The Evidence Behind
What we actually know about looking after your teeth
Preface
In October 2023 I wrote a short blog post listing five things you can do to keep your teeth healthy, along with the evidence behind each one. I expected it to be a dull piece of public service writing. Instead, while I was tracking down the studies, I found something that bothered me enough to keep going for four more posts, and eventually for this book.
Here is what bothered me.
Britain’s national oral health guidance is a document called Delivering Better Oral Health, and it is a good document. It is careful, it is written by serious people, and unlike most dental advice you encounter it tells you how confident it is in each thing it says. Some advice is labeled Strong. Some is Conditional. Some is Good practice, of which its own definition says: “No robust underpinning research evidence exists.” Such advice rests instead on extrapolation from related research and on clinical consensus. That transparency is rare and it deserves credit.
So I took a bullet from one of the Strong recommendations, “last thing at night (or before bedtime) and on one other occasion,” and I followed the citation to see the trial that proved it.
There is no trial. There is a one-page conference abstract from 2001, presented at an ORCA congress by two researchers at Unilever Dental Research, which measured how much fluoride was left in people’s saliva overnight (1). It did not compare brushing at bedtime against brushing earlier. Everyone in it brushed at bedtime. What it compared was two toothpastes.
The Scottish guideline that first cited it describes it, correctly, as an observational study, and marks the recommendation it supports with the symbol its own key defines as a Good Practice Point: “Recommended best practice based on the clinical experience of the guideline development group” (2). What happened next is not that somebody upgraded that judgment. DBOH embedded it as one bullet inside a multi-component recommendation labeled Strong, and its own methods say the strength of such a recommendation follows the main component. Its evidence table still rates timing as low certainty. The label a reader sees says Strong; the sentence explaining what that label was earned by is in another chapter.
I checked another one. Then another.
What this book is
This book takes the recommendations in Delivering Better Oral Health, one at a time, and asks a single question of each: does the evidence the guideline cites answer the question the advice asks?
That sounds like a small question. It turns out to be the one that matters, because a recommendation and its evidence can fail to line up in ways that are invisible unless you go and look. The study can be in children when the advice is for adults. It can measure fluoride in saliva when the advice is about tooth decay. It can be observational when the advice is causal. It can compare A with B when the advice is about C. In every one of those cases the guideline can honestly say it is “based on evidence,” and in every one of those cases you would be wrong to read that as “this was tested.”
For each recommendation I do the same six things. I quote the advice exactly as the guideline words it. I quote what the guideline says its evidence is. I go and read that evidence. I appraise it with the standard tools, and I show you my working rather than asking you to trust me. I search for anything better that has been published since. And then I give a verdict in a fixed format, which includes a line I think every critic owes their reader: what would change my mind.
What this book is not
It is not an argument that you should stop brushing your teeth.
I want to be blunt about this, because a book with this argument in it will be quoted by people who want to use it for something else. Weak evidence for a practice that is cheap, pleasant, and almost impossible to be harmed by is not a reason to abandon that practice. It is a reason to be honest about the label on the tin. Those are different claims and I will keep them separate throughout.
I brush twice a day. I brush last thing at night. I spit and I do not rinse. I have done all three since 2017. Chapter 31 What I actually do, and why explains why I do things the evidence has not established, which is not a contradiction so much as the ordinary condition of being a person who has to decide what to do this evening.
It is also not an attack on the people who wrote the guideline. Chapter 13 of the 2025 edition publishes an evidence statement for most recommendations and a certainty rating for many of them, in plain language, and by doing so it hands critics like me the rope. Very few guidelines anywhere are that open. My argument, developed in Chapter 3 The bundling problem, is not that the authors were careless. It is that a specific and defensible drafting convention, stated plainly in the guideline’s own methods, has a consequence I do not think anyone intended.
How to read this
The main text is written to be read straight through by anyone, with or without a dental or research background. Where I need to show statistical working, it goes into a marked box you can skip without losing the thread. If you are a dentist, a dental student, or a methodologist, the boxes are the part you want; if you are a person who would just like to know whether to buy the expensive toothpaste, they are not.
Two conventions run throughout. Strength is a property of a recommendation: it is how firmly you are being told to do something. In GRADE it takes exactly two values, Strong and Conditional; DBOH adds a separate Good practice category for advice it says is not the kind of thing certainty can be rated on at all. Certainty (high, moderate, low, very low) is a property of the evidence: it is how sure we are about the effect. These two are constantly confused, including by me in the original blog posts, and Chapter 2 Certainty is not strength is entirely about keeping them apart.
A note on my own position
I trained as a dentist and I work as a health research methodologist. I have received no funding for this book from any toothpaste manufacturer, professional body, or public health agency, and I have no financial interest in anything discussed in it.
Two files carry the working. appraisals/dboh-2025.csv holds one row per recommendation, extracted from the guideline by script rather than typed by hand: its wording, its strength label, its chapter 13 evidence statement, its citations, and how confidently the two chapters could be joined. appraisals/searches/ holds one file per chapter that reaches a verdict, giving each search as it was run, with its date and what it returned, including the searches that failed.
What the CSV does not yet hold is my own adjudicated judgments. It reserves columns for them, and they are empty. My certainty ratings, directness judgments and verdicts live in the verdict box at the end of each chapter, which is where you can hold me to them; getting them into the same file as the guideline’s own values, so the two can be compared row by row, is work I have not done. I would rather say that than let a reserved column stand in for a completed one.
Two of those files deserve a word, because this book asks a lot of other people’s citations and should be clear about its own.
appraisals/dboh-2025.csv is generated by a script from the archived guideline, not typed by hand, so the counts quoted in these pages can be regenerated and checked. Its fuzzy matching is not perfect, and at least one chapter was written from a row the script had failed to match before that was caught; the correction is recorded in Chapter 4 How I did this and in the relevant search file.
appraisals/searches/ holds one file per chapter that reaches a verdict. Three of them record a search that failed and had to be re-run, including two where an update search missed a newer Cochrane review that changed what a chapter could say. Those are kept rather than tidied away, because a book that treats an undocumented null as indistinguishable from not having looked cannot then hide its own.
This book examines research evidence. It is not medical or dental advice, it is not a substitute for examination by a qualified clinician, and nothing in it should be used to diagnose or treat a condition. Where I conclude that the evidence for a piece of advice is weaker than its official label suggests, that is a statement about the research, not an instruction to stop following the advice.
This book quotes and analyses Delivering Better Oral Health: an evidence-based toolkit for prevention (Department of Health and Social Care, Welsh Government, Department of Health Northern Ireland and NHS England), as updated on 10 September 2025. That material is Crown copyright and is reproduced under the Open Government Licence v3.0. Neither the Crown nor any of the issuing bodies endorses this book or its conclusions. Passages from SIGN 138 and other guidelines are quoted for the purposes of criticism and review. Every figure has been redrawn by me from the numbers reported in the cited source; the drawing code is in the repository.
Ahmad Sofi-Mahmudi