26 Should children’s baby teeth be filled?
Baby teeth fall out. So when a three-year-old has decay in a molar that will be gone by age eleven anyway, what should a dentist do?
For most of the twentieth century the answer was obvious: drill out the decay and put in a filling, as you would in an adult. Then somebody ran the trial.
The trial
FiCTION was a multicentre, three-arm, patient-randomized controlled trial in UK primary dental care (1). Children aged 3 to 7 with at least one primary molar with a dentinal carious lesion were randomized 1:1:1 to:
- C+P, conventional management: complete removal of carious tissue and a filling, plus prevention;
- B+P, biological management: sealing the decay in rather than removing it, plus prevention;
- PA, prevention alone: diet advice, plaque removal, fluorides and fissure sealants, with no restorative treatment at all.
1,144 children were randomized, by 72 general dental practitioners, with a median follow-up of 33.8 months. The co-primary outcomes were the right ones: the proportion of children with at least one episode of dental pain or infection, and the number of such episodes. Not surface counts. Not radiographic scores. Pain and infection, which is what a parent actually cares about.
What it found
| Arm | Proportion with ≥1 episode of pain or infection | Mean episodes (SD) |
|---|---|---|
| Conventional filling + prevention | 42% | 0.62 (0.95) |
| Biological (seal-in) + prevention | 40% | 0.58 (0.87) |
| Prevention alone | 45% | 0.72 (0.98) |
Against conventional treatment, the adjusted risk difference was −2% (97.5% CI −10% to 6%) for biological management and 4% (−4% to 12%) for prevention alone. For the number of episodes, adjusted incident rate ratios were 0.95 (0.75 to 1.21) and 1.18 (0.94 to 1.48).
The trial’s conclusion, quoted in full because the two clauses I trimmed in an earlier draft both matter: “there was no evidence of a difference among the 3 treatment approaches for incidence or number of episodes of dental pain and/or infection experienced by these participants with high caries risk and established disease.”
Who was eligible matters more than any of this. The trial excluded children who “presented with dental pain and/or sepsis” at recruitment, along with children under 3 or over 7 and those whose dental management needed special consideration. So FiCTION is a trial of what to do about decay that has reached the dentin and is not yet hurting. It says nothing about a child who is already in pain or has an infection, and nothing about a tooth that has been left until it is. Read the next section with that boundary in mind.
Two things about that result deserve emphasis.
Roughly 40% of children in every arm had pain or infection. Whatever you do, including drilling and filling, a large minority of children with decayed baby teeth end up in pain. That is a sobering number and it is the finding that gets least attention.
Prevention alone was not clearly worse. The confidence interval for prevention alone against filling runs from 4% better to 12% worse. That does not establish equivalence, and the point estimate does favor treating. But it is a long way from the confident assumption that filling baby teeth prevents suffering.
Chapter 19 drew a distinction that matters again here. A precise estimate centered on zero is a finding. A wide interval containing zero is an absence of a finding.
FiCTION is closer to the second. The risk difference interval for prevention alone spans −4% to +12%, which is compatible with prevention alone being slightly better and with it being meaningfully worse. The incidence comparison was the trial’s powered analysis; episode count was added as a co-primary outcome in 2017, near the end of follow-up, and its comparison is explicitly exploratory and unpowered. Either way the trial did not detect a difference, and did not demonstrate equivalence; the authors do not claim it did. They say superiority “could not be concluded”, which is careful and correct.
Risk of bias. The trial states plainly: “Parents, children, and dentists were not blind to allocated arm.” Blinding is impossible when one arm involves drilling. For a subjective outcome like reported pain, that matters. Against it: this was a large, registered, multicentre trial in real general practice with a patient-important primary outcome, which puts it among the best-designed studies anywhere in this book.
Attrition, and a figure I got wrong. 1,144 were randomized and 1,058 attended at least one trial visit and formed the intention-to-treat set. An earlier draft called that ratio “92.5% retention over nearly three years, which is good.” It is not retention. It is the proportion who turned up at least once, which is an entry criterion for the analysis rather than a measure of follow-up.
The retention figure is in the NIHR report and is much lower: “There was no evidence of differential attrition, with 67% attending a final trial visit” (1). A third of the analysed children did not reach the end.
I first wrote that the loss being even across arms “protects the comparison”, which is more reassurance than equal percentages can buy. Equal attrition rates are not equal missing information. Two arms can lose the same number of children while losing very different children: if one arm happens to lose those who went on to have pain and the other loses those who did not, the observed difference moves even though the true risks are identical and the percentages match.
What the authors actually report is no evidence of differential attrition, which is worth having and is not the same as no missing-data bias. Whether the missingness matters depends on why those children left, and that is not something the report can fully establish. It widens the uncertainty around every estimate in Table 26.1 rather than leaving it undisturbed.
What the guideline does with this
Here is the interesting part. Search Delivering Better Oral Health’s summary tables for a recommendation about restoring carious primary teeth, and you will not find one.
That is not an oversight. DBOH is a prevention toolkit; the management of existing decay belongs to SDCEP’s clinical guidance and to NICE. So this chapter, like Chapter 24, audits a question rather than a recommendation.
But the absence is worth noticing for a different reason. Every chapter of this book has been about prevention advice whose evidence is thinner than its label suggests. FiCTION is the reverse case: a large, well-conducted, patient-randomized trial of a treatment, with a patient-important outcome, that found the long-standing default was not clearly better than the alternatives, including a prevention-first strategy with no planned restoration of the eligible lesions.
The dental profession has been considerably slower to change practice in response to that than it has been to repeat prevention advice with much weaker support.
PubMed, 21 August 2026. I have the FiCTION main report, the protocol, the NIHR Health Technology Assessment monograph, and Freeman’s associated qualitative work, all supplied during writing.
I have relied primarily on the main trial report for the effect estimates above. The HTA monograph contains the full economic analysis, which I have not audited and which is relevant to any policy conclusion.
Full record in appraisals/searches/ch26-fiction.md.
Verdict
- Certainty of evidence
-
An earlier draft of this verdict graded “no difference was detected”, which is a fact about whether an analysis crossed a threshold rather than a statement about an effect. GRADE rates certainty in an estimate, so the estimate has to come first.
For prevention alone against conventional management, the adjusted risk difference for at least one episode of pain or infection is 4 percentage points worse, 97.5% CI 4 points better to 12 points worse. Certainty in that estimate is low: one large, well-conducted, pragmatic randomized trial, downgraded once for the unavoidable lack of blinding with a subjective primary outcome and once for imprecision, because the interval spans differences that would change what a reasonable parent chose. A 12 percentage point disadvantage would matter; so would 4 points of benefit.
The trial did not establish superiority of conventional management, and it did not establish that the strategies are interchangeable. Those are different findings and only the first was tested.
Stated that way on purpose. An earlier draft said the three approaches “produce similar rates,” which converts a failure to show superiority into a demonstration of equivalence. No equivalence margin was set, and the interval for prevention alone against filling still admits a 12 percentage point excess. The trial’s own report says it plainly: “‘no evidence of a difference’ does not mean that there was no difference, or that the three treatment approaches were equivalent.”
- Directness to the advice as worded
- Excellent. Right children, right teeth, delivered in ordinary general dental practice rather than a university clinic, with pain and infection as the primary outcomes rather than a surrogate. This is what the trials in Part II should have looked like.
- Is the strength label defensible?
- Not applicable; DBOH makes no recommendation here. My conclusion is that the evidence does not support a confident default in favor of conventional filling, and does support a genuine discussion with parents about the options.
- What would change my mind
- Longer follow-up. Three years takes a five-year-old to eight, and some primary molars are retained until eleven or twelve. A difference could emerge later. Also: the trial reports pain and infection, but a child’s experience of repeated treatment visits, and of a general anaesthetic if things go badly, is part of the comparison and needs the qualitative and economic work to be read alongside the effect estimates.
What this means for you
If your child has decay in a baby tooth, there is more than one defensible answer. If a dentist presents filling as the only option, it is reasonable to ask about sealing the decay in and about a prevention-first approach, and what FiCTION found about pain and infection under each.
First, the boundary. If your child’s tooth already hurts, or the gum around it is swollen, none of what follows applies. Children who arrived in pain or with an infection were excluded from this trial, so it has nothing to say about them. That is a case for being seen, not for waiting.
Within that boundary, the trial found no evidence of a difference in the rate of pain and/or infection between drilling and filling, sealing over, and a prevention-first strategy, in children aged 3 to 7 at high caries risk whose lesions were not yet causing symptoms. Around 40% of children had pain or infection in every group. What it did not establish is that the three are equivalent: the interval for prevention alone against filling still runs from 4 percentage points better to 12 points worse, and 12 points would matter.
That 40% is the figure to take seriously. Once decay has reached the dentin in a baby tooth, none of the approaches tested reliably prevented trouble, which is an argument for everything in Part II of this book rather than an argument about fillings.
Ask what the options are, ask what happens if you wait, and expect a real conversation rather than an automatic appointment for a filling.