33 How dental research should change
Thirty-two chapters of complaint earn one chapter of proposal.
The pattern this book found is not that dental researchers are careless. It is that the field has, over decades, optimised for a kind of study that is cheap, publishable and fast, and that answers a question adjacent to the one patients have. Almost every gap in this book has the same shape.
Five things would fix most of it.
1. Measure what patients feel
The single most common failure in this book is the surrogate outcome.
Chapter 6: fluoride concentration in saliva. Chapter 17: gingival index over one month. Chapter 16: plaque scores after a single brushing. Chapter 19: calculus levels.
Every one of those is a step removed from what a person cares about, which is pain, keeping their teeth, and being able to eat and speak and smile without thinking about it.
The counter-example is instructive. FiCTION (Chapter 26) used episodes of dental pain and infection as its co-primary outcomes, and the result is the most directly useful trial in this book. The INTERVAL trial (Chapter 28) used oral-health-related quality of life alongside caries, and produced one of the two high-certainty answers in this book to a question the public actually asks. The other is scale and polish.
It can be done. It is done, occasionally, usually when a funder insists.
2. Adopt and enforce a core outcome set
The reason Chapter 5 reports a standardized mean difference, which nobody can interpret, is that the included studies measured caries on different indices at different thresholds. Pooling required abstracting away the units.
A core outcome set is an agreed minimum list of outcomes, measured in agreed ways, that every trial in a field reports. Other specialties have them. Cochrane’s oral health reviews repeatedly note the absence of one here: recall that Worthington and colleagues asked future interdental trials to “report participant periodontal status according to the new periodontal diseases classification, and last long enough to measure interproximal caries and periodontitis” (1). That is a core outcome set request in all but name.
Journals and funders could require it tomorrow.
3. Run trials long enough to see the disease
Caries and periodontitis develop over years. A large part of the dental literature runs for weeks.
Chapter 17 is the clearest case: 35 randomized trials of interdental cleaning, and not one measured interproximal caries, because a one-to-six month trial cannot. The evidence gap is a direct consequence of the trial duration, and the trial duration is a consequence of what is affordable and publishable.
This is expensive to fix and there is no way around the expense. But the current arrangement spends a great deal of money on studies that are individually cheap and collectively unable to answer the question, which is not obviously a saving.
4. Fund the boring comparisons
Here is a list of trials that do not exist, each of which would settle a question this book found open, and none of which is scientifically difficult:
- brushing at bedtime against brushing two hours earlier, caries at three years (Chapter 6);
- rinsing against not rinsing after brushing, everything else identical, caries at three years (Chapter 7);
- once against twice daily brushing, with adherence measured (Chapter 5);
- fluoride varnish twice a year against once and against three times (Chapter 10);
- supervised brushing against unsupervised, with toothpaste supplied to both arms (Chapter 15).
Every one of these compares an existing practice against a slightly different existing practice. None involves a new product. And that is precisely why none has been done: there is no commercial sponsor for a trial whose result cannot be put on a package.
Chapter 16 found that the powered toothbrush literature is large and substantially industry-financed. The contrast is the argument. We know a great deal about which brush to sell you and remarkably little about how to use one.
Public funders exist for exactly this gap. FiCTION and INTERVAL were both publicly funded, both answered questions no company would ask, and both are the best studies in this book.
5. Publish the reasoning, not just the rating
This is the recommendation aimed at guideline developers, and it is the smallest of the five.
Delivering Better Oral Health already does more than most: chapter 13 publishes an evidence statement and a certainty rating for most recommendations, which is the reason this book could be written at all. Three further steps would close nearly every gap it found.
Label components, not bundles. Chapter 3 found 16 recommendations containing two or more separable instructions, 7 of them Strong. The certainty information already exists per component in chapter 13. Carrying it into the table people read would cost a column.
Say which component the strength label came from. The methods state that strength is set “on the main component”. Naming that component in each row would remove the need for a reader to guess, and would have removed most of Chapter 3.
Show the evidence-to-decision reasoning where a recommendation is discordant. Chapter 21 argued that a Strong recommendation with no direct evidence can be entirely legitimate. What is missing is one sentence explaining why. “No specific evidence regarding asking about smoking; however, it is an integral part of the Ask, Advise, Act pathway” is half an argument; the half that would let a reader check it is the half omitted.
The trial I want to run
I should declare an interest, since Chapter 4 says to.
When I took a course in the design of randomized controlled trials, I chose as my project the trial that Chapter 17 shows still does not exist: a randomized comparison of interdental cleaning plus toothbrushing against toothbrushing alone, long enough and powered to measure interproximal caries rather than a gingival index at one month.
It remains, as far as these searches found, the largest single hole in everyday oral hygiene advice. Thirty-five trials, nearly four thousand participants, and nobody has measured the outcome that the advice is actually about.
The trial design I worked up for that course is not reproduced here. It should be, and it will be in a later revision: a book that spends a chapter demanding that other people run the missing trials should show its own protocol rather than allude to it.
That is an outstanding item rather than a completed one, and this note is here so that the omission is visible rather than quietly absent.
The shape of the problem
Step back and the pattern across all thirty-two chapters is consistent.
We have excellent evidence about substances: fluoride toothpaste, fluoride varnish, sealants. Three of the strongest recommendations in the book, all things that can be manufactured and sold.
We have weak evidence about behaviors: when to brush, how long, how often, whether to rinse. All free, all unsellable, all universally recommended.
Interdental cleaning sits awkwardly across that line and I should not pretend otherwise, since floss and interdental brushes are manufactured products with a commercial literature behind them. What that literature measures, though, is gingivitis over weeks rather than caries over years, which is the same pattern in a different form.
And the things clinicians do routinely went untested until somebody publicly funded a trial. When that happened the answers were not uniform, and it is worth being exact rather than tidy. Recall intervals and scale and polish produced high-certainty findings of little or no difference. Restoring baby teeth produced a trial that failed to show one approach superior to another, which is not the same as showing that filling does not work. Supervised brushing has no conclusive independent effect established, which is not the same as either.
That distribution is not what you would design if you were allocating research effort to what patients need to know. It is what you get when the market pays for research and the questions with no product attached wait their turn.
Fixing it does not require a methodological breakthrough. It requires deciding that the boring questions are worth funding, and then funding them.