that are true, and it depends on how common the disease is.

Take a hypothetical test with 50% sensitivity and 99% specificity, used on 100,000 people of whom 1% have the cancer. That is 1,000 people with cancer and 99,000 without. The test flags 500 of the 1,000. It also wrongly flags 1% of the 99,000, which is 990 people. Of 1,490 positive results, only 500 are real, a PPV of about 34%. Two out of every three alarms are false, even though 99% specificity sounds excellent.

An airport smoke detector that never misses a fire but also trips whenever someone makes toast has the same problem, and the cost grows when real fires are rare. Machine-learning evaluation hits an identical wall, as explained in why 99% accuracy can be useless. Negative predictive value works in the opposite direction: when disease is uncommon, a negative result is reassuring mostly because most people are healthy, and an insensitive assay adds little to that reassurance.

Circulating tumor cells: rare cells with prognostic weight

Circulating tumor cells are intact cells that detach from a tumor and travel in blood. They are scarce, so most methods first enrich for them, for instance by capturing cells that carry certain surface proteins, and then count or analyze them. Because capture depends on markers, a method can miss cells that do not display them.

The best-known evidence comes from metastatic breast cancer. In a 2004 NEJM study, patients with five or more CTCs per 7.5 milliliters of blood (about one and a half teaspoons) before starting a new therapy had a median progression-free survival of 2.7 months and a median overall survival of 10.1 months. Patients with fewer than five had 7.0 months and more than 18 months, respectively (Cristofanilli et al., NEJM 2004). The difference persisted at the first follow-up after treatment began.

CTC counts therefore carry prognostic information, meaning they predict the course of the disease. Whether using the count to change treatment improves outcomes is a different question, and one that a prognostic association cannot answer by itself. Intact cells can offer information DNA fragments cannot: protein expression, cell shape, and sometimes clusters of cells traveling together. That is the reason the field keeps working on them despite the technical difficulty.

Extracellular vesicles and other emerging signals

Cells release membrane-bound vesicles that carry RNA, proteins, lipids, and DNA. Because the cargo reflects the cell that made it, tumor-derived vesicles are of interest as markers, and they may influence neighboring cells. Researchers also study circulating RNA, proteins, and metabolites.

The difficulty is specificity. Normal tissue, infection, and inflammation produce overlapping signals, and different isolation methods collect different particles. For that reason, most vesicle and RNA markers remain in the discovery phase, with validation in defined populations still ahead. Teams working on pattern-finding across many weak markers often turn to machine learning, a topic taken up in machine learning for biomarker discovery.

From tube to report: the laboratory workflow

A typical pipeline runs through blood collection, plasma separation, DNA extraction, library preparation, sequencing or amplification, bioinformatic analysis, and a written report. Each step can add noise.

Before the lab ever touches the sample, tube type, time before processing, centrifugation, storage temperature, and freeze-thaw cycles can all change the result. Cells that break open in the tube release their own DNA into the plasma and dilute the tumor signal. Collection protocols are therefore part of assay validation, not a footnote.

Sequencing errors are a major challenge, because the signal being sought can be rarer than the error rate of the sequencer. A common countermeasure is to tag each original DNA molecule with a short unique barcode before amplification. Reads that share a barcode can then be compared and a consensus built, much as a stack of numbered carbon copies lets you spot which smudge came from one copy and which was on the original. A variant seen in only one copy of the molecule is likely an error, while one present across copies is likely real.

The bioinformatics pipeline then decides what counts as a variant: thresholds, filters for known artifacts, and rules for blood-cell clones. Versioning and reproducibility matter, because changing a filter can change a report.

Matching a claim to its evidence

Three questions separate a good assay from a useful one. Analytical validity asks whether the test measures what it claims, accurately and repeatably. Clinical validity asks whether the result is associated with the disease or outcome. Clinical utility asks whether using the result improves decisions or health outcomes. A test can pass the first two and fail the third.

Intended use	Main signal	Key question	Evidence needed
Research discovery	DNA, RNA, protein, cell, or vesicle	Is the signal biologically meaningful?	Analytical and biological validation
Molecular profiling	Mutation or biomarker	Does it guide a defined clinical pathway?	Assay validation and concordance with tissue
Monitoring	Change over time	Does the change predict a useful outcome?	Longitudinal clinical evidence
Residual disease	Low-level tumor signal	Does acting on the result improve outcomes?	Prospective utility studies, as in DYNAMIC
Early detection	Multi-marker pattern	Do benefits outweigh false-positive harms?	Screening-population trials
Resistance analysis	Emerging alteration	Is it confirmed and actionable?	Orthogonal confirmation and guidance

Regulatory status adds another layer. A research-use assay, a laboratory-developed service, and an FDA-authorized diagnostic are not interchangeable, and authorization is tied to a specific test, intended use, and date. The FDA companion diagnostics list is a useful place to see that specificity, since each row names a device, a biomarker, a cancer type, and a drug. A trial design that randomizes patients, as DYNAMIC did, is what ultimately separates a helpful test from a merely interesting one, a logic laid out in why randomized trials are powerful.

Press releases, conference abstracts, and online test advertisements are not proof of clinical utility.

False negatives and false positives

A false negative happens when disease is present but the test does not flag it. Causes include a tumor that sheds little, a sample too small or degraded, a target not covered by the assay, a threshold set too high, or technical failure. In the Tie study, a negative ctDNA result still left a 9.8% relapse rate, and in screening, ECLIPSE's 13.2% sensitivity for advanced precancerous lesions means many such lesions are missed.

A false positive happens when the test flags a signal that is not tumor. Sources include contamination, sequencing errors, clonal hematopoiesis, benign biology, inflammation, and classifiers that were trained too broadly. The consequences depend on use. In monitoring, an unconfirmed change can cause anxiety and extra scans. In screening, it can lead to imaging or invasive procedures, as the PATHFINDER data illustrate.

Access, cost, and equity

A blood draw sounds universal, yet access depends on much more: where the test is available, who pays, how samples travel to a specialized laboratory, who interprets the report, and what follow-up care exists if a result is positive. Cost and infrastructure are uneven across and within countries.

Performance can also differ across populations. A test validated in one group may behave differently where disease prevalence, genetics, other health conditions, or sample handling differ. Fair evaluation reports who was enrolled, which cancers and stages, who was excluded, and what data were missing, rather than a single accuracy number that can hide subgroup differences.

What people get wrong about liquid biopsy

"It is a blood test for cancer." There is no single test. Each assay targets a particular analyte for a particular purpose, and its performance in one setting says little about another.

"A negative result means the cancer is gone." A negative result means the assay did not detect what it was built to detect, in that sample, at that threshold. The Tie study's 9.8% relapse rate among ctDNA-negative patients is a concrete reminder.

"More genes tested means a better test." Broader panels find more variants, but also more of the variants whose meaning is uncertain, and more chances to pick up blood-cell clones.

"It will replace tissue biopsy." For first diagnosis in most settings, tissue remains the reference. Liquid biopsy complements it when tissue is scarce, when repeat sampling matters, or for monitoring.

"Any ctDNA signal is a tumor." Razavi's data are the counterexample: in cancer-free controls, 81.6% of cfDNA variants were attributed to clonal hematopoiesis.

Where the field is heading

Research is combining ctDNA with methylation, fragment patterns, proteins, imaging, and clinical records, often with machine-learning models on top. Combining signals can lift sensitivity, but it also raises questions about calibration, explainability, and reproducibility, and about whether a score trained in one population holds in another. The related field of imaging-based detection faces similar questions, discussed in AI cancer detection with deep learning.

The next step is not simply more markers. It is evidence that a combined system is reproducible, interpretable enough for its use, safe to act on, and better than current care. Randomized trials that measure outcomes, not only detection, are the way to get that evidence, and the NCI's planned trial of multi-cancer detection tests is one example of the kind of study still needed.

Limitations of this article

This article summarizes published studies and regulatory records as of the dates cited. Test approvals, indications, and trial results change, and the figures above come from specific studies in specific populations, so they should not be applied to an individual. It does not recommend screening, testing, treatment, or clinical-trial participation. Personal results must be interpreted by qualified healthcare professionals using current guidelines.

Frequently asked questions

Q: What is a liquid biopsy? A: It is an analysis of tumor-related material in blood or another body fluid, instead of a direct tissue sample. The material can be circulating tumor DNA, circulating tumor cells, extracellular vesicles, RNA, proteins, or methylation patterns, depending on the assay.

Q: What is ctDNA? A: Circulating tumor DNA is the portion of cell-free DNA that comes from tumor cells. It is usually a small fraction of the total, because most cell-free DNA in plasma comes from normal blood cells.

Q: Can a liquid biopsy replace a tissue biopsy? A: Not in most settings. Tissue shows architecture and is the standard way to confirm a first diagnosis. Liquid biopsy is easier to repeat and can help when tissue is limited, for monitoring, or for genotyping with a validated assay.

Q: Can a negative liquid biopsy rule out cancer? A: No. In one colon cancer study, 9.8% of patients with a negative post-surgery ctDNA result still relapsed. Tumor shedding, stage, assay sensitivity, sample quality, and the targets covered all affect what a negative result means.

Q: Is there an approved blood test for cancer screening? A: One example is Shield, which FDA approved on July 26, 2024 for colorectal cancer screening in average-risk adults aged 45 or older. In the ECLIPSE study it had 83.1% sensitivity for colorectal cancer but 13.2% for advanced precancerous lesions. Multi-cancer detection tests are being studied, and the NCI says it is still unknown whether treating the cancers they find reduces deaths.

Q: Why can a positive screening result be wrong so often? A: When a disease is uncommon, even a very specific test produces many false positives compared with true ones. In the worked example above, 99% specificity and 50% sensitivity at 1% prevalence gave a positive predictive value of about 34%.

Q: What is clonal hematopoiesis and why does it matter here? A: It is an expansion of blood cell lineages carrying acquired mutations, and its DNA enters plasma. In a 2019 study, it accounted for most cfDNA variants, which can lead to false positives unless white blood cells are sequenced alongside the plasma.

The takeaway

A liquid biopsy is a window onto a tumor, and the glass is frosted. For advanced cancers that shed a lot of DNA, the view can be sharp enough to pick a drug, spot resistance, or flag leftover disease after surgery, and for one cancer a randomized trial showed that using it can spare some patients chemotherapy. For early disease, quiet tumors, and screening healthy people, the same physics works against it: faint signals, healthy blood cells that imitate cancer, and base-rate arithmetic that turns accurate tests into frequent false alarms. The right question is never whether liquid biopsy works, but which test, for which decision, shown to help in which patients.

Sources

• National Cancer Institute, Tumor markers fact sheet (reviewed December 7, 2023).
• U.S. FDA, List of cleared or approved companion diagnostic devices.
• U.S. FDA, PMA record for Shield (P230009).
• Bettegowda C, et al. Detection of circulating tumor DNA in early- and late-stage human malignancies. Sci Transl Med 2014. PMID 24553385.
• Tie J, et al. Circulating tumor DNA analysis guiding adjuvant therapy in stage II colon cancer. NEJM 2022. PMID 35657320.
• Tie J, et al. Circulating tumor DNA analysis detects minimal residual disease and predicts recurrence in patients with stage II colon cancer. Sci Transl Med 2016. PMID 27384348.
• Chung DC, et al. A cell-free DNA blood-based test for colorectal cancer screening. NEJM 2024. PMID 38477985.
• Razavi P, et al. High-intensity sequencing reveals the sources of plasma circulating cell-free DNA variants. Nat Med 2019. PMID 31768066.
• Mouliere F, et al. Enhanced detection of circulating tumor DNA by fragment size analysis. Sci Transl Med 2018. PMID 30404863.
• Cristofanilli M, et al. Circulating tumor cells, disease progression, and survival in metastatic breast cancer. NEJM 2004. PMID 15317891.
• Abbosh C, et al. Phylogenetic ctDNA analysis depicts early-stage lung cancer evolution. Nature 2017. PMID 28445469.
• Lo YMD, et al. Presence of fetal DNA in maternal plasma and serum. Lancet 1997. PMID 9274585; Rapid clearance of fetal DNA from maternal plasma. Am J Hum Genet 1999. PMID 9915961.
• Liquid biopsies come of age: towards implementation of circulating tumour DNA. Nat Rev Cancer 2017. PMID 28233803.
• Current and future perspectives of liquid biopsies in genomics-driven oncology. Nat Rev Genet 2019. PMID 30410101.

This article is educational biomedical information and is not medical advice.

---

Read next:

• Epigenetic Clocks Explained: What DNA Methylation Can and Cannot Say About Biological Age
• Machine Learning for Biomarker Discovery: Finding Disease Signals in Biological Noise
• What Causes Cancer? The Science of Uncontrolled Cells
• Adenovirus Explained: The Hardy Virus Behind Many Diseases