Background

Community water fluoridation (CWF) is a widely used public health intervention to reduce tooth decay and oral health disparities. However, evidence on its effectiveness is largely outdated and previous epidemiological studies lack quantified fluoride exposure levels. Most studies use binary definitions (fluoridated vs non-fluoridated), which assume consistent exposure at recommended levels (0.7-1.0 mg/L). We examined how moving beyond binary exposure definitions influences estimated associations with severe early childhood caries.

Methods

We conducted a national cross-sectional study of 584,381 New Zealand children undergoing a B4 School Check (2010-2021), linked to weekly water fluoride concentrations based on residential meshblock at assessment. Individual exposure was summarised as mean fluoride concentration from birth to assessment and modelled using categorical definitions and as a continuous measure, using restricted cubic splines. Associations with severe caries were estimated using Poisson generalised estimating equations clustered by meshblock.

Results

Substantial variability in fluoride exposure was observed within areas classified as always fluoridated. Individual mean exposure ranged from 0.06-0.95 mg/L and children spent on average only 68% of early childhood within the recommended range. Categorical models, using <0.3 mg/L as the reference, showed lower risk at moderate exposure (0.3-0.69 mg/L) (RR 0.83, 95% CI 0.81-0.85) but attenuation at levels within the recommended range (≥0.7 mg/L) (RR 0.93, 95% CI 0.92-0.95). Spline models similarly showed a minimum-risk region at moderate concentrations (~0.5 mg/L), with an upward trend at higher concentrations. Higher-risk children, including non-European children and those in more deprived areas, were disproportionately represented at higher fluoride exposure levels.

Conclusion

Binary fluoridation measures mask substantial exposure variability, but alternative exposure definitions introduce new challenges. In observational settings where exposure is non-random and concentrated in higher-risk populations, estimated associations may reflect population structure rather than a true dose-response relationship, potentially leading to misleading inferences about CWF effectiveness.