Tumor marker blood tests are not diagnostic tests for cancer. Their accuracy depends entirely on context: a single marker like CA-125 or PSA can be normal in confirmed cancer and elevated in benign disease, so clinicians interpret results as probabilities within a defined population, not as yes-or-no answers. In screening settings, low prevalence makes false positives dominate; in monitoring settings, the same assay becomes far more useful because each patient serves as their own baseline.

That distinction explains most of the confusion around these tests. A marker's "accuracy" is not a fixed property of the assay. It is a property of the assay plus the question being asked plus the population being tested. Understanding why requires looking at what the molecules actually are, how thresholds get set, and where the biology breaks down.

What tumor markers actually measure

Most tumor markers are glycoproteins, hormones, or shed cell-surface fragments that tumor cells produce in excess, or that normal tissues produce when inflamed, regenerating, or stressed. Alpha-fetoprotein (AFP) is a fetal serum protein re-expressed by hepatocellular carcinoma and germ cell tumors. CA-125 is a mucin-like glycoprotein expressed by ovarian surface epithelium and also by mesothelium irritated by menstruation, endometriosis, or ascites. PSA is a serine protease made by prostate epithelium, and its blood level rises with prostate volume, inflammation, and instrumentation as well as with cancer.

The key mechanistic point is that none of these molecules is tumor-specific. They are quantified by immunoassay, typically a sandwich format in which two antibodies bind different epitopes and generate a signal proportional to concentration. That design is precise but not discriminating: the antibody cannot tell whether CA-125 came from a malignant ovary or an inflamed peritoneum. A plain-language version: the test counts the molecule accurately while remaining ignorant of its origin.

This is why marker results behave like a dimmer switch rather than an on-off switch. Malignancy shifts the distribution of values upward, but the distributions for cancer and benign conditions overlap substantially. Assay accuracy (how correctly the machine measures the concentration) and clinical accuracy (how well the value separates disease from health) are different things, and the second is almost always the limiting factor.

Why thresholds trade sensitivity against specificity

Because distributions overlap, any cutoff forces a trade. Lowering a threshold catches more cancers (higher sensitivity) at the cost of more false alarms (lower specificity). Raising it does the reverse. This trade-off is captured by the receiver operating characteristic curve, and the area under that curve summarizes discrimination across all cutoffs. For most tumor markers used in screening populations, discrimination is modest, which is why guidelines rarely endorse them for that purpose.

The arithmetic of prevalence is the part most thin explainers omit. A test with 90% sensitivity and 95% specificity sounds strong. Applied to 100,000 people in whom cancer prevalence is 0.1% (100 cases), it detects 90 cancers and misses 10. Among the 99,900 without cancer, 5% (4,995) test positive falsely. So roughly 98% of positive results are false positives. The same test in a monitoring population with 50% prevalence produces positives that are mostly true. Sensitivity and specificity are stable; predictive value is not.

Explicit threshold reconciliation matters here. CA-125 is commonly referenced against 35 U/mL, a cutoff derived decades ago and applied across all ages and menstrual states. PSA thresholds have shifted: 4.0 ng/mL was the traditional reference, 2.5 ng/mL was proposed for younger men, and age-specific ranges followed. CEA is often referenced near 3 ng/mL in nonsmokers and 5 ng/mL in smokers. These numbers are not interchangeable across laboratories, because assay platforms, calibrators, and antibodies differ. A result of 40 U/mL on one CA-125 assay is not guaranteed to equal 40 U/mL on another. Clinicians therefore track trends on the same platform rather than comparing absolute values across labs.

Accuracy by clinical setting

The same marker performs very differently depending on why it was ordered.

In screening asymptomatic people, accuracy is generally poor because prevalence is low and benign mimics are common. CA-125 rises with menstruation, fibroids, endometriosis, and cirrhosis. PSA rises with benign prostatic hyperplasia and prostatitis. AFP rises with hepatitis and cirrhosis. Guidelines from major bodies generally do not recommend these tests as standalone screening tools in average-risk adults.

In differential diagnosis of a symptomatic patient, markers add modest value. A markedly elevated value in the right clinical picture raises suspicion; a normal value does not exclude cancer. This asymmetry is the practical core of marker interpretation.

In monitoring known disease, accuracy improves sharply. Serial measurements on the same assay let each patient act as their own control. A rising CEA after colorectal cancer resection, or a rising CA-125 in known ovarian cancer, can precede radiographic change by weeks to months. Here the biological noise that ruins screening becomes manageable because the comparison is internal.

In detecting recurrence after curative treatment, markers are useful but not sufficient. They trigger imaging and clinical correlation rather than standing alone. A confirmed rising trend on repeat testing carries more weight than a single elevated value.

Comparison table: tumor markers against alternatives

Marker Typical clinical use Main benign confounders Standalone diagnostic? Common reference point Key limitation
PSA Prostate cancer monitoring and shared decision-making BPH, prostatitis, recent biopsy or catheter No Around 4.0 ng/mL, age-adjusted variants Overlap with benign enlargement
CA-125 Ovarian cancer monitoring Menstruation, endometriosis, cirrhosis, ascites No 35 U/mL Poor sensitivity in early-stage disease
CEA Colorectal cancer monitoring Smoking, colitis, liver disease No About 3 ng/mL nonsmokers Not for screening
AFP Hepatocellular carcinoma and germ cell tumor context Hepatitis, cirrhosis, pregnancy No Around 10 ng/mL Nonspecific in liver disease
CA 19-9 Pancreatic and biliary monitoring Cholestasis, pancreatitis, Lewis antigen status No 37 U/mL Unreliable in Lewis-negative individuals
CA 15-3 / CA 27.29 Breast cancer monitoring Liver and benign breast disease No Assay-specific Not for early detection

The Lewis antigen detail is a genuine mechanistic caveat. Roughly 5-10% of people are Lewis antigen negative and cannot mount a CA 19-9 elevation at all, so a normal result in such a patient is uninformative rather than reassuring. This is a property of the patient's glycosylation genetics, not the assay, and it is invisible on a standard report.

Timeline table: how a marker result is actually worked up

Step Typical timing What happens Interpretation rule
Initial draw Day 0 Marker measured, often alongside imaging Single value alone rarely diagnostic
Repeat confirmatory draw Days to weeks Same assay, same laboratory Reproducible rise matters more than absolute value
Correlation Same visit Symptoms, exam, imaging, other labs Discordant results favor imaging and clinical judgment
Trend assessment Weeks to months Serial values plotted over time Slope and doubling pattern guide concern
Confirmatory imaging After trend or red flag CT, MRI, ultrasound, PET as indicated Tissue or imaging defines diagnosis
Multidisciplinary review After workup Oncology, radiology, pathology input No treatment decision rests on a marker alone

The counterintuitive detail: normal does not mean clear

The most important professional caveat is that a normal tumor marker level never excludes cancer. Small or early tumors may not shed enough antigen to cross any threshold. Some tumors never express a given marker at all, which is why panels rather than single tests are sometimes used in monitoring. Conversely, a treated patient can have a falling but still elevated value and be in remission, or a rising value within the reference range that nonetheless signals recurrence.

This means the reference range, which is derived from a healthy population distribution, can be misleading. A patient whose baseline CA-125 was 8 U/mL and now reads 30 U/mL is still "normal" by the 35 U/mL cutoff, yet has nearly quadrupled. That is why oncologists document individual baselines and interpret deltas. The threshold is a population concept; the patient is a longitudinal individual.

A second caveat concerns assay drift and interference. Heterophile antibodies, biotin supplementation, and high-dose biotin-containing products can distort immunoassays in either direction. The FDA has issued communications about biotin interference with certain immunoassays, and the practical response is to disclose supplements before testing and repeat unexpected results on a different platform.

Practical actions and red flags

Concrete steps improve interpretation more than any single number.

Document a baseline before treatment when a marker is likely to be monitored. Always repeat an unexpected abnormal result on the same assay before acting. Record the laboratory and platform, because absolute values do not transfer across assays. Interpret values against the clinical picture and imaging, never in isolation. Disclose biotin supplements, recent procedures, and infections before the draw.

Red-flag warnings:

A markedly elevated marker in a symptomatic patient warrants prompt specialist evaluation, not reassurance. A rising serial trend on the same assay should trigger imaging and clinical review even if each value is within the reference range. A normal marker does not cancel abnormal imaging or persistent symptoms. Discordance between a striking marker value and a benign clinical picture should prompt repeat testing and interference review.

Where the numbers stop being useful

What can be concluded is bounded. Tumor markers are quantitative, reproducible, and genuinely useful for monitoring and for prompting investigation. They are not diagnostic, not screening tools in average-risk people, and not interchangeable across laboratories. The accuracy question has no single answer because accuracy is a function of prevalence, threshold, assay, and the question being asked. The responsible use is trend-based and context-bound, with a clinician integrating the number rather than obeying it.