Written and medically reviewed by Dr. Daniel Chong. Last reviewed 2026-08-22. 10 min read
What decides whether a screening test is useful
Four quantities describe any screening test, and confusing two of them is the source of most misunderstanding about blood tests for cancer.
Sensitivity is the proportion of people with disease who test positive. Specificity is the proportion of people without disease who test negative. Both are properties of the assay and stay stable across populations.
Positive predictive value is the proportion of positive results that represent true disease. Negative predictive value is the equivalent for negative results. Neither is fixed. Both move with how common the disease is in the group being tested.
The practical consequence runs through everything below. An identical test, unchanged in every respect, produces mostly false positives in a low-prevalence population and mostly true positives in a high-prevalence one. Choosing a test is therefore inseparable from choosing who to test.
Conventional tumour markers
Tumour markers are macromolecules, predominantly glycoproteins, produced either by tumour cells or by host tissue responding to a tumour. They are measured in serum and reported against a reference interval.
| Marker | Principal association | Documented non-malignant elevation |
|---|---|---|
| CEA | Colorectal, some lung and breast | Tobacco smoking, inflammatory bowel disease, cirrhosis, pancreatitis, peptic ulceration |
| CA 19-9 | Pancreatic, biliary tract | Pancreatitis, cholangitis, biliary obstruction, cirrhosis, diabetes |
| CA 125 | Epithelial ovarian | Endometriosis, fibroids, menstruation, pregnancy, pelvic inflammatory disease, cirrhosis, cardiac failure |
| CA 15-3 | Breast | Benign breast disease, hepatic disease, other adenocarcinomas |
| AFP | Hepatocellular, germ cell | Viral hepatitis, cirrhosis, pregnancy |
| PSA | Prostate | Benign prostatic hyperplasia, prostatitis, urinary retention, recent ejaculation |
| EBV DNA | Nasopharyngeal carcinoma | Recent or reactivated Epstein-Barr virus infection |
Their validated uses all occur after diagnosis. Tracking response during treatment, watching for recurrence in someone previously treated, and supporting diagnosis alongside imaging and histopathology. In each case the marker is a trend within a person whose disease status is already known.
Why markers perform poorly as screening tests
Three separate failure modes apply.
Specificity is insufficient. Every marker in routine use is raised by common benign conditions. CA 125 is elevated in a substantial proportion of premenopausal women through ordinary gynaecological causes. CEA is elevated in smokers as a baseline finding.
Sensitivity is low in early disease. Marker elevation broadly tracks tumour bulk, so stage I disease frequently produces values inside the reference range. CA 19-9 carries an additional constraint: roughly 5 to 10 per cent of people have a Lewis-negative phenotype and cannot produce the antigen at all, whatever their disease burden.
Predictive value collapses at low prevalence. Take a marker with 95 per cent specificity and 70 per cent sensitivity, applied to a population with 0.5 per cent annual cancer incidence. The positive predictive value is approximately 6.6 per cent. Around 93 of every 100 positive results would be false.
That last figure is arithmetic rather than a fault in any particular assay, and it is why guideline bodies do not recommend unselected tumour marker panels for screening people without symptoms.
Where markers are used before diagnosis it is always within a defined high-risk group, precisely because enriching prevalence restores predictive value. AFP with six-monthly ultrasound in cirrhosis or chronic hepatitis B. PSA through shared decision-making. Plasma EBV DNA for nasopharyngeal carcinoma in endemic populations, which is relevant across Southeast Asia.
What MCED tests actually measure
Cells release DNA fragments into the circulation as they die. In someone with a malignancy, a small fraction of that circulating cell-free DNA comes from tumour cells.
Tumour-derived DNA carries epigenetic modifications that differ from normal tissue, principally aberrant methylation at CpG sites. Those patterns are both tumour-associated and tissue-specific, because methylation signatures differ systematically between organs.
MCED assays sequence large panels of these regions and classify the patterns using models trained on tens of thousands of clinical samples. Some platforms add protein markers or fragment-length features.
The output has two parts. A binary cancer signal detected or not detected, and where a signal is present, a ranked prediction of the tissue of origin, which directs the diagnostic pathway rather than prompting undirected whole-body imaging.
The rationale is coverage. Population screening programmes address a small number of malignancies: breast, cervix, colorectum, and lung in eligible smokers. A substantial share of cancer deaths arise from tumour types with no recommended screening pathway at all, including pancreatic, hepatic, ovarian and oesophageal, which are predominantly found symptomatically at advanced stage.
What the trial evidence shows
NHS-Galleri is the first and only randomised controlled trial of an MCED test at population scale. 142,250 participants aged 50 to 77 in England, randomised to annual testing or standard care, with primary results presented at ASCO in May 2026.
The primary endpoint was not met. Across three screening rounds the incidence rate ratio for stage III or IV cancer was 1.03, with a 95 per cent confidence interval of 0.92 to 1.14 and a p value of 0.63.
Secondary findings were more encouraging. Three years of screening reduced the number of cancers diagnosed at stage IV, and reduced the number diagnosed through symptomatic or emergency presentation, routes associated with worse outcomes. Just over half of participants with a positive test were subsequently diagnosed with cancer.
Independent commentary compiled by the UK Science Media Centre concluded that these results provide no evidence base for population-scale implementation, noting that rollout would divert finite CT and biopsy capacity away from symptomatic patients. Mortality outcomes are expected in roughly two years.
PATHFINDER 2, in 35,878 adults aged 50 and over, reported that adding MCED to standard screening produced a more than sevenfold increase in cancer detection, with over half of newly detected cancers at early stage, at 99.6 per cent specificity.
Cancerguard reports 68 per cent sensitivity for the six malignancies with lowest five-year survival, at 97.4 per cent specificity.
No MCED test holds FDA approval. MD Anderson, Fred Hutch, Dana-Farber and the American Cancer Society state consistently that these assays are not diagnostic, that positive results require confirmatory investigation, and that they do not substitute for recommended screening.
The two side by side
| Conventional tumour markers | MCED | |
|---|---|---|
| Analyte | Serum glycoproteins and antigens | Cell-free DNA methylation, plus protein or fragment features on some platforms |
| Validated for | Monitoring diagnosed malignancy | Under investigation for screening people without symptoms |
| Coverage per test | One marker, one or two tumour types | Multiple tumour types from one sample |
| Reported specificity | Variable, frequently low in asymptomatic use | 97.4 to 99.6 per cent in published studies |
| Sensitivity | Tracks tumour burden, low in stage I | Moderate overall, lowest in stage I |
| Output | A value against a reference interval | Binary signal plus predicted tissue of origin |
| Randomised mortality evidence | Not established for screening | Not yet available |
| Replaces standard screening | No | No |
Why who is tested matters as much as what is tested
Predictive value is a function of prevalence. That is the scientific basis for restricting MCED to people with elevated baseline risk, and it is worth seeing in numbers.
Applying the same assay, at 99.5 per cent specificity and 50 per cent sensitivity:
| Annual cancer incidence in the group tested | Approximate positive predictive value |
|---|---|
| 0.5 per cent, general population in the sixth decade | 33 per cent |
| 2 per cent, enriched risk | 67 per cent |
| 5 per cent, substantially enriched risk | 84 per cent |
The assay is identical in all three rows. Only the population changes.
Characteristics associated with elevated baseline risk include documented pathogenic germline variants such as BRCA1, BRCA2 and the Lynch syndrome mismatch repair genes; significant family history, particularly first-degree relatives diagnosed young or several affected relatives in one lineage; chronic hepatitis B or C or established cirrhosis; substantial cumulative tobacco exposure; advancing age, which is the single largest non-modifiable determinant; and chronic inflammatory conditions with recognised malignant potential.
What risk enrichment does not fix
Enriching prevalence improves predictive value. It does not change sensitivity, so two limitations persist.
Stage I sensitivity remains the principal constraint. Small tumours shed proportionately less DNA, so detection probability falls. A negative result does not exclude early-stage disease.
The randomised evidence comes from average-risk populations. NHS-Galleri and PATHFINDER 2 enrolled participants by age, not by risk. The Bayesian argument for targeting higher-risk groups is mathematically sound and has not itself been tested in a randomised trial with mortality endpoints. Established surveillance for defined high-risk groups, such as six-monthly ultrasound with AFP in cirrhosis, retains its own evidence base and is not displaced by MCED.
Where this leaves the decision
Conventional tumour markers are validated for monitoring diagnosed disease and perform poorly when applied to unselected asymptomatic people, because imperfect specificity meets low prevalence.
MCED is a different analytical approach with substantially higher specificity, addressing tumour types for which no screening pathway exists. Its principal limitation is sensitivity in stage I, and randomised mortality evidence is not yet available.
Because predictive value scales with prevalence, the scientific case is strongest in people with defined risk-elevating characteristics. Establishing whether you are one of them requires structured assessment of personal history, family history, infection status and exposure history, with a doctor, before any test is chosen.
That assessment is the part worth paying for. The test is downstream of it.
References
- American Cancer Society, Multi-cancer Detection Tests
- ASCO, Galleri Early Detection Test May Shift Timing of Cancer Detection
- Journal of Clinical Oncology, NHS-Galleri primary results (LBA100)
- GRAIL, NHS-Galleri ASCO 2026 factsheet
- UCL, First results from the NHS-Galleri trial announced
- Science Media Centre, expert reaction to NHS-Galleri full results
- MD Anderson, Multi-cancer detection tests
- Fred Hutch, Multi-cancer early detection tests
- Exact Sciences, Cancerguard multi-cancer early detection
- National Cancer Institute, Uncertainty around tests that screen for many cancers
Frequently asked questions
No. Not all malignancies shed detectable cell-free DNA into the circulation, and no assay covers all tumour types. False positive and false negative results occur with all MCED platforms.
Tumour markers quantify specific serum proteins associated with particular organs. MCED analyses methylation patterns in circulating cell-free DNA that are common across many tumour types and simultaneously tissue-specific, enabling both detection and origin prediction from a single sample.
No. Sensitivity is moderate and lowest in stage I disease, so a negative result does not exclude early malignancy. It does not change the indication for mammography, cervical screening, colorectal screening or low-dose CT, all of which retain their own evidence base.
A positive result is not a diagnosis. It starts a diagnostic pathway, usually imaging directed by the predicted tissue of origin, followed by tissue biopsy where indicated. In the NHS-Galleri trial, just over half of participants with a positive result were subsequently diagnosed with cancer.
Positive predictive value rises with disease prevalence in the population tested. Testing groups with elevated baseline risk increases the proportion of positive results that represent genuine disease and reduces unnecessary investigation, without any change to the assay itself.
No MCED test currently holds FDA approval. Availability is through laboratory-developed test pathways and private providers rather than approved population screening programmes.
Its primary endpoint, a reduction in stage III or IV cancer incidence, was not met across three screening rounds. Secondary findings showed fewer cancers diagnosed at stage IV and fewer diagnosed through symptomatic or emergency presentation. Mortality outcomes are expected in approximately two years.
That depends on your personal and family history, infection status and exposure history, and it is a question for structured risk assessment with a doctor rather than a decision made from a price list. For many people the honest answer is that standard screening is the better use of their money.
This article is general educational information about the scientific basis of laboratory tests. It is not medical advice, a diagnosis, a screening recommendation, or an offer of treatment, and it makes no claim that any test detects, prevents or treats cancer. No multi-cancer early detection test currently holds FDA approval, none is diagnostic, and none replaces mammography, cervical screening, colorectal screening or low-dose CT. Screening decisions should be made with a doctor who knows your personal and family history.