ANÁLISIS EXTERNO
ENCUESTAS REALIZADAS EN LOS DIFERENTES ACUEDUCTOS.
Based on 28 studies included in the topical fluoride reviews and randomly selected for assessment of reproducibility and agreement between two reviewers, interrater reliability was excellent (89%) for both allocation concealment and blinding, and agreement was good for allocation (kappa = 0.61) and very good for blinding (kappa = 0.73). There was a considerable difference in the quality of the studies in this review (using the reported information and additional information obtained from investigators).
ALLOCATION CONCEALMENT
Eighteen included trials were described as randomized and assigned code B as no description was provided on how the 'random' allocation was done, and seven trials were considered to be quasi randomized and assigned code C. None of the trials which described the randomization process or whose investigators provided further information in answer to our enquiry could be assigned code
Fluoride gels f o r preven tin g dental caries in children and adolescents (THESIS C 19
A (adequate concealment of allocation fully described).
BLINDING
Double-blinding was described in 14 trials (score A), single-blinding (blind outcome assessment but no placebo used) was described in six trials (score B) and blind outcome assessment likely/unclear was indicated in five trials (score C). Eleven of the 14 trials described as double-blinded were scored B for allocation concealment, and two of the five trials where blind outcome assessment was likely/unclear used a placebo control group.
FOLLOW-UP
All the participants included in the final analysis/present at the end o f each study, as a proportion of all the participants present at start in all studies was 72% (6286 analysed out of 8739 randomized), excluding the six studies with no data on participants randomized to relevant groups. Dropout rates could be obtained from all but two of the 25 included studies. There was considerable variation in dropout rates ranging from 8% at one year to 55% at three years.
Where this information was supplied, the most common reason for attrition was that participants were not available for follow-up examination at the end of the study; exclusions based on presence in all follow-up examinations were reported in nine trials, and exclusions based on compliance were reported in three trials. Other reasons for exclusions (when given) included characteristics of participants that should have been used as eligibility criteria before randomization (use of orthodontic bands, lifetime exposure to fluoridated water). Only one trial reported the numbers excluded according to reason for attrition.
OTHERS
Type of randomization: stratified randomization was used in 12 trials (but there was no mention of use of blocking).
Baseline comparisons and handling of any differences: two of the trials described as 'balanced' (for which randomization may have succeeded to produce nearly exact balance) did not report the actual values for the baseline characteristics (such as initial caries levels). Some degree of
imbalance was reported in four trials (for characteristics considered most influential, usually initial caries levels) and either described as not significant or adjustment mentioned to have resulted in trivial differences in effect estimates.
Objectivity/reliability o f primary outcome measurement: diagnostic methods used (clinical or radiographic) were described in all studies, but thresholds/definitions used for caries and monitoring of diagnostic errors were not always clearly described (see Notes' in the 'table of included studies' for methodological features assessed).
Results
EFFECT OF FLUORIDE GEL ON DENTAL CARIES INCREMENT
The effects of fluoride gels on dental caries increment were reported in a variety of different ways in the included studies. Where appropriate and possible these have been combined to produce pooled estimates as described in the methods section. The results are reported separately here for: (1) Decayed, Missing and Filled Surface Prevented Fraction (D(M)FS PF); (2) Decayed, Missing and Filled Teeth Prevented Fraction (D(M)FT PF); (3) D(M)FS and D(M)FT pooled using a standardised mean difference (SMD); (4) effects on deciduous dentition.
Fluoride gels f o r preven tin g dental caries in children an d adolescents (THESIS C 20
Standard deviations of mean caries increment data (new D(M)FS) were (partly) missing in four of the 23 studies which contributed data (Abadia 1978; Bijella 1981; Mestrinho 1983; Ran 1991). From the analysis of the 179 available treatment arms for the topical fluoride reviews with complete information (as of October 1999) we derived a regression equation log (SD caries increment) = 0.64 + 0.55 log (mean caries increment), (R-squared = 77%). This equation was used to estimate missing standard deviations from mean D(M)FS increments for the meta-analyses. Similarly, this same regression equation was used to estimate missing standard deviation data for two of the 10 trials reporting D(M)FT data (Bijella 1981; Mestrinlio 1983). No pooled estimate of the effects of fluoride gels on caries increment in deciduous teeth could be produced as neither of the two studies contributing data provided standard deviations of mean caries increment (new defs).
Effect on tooth surfaces; D(M)FS prevented fraction
For all 23 trials combined, the D(M)FS prevented fraction pooled estimate was 0.28 (95% Cl, 0.19 to 0.37; p<0.0001), suggesting a substantial benefit from the use of fluoride gel. Substantial
heterogeneity in results could be observed graphically and statistically (Q= 135 on 22 degrees of freedom, p<0.0001).
Meta-regression, subgroup and sensitivity analyses: D(M)FS prevented fraction
Univariate meta-regression suggested no significant association between estimates of D(M)FS prevented fractions and the pre-specified trial characteristics: baseline levels of caries, background exposure to other fluoride sources, background exposure to fluoridated water, background exposure to fluoride toothpaste, gel application mode (operator/self), gel application self-applied method (tray/ brush), and fluoride concentration. There was an association of frequency of
application as well as of'total intensity of application per year' (frequency x concentration) with the prevented fraction, but both became non-significant when the trial of Englander 1967, a study with high influence, was excluded from the analysis.
Further univariate meta-regression analyses showed no significant association between estimates of D(M)FS prevented fractions and length of follow-up (duration of study), allocation concealment (random/quasi-random), blinding of outcome assessment (blind/blind likely or unclear) or drop-out rate. However, the pooled estimated treatment effect was 19% (95% Cl, 5% to 33%; p=0.009) greater in trials with no treatment rather than placebo control groups. This association between type of control group and D(M)FS prevented fraction remained significant after excluding the trial of Englander 1967. The pooled estimate of treatment effect on D(M)FS PF from the nine trials with a no treatment control group was 0.38 (95% Cl, 0.24 to 0.53; p< 0.0001), while that from the
14 placebo-controlled trials was 0.21 (95% Cl, 0.14 to 0.28; p< 0.0001). There remained
statistically significant heterogeneity in the analysis of the 14 trials with placebo control groups (Q= 22.52 on 13 degrees of freedom, p= 0.05) but this was substantially less than that observed when all trials were included in the meta-analysis. Although this was a post hoc analysis and thus should be viewed with caution, we have decided to present the results of the D(M)FS PF meta-analyses subgrouped by type o f control group.
For each study, the D(M)FS prevented fraction and 95% CIs can be viewed in the Tables of other data; the results o f the random effects meta-analyses of D(M)FS prevented fractions (performed in Stata) are given in Additional Table 01: Meta-analyses of prevented fractions. Forest plots showing the effects of fluoride gel (prevented fractions and 95% Cl) on D(M)FS increments resulting from these meta-analyses, subgrouped by type of control group, are available on the Cochrane Oral Health Group web site (www.cochrane-oral.man.ac.uk). In addition, forest plots showing the effects o f fluoride gel (PFs and 95% CIs) on D(M)FS increments resulting ffom a meta-analysis
Fluoride gels f o r preven tin g dental caries in children and adolescents (THESIS C 21
stratified by type of control group are available in the 5th review in this series (whose results invigorate the discussion of the influence of this factor on effect estimates): Topical fluoride (toothpastes, mouthrinses, gels or varnishes) for preventing dental caries in children and
adolescents (Comparison 01, Sub-categories 02 for placebo-controlled gel trials, and 05 for non placebo control gel trials).
The association between type of control group and D(M)FS prevented fraction also remained significant when each one of the above investigated potential effect modifiers were included in bivariate meta-regression analyses, with or without the trial of Englander 1967; the exceptions occurred for adjustment by self-applied method of application, fluoride concentration and intensity of application, when the association between prevented fraction and type of control group became non-significant, irrespective of exclusion or not of the trial of Englander 1967, and for adjustment by blinding of outcome assessment if the trial of Englander 1967 was excluded. While an
association between blind outcome assessment and prevented fraction not shown when adjusting for type of control group appeared after excluding the trial of Englander 1967 from the analysis, associations between (operator/self applied) mode of gel delivery and prevented fraction and between application frequency and intensity and prevented fraction were shown when adjusting for type of control group, but did not remain after excluding the trial o f Englander 1967 from the analysis.
Meta-regression results for all potential effect modifiers, including the Englander study and in each case adjusting for type of control group are given in Additional Table 02: Random effects meta regression analyses of prevented fractions: D(M)FS. These results must be interpreted with caution given the observational nature of the comparisons and the large number of comparisons made. Note that differences between subgroups from meta-regression may differ from differences
between separate meta-analyses in separate subgroups (Additional Table 01) due to an assumption of similar residual heterogeneity in the meta-regression.
In order to illustrate the magnitude of the effect, numbers of children needed to treat (NNT) to prevent one D(M)FS were calculated based on the pooled D(M)FS prevented fraction and on the caries increments in the control groups of the placebo-controlled trials, which ranged from 0.17 to 5.4 D(M)F S/year. The overall caries-inhibiting effect (%PF) derived from the pooled results of the 14 trials with placebo control groups was 21% (95% Cl, 14% to 28%). In populations with a caries increment of 0.2 D(M)FS per year (at the lowest end of the results seen in the included studies), this implies an absolute caries reduction of 0.042 D(M)FS/year, equivalent to an NNT of 24 (95% Cl, 18 to 36). In populations with a caries increment of 2.2 D(M)FS per year (at the mid range of the results seen in the included studies), this implies an absolute caries reduction of 0.46 D(M)FS/year, equivalent to an NNT of 2 (95% Cl, 1 to 3).
Funnel plot: D(M)FS prevented fraction
A funnel plot of the 23 included studies reporting D(M)FS PFs indicated a relationship between prevented fraction and precision (related to sample size). The funnel plot is asymmetrical and the Egger formal test for asymmetry (Egger 1997) suggests statistically significant asymmetry
(intercept (95% Cl) = -2.7 (-5.08 to -0.31) (p= 0.03)). If this was a reflection of publication bias it would imply that small studies with especially large beneficial effects of fluoride gel were missing. The clinical significance of this result is unclear and appears to be related to the effects of two outliers, a small study suggesting a large deleterious effect of fluoride gel and a large study suggesting the largest positive effect.
Fluoride gels f o r preventing dental caries in children an d adolescents (THESIS C 22
oral.man.ac.uk). In addition, a funnel plot of the 23 trials comparing fluoride gel with placebo/no treatment is available in 'Additional Figure 01', where D(M)FS standardized mean differences are ploted against standard errors (see 'Alternative treatment effect measure' below).
Effect on whole teeth; D(M)FT prevented fraction
Ten trials reported data which allowed the calculation of the D(M)FT prevented fraction. All of these trials are also included in the analysis of D(M)FS PF. The pooled estimate of D(M)FT prevented fraction was 0.32 (95% Cl, 0.19 to 0.46; p< 0.0001). There was substantial heterogeneity between trials (Q= 102.50 on 9 degrees of freedom, p< 0.0001).
As with the estimates o f the effects of gels on D(M)FS PF, a meta-regression suggested that the estimates from trials using no treatment as opposed to placebo controls were substantially higher, in this case by 26% (95% Cl, 4% to 48%; p=0.02). The pooled estimate of the D(M)FT prevented fraction from the six trials with a no treatment control group was 0.43 (95% Cl, 0.29 to 0.5; p< 0.0001), while that from the four placebo-controlled trials was 0.18 (95% Cl, 0.09 to 0.27; p< 0.0001). When the meta-analysis was confined to those trials with placebo controls, there remained no statistically significant heterogeneity (Q= 3.18 on 3 degrees of freedom, p= 0.37).
For each study, the D(M)FT prevented fraction and 95% Cl can be viewed in the Tables of other data. The results of the random effects meta-analyses of D(M)FT prevented fractions performed in Stata are also presented in Additional Table 01 : Meta-analyses of prevented fractions.
Alternative treatment effect measure: Standardized mean difference (SMD)
Due to the character of D(M)FS data, mean caries increments are closely related to their standard deviations (they are about the same). Thus, meta-analyses using standardised mean differences (the difference between two means divided by an estimate of the within group standard deviation) yielded materially similar results to those using prevented fractions (the difference in mean caries increments between the treatment and control groups divided by the mean increment in the control group). We therefore decided to present D(M)FS and D(M)FT SMDs in RevMan/Metaview, grouped by type of control group (placebo/no treatment) since it was not possible to present the main outcome analyses with prevented fractions. The results of these analyses are very similar to and consistent with those reported above.
For the 14 placebo-controlled trials the pooled D(M)FS SMD estimate was 0.20 (95% Cl, 0.10 to 0.29; p< 0.0005). There was heterogeneity between trials (chi-square = 29.55 on 13 degrees of freedom, p< 0.0055). The pooled estimate in the nine trials with no treatment control groups was 0.46 (95% Cl, 0.27 to 0.65; p< 0.0001) with substantial heterogeneity (chi-square = 44.42 on 8 degrees of freedom, p< 0.0001).
The pooled estimate of D(M)FT SMD based on the four placebo-controlled trials was 0.19 (95% Cl, 0.09 to 0.29; p< 0.0002). There was no statistical evidence of heterogeneity (p= 0.43). The pooled estimate o f D(M)FT SMD based on the six trials with no treatment controls was 0.73 (95% Cl, 0.32 to 1.13; p< 0.0004), and heterogeneity was substantial (chi-square = 76.81 on 5 degrees of freedom, p< 0.0001).
Effect on deciduous dentition (defs increments)
Neither of the two trials reporting caries increment in deciduous tooth surfaces provided standard deviations or data from which these could be derived and their results were therefore not pooled. The trial of Englander 1978 reported equivocal results (no demonstrated effect of fluoride gel);
Fluoride gels f o r preventing dental caries in children and adolescents (THESIS C 23
while Treide 1988 reported an effect (defs PF) of fluoride gel of 0.39 (Cl not obtainable).
EFFECT OF FLUORIDE GEL ON OTHER OUTCOMES
Few trials report data for other relevant outcomes.
Proportion developing new caries and proportion not remaining caries-free
The only trial reporting on the proportion of children developing one or more new caries
(Gisselsson 1999) reported a risk ratio (RR) of 0.82 (95% Cl, 0.68 to 0.99). This is equivalent in the study population to an NNT to prevent one child from developing caries of 8 (95% Cl, 4 to
100). Two trials reported results on the proportion of children not remaining caries-free
(Gisselsson 1999; Englander 1978). The pooled estimate (random effects meta-analysis) of the risk ratio was 0.66 (95% Cl, 0.45 to 0.97; chi-square for heterogeneity 0.55 on 1 degree of freedom, p= 0.46). This corresponds to an NNT to prevent one child ceasing to be caries-free of 7 (95% Cl, 4 to 50) in a population with a caries risk the same as that found in the control groups in these trials.
Adverse symptoms (nausea/ vomiting)
Only two trials reported usable data on adverse events (Mestrinho 1983; Hagan 1985), but one of these had no events in both arms. The pooled estimate of the risk difference between the gel and placebo arms was 0.01 (95% Cl, -0.01 to 0.02; chi-square for heterogeneity 0.8 on 1 degree of freedom, p= 0.37), ie. marginally favouring the Placebo/No-treatment (PL/NT) arm, although the results were consistent with no difference.
Unacceptability of treatment (dropouts/exclusions)
The pooled estimate of the relative risk of dropping out from the gel as opposed to the non treatment arm in the six non-placebo controlled trials that reported dropouts was 1.09 (95% Cl, 0.70 to 1.72). There was substantial heterogeneity in these results (chi-square = 28.19 on 5 degrees of freedom, p< 0.0001).