What the flossing evidence actually measured

Across 35 randomised controlled trials and 3,929 adult participants, not one measured decay on the surfaces where two teeth touch. That sentence is not a critic’s summary of the Cochrane review on interdental cleaning — it is in the review’s own Main results, in eleven words: “No trials assessed interproximal caries, and most did not assess periodontitis.”

Interproximal caries is the thing floss is supposed to prevent. It is the reason a dentist hands you a length of string rather than telling you to brush harder. And the largest and most frequently cited body of evidence on the habit does not contain a single measurement of it.

This page is about what that review did measure, what it found, and what you are and are not entitled to say afterwards. It is also, in the last section, a correction to one of our own pages.

The review, and what it compared

The paper is Worthington and colleagues, Home use of interdental cleaning devices, in addition to toothbrushing, for preventing and controlling periodontal diseases and dental caries, published in the Cochrane Database of Systematic Reviews in April 2019. Cochrane reviews are the most careful thing in this literature, and this one is careful about its own limits in a way almost nobody quoting it has been.

It pooled 35 trials in 3,929 adults, comparing toothbrushing plus an interdental device against toothbrushing alone, and then comparing devices with each other. The split by device is worth seeing, because “the flossing review” is mostly not about floss:

Fifteen trials, then, on the specific question most readers think this review settled. The remaining comparisons set the devices against each other, which turns out to produce the most useful finding in the whole paper — and the one that never makes a headline.

What it measured instead

Two things, and both are surrogates.

• Gingivitis, mostly by the Löe-Silness gingival index — a nought to three scale a clinician scores by eye at the gum margin — and by the proportion of sites that bleed on probing.
• Plaque, mostly by the Quigley-Hein index, a nought to five scale for how much deposit is visible on a tooth surface after disclosing.

Neither is a bad measurement. Both are genuinely informative about inflammation and deposits this month. Neither tells you anything about a hole in a tooth in four years, and the review does not pretend otherwise.

The reason they were used is not laziness. Gum inflammation responds within weeks and can be scored in a dental chair in a few minutes. Decay between two teeth needs years to become detectable and a properly powered trial to detect it. One of those studies is affordable and the other is not, so the affordable one is what exists.

What it found, in the review’s own words

Here the wording matters more than the numbers, because the wording is where the certainty lives.

Flossing in addition to brushing may reduce gingivitis at one month (standardised mean difference −0.58, 95% CI −1.12 to −0.04, from 8 trials and 585 participants), and at three and six months — all on low-certainty evidence. Results for bleeding sites and for plaque were inconsistent, on very low-certainty evidence.

And then the sentence that gets left out of almost every article citing this review:

Two more details from the same page are worth carrying. Only two of the 35 trials were at low risk of bias once you set aside the blinding problem — which is unavoidable, because you cannot blind someone to whether they are holding a piece of string. And many participants started with a low level of gum inflammation, meaning the trials were frequently looking for improvement in people who did not have much room to improve.

An independent commentary in Evidence-Based Dentistry the same year reached the blunter version: the question of whether interdental cleaning aids improve oral health cannot be answered by this systematic review.

Three different things get called “weak evidence”

This is the part that transfers to every other health claim you will read, so it is worth separating properly. When someone tells you the evidence for something is weak, they could mean one of at least three quite different situations.

1. It was measured properly and it did not work That is a finding. It should change what you do. This is the only one of the three that licenses the sentence “it does not work”.
2. It was measured badly Small trials, short follow-up, self-reported habits, no blinding. You cannot tell either way from this. It is a statement about the researchers, not about the intervention.
3. The outcome you care about was never measured at all Something adjacent was measured instead, and the result got attached to the original question somewhere on the way out of the building.

Flossing and cavities is the third case. The blurring happens downstream, and it happens in both directions: one kind of article uses this review to announce that flossing is pointless, another uses it to announce that flossing is proven. Neither claim is available from a set of trials in which the outcome was never collected.

The three cases also call for different responses from you. If a thing was measured and failed, stop. If it was measured badly, weigh it by the mechanism and by what it costs you. If it was never measured, say so, and stop treating the file as closed.

The same test is worth running on any product page that cites research at you — it is the first thing we do when checking the science behind a supplement, and it is usually enough on its own.

So should you floss?

Yes, and the reasons have almost nothing to do with the trials.

A toothbrush physically cannot reach between two teeth that are touching. That is not in dispute and does not need a trial. Gum disease starts disproportionately in exactly those sheltered, oxygen-poor gaps, and plaque left undisturbed there hardens into something you cannot remove at home. The habit costs a minute, costs almost nothing, and the main way to hurt yourself with it is to be rough.

A direct mechanism, negligible cost, negligible risk, and a body of trials nobody is proud of. That is a good enough reason to do it daily. It is not a good enough reason for anyone to tell you the evidence is overwhelming — in either direction.

The finding nobody quotes

The same review contains something more immediately useful than the argument it is famous for: interdental brushes may be more effective than floss at reducing gingivitis, on low-certainty evidence, at one and three months. Interdental brushes may also reduce plaque more than brushing alone, where flossing’s plaque results were inconsistent.

If the gaps between your teeth are wide enough to take a small tapered brush, that is probably your better tool, and most people find it easier to use properly than a length of string. Tight contacts still need floss, because nothing else fits. The practical version of this, including how to size them, is on how often should you floss.

It is a little bleak that the most actionable sentence in the review is the one that got no coverage at all, while the sentence about weak evidence became a decade of headlines.

The correction we owe our own page

Our flossing page said that when the evidence was publicly called weak, the finding was about the quality of the studies rather than about whether flossing does anything. That is true, and we stand by it. It is also incomplete: it describes the second case above, when the more specific problem is the third one.

The page now says that no trial in the review measured decay between the teeth at all, and that the review’s own authors have asked for trials that do. It is a small edit, and it is the difference between telling a reader the evidence is shaky and telling them exactly what is missing from it. Only the second version gives them anything to work with. Our corrections policy is to say so on the page rather than quietly re-date it.

How to do this yourself, in about ten minutes

1. Find the actual paper, not an article about it The title is usually in the news story. PubMed or Europe PMC will have the abstract of record free, which is enough for everything below.
2. Read the Selection criteria before the Conclusions How long did a study have to run to be included? That single number often tells you which outcomes were physically possible to observe.
3. Find the outcome list in Main results Then ask whether the thing you care about is in it. In this review, it is not, and the authors say so in the second paragraph.
4. Keep the certainty wording attached to the finding “May reduce” on low-certainty evidence is not “reduces”. Reviewers choose those words carefully; everyone downstream throws them away.
5. Check who paid, in the same breath Funding and competing-interest statements are at the end of every paper. Read them next to the finding, not afterwards — that is our standing rule.

Ten minutes, and often enough the answer is not the one the headline gave you.

Our illustrations, including Fig. 1 above, are free to reuse under CC BY 4.0 — the collection is on the diagrams page.

References

1. Worthington HV, MacDonald L, Poklepovic Pericic T, et al. Home use of interdental cleaning devices, in addition to toothbrushing, for preventing and controlling periodontal diseases and dental caries. Cochrane Database Syst Rev 2019;4:CD012018. DOI · PubMed · Funding: see ledger
2. Gallie A. Home use of interdental cleaning devices and toothbrushing and their role in disease prevention. Evid Based Dent 2019;20(4):103–104. DOI · PubMed · Funding: see ledger