Water fluoridation schemes in England
PWS are delivered through distinct water supply zones (WSZs). Each WSZ is defined by either a single point of water supply or, if there are multiple water supply sources of a similar nature and treatment, WSZs are ascribed to permanent resident populations of 100,000 or fewer (Drinking Water Inspectorate 2016b). The number and size of WSZs are reviewed annually by water undertakers and the Drinking Water Inspectorate (DWI, the water quality regulator for England and Wales) and, from 2004, the permitted number of residents per WSZ was increased which contributed to a subsequent reduction in the number of WSZs (forming larger zones with greater populations).
Water companies have a duty to monitor the fluoride concentration of PWS in all of the WSZs they supply and provide these monitoring data to the DWI. In addition to the upper limit, prescribed concentration value (PCV) of 1.5 mg/L fluoride concentrations in fluoridated areas should reach the 1.0 mg/L target (Drinking Water Inspectorate 2016a). Sampling points to monitor fluoride concentrations within the WSZs are randomly chosen (typically consumers’ taps) and must be representative of the WSZ as a whole (Statutory Instrument 2018 no. 647 2018). Samples may also be taken from ‘water supply points’, such as service reservoir outlets, treatment works outlets or blending points. These supply points may supply more than one zone, provided there is no substantial difference in the concentration between the supply point and consumers’ taps in the zone (Drinking Water Inspectorate 2016a). Sampling frequency is determined by WSZ factors such as population size and daily volume of water supplied (for supply point samples) (Statutory Instrument 2018 no. 647 2018). Regulations establish mandatory minimum standards for accuracy and precision of fluoride concentration testing (Statutory Instrument 2018 no. 647 2018). The limit of detection must be within 10% of the prescribed concentration (1.5 mg/L), which would therefore be 0.15 mg/L. However, this is only a minimum standard which we hypothesised real test performance would outperform, because a prior description of water fluoride concentrations in England and Scotland revealed 48% of areas had fluoride concentrations < 0.1 mg/L (with a minimum value of 0.04 mg/L) (Blakey et al. 2014). Though the DWI has maintained routine fluoride PWS monitoring data since 1995, digitised geographic WSZ-boundary data (shapefiles) have only been collated since 2004.
Rationale for using PWS fluoride concentration as a proxy for total fluoride intake in population exposure models
Fluoride is readily and predictably absorbed into the body via the gastrointestinal tract, and this is the main mode of absorption (International Programme on Chemical Safety (Environmental Health Criteria 227) 2002). Thus, fluoride in drinking water is generally bioavailable. This is unlikely to be affected by water hardness at concentrations of around 1 mg/L (Maguire et al. 2005). Available evidence, though limited in extent, strongly suggests, in terms of chemistry and bioavailability, there is no important difference between added and ‘natural’ fluoride occurring from geological sources (Maguire et al. 2005; Jackson et al. 2002). Drinking water with more than 0.3 mg/L of fluoride is amongst the main sources in human total fluoride intakes, particularly at higher fluoride concentrations (e.g. > 0.7 mg/L) typically seen in fluoridated supplies (‘Opinion on critical review of any new evidence on the hazard profile, health effects, and human exposure to fluoride and the fluoridating agents of drinking water’ 2011; Zohouri et al. 2006a). Fluoride water concentrations therefore correlate with human biomarkers of exposure such as urine (Till et al. 2018; Zipkin et al. 1956) and blood plasma or serum (Rafique et al. 2012). For example, maternal urinary fluoride (MUF) concentrations, used as a biomarker of fluoride intake, were almost twice as high in pregnant women living in fluoridated communities than those in non-fluoridated communities in Canada (Till et al. 2018). Log MUF concentrations increased linearly with increasing water fluoride concentration, which was the strongest predictor of urinary fluoride concentration (Till et al. 2018). This study relied upon spot urinary fluoride measurements; these have been validated against 24 h urinary fluoride measurement (recommended for definitive estimation) in children (Zohouri et al. 2006b), but not extensively used in adults. However, there was consistency between repeated spot samples (Till et al. 2018). Living in communities with CWF has also been demonstrated to be a strong determinant of spot urinary fluoride concentrations in the 3- to 79-year-old population in Canada on a population survey, with age having little effect, supporting the validity of spot urinary assessment across age groups and the use of CWF as a proxy for total fluoride intake (McLaren 2016). Finally, the prevalence and severity of dental fluorosis (a relatively specific indicator of increased fluoride consumption) are higher amongst children and young people in fluoridated regions (Pretty et al. 2016; Beltran-Aguilar et al. 2010; Iheozor-Ejiofor et al. 2015), and dental caries prevalence and severity lower (Public Health England 2018; Iheozor-Ejiofor et al. 2015), meaning there is some face validity to using PWS fluoride concentrations as a proxy for total intake.
Some potential limitations to our rationale should also be considered. Alternative fluoride sources, for example from tea or fluoridated dentifrices, may also be important determinants of total fluoride at high intakes (particularly dentifrices in children (Zohouri et al. 2013)), but their relative contribution to total fluoride intake is generally lower when water fluoride concentrations reach levels typically targeted by fluoridation schemes (‘Opinion on critical review of any new evidence on the hazard profile, health effects, and human exposure to fluoride and the fluoridating agents of drinking water’ 2011; Zohouri et al. 2006a). Additionally, globalised production and importation of foodstuffs may modify fluoride intake independent of local water fluoride concentration (a phenomenon referred to as the ‘halo effect’) (Griffin et al. 2001). This would presumably be of most relevance to population groups who consume more manufacturer-added than consumer-added water such as from soft drinks in children. For example, estimates of water intake in 11–12 years olds in North East England estimated carbonated drinks provided 17% of water intake (Zohouri et al. 2004). Even so, consumer-added water still accounted for a much higher percentage of total water intake (35%) than manufacturer-added water (24%), meaning water fluoridation is still highly likely to be a major determinant of total fluoride intake (and is supported by recent biomarker studies in children and adults referenced above). However, we recognise that the validity of PWS fluoride concentration as a proxy for total intake may vary by differing dietary habits and age groups.
Data sources and management
Data cleaning, management and analysis were executed in Microsoft Excel 2010 (Microsoft, USA) and STATA version 14 (StataCorp, USA). Geo-referencing was performed using ArcGIS ArcMap version 10.2 (Esri, USA). Exposure indicators were estimated by combining fluoride concentration obtained from routine fluoride monitoring data from 1995 to 2015, provided by the DWI and population data obtained from the Census and related mid-year estimates computed by the Office for National Statistics (ONS). The DWI supplied copies of water company WSZ boundary files in digital format for 2004–2015, of which we were able to prepare 2005–2015 for analyses. We used Geographic Information Systems (GIS) point-in-polygon (PIP) methods to assign statistical areas to WSZ boundaries using the population weighted centroid (Chapter 7—Using GIS for Environmental Exposure Assessment: Experiences from the Small Area Health Statistics Unit 2004). The smallest geographical unit of analysis was the 2011 Lower layer Super Output Area (LSOA); analyses at larger geographical areas were performed by using LSOA level fluoride, health and population data as ‘building blocks’, aggregated to form their larger ‘parent’ 2011 Middle layer Super Output Area (MSOA) and 2011 Lower Tier Local Authority (LTLA) areas (with which their borders match). See Supplementary Table 1 for population and count information on these geographic units. The population-weighted centroid of each LSOA (‘point’), which assigns a single geographic point to each LSOA based on the largest aggregation of its population, was overlaid onto WSZs (‘polygons’), thereby allocating a LSOA (and their populations) to a WSZ. As WSZs may be aggregated or dis-aggregated to form new WSZs, the number and geographic boundaries of WSZs are not fixed over time, though this occurs at (at least) annual intervals. WSZs will retain a unique identifier (consisting of the concatenated water company name and water company WSZ reference) whilst under the ownership of a single company, which can be used to track monitoring data over the time period of the WSZ’s ownership by that particular water company. To overcome the issue of WSZs changing shape and size over time, PIP analysis was repeated for each year of available (mapped) WSZ data (2005–2015). The linked LSOA-WSZ pairs were then linked with the DWI fluoride concentration and fluoridation scheme flagging dataset, using concatenated site reference and water-company coding by year to identify common WSZ years. Arithmetic mean period fluoride concentrations for the exposure period of interest were then aggregated from LSOA to higher geographic levels, weighted by the exposed population. We deliberately selected the arithmetic mean rather than other measures of central tendency less effected by extreme values as this would most accurately reflect population exposure in the whole period of interest, which would include ‘outlier’ periods.
Area-level fluoride concentration in water supply, regardless of source, was then categorised into the following: 0.0–< 0.1 mg/L, 0.1–< 0.2 mg/L, 0.2–< 0.4 mg/L, 0.4–< 0.7 mg/L and ≥ 0.7 mg/L. Such an approach was chosen to examine possible biologic gradient in the form of presence of a dose–response, while not assuming it would follow a specific model form (Rothman et al. 2008). Rather, we intended to break the range of the study exposure into categories that could be used to look for trends in the category-specific coefficients or relative risks, while avoiding a mechanical algorithm such as the percentile method, which could lack power to detect exposure effects stronger at extreme ends of the exposure scale (Greenland 1995). These categories were selected because, from international evidence, the association between fluoride concentration and decreasing caries prevalence was thought to increase linearly with increasing fluoride concentration, with reductions in dental caries prevalence tailing off above 0.7 mg/L (U.S. Department of Health Human Services Federal Panel on Community Water Fluoridation 2015). Further categorising the fluoride concentration of the population receiving water supplies with fluoride < 0.7 mg/L would give the ability to detect a dose–response, plateau, and threshold effect at higher fluoride levels. Given the relatively low water fluoride concentrations in England, the population receiving supplies at concentrations > 0.4 mg/L was thought unlikely to be large enough for division into more than two further categories and still allow meaningful examination of associations with less common health outcomes.
Missing data, outliers and unexpected values (e.g. high fluoride concentrations in zones not flagged as fluoridated, and vice versa) were investigated. Since 2006, the DWI has retained annual records that identify, via a flag, those WSZs that have fluoridation schemes. As no new CWF schemes have been initiated since 1995, flagged WSZs were considered to have been fluoridated from at least 1995 continuously to 2015, and we checked whether all zones ever fluoridated had consistent CWF flagging. Where this was not the case, we asked the DWI to check their supporting databases as to whether the zone was truly fluoridated or not and whether inconsistent flagging may have been the result of disruption in operation of fluoride plant. If there was known disruption of fluoridation, then the zone was re-flagged as being not fluoridated for the years during which fluoridation was disrupted.
Exposure indicator descriptive analysis
The number of water supply zones, number of samples and average number of years of monitoring data per WSZ were described from 1995 to 2015 stratified by time period (1995–2004 or 2005–2015), reflecting the variation in availability of WSZ mapping data and approach to fluoridation and its monitoring over different time periods. We then described the annual mean fluoride concentrations in each zone for these two time periods, stratified by presence of a fluoridation scheme, using histograms and box-plots. Stability of fluoride concentrations within unique zones was further described by creating scatter plots and calculating Spearman rank coefficients (stratified by presence or absence of a fluoridation scheme) for the WSZ-level period mean fluoride concentrations from 1995 to 2004 compared to 2005–2015. Spearman rank coefficients were calculated due to the skewed distribution of the data in fluoridated areas. Fluoridation schemes known to have significant disruption to operation were excluded from the creation of scatter plots and correlation analysis. MSOA-level public water supply grand mean (of the annual means) fluoride concentrations for 2005–2015 and location of fluoridation schemes were then described spatially by mapping the 2005–2015 grand mean fluoride concentrations and the distribution of fluoridation schemes onto 2011 MSOA boundaries. In order to map grand mean fluoride concentrations onto MSOA boundaries, we first calculated the grand mean fluoride concentrations for each LSOA and then aggregated to MSOA level by weighting the means of each constituent LSOA by its 2005–2015 population, using ONS mid-year population estimates. We then tabulated the MSOA-level count of population supplied (taking the period average) for categorised levels of fluoride concentration in milligrams per litre.