5 Brush twice daily
Everyone knows this one. It is on the tube of toothpaste, it is in the waiting room, and it is the first thing anybody says about looking after teeth. It is also, according to the systematic review that looked hardest at it, a “social norm” whose “evidence base for this frequency is weak” (1).
Those are the review authors’ words, not mine.
The advice
Brushing frequency appears in each of the four core age-specific brushing recommendations in Delivering Better Oral Health, always bundled with other instructions. The adult version is the shortest:
Brush teeth at least twice daily:
- last thing at night (or before bedtime) and on at least one other occasion
- with toothpaste containing 1,350 to 1,500ppm fluoride
- spitting out after brushing rather than rinsing with water, to avoid diluting the fluoride concentration
Strength of recommendation: Strong
— Delivering Better Oral Health, chapter 2, table 1f (2)
The component under examination here is the first three words: at least twice daily. The other bullets get Chapter 6, Chapter 8 and Chapter 7.
What the guideline says its evidence is
Recommendation based on moderate certainty evidence for value of toothbrushing with fluoride toothpaste. Moderate certainty from studies with children and adolescents for spitting versus rinsing. Low-certainty evidence from children and adolescents for frequency and timing. Evidence for the concentration is based on studies on immature permanent dentition in children and adolescents.
— Delivering Better Oral Health, chapter 13, table 6 (2)
So DBOH already says the frequency evidence is low certainty, and already says it comes from children when the advice is for adults. There is no dispute to have about that. The question is what the low-certainty evidence actually is, and whether “at least twice daily” should be inheriting a Strong label from the fluoride toothpaste sitting next to it.
The citation is SIGN 138 (3).
Following the citation
SIGN 138 covers brushing frequency in §5.5.1. It cites two things.
First, a Cochrane review of fluoride toothpaste (4). Its randomized finding, which is not in dispute and is not the point at issue, is a pooled prevented fraction of 24% (95% CI 21% to 28%) for DMFS with fluoride toothpaste against none, from 70 studies and 42,300 children. What SIGN cites it for here is something else:
A Cochrane review found the effect of fluoride toothpaste increases with higher frequency of use. There were statistically significant associations between estimates of DMFS prevented fractions and frequency of use, with a 14% increase in PF (95% CI 6% to 22%) with twice daily brushing as opposed to once daily.
— SIGN 138, §5.5.1 (3)
Second, ten cross-sectional studies, which SIGN describes in its own words as “a small number of cross-sectional studies of variable quality.”
On the strength of those two things, SIGN grades the recommendation A, its highest, and the margin of that page carries the evidence level 1++: “high quality meta-analyses, systematic reviews of RCTs, or RCTs with a very low risk of bias.”
Here is the problem, and it is not a small one.
That 14% figure does not come from a meta-analysis of brushing frequency. It comes from a meta-regression inside a review whose randomized question was something else entirely. Marinho and colleagues pooled trials comparing fluoride toothpaste against non-fluoride toothpaste or none. That comparison was randomized, and the answer, a 24% prevented fraction, is trustworthy. Then, as a secondary analysis, they asked whether the size of that fluoride effect varied across trials according to trial-level characteristics, one of which was how often participants were asked to brush.
Three things follow, and none of them is a technicality.
Nobody was randomized to a brushing frequency. Trials were randomized to toothpaste. Frequency is a characteristic of the trial, so comparing high- and low-frequency trials is an observational comparison of studies, not an experiment. Trials that instructed twice-daily brushing may differ from trials that instructed once-daily brushing in country, decade, supervision, baseline caries, and follow-up.
The comparison is at the wrong level. This is an association between an aggregate feature of a trial and that trial’s average result. Inferring from it what happens when an individual brushes more often is the ecological fallacy. It may well be true. This design cannot show it.
It answers a different question. The finding is that fluoride toothpaste works better when used more often. That is a claim about fluoride delivery. The recommendation is a claim about what a person should do. They are close enough to be confused and far enough apart to matter.
The second strand of evidence, the ten cross-sectional studies, has the problem cross-sectional studies always have: people who brush twice a day differ from people who do not, in income, education, diet, dental attendance and much else, all of which independently affect tooth decay. SIGN reports that these studies were of variable quality, and the margin of that same page carries the evidence level 3, which in SIGN’s scheme means non-analytic studies such as case reports and case series, one tier above expert opinion. So a recommendation graded A carries, in its own margin, both the highest evidence level and one of the lowest.
Is there better evidence?
Yes, and it is not cited.
In 2016 Kumar, Tadakamadla and Johnson published a systematic review and meta-analysis whose stated aim is precisely the question at issue: “to assess the effect of toothbrushing frequency on the incidence and increment of carious lesions” (1). They screened 4,305 records and included 33 studies.
PubMed, 15 August 2026, and then again after a reviewer looked at what I had actually typed.
My original queries were a known-item search naming the journal and year and capped at 2018, plus a search about brushing timing. Neither could have returned a systematic review of brushing frequency published after 2018, and I nonetheless wrote that “nothing more recent came back.” A search that was not looking is not evidence of absence, and this is the third time in this book that an update search has failed because of how it was written rather than because of what exists.
Re-run properly, one thing comes back that I had missed. Al-Sharani and colleagues published a systematic review and meta-analysis of brushing initiation age and frequency in December 2025 (5). It is narrower than Kumar, covering early childhood caries in children under six from 32 observational studies, and it points the same way: brushing fewer than twice a day was associated with 2.11 times the odds of early childhood caries.
That does not change this chapter’s argument, because it is observational too and because it postdates the guideline. It does mean the sentence “Kumar 2016 is the only such review” was wrong, and it means a chapter complaining that DBOH failed to cite the relevant review had itself failed to cite a newer one, for exactly the reason it accuses the guideline of: nobody ran the search.
I also checked whether Delivering Better Oral Health cites Kumar. Searching the full text of all thirteen chapters and the appendix for the name returns one hit, in chapter 6, on oral cancer: a different Kumar, on the oral management of oncology patients. The only brushing-frequency systematic review DBOH cites is Zimmermann and colleagues, which is about periodontitis rather than caries and appears in chapter 8, on oral hygiene. Chapter 8 asserts “Frequency of brushing is important. Brushing should occur twice daily as a minimum” without citing anything for the caries claim at all.
Full strategy, including the failed queries, in appraisals/searches/ch05-brushing-frequency.md.
The results:
| Comparison | Effect | 95% CI |
|---|---|---|
| Infrequent vs frequent brushers, caries incidence | OR 1.50 | 1.34 to 1.69 |
| Infrequent vs frequent brushers, caries increment | SMD 0.28 | 0.13 to 0.44 |
| Fewer than 2/day vs 2 or more per day | OR 1.45 | 1.21 to 1.74 |
| Fewer than 1/day vs 1 or more per day | OR 1.56 | 1.37 to 1.78 |
| Deciduous dentition | OR 1.75 | 1.49 to 2.06 |
| Permanent dentition | OR 1.39 | 1.29 to 1.49 |
So brushing less often is associated with more decay, consistently, in 33 studies. That is a real finding and it points the right way.
One thing it is not, and I said otherwise in the blog version: it is not direct evidence of a causal effect. Kumar’s included studies are observational, with brushing frequency self-reported and self-selected, exactly like SIGN’s ten cross-sectional studies. The difference between the two bodies of evidence is real but narrower than I made it: Kumar asks the right question, of a synthesised and much larger literature, with a formal appraisal of what it finds. SIGN’s meta-regression asks a different question at the wrong level of analysis. Neither tells you what happens when a person is assigned a second daily brushing.
Now look at the two middle rows, and then at what they cannot tell us.
The review reports two different dichotomies of the same underlying variable: at least once a day against less (OR 1.56), and at least twice against less (OR 1.45). It is tempting to read those as the effect of the first brushing and the effect of the second, and to conclude that the second adds less. I made something close to that argument in an earlier draft, and it does not hold.
The two contrasts are cut at different thresholds, drawn from overlapping but not identical sets of studies and populations, and the review reports no test for trend across ordered categories and no interaction test between subgroups. Two subgroup estimates whose confidence intervals overlap have not been shown to differ, and equally have not been shown to be the same. Reading a marginal effect of the second daily brushing out of them is exactly the kind of over-reading this book objects to elsewhere.
What can be said is narrower and still useful: the review reports no analysis capable of showing a gradient across ordered frequency categories. That absence matters for a specific technical reason.
Kumar 2016 pools observational studies, so under GRADE the body of evidence starts at low certainty.
In my original blog post I stopped there, noted no serious additional problems, and concluded low. That was a misapplication, because GRADE also permits rating observational evidence up, for three reasons, and I never asked whether any applied. Doing it properly:
Large magnitude of effect. An odds ratio of 1.5 is not large by GRADE’s convention, which looks for something in the region of 2 before upgrading. And because caries is a common outcome, the odds ratio materially overstates the risk ratio: in a population where a third of children develop new lesions, an odds ratio of 1.50 corresponds to a risk ratio nearer 1.3. No upgrade.
Dose-response gradient. This is the one that could have applied, and the reason I flagged my earlier assessment as wrong. GRADE wants a gradient across ordered exposure categories. The review does not provide one: it provides two separate dichotomies, cut at different thresholds, with no test for trend and no interaction test, from overlapping but non-identical sets of studies, and its own summary of them is that the odds “differed little” between the two cuts. No upgrade is available on this criterion, because a dichotomy is not a gradient however many of them you report.
Plausible confounding would reduce the observed effect. The opposite is true here. People who brush twice daily are, on average, wealthier, better educated, more likely to attend a dentist and more likely to eat less sugar. Every one of those independently reduces caries. Residual confounding would inflate the apparent benefit of brushing frequency, not suppress it. No upgrade.
Rating down then has to be considered on its own terms, and low is not the floor: GRADE runs to very low, and saying “low is already low” would be a second error on top of the first.
Risk of bias. Exposure is self-reported in every included study, and self-reported brushing is over-reported in a way that is likely to be related to the same social and educational factors that predict caries. That is serious. The review’s meta-regression found methodological quality did not explain the effect estimate, which speaks to consistency across studies but not to a bias they all share.
Inconsistency, imprecision, publication bias. The direction is consistent across 33 studies, the pooled intervals are tight, and I found no strong signal of publication bias in a literature where the finding is the expected one.
An earlier draft stopped at low, on the grounds that rating down for risk of bias after already declining to upgrade for confounding direction would double-count the same problem. That reasoning is wrong and I am replacing it rather than softening it.
Declining an upgrade is not a downgrade already taken. The three upgrade criteria are a route by which observational evidence can rise above its starting level. Not qualifying for that route leaves the body where it started; it spends nothing. There is no penalty banked against which a later downgrade can be double-charged.
The two concerns are also not the same problem. Confounding is that brushing frequency travels with income, education, diet and dental attendance. Measurement error is that the exposure is self-reported, and over-reported in a way likely correlated with those same characteristics. They can each bias the estimate on their own, in the same direction, and a body of evidence can suffer both. Naming confounding under the upgrade question does not immunise the measurement problem from the downgrade question.
Taken properly, exposure misclassification across every included study is a serious limitation beyond what “observational” already assumes, and the honest rating is very low.
Certainty: very low, one level below both my earlier assessment and DBOH’s. That does not change what this chapter concludes about the recommendation, which turns on bundling rather than on the frequency evidence being weak, and it does not change the practical advice. It is recorded here because a book that spends its time on other people’s GRADE domains has to accept a correction to its own.
The verdict
The direction of this evidence is not in doubt. People who brush their teeth more often have less tooth decay, in 33 studies across many countries and age groups. I would be astonished if brushing twice a day were not better than brushing once.
But that is a statement about my expectations, not about the evidence, and the two things this chapter set out to check both come up short. The evidence the guideline cites does not answer the question the advice asks: it is a trial-level meta-regression about fluoride delivery plus ten cross-sectional studies. And the evidence that does answer the question is not cited at all, even though it is a decade old, is in the Journal of Dental Research, and says in its own abstract that the evidence base for twice-daily brushing is weak.
Verdict
- Certainty of evidence
-
Very low, for the frequency component specifically. Pooled observational data starting at low, with no grounds to upgrade on magnitude, dose-response or confounding direction, then downgraded once for self-reported exposure misclassified across every included study.
This is one level below DBOH’s own rating of low, and one level below the rating an earlier draft of this chapter gave. The reason for the change is set out in the GRADE box above: declining an upgrade is not a downgrade already taken, and confounding and measurement error are separate domains.
- Directness to the advice as worded
-
Poor for the cited evidence, better for the uncited evidence. SIGN 138’s meta-regression is indirect on comparison and on level of inference: it compares trials, not people. Kumar 2016 asks the recommendation’s question with the right outcome, and remains observational, so it speaks to association rather than to the effect of assigning a second daily brushing. It is not cited by DBOH.
One qualification on population, since the recommendation quoted at the top of this chapter is the adult one. All but eight of Kumar’s 33 included studies are in infants or children. So the uncited evidence repairs the directness problem on question and outcome while leaving the child-to-adult mismatch that DBOH itself flags for frequency. Al-Sharani 2025, being confined to children under six, does not help with that either.
- Is the strength label defensible?
-
Not on the published reasoning. Using the categories from Chapter 2, this looks like one that “should have been Conditional”: the certainty is low, the cited evidence does not address the component, and no rationale for a discordant Strong recommendation is given.
Two caveats, per Chapter 4. I have not run an evidence-to-decision assessment, so I cannot rule out that a panel weighing values, burden and resources would still land on Strong; a case could be made, since the burden is small and widely accepted. And DBOH does not say which component set the strength for this row, so “the frequency is riding on the fluoride toothpaste” is my inference from the certainty ratings, not the guideline’s statement.
- What would change my mind
- A trial randomizing participants to brush once versus twice daily, holding toothpaste, technique and supervision constant, with caries increment or tooth pain at two years or more. Failing that, a cohort study with objectively measured brushing frequency rather than self-report, adjusted for socioeconomic position and dental attendance, showing a monotone gradient across three or more frequency categories. If either exists and I have missed it, I will correct this chapter.
What this means for you
Brush twice a day. I do, and nothing in this chapter is a reason to stop: the evidence points the right way, the burden is small, and no harm from it has been identified. What the evidence has not established is how much the second brushing adds over the first, and the claim that it has been established is what this chapter disputes. If you are ever in a position to do only one thing, the fluoride toothpaste is the part that has been tested in randomized trials against tooth decay. The frequency is what we have long assumed alongside it.