What a Blood Test Can and Cannot Tell You About Cancer

Oak
cancer cfdna biomarkers screening liquid-biopsy
A lab instrument draws plasma from a vial into a glowing glass column where short blue filaments sort into bands, a few glowing orange.

There is no blood test that tells you whether you have cancer. There are blood tests that shift the probability, some of them substantially, and the amount of shift depends almost entirely on what your probability was before the draw. A multi-cancer early detection (MCED) test run on an asymptomatic 45-year-old has a low positive predictive value not because the assay is bad but because the prior is roughly 0.5% per year across all cancers combined. The same test in someone with unexplained weight loss and a rising CA 19-9 is a different instrument. Understanding blood-based cancer detection means understanding the measurement, the analyte, and the arithmetic that connects them.

What is in the blood that comes from a tumor

Four analyte classes, in rough order of how much signal each carries.

Cell-free DNA (cfDNA). Every cell that dies sheds fragments of its genome into plasma, mostly 160–180 bp mononucleosomal pieces. In a healthy person the total is a few nanograms per milliliter of plasma, almost all from hematopoietic cells. A tumor contributes circulating tumor DNA (ctDNA) carrying its somatic mutations and, more usefully, its aberrant methylation. The fraction matters enormously: late-stage metastatic disease can push tumor fraction above 10%, while a stage I tumor under 1 cm may contribute under 0.01% of cfDNA, which is below the detection floor of most assays no matter how deeply you sequence. This shedding gradient, not assay chemistry, is why MCED sensitivity climbs steeply with stage.

Circulating tumor cells (CTCs). Intact cells in circulation, present at roughly one per billion blood cells when present at all. Detection requires enrichment (EpCAM capture, size filtration, imaging flow cytometry) and yields are low and variable across tumor types, which has kept CTCs largely in research and prognostic use rather than screening 1.

Proteins. The classical tumor markers: PSA, CA 125, CA 19-9, CEA, AFP, CA 15-3. These are secreted or shed proteins, most of them not tumor-specific. CA 19-9 rises in pancreatitis, cholangitis, and biliary obstruction, and about 5–10% of people are Lewis-antigen negative and cannot produce it at all, so their CA 19-9 stays near zero even with a large tumor 2. Protein panels perform far better as trend-monitoring tools in someone with known disease than as screens.

Metabolites and lipids. Tumors reprogram metabolism, and that shows up in plasma. Mass-spectrometry metabolomics has separated pancreatic cancer cases from controls using panels of amino acids and lipid species 3, and lipidomic profiling has distinguished prostate cancer plasma from controls 4. Metabolomic work in bladder cancer using paired in vitro and ex vivo systems has helped show which signals originate in tumor tissue rather than downstream physiology 5. These are promising and mostly pre-clinical. Metabolite levels move with fasting, sleep, exercise, and gut flora, and cross-platform reproducibility is the standing problem.

Why cfDNA methylation beat mutation calling

The first generation of liquid biopsy tried to find tumor mutations in plasma. Two problems killed it as a screening approach. First, clonal hematopoiesis: mutations in DNMT3A, TET2, ASXL1, and TP53 arise in expanded blood-cell clones in a large fraction of people over 60, and they land in plasma cfDNA looking exactly like tumor signal. Controlling for this requires sequencing matched white-cell DNA, which most workflows now do. Second, a mutation tells you cancer is somewhere, not where. Tissue of origin matters because it determines the diagnostic workup.

Methylation solves both. CpG methylation patterns are tissue-specific, so a methylation signature both flags malignancy and predicts tissue of origin. Commercial MCED tests use targeted bisulfite or enzymatic conversion sequencing across ~100,000 informative regions, then a classifier trained on case-control cohorts. Published performance for the best-characterized test: specificity around 99.5%, overall sensitivity across cancer types near 50%, with tissue-of-origin prediction correct roughly 88% of the time when a signal is detected. Sensitivity by stage is the number to hold onto: roughly 17% at stage I, 40% at stage II, 77% at stage III, 90% at stage IV in the validation cohorts.

Run the arithmetic. At 99.5% specificity and an annual all-cancer incidence of about 1% in a 60-year-old, positives are roughly half true, half false. At 45, with incidence near 0.3%, most positives are false. A false positive triggers a whole-body imaging workup with its own costs and incidental findings. A negative result moves your probability very little, because a test that misses 83% of stage I disease cannot rule anything out. If you take one thing from an MCED result, take this: negative is close to uninformative, positive requires a clinician and imaging, and neither replaces colonoscopy, mammography, low-dose CT, or cervical screening, all of which have randomized mortality evidence that MCED currently does not.

What routine blood work shows

A CBC and comprehensive metabolic panel are not cancer tests, but they are where a surprising number of cancers first announce themselves. Leukemias show up directly: blast cells on the differential, white counts in the hundreds of thousands, or unexplained cytopenias. Lymphoma often shows an elevated LDH. Multiple myeloma shows an anion-gap narrowing, elevated total protein with low albumin, hypercalcemia, and renal impairment, and is confirmed by serum protein electrophoresis with immunofixation. Colorectal cancer classically presents as iron-deficiency anemia (low MCV, low ferritin, low transferrin saturation) from chronic occult bleeding. Liver and bile-duct tumors show up as a cholestatic pattern: alkaline phosphatase and GGT rising out of proportion to ALT/AST.

Chronic inflammation is a weaker, slower signal. Prospective cohort work has found that circulating inflammatory markers including IL-6 and CRP associate with subsequent cancer incidence at the population level 6. That association is real and it is also nearly useless for an individual, because the effect sizes are small and the causes of an elevated CRP are legion. Treat a high-sensitivity CRP as a prompt to look for a reason, not as a cancer signal.

What we would do with our own data

Track trajectories, not thresholds. A single tumor-marker value sits inside a wide reference interval with substantial biological and assay variation. Six values over three years, run on the same platform, tell you something a single value cannot. The same logic drives the interest in serial cfDNA and in proteomic panels for lung cancer, where single markers have consistently failed and multi-analyte longitudinal approaches are where the field went 7. The general problem of biomarker discovery producing many candidates and few clinically validated tests has been well documented, and the failure mode is almost always overfitting to a case-control cohort that does not resemble a screening population 8.

Concretely, if you are building your own baseline: get a CBC with differential, CMP, ferritin, LDH, and hs-CRP at consistent times of day under consistent fasting conditions, at least annually, and keep the raw values in a table rather than in a portal you cannot query. Add whatever imaging and endoscopic screening your age and family history warrant, on schedule. If you have a germline whole-genome sequence, check the ACMG-recommended cancer genes (BRCA1/2, the Lynch syndrome mismatch-repair genes MLH1/MSH2/MSH6/PMS2, TP53, PALB2, APC) with a caller you trust and get any candidate pathogenic variant confirmed in a clinical lab and interpreted by a genetic counselor. Germline risk changes screening schedules, which is a decision that belongs with a physician. Do not act on a research-grade variant call.

Where a blood test flags something, the next step is imaging and tissue, not another blood test. Nothing in plasma substitutes for a biopsy.

Questions people also ask

What is the hardest cancer to detect? Pancreatic, ovarian, and biliary tract cancers, for the same reason: no effective screening test, deep anatomic location, vague early symptoms, and low cfDNA shedding until the tumor is large. Pancreatic cancer is the canonical case, where CA 19-9 lacks the specificity to screen and a meaningful subset of people cannot produce it at all 2.

What 50 types of cancer can be detected by a blood test? MCED tests detect a shared methylation signal rather than 50 separate assays, and the “50+ types” figure refers to the range of cancers represented in the training and validation cohorts. Sensitivity varies enormously across those types and is driven mostly by shedding and stage.

What is the most accurate test to detect cancer? Tissue biopsy with histopathology. Every blood test is a triage step whose output is a probability.

Would a full blood count show anything serious? Yes, sometimes. Leukemias, myelodysplasia, and the anemia of an occult GI bleed can all appear on a CBC. A normal CBC excludes very little, since most solid tumors leave it untouched until late.

What cancers cannot be detected by a blood test? Early brain tumors, in large part because the blood-brain barrier limits cfDNA release into plasma, and most early-stage cancers of any type. Small, non-shedding, low-grade tumors are invisible in plasma by current methods.

What is the biggest indicator of cancer? Persistent unexplained symptoms, especially weight loss, new pain, a palpable mass, or bleeding. Those warrant a physician’s evaluation, and no blood test should delay one.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Laura F. Ogle, James G. Orr, Catherine E. Willoughby, et al. Imagestream detection and characterisation of circulating tumour cells – A liquid biopsy for hepatocellular carcinoma?. Journal of Hepatology, 2016. https://doi.org/10.1016/j.jhep.2016.04.014 ↩

  2. Christina Zhang. Examining the Use of Serum Biomarkers to Guide Early Detection of Pancreatic Cancer. Clinical Journal of Oncology Nursing, 2025. https://doi.org/10.1188/25.cjon.151-156 ↩ ↩2

  3. Guoxiang Xie, Lingeng Lu, Yunping Qiu, et al. Plasma Metabolite Biomarkers for the Detection of Pancreatic Cancer. Journal of Proteome Research, 2014. https://doi.org/10.1021/pr501135f ↩

  4. Xinchun Zhou, Jinghe Mao, Junmei Ai, et al. Identification of Plasma Lipid Biomarkers for Prostate Cancer by Lipidomics and Bioinformatics. PLoS ONE, 2012. https://doi.org/10.1371/journal.pone.0048889 ↩

  5. Daniela Rodrigues, Carmen Jerónimo, Rui Henrique, et al. Biomarkers in bladder cancer: A metabolomic approach using in vitro and ex vivo model systems. International Journal of Cancer, 2016. https://doi.org/10.1002/ijc.30016 ↩

  6. Dora Il’yasova, Lisa H. Colbert, Tamara B. Harris, et al. Circulating Levels of Inflammatory Markers and Cancer Risk in the Health Aging and Body Composition Cohort. Cancer Epidemiology, Biomarkers & Prevention, 2005. https://doi.org/10.1158/1055-9965.epi-05-0316 ↩

  7. Alexios Matikas, Konstantinos N. Syrigos, Sofia Agelaki. Circulating Biomarkers in Non–Small-Cell Lung Cancer: Current Status and Future Challenges. Clinical Lung Cancer, 2016. https://doi.org/10.1016/j.cllc.2016.05.021 ↩

  8. Ed Yong. Cancer biomarkers: Written in blood. Nature, 2014. https://doi.org/10.1038/511524a ↩