Tumour Markers: What They Can and Cannot Tell You

A tumour marker result arrives looking like a verdict. There is a number, a reference range, and often a flag. The instinct is to read it the way a blood glucose or a potassium is read: inside the range means fine, outside means a problem. For this particular family of tests, that instinct is close to the opposite of correct.

Tumour markers are substances, almost always proteins, that can be measured in blood and that tend to be present at higher concentrations when certain cancers are present. The word “tend” is doing enormous work in that sentence. Almost every established marker is also made by healthy tissue, is raised by a list of harmless conditions, and is normal in a substantial fraction of people who genuinely have the cancer it is associated with. None of that makes the tests useless. It makes them useful for a narrower job than the one they are frequently asked to do.

The job they do well is watching. If a cancer is known to be present and known to produce a particular marker, the concentration of that marker becomes a proxy for how much cancer is there, and changes in it over time track response to treatment or the return of disease, often earlier than imaging. The job they mostly do badly is finding. Applied to a general population where the disease is rare, the same test that performs respectably in a clinic produces far more false alarms than true findings, and it does so for reasons that are purely arithmetic rather than a failure of the assay.

This piece works through the mechanism, then through the arithmetic, and then through what a raised result actually warrants.

What a Tumour Marker Physically Is

Almost all clinically used tumour markers are proteins or glycoproteins that circulate in blood, and almost all of them have an ordinary physiological existence. Prostate specific antigen is an enzyme secreted by prostate tissue into semen, where it liquefies the ejaculate; it appears in blood because a small amount leaks across from the gland. CA-125 is a large glycoprotein expressed on cells derived from the same embryonic tissue that forms the lining of the abdominal cavity, the pleura and parts of the reproductive tract. Carcinoembryonic antigen is, as the name suggests, a protein produced abundantly in fetal development and at low levels in adult gut tissue.

None of these is a cancer molecule. What cancer changes is quantity and access. Tumour tissue often produces more of the protein per cell, contains far more cells than the original tissue, and disrupts the normal barriers that keep the protein where it belongs, so more of it reaches the bloodstream. Three separate mechanisms therefore raise a marker: more production, more tissue, and more leakage. Anything else that produces the same three effects raises it too, and inflammation is extremely good at producing all three.

The measurement itself is an immunoassay. Antibodies raised against the target protein capture it from the sample, a second labelled antibody binds to a different part of the same molecule, and the amount of label detected is proportional to how much of the target was present. This sandwich arrangement is sensitive and specific at the molecular level, but it introduces its own quirks. Different manufacturers use different antibody pairs that recognise different regions of the protein, which means results from two platforms are not interchangeable. A patient followed across a change of laboratory can appear to have a change in disease when what changed was the assay.

Why Sensitivity and Specificity Are Not Enough

Sensitivity is the proportion of people with the disease whom the test correctly identifies. Specificity is the proportion of people without it whom the test correctly clears. Both are properties of the test measured against a known answer, and both are quoted in test literature as though they settle the question of usefulness. They do not, because a patient does not want to know how the test performs on people whose diagnosis is already established. They want to know what their own result means.

That is a different quantity, and it depends on something the test itself knows nothing about: how common the disease is in the population being tested. This is the pivot on which the whole subject turns. A test’s sensitivity and specificity stay roughly constant as you move it from a clinic full of symptomatic patients to a general population. Its predictive value collapses.

The reason is that false positives come from the healthy group, and the healthy group grows as the disease becomes rarer. Even an excellent specificity generates a large absolute number of false positives when applied to a very large number of people without the disease, while the number of true positives is capped by how few people actually have it. At some prevalence, the false positives outnumber the true ones, and then outnumber them heavily.

Predictive Value in Low-Prevalence Populations

It helps to work this through with round numbers, chosen for illustration rather than taken from any particular test. Imagine a marker with high sensitivity and a specificity of ninety-five per cent, meaning one healthy person in twenty is flagged.

The test has not changed between those rows. Only the population has. In the bottom row, a person with a positive result is overwhelmingly likely not to have the disease, and yet they now enter a pathway of repeat testing, imaging and sometimes biopsy, each step of which carries its own cost, delay and risk.

This is why screening programmes are so much harder to justify than they appear. A screening test must not merely detect disease; it must detect it early enough to change the outcome, in enough people, to outweigh the harm done to the much larger group who are investigated unnecessarily. Several tumour markers detect disease reliably and still fail that test, because the cancers they find early were either going to be found in time anyway or were never going to cause harm.

The mirror image of this problem is under-detection. A normal marker does not rule cancer out. A meaningful proportion of tumours of the relevant type simply do not secrete enough of the marker to raise the blood level, particularly when small. A reassuring number in someone with genuine symptoms is one of the more dangerous results in laboratory medicine, because it can delay the imaging or biopsy that would have found the answer.

Benign Conditions That Raise Common Markers

Every widely used marker has a substantial list of non-malignant causes of elevation, and in routine practice those causes are far more common than the cancer.

Prostate specific antigen rises with benign prostatic enlargement, which is nearly universal with age, and with prostatitis, urinary infection, urinary retention, recent instrumentation of the urinary tract, and mechanical pressure on the gland including from cycling. It also falls, sometimes substantially, on medications used for prostate enlargement, which can mask a rise that would otherwise have prompted investigation.

CA-125 is raised by anything that irritates the tissue lining the abdomen and pelvis. Endometriosis, fibroids, pelvic inflammatory disease, ovulation, menstruation, pregnancy, liver disease with fluid in the abdomen, heart failure, pancreatitis and recent abdominal surgery all do it. In premenopausal women this list is long enough that the marker performs poorly as a discriminator, which is precisely the group in whom it is most often ordered out of anxiety.

Carcinoembryonic antigen rises in inflammatory bowel disease, liver disease, pancreatitis, peptic ulceration and, notably, in smokers, whose baseline sits higher than that of non-smokers. Alpha-fetoprotein is raised in pregnancy and in chronic liver disease of any cause, which is awkward given that chronic liver disease is also the main risk factor for the cancer it is used to detect. CA 19-9 is not produced at all in the small proportion of people who lack a particular blood group enzyme, so in them it remains undetectable regardless of disease.

Kidney function matters across the board. Many markers are cleared renally, so impaired clearance raises the measured level without any change in production.

Marker Trends Versus Single Values

If there is one practical idea worth carrying away, it is that these tests were designed to be read as a series. A single value answers the question “how much marker is in this tube”, which is rarely the clinical question. A sequence of values answers “is the amount changing, in which direction, and how fast”, which usually is.

The reason trends work better is that most of the confounding factors are stable within a person. Someone with an enlarged prostate has a higher baseline PSA than they would otherwise, but that baseline is theirs, and a value that climbs steadily above it means something even if every individual reading sits inside a population range. Conversely a value slightly above the reference limit that has been sitting there unchanged for years is far less concerning than a value inside the limit that has doubled since the previous sample.

Two technical concepts support this. Half-life describes how quickly a marker falls once its source is removed; markers with short half-lives fall rapidly after successful surgery, and a failure to fall as expected suggests residual disease. Doubling time describes how fast a rising marker is rising, and for several markers a short doubling time carries worse implications than the absolute value.

The practical requirement is consistency. Serial values should come from the same assay platform where possible, ideally the same laboratory, and be interpreted against the patient’s own previous results rather than only against a printed range. When a marker is followed across a laboratory change, a fresh baseline is usually needed.

Where Markers Genuinely Earn Their Place

Set against all of the above, there is a clear set of situations in which these tests are genuinely valuable and sometimes indispensable.

The strongest use is monitoring known disease. Once a cancer is confirmed and shown to produce a given marker, the marker becomes a quantitative measure of disease burden that can be sampled as often as needed, without radiation, at low cost. Falling values during treatment support that the treatment is working. A rise after treatment can precede radiological evidence of recurrence, buying time.

The second use is in a small number of cancers where markers are exceptionally informative. Certain germ cell tumours produce markers with such reliability that the values contribute directly to staging and to assessing response, and their behaviour after treatment is a formal part of how those cancers are managed. Thyroglobulin after complete removal of the thyroid gland is another good case: with the gland gone, there is no legitimate source, so detectable thyroglobulin is meaningful in a way that no marker measured against an intact organ can be.

The third use is risk stratification in defined high-risk groups, where the prevalence arithmetic is more favourable. Surveillance of people with established chronic liver disease, or with a strong inherited predisposition, sits in a different part of the predictive value table than population screening, and the same test can be reasonable in one and unreasonable in the other.

What unites these is that a specific question was asked before the blood was drawn. Markers ordered as part of a general check-up, with no question attached, generate results that no one is well placed to interpret.

Questions Worth Asking About a Flagged Result

A flagged marker on a report is a starting point for a conversation, not a diagnosis, and there are a handful of questions that reliably move that conversation forward.

The first is why the test was done. A marker ordered to follow known disease means something quite different from the same marker ordered speculatively during a health screen. The second is what the previous values were. A single result with no history is the least informative version of this test; if earlier samples exist, the trajectory is the finding.

The third question is how far outside the range the value sits. Marker distributions have long tails, and values marginally above a cut-off are common, frequently transient, and often explained by benign conditions. Very large elevations behave differently and warrant a different urgency. The fourth is whether anything on the benign list applies: recent infection, inflammation, a procedure, pregnancy, smoking, impaired kidney or liver function, or in the case of PSA anything that pressed on the prostate in the preceding days.

The fifth is the most useful of all: what would change depending on the answer? If a repeat sample in a few weeks would be the next step regardless, that is often the right step. If imaging is warranted on the strength of symptoms alone, the marker is not the deciding factor.

None of this argues that tumour markers are bad tests. It argues that they are measurements of a quantity that correlates imperfectly with something we care about, and that the correlation is strongest when we already know what we are looking at. Read as a series, in a person whose situation is defined, they are among the more useful tools in oncology. Read as a single number in a healthy person, they mostly generate anxiety and further tests, and the arithmetic says that is not a fixable problem with a better assay.