Investigation Into the Effectiveness of Community Water Fluoridation and Dental Caries Experience in 4-Year-Old Aotearoa New Zealand Children: A National Repeated Cross-Sectional Study From 2010 to 2022.

2.1 Study Design

This was a national repeated cross-sectional study. NZ’s Health and Disability Ethics Committee defined this study as minimal-risk observational research not needing formal ethics committee review or further consents for use of data. Use of the B4SC data was approved by the NZ Ministry of Health (MoH).

2.2 Population Studied

This research used data for NZ children aged 4 years who had their B4SC assessment from July 2010 to June 2022. To be eligible for a B4SC, the age of assessment is meant to be between 4 years and 5 years 7 days, so there is some variation in age of assessment.

2.3 Outcome Measurement

Our primary outcome was dental caries as identified in the “lift the lip” oral health screen conducted as part of the B4SC [19]. This screening is conducted by trained nurses in accordance with the Ministry of Health’s standardised protocol [19]. Nurses receive training in the lift the lip assessment, including the use of clinical guidance and photographic reference examples. Children’s teeth are classified on a scale from one to six, where one represents no visible caries, while any score two or higher meets the criteria for early childhood caries (ECC) at age 4 years [19], and the child is referred for treatment (approximately 15% of NZ 4-year-olds) [14]. Note, this is substantially lower than the prevalence reported from NZ’s Community Oral Health Service at age 5 years (43% in 2024), suggesting it is likely that considerable early or uncavitated caries remains undetected in the age 4 B4SC [1].

2.4 Exposure Assessment

The key assumption in this repeated cross-sectional study design is that fluoride exposure at age four (B4SC assessment) is representative of lifetime exposure to that point. Data were obtained from all TAs about the start and end dates for CWF and any interruptions to CWF greater than 6 months. Any year with an interruption of greater than 6 months was treated as non-fluoridated.

Three categories of fluoridation status were created: (1) entirely fluoridated (where the assessment year and previous 3 years were fluoridated); (2) entirely non-fluoridated (where the assessment year and previous 3 years were not fluoridated); and (3) mixed fluoridation (where at least 1 year in four (assessment year plus three previous years) was discordant (either fluoridated or non-fluoridated)). In the main analysis, mixed fluoridation was recoded as “entirely fluoridated”, while in a sensitivity analysis, all mixed were included as entirely non-fluoridated as well as excluding mixed supplies.

For each year from 2010 to 2022, a fluoridation status variable was generated for each WDZ (based on data from 2007 to 2022). The development of a national spatial dataset of all publicly owned WDZ in NZ has recently been documented [18]. This dataset covers ~88% of the total population, with the remaining population drawing water from their own private supplies [18] In validation analyses, this dataset has demonstrated a high positive predictive value (accurately identifying an individual as being on a public water supply) [20].

Each child was linked to a WDZ based on their meshblock of residence taken as part of the B4SC assessment and used this as a proxy for exposure from age 0 to the year of assessment. A meshblock is the smallest geographic unit used by Statistics New Zealand, which usually includes between 30–60 dwellings and is no larger than 120 dwellings (representative of a city block).

2.5 Sociodemographic Variables

Sociodemographic variables derived from the B4SC data set included sex, age, and ethnicity. Sex was categorised as female and male. Age (in months) was calculated from B4SC assessment and birth dates. Ethnicity was based on MoH protocols where children with multiple identifications were assigned a single ethnicity via the prioritisation hierarchy: (1) Maori; (2) Pacific; (3) Asian; (4) Middle Eastern, Latin American, and African; and (5) European/other.

A meshblock defined level of neighbourhood deprivation was measured using the New Zealand Deprivation Index 2013 (NZDep2013) [21] and NZDep2018 [22] linked based on the child’s recorded residential meshblock at their B4SC assessment via the 2013 and 2018 censuses. The NZDep combines census data relating to income, home ownership, employment, qualifications, family structure, housing, access to transport, and communications into a single measure [21, 22]. Each meshblock is assigned a score from 1 (least deprived) to 10 (most deprived) based on decile splits, which were aggregated into quintiles for our analysis.

Rural/urban classification was determined by linking a child’s residential meshblock to the New Zealand Urban Rural Standard Classification: (1) main urban (>100,000 people); (2) large urban (30,000–99,999 people); (3) medium urban (10,000–29,999 people); (4) small urban (1000–9999 people); (5) rural (200–1000 people); and (6) other rural (< 200 people). Any assessment between 2010 to 2017 was linked to 2013 NZDep or Urban/Rural classifications, while assessments from 2018 onwards were linked to the 2018 classification.

2.6 Statistical Analysis

The Reporting of studies Conducted using Observational Routinely-collected Data (RECORD) guidelines were used to inform reporting of this study [23]. Comparisons between the exposure groups and the sociodemographic characteristics were made using Pearson’s X² test. For the primary analyses, fluoridation exposure was dichotomised to include entirely fluoridated vs. entirely non-fluoridated groups, where WDZ classified as mixed fluoridation were recoded to entirely fluoridated. A secondary sensitivity analysis followed, with exposure dichotomised as entirely fluoridated vs. entirely non-fluoridated groups, where mixed fluoridated WDZ were recoded as entirely non-fluoridated. Because ECC is not a rare outcome, and logistic regression models are known to provide biased estimates of relative risk in such instances, modified Poisson regression models were instead employed [17]. Further, as children were nested within meshblocks, multilevel mixed-effects models were employed with children nested within meshblocks. In these models, meshblocks were treated as random intercepts. Mirroring procedures employed elsewhere [14], sociodemographic and potentially confounding variables were initially considered and sequentially added to develop a main effects model. Next, all 2-factor interaction terms, including those with fluoridation status, were considered and sequentially added to the main effects model until there was no improvement in the Bayesian Information Criterion (BIC) [24]. In the regression models, age was centred at its median (51 months). Reference categories were: female (sex), European/Other (ethnicity), NZDep quintile 1 (least deprived), urban—major (rural/urban classification), and mid-study year (2016) for assessment year. All analyses were performed using Stata SE, version 18.0 (StataCorp LLC, College Station, Texas, USA).

3.1 Study Sample

The raw B4SC file contained assessment data for 732,622 children from 2010 to 2022. Figure 1 shows the inclusion flow chart and exclusion criteria for the study cohort. Children were excluded from the cohort due to not having a valid oral health assessment (n = 77,236, 10.5%); missing address information (n = 22,350, 3.1%); and missing other covariate information (n = 2110, 0.3%). The final analytic sample included 630,926 4-year-old children (86.1% of initial sample).

Overall, included children had a median age of 51 months (4.25 years) at their B4SC assessment and 307,481 (48.7%) were female (Table 1). European/Other was the primary ethnic identity for 50.7% of the sample, 25.9% lived in the most deprived areas, and 84.9% resided within an urban area.

3.2 Fluoridation Exposure

In total, 628 water distribution zones (WDZ) were included in the nationwide spatial dataset. Of these, 483 (76.9%) WDZ were never fluoridated at any point in time (Table S1), while a further 8 (1.3%) WDZ had been fluoridated in the past but had ceased at some point prior to 2007 during the observation period used in this analysis. An additional 106 (16.9%) WDZ were fluoridated for the whole study period. Finally, 31 (4.9%) WDZ had a mixed fluoridation status.

Across these WDZ’s 288,770 (45.8%) children were classified as living in unfluoridated areas, 317,371 (50.3%) in fluoridated areas, and 24,785 (3.9%) in areas of mixed fluoridation status. A greater proportion of children classified within the mixed or unfluoridated exposure groupings were Maori children or living in small urban, rural, or remoted regions, whereas a greater proportion of children classified in the fluoridated or mixed exposure groups lived in the most deprived regions of NZ (Table 1).

3.3 Oral Health

In total, 89,591 (14.2%) of children experienced ECC. Table 2 gives the distribution of ECC status across sociodemographic characteristics overall and partitioned by the fluoridation exposure groups. Clear differential patterns can be identified, with greater rates of ECC associated with higher area-level deprivation, and with Pacific and Maori children experiencing higher rates than European/Other children.

3.4 Examining the Association Between CWF and Oral Health

Crude analysis showed that 48,704 (15.3%) children residing in fluoridated areas had ECC higher than the 37,707 (13.1%) and 3180 (12.8%) seen in the unfluoridated and mixed fluoridated areas, respectively. A crude multi-level mixed effect analysis comparing fluoridated/mixed group against the unfluoridated group yielded an incidence rate ratio (IRR) of 0.92 (95% confidence interval [CI]: 0.91, 0.94) which suggests that living in unfluoridated areas was associated with less caries experience. However, variations in ECC patterns across sociodemographic characteristics between the fluoridation exposure groups are also evident from Table 2. The crude analysis ignores these differing fluoridation exposure and caries experience patterns, behoving statistical adjustment.

3.5 Adjusted Analyses

Initially, a main effects model was derived. Forward selected variables minimising the BIC statistic consisted of ethnicity (BIC = 495368.4), followed by area-level deprivation (BIC = 490650.8), age (BIC = 488908.7), B4SC assessment year (BIC = 488606.9), sex (BIC = 488488.2), and rural/urban classification (BIC = 488456.1). A quadratic term for age was also considered, but resulted in a worsening of the BIC statistic, and was thus not included (BIC = 488458.7). All included variables were associated with caries experience in this main effects model (all p < 0.001).

Next, all 2-factor interactions were considered. Interactions minimising the BIC statistic included ethnicity × NZDep (BIC = 488110.9), followed by fluoridation exposure × ethnicity (BIC = 488081.7), and ethnicity × age (BIC = 488070.3). No other 2-factor interactions yielded a lower BIC statistic, and thus model building terminated. The included interaction terms were also significantly associated with caries experience (all p < 0.001). While all identified interaction terms were retained in the final model, only the fluoridation exposure × ethnicity interaction is presented (Figure 2), as this was the primary interaction of interest.

Figure 2 presents the IRR effect size estimates and associated 95% CI from the final adjusted multilevel mixed-effects modified Poisson regression models (figures used to populate Figure 2 available in Table S2). Due to the significant fluoridation exposure × ethnicity interaction, this implies that the association between CWF and oral health outcomes differed by ethnicity.

For the primary exposure, characterising mixed supplies as entirely fluoridated, the largest estimated effect size was European/Other children (IRR = 1.25; 95% CI: 1.21, 1.29), followed by Asian (IRR = 1.14; 95% CI: 1.10, 1.18) and Maori (IRR = 1.11; 95% CI: 1.08, 1.13) children. However, the impact of fluoridation appeared negligible for Pacific (IRR = 1.03; 95% CI: 1.00, 1.07) and MELAA (IRR = 0.94; 95% CI: 0.84, 1.05) children, once differences in measured sociodemographic variables were accounted for. In the second model, where mixed supplies were treated as entirely nonfluoridated, the same patterns emerged—except the estimated IRR were generally dampened (e.g., European/Other children IRR = 1.21; 95% CI: 1.17, 1.24).

3.6 Population Attributable Fraction

In total, 89,591 children had a positive ‘lift-the-lip’ B4SC screening over the study period. The ethnicity-weighted population-attributable fraction was 4.4% (95% CI 3.7–5.2) suggesting 3942 (95% CI 3315–4659) cases were attributable to non-fluoridation of community water supplies, or 282 cases of ECC detected by ‘lift-the-lip’ in the B4SC screening at age 4 years annually.

4.1 Strengths and Limitations

A major strength of this study is its near whole-of-population coverage, drawing on 13 years of data from a national oral health surveillance programme. The analysis provides additional contemporary evidence on the effectiveness of CWF while controlling for key confounders. The large sample size also enabled investigation of potential interaction effects between CWF and important covariates. The availability of individual-level data also mitigates possible ecological fallacy bias, an issue highlighted by analyses of oral health data and CWF status presented above. The study also used an updated exposure assessment that leveraged national dataset of the spatial extent of WDZ, dates for start and finish of and interruptions to CWF and considered fluoridation status in years prior to the B4SC assessment date.

The primary limitation of this study is its cross-sectional design. The analysis assumes that CWF exposure remains constant from birth to age four, meaning reverse causation cannot be excluded. In a prospective longitudinal cohort study of children in NZ, 34% of the cohort did not move between birth and age four, while 26% had one move and 40% two or more relocations [26] However, only 9% of all relocations in the same cohort at age two were over 50 km, suggesting most children likely retain their CWF exposure as fluoridation status is represented at the city-scale [27] Nevertheless, this mobility introduces potential misclassification bias, likely attenuating effect estimates toward the null, as children in the entirely non-fluoridated group may have received the intervention in part (e.g., CWF from age 0 to 3) and children in the entirely fluoridated group may not have received the intervention until age 4 (e.g., non-fluoridated until age 3). As stated earlier in the discussion, this bias is likely to have affected children of Maori and Pacific ethnicity the most.

Another limitation is the use of fluoridation status as a proxy for fluoride exposure. This approach assumes water supplies are consistently fluoridated at optimal concentrations (0.7–1.0 ppm). While this assumption holds in countries with high compliance, in NZ, water is underdosed approximately 40% of the time, with variation by ethnicity and area-level deprivation [5].