Nine hundred and eleven (911) subjects from eleven (11) schools were approached to participate in this survey. Seventy three (73) subjects did not provide consent to participate. Eight hundred and thirty eight (838) subjects consented to participate, following screening for suitability seven hundred and eight subjects (708) were enrolled onto the study. Subject accountability is detailed in Figure
. Photographic examinations and interviews took place between December 2007 and September 2008. Six hundred and thirty four subjects (634) were included in the study following completion of photographic examinations and interviews. Subjects were excluded from the examinations if they were deemed to be non-lifetime residents, had unsuitable dentition or if inclusion based on the water fluoride analysis results would have created imbalance in the population groups. Additional subjects were removed from the analysis during data checking and are described in Figure
. Subjects were excluded if information from the interview conflicted with demographic data relating to lifetime residency and age at time of examination. Subjects were also excluded if the upper maxillary teeth could not be ascribed a TF score from the photographs – this would have resulted from the presence of restorations, loss of tooth tissue owing to trauma and presence of extrinsic stain. In total five hundred and sixty (560) subjects were available for analysis. There were 298 males (mean age at exam 10.44, range 8–13) and 262 females (mean age at exam 10.48, range 8–13).
Subject accountability illustrating flow of subjects through each stage of the study.
Reproducibility for the photographic image scores was performed on sixty five (65) randomly selected images five (5) months after the original assessments. A weighted Kappa value of 0.80 was obtained (SE 0.05, 95% CI 0.71, 0.89) demonstrating good agreement with the examiners using the full range for TF scores for the images presented. The repeat consensus score for TF was never more than one unit different to the initial assessment.
Descriptive statistics are presented in Table
for the distribution of each independent variable for each of the TF score categories. The data illustrates as the mean values of fluoride concentration in current drinking and current cooking water increase the fluorosis severity increases. For subjects with a TF score of 0 the mean fluoride concentration for drinking and cooking water was 0.35
ppm (SD 0.37) and 0.65
ppm (SD 0.84) respectively. For subjects presenting with TF scores of 4 or higher the mean fluoride content increased to 0.83
ppm (SD 0.90) and 2.23
ppm (SD 1.52) respectively.
Distribution of independent variables for each fluorosis category
The results of the water fluoride analysis provided a more complex range of data than anticipated in the study planning. This necessitated the creation of arbitrary water intervals based upon the distribution of the water fluoride data. Allocation of subjects to water fluoride intervals based on the frequency distribution of cooking water fluoride content resulted in the creation of five (5) intervals cooking water and four (4) corresponding intervals for drinking water. The details of these intervals and the distribution of subjects are illustrated in Table
The variables associated with water interval data demonstrated as the fluoride content of the water increased, greater numbers of subjects presented with fluorosis of increasing severity. This was true of the interval data for current drinking and cooking water derived from the water analysis data and also for the variables created from the interview data. These variables were drinking and cooking water at age three (Drinking water age 3, Cooking water age 3), water used for preparing infant food (Water Infant Food) and water used to reconstitute infant formula (Water formula). This pattern was less clear for the variables relating to oral hygiene practices. Insufficient reliable data were available for the reported history of swallowed dentifrice and was excluded from the analysis. This was largely due to a lack of recall. Where this data were available exploratory analysis suggested no pattern associated with the presentation of fluorosis in this population.
There appeared to be no clear pattern in this population between the severity of fluorosis presentation, the age at which tooth brushing commenced, the frequency of toothbrushing and the fluoride content of toothpaste. This was also true of infant feeding practises.
The overall prevalence of fluorosis in the study population was 70.9% (Table
) with a prevalence of aesthetically significant fluorosis (TF 3+) of 16.8%. To evaluate the effect of differing fluoride levels of both drinking and cooking water on fluorosis severity, data were combined into <0.9
ppm fluoride and >0.9
ppm fluoride categories i.e. grouping together water intervals to produce dichotomous variables. The rationale for these arbitrary cut offs was based upon both the study water interval cut offs (derived from data distribution) and the approximation to historical values for community water supply fluoride levels for caries prevention incorporating climate and fluorosis risk [12
]. The prevalence of fluorosis among subjects consuming drinking and cooking water <0.9
ppm fluoride was 60.6% (10.1% for TF 3+). The prevalence of fluorosis among subjects consuming drinking and cooking water >0.9
ppm fluoride was 85.1% (16.8% for TF 3+).
Prevalence data for fluorosis (accounting for combined drinking and cooking water sources)
Results of the bivariate analysis of each explanatory variable and TF score are presented in Table
.This was for both the TF score (5 categories) and a dichotomous variable based on the presence or absence of aesthetically significant fluorosis (TF 0–2 versus TF 3+).
Bi-variate analysis of each risk factor and TF score (as five categories and dichotomised)
Variables for fluoride content of current drinking and cooking water (obtained from water analysis), content of cooking and drinking water at age 3 (obtained from interview data), water used for infant formula, cooking infant food (all obtained from interview data) were all found to have a significant association with the presentation of fluorosis. This was reflected in the unadjusted odds ratios. For current drinking water interval data the odds ratio for the presentation of aesthetically significant fluorosis was 4.02 (p
0.001; 95% CI 2.12, 7.63) for subjects consuming drinking water with a fluoride content ≥0.9
ppm relative to subjects consuming drinking water <0.2
ppm fluoride. For current cooking water interval data the odds ratio for the presentation of aesthetically significant fluorosis was 6.77 (p
0.001; 95% CI 2.86, 16.02) for subjects using cooking water with a fluoride content ≥1.6
ppm relative to subjects using cooking water <0.2
All of the remaining explanatory variables demonstrated no significant association with the presentation of fluorosis. The variables for toothbrushing frequency, age at which toothbrushing commenced and infant feeding pattern were found not to have significant association with fluorosis score in this population. The one exception was fluoride content of toothpaste which actually demonstrated a decrease in fluorosis with fluoride content of 1000
ppm when compared to fluoride content <1000
ppm. However, this did not achieve statistical significance (p
When all of the variables were entered into a forward stepwise regression analysis the model yielded contained two variables that were the best indicators for the presence of aesthetically significant fluorosis: the fluoride content of the current drinking and current cooking water. However, the attempt to fit a logistic regression model with the continuous variables resulted in the assumptions underlying logistic regression not being upheld. The residuals were strongly related to the fluoride levels for both variables and increased as the water fluoride level increased.
The data were exported to Stata (release 11, StataCorp, TX, USA) for further analysis. A logistic regression model for dichotomised threshold of fluorosis (presence or absence of aesthetically significant fluorosis) with the independent variable for the current drinking water fluoride content coded as water interval data were fitted. The fit improved significantly when the water interval data for current cooking water was added to the model (Likelihood-ratio test, LR χ2
30.09, <0.001). The clustering of the children within schools was also taken into account by using the robust standard errors. This data is presented in Table
. The odds ratio for the presentation of aesthetically significant fluorosis was 3.34 (robust SE 1.22; 95%CI 1.52, 7.04) for subjects consuming drinking water with a fluoride content equal to or greater than 0.9pmm relative to drinking water consumption with less than 0.2
ppm fluoride. The odds ratio for the presentation of aesthetically significant fluorosis was 5.54 (robust SE 3.01; 95%CI 1.91, 16.04) for subjects consuming cooking water with fluoride content equal to or greater than 1.6
ppm relative to cooking water consumption with less than 0.2
Final Logistic regression model for predicting presence or absence of aesthetically significant fluorosis (TF3+), including the clustering of the children in 11 schools
The presence of any interaction between the fluoride level in the drinking and cooking water was investigated. The overall p-value for this was 0.28 and many of the categories were excluded due to collinearity and small numbers of subjects. Table
presents the probability estimates and numbers of subjects for each category when these two variables are cross classified. It can be seen the probability of aesthetically significant fluorosis rises to 0.53 if there is exposure to high levels of fluoride in both drinking (≥0.9
ppm) and cooking water (≥1.6pmm). There was no evidence of an interaction from the probabilities shown here.
Cross-tabulation of the predicted probabilities of having aesthetically significant fluorosis (TF3+) for the fluoride levels in the drinking and cooking water (number of subjects)