32 When was this last checked?
Every recommendation in this book carries two dates, and the guidance prints neither of them clearly: the date its evidence was last reviewed, and the date the page you are reading was last touched. They are not the same date, and an earlier draft of this chapter confused them badly enough to invent a finding.
That correction is the chapter. It is worth watching it happen, because this mistake is easy to make and it flatters whoever makes it.
The mistake I made
Every chapter of Delivering Better Oral Health displays a line at the top: “Updated 10 September 2025.” I treated that as the date a panel finished reviewing the evidence, and then asked what that panel should have known. On that basis I calculated that the median recommendation rested on evidence published seven years earlier, and that sixteen rows rested on citations a decade old or more.
The publication’s own change log says otherwise:
10 September 2025. Updated to add ‘Appendix: clinical case studies’ and to make improvements to the layout and formatting.
21 September 2021. Reviewed and updated guidance in full. Update published as 4th edition.
12 June 2014. First published.
— GOV.UK, publication history for Delivering Better Oral Health
The 2025 change added case studies and improved the formatting. The last time the evidence was reviewed in full was September 2021. There was no 2025 panel to have missed anything.
Measured against the date the evidence was actually reviewed, the picture reverses:
| Age of the newest citation, at the 2021 evidence review | Recommendations |
|---|---|
| Same year | 2 |
| 5 years or older | 23 of 74 |
| 10 years or older | 0 of 74 |
| Median | 3 years |
| Oldest | 9 years |
Not one recommendation rested on a citation ten years old or more. The median was 3 years, which for a guideline assembled by volunteer panels over a multi-year process is unremarkable and arguably good.
My seven-year figure was not a finding about the guideline. It was an artifact of subtracting publication years from the wrong date, and it happened to make the guideline look worse, which is the direction a person writing a book of criticism should be most suspicious of.
Even corrected, this is a descriptive statistic about the age of cited publications. It is not a measure of certainty, of clinical validity, of when each row was searched, or of how long a recommendation will stay right.
Three specific limits. A review published in 2013 closed its own search before 2013, so the evidence underneath is older than its citation year. The 17 recommendations citing nothing dated are excluded rather than counted as infinitely stale. And because DBOH repeats advice across age groups, the same source can be counted several times.
An earlier draft went further and predicted that this book would stay current for about seven years and then start being wrong. That was not a calculation, it was a guess wearing a calculation’s clothes. Nothing in a snapshot of citation ages observes a recommendation becoming obsolete, and two fields with identical citation ages can change at completely different rates.
Old is not the same as decayed
Age alone is not a fault. There are three situations here and only one is anybody’s mistake.
Settled. The evidence is old because the question is answered. Chapter 28 is the clearest case: the Cochrane review of recall intervals states that “further studies comparing dental recall intervals for adults in primary care seem unnecessary.”
Quiet. The evidence is old and no update to the cited review has appeared. Chapter 10 is this: the varnish review’s newest included trial is from 2012 and its search closed in May 2013.
An earlier draft called this category “dormant” and explained it as trials having stopped because nobody funded them. I have removed that, because I did not investigate it and it is not true as stated. The absence of a new version of one review is not the absence of new trials: Protecting Teeth @3, a randomized nursery-school varnish study, was published in 2020. My own search record for that chapter says in terms that post-2013 primary trials were not searched. Turning an acknowledged gap in my own retrieval into a claim about the research economy was not evidence. It was a story I preferred.
Decayed. A newer version of the cited review existed, and the guidance quotes the old one. This is the only one of the three that is a mistake.
What actually decayed, and when
Two of the twelve Cochrane reviews DBOH cites had been superseded by the time of the 2021 evidence review:
| Recommendation | Version cited | Available at the 2021 review | Published |
|---|---|---|---|
DBOH-084, oral cancer adjuncts (Chapter 20) |
CD010276.pub2, 2015 |
CD010276.pub3 (1) |
July 2021 |
DBOH-074, e-cigarettes (Chapter 21) |
CD010216.pub3, 2016 |
CD010216.pub6 (2) |
14 September 2021 |
Two more have been superseded since, and that is a different and much lesser charge. CD006103 and CD013308, both supporting DBOH-072, were updated in 2023, two years after the panel reported. No panel failed there. What happened is that the guidance has been displayed, unchanged in substance, for long enough that its evidence base moved underneath it.
That distinction matters, and I did not draw it in the previous draft, where all four appeared as a single indictment.
The e-cigarette row, correctly dated
Chapter 21 works through this row in detail. The short version belongs here because the dates are the whole point.
DBOH reports that its e-cigarette recommendation rests on “low certainty evidence from one systematic review,” citing CD010216.pub3 from 2016, whose search closed in January 2016. One week before the fourth edition was published, .pub6 reported moderate certainty that nicotine e-cigarettes increase quit rates compared with nicotine replacement therapy, RR 1.53 (95% CI 1.21 to 1.93), from 61 studies in 16,759 participants (2).
A week is not a fair deadline and I am not going to pretend it is. But .pub4 appeared in October 2020 and .pub5 in April 2021, so a more current version had been available for most of a year. And .pub6 already carried this sentence in its abstract:
To ensure the review continues to provide up-to-date information to decision-makers, this review is now a living systematic review. We run searches monthly, with the review updated when relevant new evidence becomes available.
— Hartmann-Boyce and colleagues, CD010216 (2)
A living systematic review is one kept continuously current rather than revised in occasional editions (3). This one is maintained by Cochrane, searches monthly, and publishes under the same review identifier the guideline’s reference list already prints. Its current version, .pub11, searched to January 2026 and reports high certainty that nicotine e-cigarettes beat nicotine replacement therapy, RR 1.61 (95% CI 1.23 to 2.12) from 11 trials in 4,114 participants (4). That version appeared while this chapter was being revised, and replaced the one an earlier draft called current.
So the criticism is narrower than the one I first made, and it survives: the certainty statement a reader sees today describes a 2016 evidence base, and the mechanism for keeping it current was not merely available but advertised in the abstract of the review being cited.
I did the same thing, three times
This would be an easier chapter to write if the book were not guilty of it.
Four of my searches failed, and they are recorded in appraisals/searches/ rather than quietly re-run. Three are the same failure:
- Chapter 9 searched for a Cochrane review by its identifier, got nothing useful back, and concluded the 2010 version was current. A 2024 update existed (5).
- Chapter 20 searched the same way and missed
CD010276.pub3(1). - Chapter 13 searched on the 2014 review’s authors and title and missed its own ten-year update (6), published in the same journal, which had raised the evidence for the 5% threshold a full level.
The common fault is a known-item query: a search built out of the terms that describe the paper you already have. It finds that paper. It is structurally incapable of telling you a successor exists, because every word in it points backwards.
I can say that the guidance’s displayed citations and my searches missed some of the same updates. I cannot say that we failed by the same method, because DBOH’s search procedures for individual rows are not published and I have not seen them. An earlier draft claimed otherwise on no evidence.
What I would actually ask for
The obvious conclusion is to run a continuously updated review for every recommendation. That is the wrong conclusion, and it is worth saying why.
Living reviews suit a specific situation, and the methods paper that named them says which: questions where “research evidence is emerging rapidly, current evidence is uncertain, and new research may change policy or practice decisions” (3). All three, together. Most questions in this book fail at least one of the three.
Three cheaper things would do more.
Consume the living reviews that already exist. Nobody needed to build one for e-cigarettes. Cochrane maintains it and searches it monthly.
Print the version and the search date, not just the certainty. Chapter 13 says “low certainty evidence from one systematic review.” A dozen more characters would let a reader check it in under a minute:
Low certainty.
CD010216.pub3, searched to January 2016.
Separate the review date from the page date. A reader currently sees “Updated 10 September 2025” on a page whose evidence was reviewed in 2021, with no way to tell the difference. That single line is what misled me, and the change log was one click away.
GOV.UK publication history for Delivering Better Oral Health, read 8 September 2026, for the change log quoted above.
PubMed, 20 August and 8 September 2026, for the version history of CD010216. Metadata retrieved and structured abstracts read for .pub3 (2016), .pub6 (September 2021), .pub8 (2024), .pub9 (January 2025), .pub10 (November 2025) and .pub11 (August 2026). Version numbers, search closing dates, study counts and certainty ratings above are from those abstracts. I did not retrieve .pub4, .pub5 or .pub7; their publication dates come from the PubMed record rather than from reading them.
The citation ages in Table 32.1 are computed by scripts/extract_dboh.py, which takes the newest publication year in each recommendation’s reference list and subtracts it from 2021. Two limits of that method are in the box above. A third: the extractor reads years out of citation strings, and an earlier version also matched page numbers, so that Journal 1999;10:2001-2008 yielded 2008 as the newest year. That is fixed and tested, but a year read out of a string is still not a verified publication date.
Full record in appraisals/searches/ch32-currency.md.
What this means for you
Nothing about your teeth. This one is about how to read the advice.
If you are told something is backed by strong evidence, the useful follow-up is “as of when?”, and the answer has two parts: when the evidence was last reviewed, and when the page was last touched. Those can be four years apart, and only one of them is usually printed.
The same applies here. This book searched the literature in August and September 2026. I am not going to tell you how long that will hold, because I tried that in an earlier draft and the number was invented. New evidence can overturn an answer within a year of publication, or leave it standing for twenty. What a book like this can honestly offer is not a shelf life. It is a date, a method, and enough working shown that you can tell when it has gone stale.