20 Looking for oral cancer, and the tests you are told not to use
Oral cancer is the reason a dentist asks you to stick out your tongue and then holds it in a piece of gauze while looking under it. It is the most serious thing they are looking for, and survival depends heavily on how early it is found.
This chapter contains the most unusual recommendation in the guideline: a Strong recommendation not to do something. It is also, on the evidence, an excellent one.
The advice
It’s not recommended to use vital staining, oral cytology or light‐based detection and/or oral spectroscopy for evaluating lesions for malignancy.
Strength of recommendation: Strong
— Delivering Better Oral Health, chapter 2, table 3e (1)
Alongside it, two Good practice points: examine everyone at every oral health assessment, and refer urgently on a specified list of signs.
Why the negative recommendation is the strongest thing here
Recommendation based on moderate certainty evidence from a well conducted systematic review of diagnostic accuracy of index tests for the detection of oral cancer and potentially malignant disorders of the lip and oral cavity, in patients presenting with clinically evident lesions.
— Delivering Better Oral Health, chapter 13 (1)
Three features make this the cleanest evidence-to-recommendation match in the whole guideline, and it is worth naming them because the rest of this book has been a catalogue of the opposite.
The evidence is the right design for the question. A diagnostic accuracy review is what you need to answer “how well does this test discriminate”. It is not a proxy, not a mechanism, not an observational association. It is also not a measure of clinical utility: sensitivity and specificity against biopsy tell you how the test performs, not whether using it changes referrals, stage at diagnosis, morbidity or survival. Nobody has measured that, and it is worth holding in mind before calling this row perfect.
The population matches. Patients presenting with clinically evident lesions, which is exactly when a clinician would be tempted to reach for an adjunctive test.
The direction matches. The tests do not perform well enough to substitute for the established pathway, supporting a recommendation not to rely on them.
The review DBOH cites has been updated, and the update splits the tests apart
DBOH cites the 2015 version of this Cochrane review. There is a 2021 update (2), published four years before DBOH’s own September 2025 edition, and I missed it too until a reviewer checked the citation. It includes 63 studies and 79 datasets covering 7,942 lesions, and its central contribution is that it stops grading the adjunctive tests as a block:
| Test | Sensitivity (95% CI) | Specificity (95% CI) | Certainty |
|---|---|---|---|
| Oral cytology | 0.90 (0.82 to 0.94) | 0.94 (0.88 to 0.97) | moderate both |
| Vital staining | 0.86 (0.79 to 0.90) | 0.68 (0.58 to 0.77) | low / very low |
| Light-based, spectroscopy | 0.87 (0.78 to 0.93) | 0.50 (0.32 to 0.68) | low / very low |
| Combined tests | 0.78 (0.45 to 0.94) | 0.71 (0.53 to 0.84) | very low both |
Two things follow that the 2015-based version of this chapter got wrong.
The tests are not equivalent. A specificity of 0.50 for light-based detection means that among people without disease, about half still test positive. In a clinic where most suspicious-looking patches turn out to be harmless, that generates a great many false alarms, each one leading to a referral, a wait and possibly a biopsy that was never needed. I called that “a coin toss” in an earlier draft, which is wrong: a test is not defined by one of its two error rates, and a test can have poor specificity while still catching most of the disease that is there. The case against this one is the false-positive burden in the pathway it would actually be used in, not a slogan about chance. Oral cytology at 0.90 sensitivity and 0.94 specificity, both at moderate certainty, is a different object. The reviewers’ own conclusion is that cytology “would appear to offer the most potential” and that combined tests involving cytology “warrant further investigation.”
The conclusion is about replacement, not adjunctive use. What the review actually says is that “none of the adjunctive tests can be recommended as a replacement for the currently used standard of a surgical biopsy and histological assessment.” DBOH’s recommendation is broader than that, and defensibly so on cost and over-referral grounds, but the review does not on its own establish the broader claim, and I previously wrote as though it did.
There is also a structural reason a negative recommendation is easier to get right. When you recommend for something, the harm side is often unmeasured (see nearly every chapter in Part II). When you recommend against a diagnostic test, the harms are the point: a false positive means an unnecessary biopsy and weeks of fear, a false negative means false reassurance about a cancer. The evidence-to-decision reasoning is doing visible work.
Devices in this category are sold to dental practices, often at meaningful cost, on the promise of catching cancers earlier. The guideline says the evidence does not support them. That is a commercially inconvenient recommendation to make, and making it plainly deserves credit.
The uncomfortable part: what supports the examination itself
Now the awkward question, which the guideline handles honestly and which most readers will not have considered.
The instruction to perform the examination is not Strong. It is Good practice, with no evidence statement in chapter 13.
That is a statement about chapter 13, not about the guideline. Chapter 6 does cite sources for the examination: it says most symptomless mouth cancers “can be detected by means of a visual examination by clinicians,” with a citation, and places the practice in line with the regulator’s recommendations. So there is supporting material; what there is not is a graded certainty statement in the place this book has been auditing. I have not appraised those chapter 6 sources, and say so rather than treating absence from one table as absence from the document.
And the referral criteria, the specific list of signs that should trigger an urgent cancer-pathway referral, carry this:
NICE guidance found no evidence with regard to the positive predictive values of different symptoms of oral cancer in primary care. The benefits of rapid referrals need balancing against the harms of over-referral.
— Delivering Better Oral Health, chapter 13 (1)
Read that again. The list of warning signs that determines who gets referred urgently for suspected oral cancer is not underpinned by evidence about how predictive those signs actually are.
This is not a scandal, for the same reasons set out in Chapter 21. An ulcer that has not healed in three weeks is a reasonable thing to investigate whether or not anyone has calculated its positive predictive value. Withholding referral until the arithmetic exists would be indefensible. And DBOH labels it Good practice, discloses the gap, and names the trade-off with over-referral in the same sentence.
But it is worth a reader knowing that the screening examination itself, one of the most routine acts in dentistry, sits on much thinner evidence than the recommendation against the gadgets.
A distinction this chapter has to keep straight.
Does opportunistic examination reduce oral cancer deaths? That is a population screening question, and it is genuinely hard. Screening programs are only justified when early detection changes outcomes enough to outweigh the harms of over-diagnosis, and for oral cancer in general populations the evidence has never been strong enough to support a national screening program.
Should a clinician who is already looking in your mouth look properly? Different question, obviously yes, no trial needed.
DBOH is asking the second question, not the first, and Good practice is the right label for it. I flag the distinction because “there is no good evidence for oral cancer screening” is a true sentence about the first question that would be badly misleading if applied to the second.
A correction of my own. An earlier draft of this chapter said I had not obtained the review DBOH cites, relied on the guideline’s characterization of it, and called this the one chapter where the central claim rested on someone else’s account. That was true of the 2015 version, and it was also a failure of the search: the 2021 update is freely available, indexed, and a title search returns it at once. I have now read the update, and the table above is from it. The lesson is the one this book keeps arriving at: an update search by review ID is not a search.
PubMed, 21 August 2026, for systematic reviews of adjunctive tests for oral cancer detection, in two passes. The first missed the Cochrane update and is recorded as a failure; the second retrieved Walsh and colleagues 2021 (DOI 10.1002/14651858.CD010276.pub3, PMID 34282854), whose own search ran to 20 October 2020. I have the abstract and summary estimates for the update, not its full text, so the QUADAS-2 judgments above are the reviewers’ summary rather than my reading.
Full record in appraisals/searches/ch20-oral-cancer.md.
Verdict
- Certainty of evidence
-
Test by test, on the 2021 update rather than the 2015 review DBOH cites. Moderate for oral cytology’s accuracy, which is the best-performing adjunct and the one the reviewers say has the most potential. Low to very low for vital staining and light-based detection, whose specificities are 0.68 and 0.50. DBOH’s single moderate rating covers all three as a block, which the update no longer supports.
None stated for the examination itself or for the referral criteria, and the guideline says so.
- Directness to the advice as worded
-
Good on test and population, limited on outcome and setting. Right tests, right patients, and accuracy measured against biopsy. But diagnostic accuracy is itself an intermediate outcome: no study here shows that using or not using these devices changes referrals, stage at diagnosis, or survival. The review also judges its applicability to be to patients attending secondary care, while the recommendation is read in general dental practice.
Not separately assessed for the Good practice points. Chapter 13 gives them no certainty statement; chapter 6 cites supporting sources I have not appraised.
- Is the strength label defensible?
-
Yes, and it is still among the guideline’s best rows, though not the flawless one I first called it. A Strong recommendation against an intervention, grounded in a diagnostic accuracy review of the right design, in the right population, where the harms of getting it wrong are explicit, is a good piece of guideline writing.
Two qualifications survive the update. The certainty is not uniform across the three named tests, and cytology in particular is better supported than the single rating implies. And the review’s conclusion is that these tests cannot replace biopsy, which is a narrower claim than not using them at all; the broader recommendation needs the cost and over-referral arguments to carry it, and those are evidence-to-decision judgments the guideline has not published.
Good practice is also right for the examination and the referral list, with the gap disclosed.
- What would change my mind
- A study of whether using an adjunct changes anything a patient experiences, rather than how it performs against biopsy: referral rates, stage at diagnosis, survival, or the anxiety cost of a false positive. On the current evidence nobody knows. Further work on cytology specifically, which the reviewers single out, could make the blanket recommendation obsolete rather than wrong. And work on the positive predictive values of oral cancer symptoms in primary care, which NICE identified as missing and which would let the referral criteria be tuned rather than asserted.
What this means for you
If a dental practice offers you a dye rinse, a special light, or a brush sample, often for a fee, as a way of checking a lesion in your mouth, the national guideline advises clinicians against relying on those tests, and the evidence behind that advice is among the better-supported things in this book. What none of them can do is take the place of a biopsy, which is the thing that actually gives an answer. If you have a lesion someone is worried about, the question to ask is about referral and biopsy, not about which gadget to buy.
What matters far more is that a dentist looks properly at every check-up, and that certain things get acted on rather than watched. Here is the guideline’s own list of what should trigger an urgent referral, in full and in its own words, because this is not a place for me to summarize:
- an unexplained ulceration in the oral cavity lasting for more than 3 weeks
- a persistent and unexplained lump in the neck
- a lump on the lip (inner or outer) or in the oral cavity consistent with oral cancer
- a red patch in the oral cavity consistent with erythroplakia
- a red and white patch in the oral cavity consistent with erythroleukoplakia
- persistent unexplained hoarseness
- persistent pain in the throat or pain on swallowing lasting for more than 3 weeks
— Delivering Better Oral Health, chapter 2, table 3e (1)
That is a list for clinicians, not a self-examination checklist, and it is reproduced here so you can see what it says rather than so you can score yourself against it. An earlier draft of this chapter gave a shortened version, which dropped the lip and oral-cavity lump entirely and merged two of the patch criteria. Abbreviating a cancer referral list is not something this book should do.
Nobody has calculated how predictive each of those signs is, at least not in the evidence NICE assembled for primary care, which is what DBOH reports. They are the list because clinicians agree they matter, which is the honest reason, and the guideline says so.