Evaluation of water fluoridation scheme in Cumbria: the CATFISH prospective longitudinal cohort study

Susceptibility of the tooth surface

The morphology of teeth influences susceptibility to caries, and the pits and fissures of teeth are more susceptible than the smooth surfaces of teeth. The presence of dental restorations or prostheses makes tooth surfaces more susceptible to caries, whereas professionally applied preventative sealants form a barrier between the tooth surface and the biofilm. Professionally prescribed preventative treatments using high concentration fluoride preparations, such as fluoride varnish, also provide a degree of protection.

Time

Caries is usually a slow-progressing disease that is characterised by periods of demineralisation and remineralisation of the tooth surface as intraoral conditions change in response to cariogenic challenges. Only when the balance of the demineralisation–remineralisation dynamic is weighted towards demineralisation does caries occur. If there is constant exposure to refined carbohydrates, then the disease can progress rapidly.

The structure and content of the dental biofilm

Disruption of the biofilm through oral hygiene-related behaviour (e.g. toothbrushing and professionally provided cleaning) reduces susceptibility. Likewise, an increase in the availability of fluoride within dental plaque helps prevent dental caries.

Diet and dietary behaviour

Reduction in the frequency of consumption of fermentable carbohydrates reduces caries risk.

Saliva

Saliva acts as a buffer and can limit falls in plaque pH. If saliva flow is impaired due to disease or, more commonly, as a side effect of medications, then the buffering action can be lost, favouring demineralisation and, therefore, increasing caries risk.

The international picture

Internationally, the prevalence of dental decay varies from country to country. Information from the World Health Organization Oral Health Data Bank shows that at a global level there has been a trend for an overall decrease in the prevalence of caries. 39 Numerous epidemiological studies3,40,41 have shown a reduction in caries in developed countries. In the immediate post-war period, the increased availability of sugar was associated with the increased levels of caries, but by the 1970s and 1980s a decrease in the prevalence of caries was being widely reported internationally in the literature. 3,41–43 Most commentators attribute this decrease to the rapid expansion in the availability and widespread use of fluoride toothpaste first introduced in the 1970s. 44 It is likely that other factors, such as government dental health polices, shifts in dietary patterns and improved oral hygiene, have also played a contributory role in this fall in disease. Primarily, the decrease in caries was reported in developed countries, with some studies41,45 conducted in a number of developing countries indicating a rise in the levels of caries. However, a recent systematic review46 looking at data from 1970 and 2004 concluded that the perception that dental caries rates are increasing in developing countries could not be supported.

UK trends in caries epidemiology

Health inequalities

In common with many other communicable and non-communicable diseases, dental caries is more common in deprived communities. From the 1990s, a new type of caries distribution among children in developed countries was being reported in the literature. 3 Most of the disease was being assigned to small, high-risk, socially and economically disadvantaged communities within the population, and this was initially described as the 80–20 phenomenon, that is 20% of the children harboured 80% of the caries in the population. 48 However, population segmentation analyses using area measures of socioeconomic status reported that about half of the population disease was confined to a minority of the population, but not to the extent of 80% of the disease in 20% of the population. These analyses also found that, although children with caries were more likely to be found in areas of social deprivation, caries was not confined exclusively to a small number of such areas, and caries prevalence exhibited a shallow gradient from poor to more affluent localities. This gradual fall in disease prevalence from most to least deprived illustrated the social gradient of disease model, as described in the Marmot review. 49

The national child dental heath surveys document the association between caries and deprivation in the UK. 50 The 2013 survey showed that children who were from lower-income families (defined as eligible for free school meals) were more likely to have caries than their more affluent peers. One-fifth (21%) of the 5-year-olds who were eligible for free school meals had severe or extensive tooth decay, compared with 11% of 5-year-olds who were not eligible for free school meals. Looking at the permanent dentition, one-quarter (26%) of 15-year-olds who were eligible for free school meals had severe or extensive tooth decay, compared with 12% of 15-year-olds who were not eligible for free school meals.

Significant inequalities are also evident in access to, and utilisation of, dental services. Individuals with the greatest need are least likely to access and utilise dental services. The 2013 child dental health survey reported that income deprivation is negatively associated with dental attendance. 50 Children aged 12–15 years from more deprived families were more likely to report their dental visiting to be triggered by symptoms, rather than attending asymptomatically for check-ups. Children from disadvantaged backgrounds were also more likely to report that they had never been to the dentist at the age of 12 years.

Caries risk

Caries risk or susceptibility to caries is largely dependent on how an individual’s lifestyle and behaviour (see Figure 3) influence the main aetiological determinants of caries (i.e. susceptible tooth surfaces, presence of cariogenic bacteria and consumption of refined sugars). Widespread availability of fluoride toothpaste and regular exposure to fluoride via toothbrushing as a social norm is believed to be behind the significant reductions in population caries prevalence described above. Toothbrushing with fluoride toothpaste both loads the biofilm with fluoride, protecting the tooth surface, and disturbs the biofilm, reducing its bacterial load. A recent behavioural study,51 conducted in parallel with the National Institute for Health and Care Research (NIHR)-funded Northern Ireland Caries Prevention In Practice (NIC-PIP) trial,52 on a population of 2- to 3-year-olds followed for 3 years provided useful insights as to why some children developed the disease and others remained caries free. The study51 reported that toothbrushing was widely adopted from a very young age and quickly became an ingrained or automatic behaviour for most families. However, frequent, between-meal sugar consumption was highly prevalent, and many parents struggled to control their child’s sugar consumption. The authors51 hypothesised that use of fluoride to prevent caries is limited in its effectiveness in the presence of unrestricted and frequent sugar consumption.

Over the last 10 years, a growing number of caries risk or caries prediction models and/or tools have appeared in the literature, such as the Caries Management By Risk Assessment (CAMBRA) and the Veterans Affairs Caries Risk Assessment (VA CRA) tool. 53,54 Despite these tools having numerous questions linked to ‘predictive factors’, the most accurate predictive risk factor for caries development at new sites is existing presence of the disease, typically over a 12-month period. A longitudinal observational study55 of caries development and progression in 7- to 16-year-olds in the north-west of England provides useful information for the disease trajectory of dentinal caries in permanent teeth as children move into adolescence (the study population55 is also geographically similar to the CATFISH study population). Clinical data were available from 6651 children over four time points. Caries prevalence was 16.7% at the first clinical examination (at ages 7–9 years), increasing to 31.0%, 42.2% and 45.7% at subsequent examinations. Children with caries in their primary dentition had a much steeper trajectory of disease in their permanent dentition than their caries-free contemporaries. The decayed, missing or filled teeth (permanent) (DMFT) count as pupils aged was significantly higher (4.49 times, 95% CI 3.90 to 5.16 times) in pupils with caries in their primary dentition than in pupils with a caries-free primary dentition. This study55 highlighted the importance of prevention in early years to reduce the risk of development of caries in late childhood into adolescence.

Caries in adults

In addition to the 10-yearly UK child dental surveys, the Department of Health and Social Care in England, the Welsh Assembly Health Department, and the Department of Health and Personal Services in Northern Ireland have commissioned 10-yearly national adult dental health surveys since 1968. Data from successive adult dental health surveys show a clear and rapid declining trend in the prevalence of edentulism (i.e. complete absence of teeth) in the population. 56 The first adult dental health survey in 1968 revealed a prevalence of edentulism of 37%, compared with just 6% in 2009, as many more people are retaining their teeth into old age. The UK has an ageing population, and over the past 40 years the proportion of population aged ≥ 65 years has increased from 13.8% to 17.7%. Furthermore, it is predicted that over the next two decades the proportion of people aged> 65 years will further increase to around 24% of the population. Retention of teeth into older age is associated with a need for more complex and, therefore, expensive dental care to maintain ageing, usually heavily filled, dentitions. This complex dental care, coupled with an expanding older population, suggests that there will be an increased burden on NHS dental services to meet the oral health needs of older people over the next 20 years. Observational studies and surveillance programmes56,57 show that, in the last four decades, dental caries in permanent teeth as a public health problem has evolved from a rapidly progressing disease of childhood, which results in early tooth loss, to a slowly progressing disease, where much of the burden is increasingly experienced by older adults.

Burden of disease

Tooth decay can cause pain, sleepless nights, sepsis, loss of function, social embarrassment, excessive use of antibiotics and the loss of productive workdays/school attendance. 9 As the disease claims progressively more tooth tissue throughout the life cycle, its effects are cumulative and can lead to complete tooth loss, and is one of the leading global causes of years lost to disability. 56 Dental treatment of caries is an uncomfortable experience and can provoke severe anxiety for some people. Thirty per cent of UK adults report that having a tooth drilled would make them very or extremely anxious. 12 Dental care is also very costly to society. Across the 28 European Union countries, treatment costs are higher than costs for Alzheimer’s disease, cancer and stroke, with only diabetes and cardiovascular disease costing more. 58

Looking at young children, a UK prospective cohort study of 3- to 6-year-olds showed that once a child develops caries it progresses rapidly and has a significant impact on their lives and their families. 7 Children with caries had a 25% risk of experiencing pain and an 11% risk of having an extraction each year. If the disease is unchecked in young children, multiple extractions under DGA are common. Dental extractions are the most common reason why young children in the UK have a DGA. Exact figures are difficult to quantify, but Hospital Episode Statistics data, which captures only a proportion of extractions carried out within a hospital setting, estimates at least 60,000 hospital episodes each year within England alone. 8 DGA extractions have a significant negative impact on young children and their families, and there is a strong association between dental extractions and dental anxiety, which can continue to affect individuals in later life. 59,60 Unsurprisingly, given the well-documented social gradient in dental caries, inequalities in extractions under DGA are also clear, with children from disadvantaged backgrounds more likely to have experience of DGA for exodontia than their more affluent peers. 61

Systemic effects

Teeth develop in utero and during infancy. 62 Within tooth germs, specialised secretory cells called ameloblasts and odontoblasts lay down enamel and dentine, respectively. Development of enamel has a secretory stage when the organic and mineral content of the enamel matrix is produced and a maturation phase when the composition of the enamel is modulated. During this phase, matrix proteins are removed from extracellular space, and mineralisation increases to form a fully mineralised enamel matrix.

The maturation stage of enamel is considered the stage most susceptible to fluoride exposure. Under chronic fluoride exposure the outer surface of the enamel is thought to progressively hypermineralise during the maturation stage, providing greater protection to the tooth surface. However, studies have shown that the differences in fluoride concentration in surface enamel between permanent teeth from individuals from areas with no or low fluoride levels and fluoridated areas are minimal, and there is a contemporary consensus that the preventative effect of fluoride is almost exclusively post-eruptive and topical rather than systemic. 63 There does, however, remain uncertainty regarding the role of systemic fluoride in the delayed eruption of the permanent teeth. It is argued by anti-fluoridation groups that such a delay, possibly resulting in teeth erupting after peak risk periods, could explain the lower caries prevalence and severity of individuals exposed to WF. 64,65

Topical effects

The preventative effect of fluoride is thought to occur at the tooth–plaque interface by influencing the dynamics of ion exchange. Fluoride impedes demineralisation of the tooth surface in several ways, namely by reducing bacterial acid production, by reducing the solubility of apatite crystals that make up the inorganic content of the enamel matrix and by the fluoridation of apatite crystal surfaces, interacting with hydroxyapatite to form fluorapatite, which is less susceptible to erosion by acid-producing oral bacteria. 66

From available evidence,67 the latter process seems to have the most important effect. The effect depends on the presence of sufficiently high fluoride concentrations in the plaque to maintain tooth surface fluoridation. 67 Fluoride also promotes remineralisation of hard tissues (even at low concentrations) and, therefore, slows or prevents overall mineral loss. The formation of intraoral fluoride reservoirs capable of supplying ions for a prolonged period is crucial to the success of topical treatments. Such reservoirs include calcium fluoride, formed mainly at tooth surfaces, and fluoride associated with organic components of plaque and oral soft tissues. Fluoride is thought to delay the point of (irreversible) cavitation at a given sugar intake, thereby slowing the progression of the disease. However, when consumption of sugar is higher than 3% of total energy intake, caries will steadily accumulate throughout life, even in populations that are widely exposed to fluoride. 68

Risks of overexposure to fluoride

Fluoride is a single highly electronegative ion that interacts with the cells and developing enamel matrix within the tooth germ at the different stages of enamel formation. Both fluoride dose and duration of exposure have important influences on outcomes. 69,70 Intermittent exposure to high concentrations of fluoride over a prolonged period is believed to have the biggest impact on the developing tooth germs, leading to dental fluorosis (Figure 4), which is a condition that ranges from mild and barely discernible mottling of the tooth surfaces to pitting and softening of the enamel surface with hard tissue loss. 70 Dental fluorosis of aesthetic concern is an established risk related to WF, but it is dose dependent. For a fluoride level of 0.7 ppm, the chance of developing dental fluorosis is 12%, and the odds increase by 2.9% with each 1-ppm increase in fluoride level. 71 Supraoptimal levels of fluoride (e.g. high concentrations in water supplies occurring naturally) have been linked not only to severe dental fluorosis, but also to skeletal fluorosis. 72,73

FIGURE 4.

An example of mild fluorosis of the type seen in English studies.

Dental fluorosis is the only scientifically recognised impact of fluoride at the levels found in England’s WF schemes. However, it should be noted that fluorosis is seen in children who do not, or have ever, lived in a fluoridated area. 74,75 Public Health England has published the WF health monitoring report every 4 years (as a requirement of the Water Act 200331), which covers dental caries, fluorosis and other conditions that have been, arguably spuriously, linked to higher levels of fluoride, and the last publication of the report was in 2022. 76 Anti-fluoridation groups will point to studies showing the impact of fluoride on intelligence quotient (IQ) levels, thyroid toxicity and kidney damage; however, many of these studies are scientifically flawed, were conducted in areas where fluoride levels are extremely high (e.g. in some southern Indian states or in provinces in China) or fail to demonstrate any link between the fluoride levels observed and the supposed adverse health outcome. 77

More recently, however, a number of well-respected groups in the USA and Canada have reported ‘associations’ between artificially provided fluoride (in salt, as well as water) and adverse health outcomes, including IQ. 78 Although the authors78 correctly state that their work is preliminary findings and, therefore, not conclusive, it would be inappropriate to disregard such work as junk science, as has been done in the past.

Concerns are also raised by dental academics regarding the levels of fluoride that individuals will be exposed to, for example with the increased use of fluoride varnish for younger children and the availability of high-concentration fluoride toothpastes for individuals at risk of caries. Contested reports from the US National Health and Nutrition Examination Survey suggest not only that the prevalence of fluorosis is rising, but also that its severity is increasing. A near doubling of the most severe levels of fluorosis is reported in the work by Neurath et al. 79 and these risks need to be placed into two contexts:

1. In most cases, fluorosis is purely aesthetic and many adolescents report preferring the appearance of mildly fluorotic teeth as teeth often look ‘whiter’. In addition, evidence suggests that the appearance of milder forms of fluorosis reduces with age.

2. The burden of risk is carried by individuals with little risk of caries, and, in England, this represents the majority of children.

The current research does not seek to explore the risks of fluorosis or any other health-related condition because of the limited follow-up period of the birth cohort (fluorosis is usually measured after the permanent incisors erupt after 9 years of age) and is, instead, focused on the dental health benefits. However, in considering these benefits, should they be proven, the reader should be aware that they are not achieved without some risk (e.g. there is a low recognised risk of fluorosis) to the population.

The York review

The aim of the York review16 was to assess the evidence on the positive and negative effects of population-wide drinking WF strategies to prevent caries. The report was critical of the then current research base for fluoridation, commenting that for such a high profile and contentious subject there was a surprising lack of high-quality research due to the high risk of bias in available studies. The York review16 had five distinct objectives, the first of which was to evaluate the effects of the fluoridation of drinking water supplies on the incidence of caries. The authors reported a median difference in the proportion of caries-free children of 14.6% (lower quartile 5.05%, upper quartile 22.1%) and a median difference in dmft/DMFT score of 2.25 teeth (lower quartile 1.28 teeth, upper quartile 3.63 teeth).

The second objective was to identify if WF has beneficial effects over and above alternative interventions. Nine studies, all completed after 1974, following the introduction of fluoride toothpaste, were included in this part of the review. 16 The quality of these studies was, again, assessed as suboptimal and limited the ability of the team to confidently answer the research question. However, the review team reported that a beneficial effect of WF was still evident, even in the presence of exposure to fluoride from other sources. The third objective was to examine the effect of WF on inequalities in caries experience. Again, the quality of studies was judged to be limited and analysis was hampered by heterogeneity in the measurement of socioeconomic status. Although the team reported that there appeared to be some evidence that WF reduces the inequalities in caries experience in 5- and 12-year-old children, using the dmft/DMFT index, this effect was not seen in the proportion of caries-free children among 5-year-olds, and the team urged caution in interpreting these results because of the small number and poor quality of available studies. 16

The fourth objective assessed negative effects of fluoridation and primarily dealt with risk of fluorosis. A significant dose–response relationship was identified: the prevalence of (any) fluorosis was estimated to be 33% (95% CI 26% to 41%) at a water fluoride concentration of 0.4 ppm, 48% (95% CI 40% to 57%) at a water fluoride concentration of 1.0 ppm and 72% (95% CI 62% to 80%) at a water fluoride concentration of 4 ppm. The review16 reported a lack of evidence for other postulated harms of WF, such as cancer and bone fractures. The final objective concerned potential differences in the effects of natural and artificial WF. This section of the review16 was substantially limited by the lack of studies comparing the effects or natural and artificially fluoridated water supplies. The team could find no major differences, but the evidence was not sufficient to make a conclusion regarding this objective.

Several areas of the evidence base were criticised by the report. 16 One major concern was the presence of observer bias in the measurement of caries and fluorosis within populations, especially as the examiners were aware of the fluoridation status of the participants. Weaknesses were also attributed to subjective instruments (visual indices) used to measure the presence or absence of caries and fluorosis. One of the most valuable outcomes of the review16 was a set of recommendations to improve the quality of future research projects. The following review study inclusion criteria have assumed a special importance as a marker of quality of research into WF (for the evaluation of impact on caries):

• at least two populations compared

• different fluoride levels in different populations

• prospective study design, assessing two points in time

• start of study < 1 year since change in fluoridation status

• measurable outcomes reported (e.g. dmft/DMFT scores).

In addition, it was recognised that it is important to adequately adjust for effect modifiers, such as socioeconomic status, frequency of sugar consumption, total exposure from all sources of fluoride, the number of erupted teeth per child and the level of spending on dental health promotion and primary care services, and to report variance data. It was recommended that blinding of observers measuring outcomes should be attempted and that standardisation of the assessment methods is essential in future studies to reduce the risk of observer bias. The need for appropriate measures of socioeconomic status and and to consider longitudinal changes in socioeconomic status was also highlighted.

The Medical Research Council Working Group’s Water Fluoridation and Health

Following publication of the York review16 and the criticism of the quality of evidence, the Department of Health and Social Care approached the MRC to review the conclusions and recommendations of the York review16 and to consider what further research might be required to improve the evidence base in fluoride and health. A working party was established with the following terms of reference:

• provide advice on current scientific evidence regarding the health effects of fluoride in the context of WF

• consider what further research in this area might be required and what priorities should apply to usefully inform public health policy in this area.

Unsurprisingly, given the MRC’s interests, there was a significant biomedical flavour to the Working Party’s recommendations. In particular, the Working Party placed importance on a better understanding of the bioavailability and absorption of fluoride from naturally fluoridated and artificially fluoridated drinking water and the impact of water hardness on bioavailability. The Working Party recommended that WF should be a priority area for research funding in the future and also picked up many of the methodological difficulties identified by the York review. 16

The possible beneficial influence of WF on health inequalities was highlighted; however, it was recommended that further research should be conducted on this issue. 18 Both the York review16 and the MRC report18 stated that the link between caries and WF should be studied further and that contemporaneous data on possible effect modifiers, such as ‘discretionary’ fluoride (primarily toothpaste) use and dietary sugar consumption, should be collected in subsequent studies and included in analyses.

The Cochrane systematic review

Fifteen years after the York review,16 the Cochrane Oral Health Group published a systematic review, Water Fluoridation for the Prevention of Dental Caries. 71 The review71 was undertaken in response to the subject matter being identified as a priority topic in the Cochrane Oral Health Group’s international priority-setting exercise. The review team also acknowledged the passing of time since the York review,16 querying whether or not the conclusions were still relevant to contemporary society and noting that many of the caries studies presented in the York review16 were conducted prior to the widespread use of fluoride toothpastes, introduced in the 1970s, and exposure to other fluoride vehicles, such as fluoride varnish, which is now used extensively in primary care and community prevention strategies.

The Cochrane review is a key document for the CATFISH project, as it is the most recent comprehensive review of the effects of WF provided by a trusted source with a peerless track record of conducting high quality systematic reviews in the dental field. The Cochrane review updated the York review and aimed to contextualise the evidence to inform current national and international guidelines.

The Cochrane review71 was narrower in focus than the York review,16 concentrating solely on the effects of WF on preventing dental caries and increasing the risk of dental fluorosis. The Cochrane review71 made no distinction between artificial and natural WF and so did not investigate differences in the effects of natural and artificial WF. The authors71 confined their assessment of potential negative effects of WF to fluorosis and did not consider other possible adverse effects, such as bone fracture and cancer, which came under the York review’s16 remit. The inclusion criteria for caries and fluorosis in the Cochrane review71 followed those of the York review,16 illustrating the importance of the York inclusion criteria in guiding the design of evaluations of WF.

A total of 155 studies met the inclusion criteria of the review and 107 studies provided sufficient data to enable quantitative synthesis. 71 For caries severity, the review reported results in reductions in dmft because of WF [mean difference 1.81 teeth (95% CI 1.31 to 2.31)] and in DMFT [mean difference 1.16 teeth (95% CI 0.72 to 1.61)], translating to a 35% relative reduction in dmft and a 26% relative reduction in DMFT compared with control group mean values. The studies included in both analyses were assessed collectively as having a high risk of bias.

When assessing the impact on caries prevalence, increases in the percentage of caries-free children were identified in fluoridated compared with non-fluoridated populations. An increase of 15% (95% CI 11% to 19%) caries free in the primary dentition and 14% (95% CI 5% to 23%) in the permanent dentition were reported. The authors71 noted that most studies (71%) were conducted prior to the introduction and widespread use of fluoride toothpaste in 1975.

The Cochrane review,71 like the York review,16 found insufficient evidence to clearly determine whether or not WF affects inequalities in caries experience and/or prevalence. Like the York review,16 the Cochrane review71 noted the paucity of studies investigating the impact of WF on adult dental health and how no studies that aimed to evaluate the effectiveness of WF in adult populations met the review’s inclusion criteria. The review71 reported an estimated 40% (95% CI 35% to 44%) prevalence of fluorosis of any level and a prevalence of 12% (95% CI 8% to 17%) of fluorosis of aesthetic concern when WF occurred at a level of 0.7 ppm. The review71 noted that over 97% of the studies reporting fluorosis outcomes were at high risk of bias and there was substantial between-study variation in outcome measurement. The studies included in the review that examined dental fluorosis were generally more recent than those that evaluated caries and, consequently, the assessment of WF’s contribution to fluorosis risk could have been influenced by exposure to other sources of fluoride. 71

As WF is a contentious topic, it was unsurprising that the Cochrane review would be criticised. Rugg-Gunn et al.,82 writing in the British Dental Journal, from a pro-fluoridation stance, provided a critique of the review, which argued that the inclusion criteria were too narrow and restrictive for a public health intervention and that the risk-of-bias assessment was also limiting, preventing inclusion of studies that collectively would provide a fuller understanding of the effectiveness of WF. The critique argued that the findings and conclusions of the Cochrane review71 are at odds with the wider literature on WF, concerning its effectiveness in adults and its effectiveness in reducing inequalities.

The critique noted that assessment of WF’s impact on inequalities was not a stated objective of the Cochrane review71 and identified the narrow inclusion criteria as the reason why some relevant reports were excluded. The York review found that the evidence about reducing inequalities in dental health was of poor quality, contradictory and unreliable. 83 A secondary analysis of English national surveillance data suggested that the preventative effect of WF is greater in the most deprived communities. 77 Importantly, the critique pointed out that the absence of evidence, or the existence of poor-quality evidence, should not lead to the conclusion or implication of an absence of effect.

The Cochrane review71 authors noted, like the York review16 and the MRC report,18 the significant shortage of contemporary evidence meeting both the York and Cochrane reviews’ inclusion criteria,16,71 and that little had changed in the 15 years between the reports. In fact, the available data for both reports came predominantly from the same studies, which were conducted prior to fluoride toothpaste introduction in 1975, and so it is not surprising that the York review16 and the Cochrane review71 findings are similar. What is surprising, however, given that both the York and MRC reports called for WF to be a priority for research funding, is that very few studies meeting the ‘York criteria’ have been commissioned since the York review16 and MRC report. 71 In the UK, this is primarily due to the political difficulties experienced in implementing new schemes (i.e. one of the key inclusion criteria of the York review), the significant costs of longitudinal studies involving clinical examinations and the difficulties of identifying matched comparator populations, particularly in countries such as the USA, Ireland and Australia with widespread WF schemes in place.

The Cochrane review71 identified several research questions, which are still to be answered 20 years after being proposed within the York review. 16 We still do not have a contemporary understanding of the health risks and benefits of WF, nor of its impact on health inequalities. As caries is thought to be increasingly more of a public health problem in older, rather than younger, populations, it is worrying that there is still very little information of its effects on adult populations. Standardisation of diagnostic criteria and reporting, as well as adequate measurement and control of potential effect modifiers, were all (once again) highlighted by the Cochrane review,71 as the implications of the findings for further research were discussed. A single study cannot hope to provide definitive answers to all of these questions; however, a contemporary study conducted in the UK that meets the York inclusion criteria is probably more helpful to UK policy-makers than another review of less robust, mostly historical, studies, most of which have little relevance to the current epidemiological, social, financial and behavioural climate.

Birth cohort

A birth cohort was chosen because this cohort provided an opportunity to investigate the ‘full’ effect of WF, comprising systemic exposure where ingested fluoride is incorporated into the tooth germ during its development in utero, coupled with topical exposure of the tooth surface to fluoride after the teeth erupt into the mouth. Participants were recruited from maternity units within the two hospitals in Cumbria by the NIHR Clinical Research Network North East and North Cumbria. All new parents were approached during pregnancy and postnatally (at the 20-week scan, immediately after delivery and at health visits within the first 3 months of birth) (Figure 5).

FIGURE 5.

Recruitment outline for the birth cohort. Reproduced with permission from Goodwin et al. 85 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

Reproduced with permission from Goodwin et al. 85 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: https://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original figure.

Trained recruiters, including clinical research assistants and research nurses, identified eligible women and began face-to-face recruitment during 20-week scans at hospital sites in May 2014. All eligible women were provided with the study information leaflet prior to their 20 weeks’ gestation scan or at the 20-week scan (see NIHR Journals Library URL: www.journalslibrary.nihr.ac.uk/phr/SHMX1584). Individuals could fully consent by providing written consent (see NIHR Journals Library URL: www.journalslibrary.nihr.ac.uk/phr/SHMX1584) at the time they were approached or leave their details to be contacted after the birth of their child. If any women were not contacted during this time, then they were approached face to face in the hospital/maternity ward after giving birth (starting from September 2014) or a consent pack was sent home. Recruitment ended in September 2015 and no children born after September 2015 were included (see Appendix 1, Figure 16).

Eligibility was assessed by the research nurses according to the following criteria.

Older school cohort

The older school cohort comprised 5-year-olds recruited from primary schools in West Cumbria (i.e. fluoridated) and a control group of 5-year-olds recruited in schools located in the non-fluoridated area of Cumbria. The older school cohort was chosen because this cohort provided an opportunity to investigate the topical effect of WF on caries to enable comparison of effect size with children who have systemic and topical exposure as the cohorts age.

We approached all primary schools in West Cumbria (i.e. the exposed population) and a comparable group across North Cumbria (i.e. the non-exposed population) and asked them to participate in the study. Parents of 5-year-old children attending participating schools were contacted over one academic year (from September 2013 to July 2014). A study information sheet was sent to each child’s home address through the school. Consent was obtained using parental written consent via a letter home through the school, with child assent gained at time of the clinical examination. For the information sheet and consent form see NIHR Journals Library URL: www.journalslibrary.nihr.ac.uk/phr/SHMX1584.

Eligibility criteria were assessed, and provision of consent was checked by the clinical examiners prior to clinical examination in school.

Birth cohort

Recruitment took place in the maternity units of West Cumberland Hospital and Cumberland Infirmary. Examination of children at age 3 years took place at dental clinics across Cumbria. Participants could choose to attend clinics in Carlisle, Wigton, Workington, Cleator Moor, Penrith or Kendal. Participants could book in on certain days to have their dental examinations and were also provided with a questionnaire when they attended. Children were examined at age 5 years in their primary school. In total, 172 schools were approached and agreed to allow access for dental teams to carry out dental examinations for children to take part in the study. An additional eight schools with a total of 38 participants stated that they did not want the team to carry out the clinical examinations at the school. Children attending these schools were to be invited to clinics.

Older school cohort

Children were examined at 5, 7 and 11 years in their primary school. One hundred and thirty-five schools were included when the study started, and nine schools refused to participate at the time. Reasons for schools not wanting to participate in study included the school was small and had few children in each year (and it would have been disruptive for the rest of the class), and schools having building works and it not being appropriate, or there was limited space, to carry out dental examinations.

Control groups

The comparator groups were children in each cohort residing in Cumbria, born during the same period as children in the intervention group, who had not lived in a fluoridated area during the 5-year follow-up period.

Choice of comparator group

The intervention was provided to a defined population based on the geography of the water supply and, therefore, randomisation of individuals to intervention and control groups was not possible. A whole population approach was taken to provide a comparator population, rather than matching individuals. It was important to identify a control population that was as socioeconomically and geodemographically similar to the intervention population (i.e. similar age, sex, deprivation and ethnic profile, similar access and utilisation of dental services). Therefore, we chose neighbouring non-fluoridated districts in Cumbria as the control population. For both cohorts, we compared the sociodemographic variables of intervention and control groups (i.e. age, sex, deprivation at baseline) to ensure that the groups were comparable, to ensure unbiased estimates and to minimise selection bias. If significant differences between the groups on key variables at baseline were found, then we planned to perform regression analysis to adjust for these variables.

Blinding

Owing to the geographical nature of the intervention, dental examiners who worked in the Cumbria community dental service were likely to be aware of the fluoride exposure status of participants from the school they attended. This potential source of bias, which was also recognised by the MRC report,18 is difficult to mitigate. There have been attempts in past studies to blind examiners to water fluoride exposure status by moving children to central examination centres or using examiners from out of area. 90,91 However, such approaches were not feasible for this study because of its size and because of the large size of Cumbria and its very rural nature. Therefore, we elected to accept that the clinical workforce would be aware of participants’ fluoride exposure and to assess the presence of any bias by taking high-resolution intraoral photographs [using the S950 Pal Sopro Care intraoral camera (Acteon Group, St Neots, UK)] and comparing examination of caries in photographs with traditional clinical caries examinations. Intraoral photos were taken following the clinical examination by the same dental examiner. For 5-year-olds, five photos were taken, namely of the two upper incisor teeth/area, upper-left molar teeth/area, upper-right molar teeth/area, lower-left molar teeth/area and lower-right molar teeth/area. For 7- and 11-year-olds, six photos were taken, namely of the upper-right incisor, upper-left incisor, upper-right molar, upper-left molar, lower-right molar and lower-left molar. Intraoral photos were examined remotely and blind to fluoride exposure by one expert examiner, who had previously scored caries from images for other large-scale caries studies. 92 Figure 6 provides an example of intraoral images taken during the study.

FIGURE 6.

An example of intraoral images from an 11-year-old participant captured during the CATFISH study. (a) Upper-right incisor; (b) upper-left incisor; (c) upper-right molar; (d) upper-left molar; (e) lower-right molar; and (f) lower-left molar.

Images were stored on a secure database prior to being scored93,94 and then assessed in random order using the same caries classification scoring system used for the clinical examinations. 95 The difference in the proportion of participants with decay between test and control groups assessed using blinded photographs was compared with the difference in the proportion of participants with decay between test and control groups assessed using traditional unblinded clinical examinations to identify any systematic bias that may have been present due to examiner knowledge of exposure status.

Diagnostic protocol

The examinations were undertaken by trained and calibrated dentists working for the Cumbria community dental service. One of the key services provided by the community dental service is to undertake regular dental epidemiological surveys, which are conducted in schools and coordinated by Public Health England. We used the same diagnostic protocol and standard operating procedures as these surveys, that is, teeth were cleaned and dried using cotton wool rolls and a Daray lamp (Daray Ltd, Swadlincote, UK) was used to provide light. 95 Caries was recorded at the caries into dentine level for each tooth surface (final analysis was carried out at tooth level) and caries was considered present if there were one or more lesions into dentine present at the examination. Box 1 provides a list of the dental codes that were utilised (the full version of data collection forms can be found in NIHR Journals Library URL: www.journalslibrary.nihr.ac.uk/phr/SHMX1584).

The same codes (see Box 1) were used for permanent teeth, but with an additional code of 7 for extracted due to ortho. Decay was recorded with any tooth, which had a code of 1, 2, 3, 4, 5, 6, 10, 11 or 12.

These codes enabled direct comparison with the Public Health England Child Dental Health Surveillance Programme outputs and ensured the schools and dentists used familiar consent, examination and data recording processes. Children’s teeth were cleaned and dried using cotton wool rolls prior to being scored clinically by trained and calibrated assessors.

Data were recorded on pseudoanonymised paper forms and were subsequently transferred to and processed within the Dental Survey Plus 2 software version 2.1 (University of Dundee, Dundee, UK; URL: www.nwph.net/dentalhealth), which is designed to record dmft. All data used unique identifiers to ensure anonymity when data were entered. These data were transferred and kept within locked cabinets at the main research centre (i.e. the University of Manchester).

Training and calibration

Calibration occurred every year there were clinical examinations, and this involved usual British Association for the Study of Community Dentistry (BASCD) practice, that is, a pre-training session was provided for any new examiners, children were screened the day before calibration to ensure sufficient numbers of children with decay experience and children who were caries free were included, and a training round occurred before calibration with at least two calibration rounds performed (for further information on calibration see the National Dental Epidemiology Programme: Training and Calibration Guide for Oral Health Surveys of Children96). Each dentist taking part and the reference examiner examined the same group of children for calibration (following the diagnostic protocol described above) and comparison of interexaminer differences were performed. Only dentists who achieved the minimum requirements for calibration set by BASCD took part in the CATFISH study.

Calibration results

To calibrate, dentists should have a sensitivity of 0.75 (minimum), a specificity of 0.90 (minimum) and a deviation from the mean of < 0.50. Tables 4–8 show the calibration results for each year of clinical examinations.

Primary outcome

Secondary outcomes

In the birth cohort, the clinical dental examinations described above were used to calculate the mean number of dmft in the primary dentition at 5 years. Likewise, in the older school cohort, the standardised clinical examinations at ages 7 and 11 years were used to calculate the mean number DMFT in the permanent dentition.

For the birth cohort, the number of erupted primary teeth could also be calculated at ages 3 and 5 years from the clinical examination.

Parents were asked at recruitment to provide consent for study researchers to collect data from relevant medical and NHS dental records. Therefore, in addition, we accessed activity data from the NHS Business Services Authority (BSA) and from the NHS Child DGA service operated by Cumbria Community Dental Service. The data from the former source were used primarily to estimate health services costs in the economic analyses (see Chapter 8), as well as to compare the numbers of dmft at 3 years old, given the low participant attendance for the 3-year-old wave of data collection within the CATFISH study.

Hospital data enabled us to identify children in both cohorts who had one or more teeth extracted under general anaesthesia during the follow-up period. Adverse reactions were measured by questionnaire in the birth cohort only. The questionnaire asked parents and guardians to detail any new chronic conditions.

Effect modifiers

Where randomisation of participants to intervention and control groups is not possible, the York review16 and the MRC report18 highlighted the importance of collecting data on potential effect modifiers to include as covariates in the analysis. These modifiers are largely behavioural factors associated with the aetiology of caries and its treatment, chiefly concerning sugar consumption, fluoride exposure from other sources and dental service use. This information was collected primarily by questionnaire.

Measures of deprivation and socioeconomic status

We used the IMD from 2010 to measure relative deprivation of participants. The IMD measures relative levels of deprivation in 32,844 small areas or neighbourhoods in England, called lower-layer super output areas. The IMD is used routinely in government reports, most notably the national 10-yearly UK children’s dental health surveys and was available for all participants via a look-up file, which matched their home postcode to the lower-layer super output area in which they lived. The IMD is converted into national quintiles (with 1 representing least deprived and 5 representing most deprived), which were the main measures used within this analysis. In addition to the small area measure of relative deprivation, we collected data on individual measures of socioeconomic status (e.g. parental income, occupation and education), as outlined in Household data, demographics, water environment, attitudes and parental health status.

Tables 9 and 10 provide summaries of the outcome and effect-modifier variables collected in the birth and older school cohorts, respectively, as well as the methods used in collecting the variables.

Recruitment and retention

Data were continuously collected on recruitment, ineligibility, examination refusals, withdrawals and loss to follow-up for intervention and control groups in both cohorts to enable compilation of STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) flow charts.

Baseline comparisons

For both cohorts, the distribution of baseline demographics was reported according to exposure group for age, sex and deprivation for the birth cohort and, additionally, dmft for the older school cohort, with 95% CIs for the between-group difference across the exposed and non-exposed groups.

Caries calibration analyses

Caries calibration was based on the BASCD guidance for caries calibration. 101 For each examiner, calibration utilised the dmft scores given to children. For examiners to successfully calibrate, the deviation from the benchmark should not be > 0.50, the sensitivity should be at least 0.75 when comparing dmft with the benchmark and specificity should be at least 0.90 when comparing dmft with the benchmark.

Evaluation of non-blinded examiners

The magnitude of difference between groups was assessed, comparing the difference between groups observed for the traditional clinical examinations (unblinded) and standardised photographs (blinded).

Caries analysis

In both cohorts, our primary outcome was presented as a cross-tabulation of presence or absence of caries with exposure or non-exposure to WF, with the risk ratio (with 95% CI), unadjusted odds ratio (OR) (95% CI) and adjusted OR (95% CI) obtained from a generalised linear model as the measure of effect. The ‘natural experiment’ in WF exposure implies an absence of confounders in this study (WF dosing in the exposed population is independent of social class, other fluoride sources, etc.). However, any significant differences between the groups on key variables identified as potential effect modifiers were accounted for by using appropriate regression models to adjust for these variables. Adjusted analysis of the primary outcome was undertaken using a generalised linear model with logit link function and adjusted for age, deprivation (quintiles) and sex as covariates. For the older school cohort, we were able to consider the baseline dmft caries levels between the two populations to identify any baseline imbalance in caries that required subsequent adjustment in the statistical analysis.

Secondary outcomes

Secondary outcome variables of dmft (for the birth cohort) and DMFT (for the older school cohort) were analysed using generalised linear models using the appropriate link function for the outcome, initially including only the exposure group and then additionally adjusting for baseline values of dmft, age, sex and deprivation (quintiles), and offset for number of erupted teeth.

Inequalities

In both cohorts, the primary outcome and secondary outcomes (dmft/DMFT) were compared across exposed and non-exposed groups by quintile of deprivation. Generalised linear models with the appropriate link function (negative binomial regression), and including an exposure by deprivation (quintiles) interaction term, were undertaken to determine the effects at different levels of deprivation. The regression examined the count of decay and the interaction between WF status and deprivation, and included the covariates age, sex and (for the older school cohort only) dmft at baseline, offset by the log of erupted teeth. This analysis was supplemented by the resulting marginal predicted probabilities for interpretability.

Approach

The framework for the economic evaluation was a within-study cost-effectiveness analysis to estimate the net cost per unit of health benefit gained by WF. A cost-effectiveness acceptability approach was used to estimate the likelihood that WF is cost-effective.

The population for the evaluation falls into two groups: (1) children from birth to age 5 years (i.e. the birth cohort) and (2) children aged 5–11 years (i.e. the older school cohort). The intervention is WF and was compared with no WF in the child’s area of residence (Cumbria).

The measure of health benefit was the QALY and the comparison of costs and benefits was generated using the ICER (i.e. cost per QALY gained). For the within-study analyses, the time horizon is 5 years for the birth cohort (i.e. ages 0–5 years) and 6 years for the older school cohort (i.e. ages 5–11 years).

The chosen perspective was that of both the health and care sector and the local authority, which incur the costs of dental treatment and of fluoridation, respectively.

Within-study analyses

Quality-adjusted life-years

Quality-adjusted life-years gained from baseline to end of follow-up were estimated as the number of days multiplied by the utility of observed survival. The utility values were estimated from the CHU9D questionnaire. The CHU9D is a generic health-related quality-of-life measure for children aged 7–17 years. 105–107 The CHU9D contains the following nine dimensions, with five levels:

1. worried

2. sad

3. pain

4. tired

5. annoyed

6. school work

7. sleep

8. daily routine

9. activities.

Children (or proxies for younger children aged 5–7 years) self-complete the questionnaire, with questions placed in the context of how the child is today/last night.

For the birth cohort, there is no baseline CHU9D utility measure and we assume that the utility value is equivalent across both fluoridated and non-fluoridated groups. A utility value of 1 (i.e. perfect health) is assumed in both groups. At follow-up, parents/carers completed the CHU9D for their child at age 5 years. Although the CHU9D was originally created and validated to be used for children aged 7–11 years, a proxy version for children aged 5–7 years has also been created and has been trialled. Early results indicate that a proxy- (parent-) reported CHU9D is appropriate. 108

For the older school cohort, parents/carers completed the questionnaire for their child at age 5 years and children self-completed the questionnaire at their final clinical assessment (at age 11 years).

The CHU9D responses are transformed to utility measures using preference weights obtained from a sample of 300 adults in the UK. 109 QALYs were then estimated as:

where U is utility value, i is individual, d is day of assessment and t is assessment point (1 = baseline, 2 = final).

Quality-adjusted life-years are discounted from year 2 at a discount rate of 3.5% in line with NICE guidance for technology appraisal. 103

Economic analyses

Descriptive analysis was used to summarise the CHU9D data, QALYs, service use and cost. Regression analysis was used to estimate incremental (net) costs and QALYs.

The following variables were selected as covariates for the analyses to estimate net costs and QALYs. The covariates used in the analysis of QALYs were:

• sex

• age

• follow-up duration (years)

• deprivation quintile of child’s residence

• CHU9D utility measure at baseline.

Sex and age were included to reflect the fact that health and, therefore, utility typically vary across sex and decline with age. Deprivation quintiles were included because of the variation in deprivation observed between fluoridated and non-fluoridated areas in the study.

The covariates used in the analysis of costs were:

• sex

• age

• follow-up duration (years)

• deprivation quintile of child’s residence

• CHU9D utility measure at baseline.

The CHU9D utility score at baseline was included as an overall measure of health and health need. For the birth cohort analyses of costs, there was no baseline CHU9D utility measure and so this is not adjusted for in the regressions for this cohort.

The primary measure for the economic analysis was the ICER. The ICER was estimated as the net cost divided by the net QALY estimates from the regression analyses.

To account for uncertainty in the parameter estimates, the estimates of net costs and outcomes from the regression were bootstrapped to simulate 10,000 pairs of net cost and net outcomes. These simulated pairs of net cost and net outcomes were used to generate cost-effectiveness acceptability curves, as recommended by NICE for health technology appraisals. 103 Based on the range of willingness-to-pay value thresholds (WTPTs) historically implied by NICE decisions,110 the simulated net QALY values from the bootstrap simulation were assigned a monetary value using a range of maximum willingness-to-pay values from £1 to £30,000 to gain 1 QALY.

A net benefit statistic for each pair of simulated net costs and net outcomes for each WTPT can then be calculated. For example, if the measure of health benefit is the QALY, then the net benefit of WF is estimated as:

This calculation was repeated for each WTPT. Cost-effectiveness acceptability curves plotted the proportion of bootstrapped simulations where the net benefit of WF is greater than zero for each WTPT111–114 and this provides a probability of cost-effectiveness that takes into account parameter uncertainty.

Missing data

Baseline data on the CHU9D were obtained via questionnaire [for the birth cohort, there is no baseline data and health-related quality of life utility is assumed to be identical in both groups, i.e. at 1 (perfect health)]. For the birth cohort, missing data can occur at follow-up via non-completion of the questionnaire or by incomplete CHU9D and/or resource (cost) data. For the older school cohort, missing data can occur via non-completion of the questionnaire at baseline or by incomplete CHU9D data at baseline and/or incomplete CHU9D data or resource (cost) data at follow-up.

For both cohorts, assessments were made of factors associated with questionnaire completion to inform whether or not there may be bias in the questionnaire data related to fluoridation group and/or IMD decile.

The approach taken to deal with missing data concerns missing data relating to incomplete CHU9D and/or resource (cost) data at baseline and/or follow-up.

Faria et al. 115 provide a guide to handling missing data. Faria et al. 115 propose identifying plausible assumptions for the missing data mechanism and applying methods that account for these reasons for missingness with sensitivity analyses that explores variations on the assumed mechanism. The suggested steps include (1) the identification of the missing data mechanism via descriptive analysis and logistic regression analyses on missing cost and QALY data against baseline measures, covariates and treatment, (2) the choice of methods to deal with the mechanism identified and (3) sensitivity analyses that relaxes/varies the assumed mechanism.

Identifying the missing data mechanism

The missing data mechanism is defined in line with Rubin’s classification. 116 Here, data are missing completely at random (MCAR) (i.e. where missing data are not associated with observed and unobserved values), covariate-dependent missing completely at random (CD-MCAR) (where missing data are not associated with observed outcomes but are associated with observed baseline covariates),117 missing at random (MAR) (where missing data are not associated with unobserved values but are associated with observed covariates and outcomes) and missing not at random (MNAR) (where missing data are associated with unobserved values).

Faria et al. 115 propose descriptive analysis of the missing data to infer the missing data mechanism and this includes:

• an assessment of missing data by treatment group at each follow-up (as equivalent missing data rates provide some support for MCAR)

• an assessment of missing data patterns (as patterns can help inform the methods to account for missing data)

• assessments of associations between missingness and baseline variables (as any variable that predicts missingness suggests that the MCAR mechanism is not valid)

• assessments of associations between missingness and observed outcomes (as any prior observed outcome variable that predicts missingness suggests that MAR is a more plausible mechanism).

Although these steps help inform the mechanism of missing data, the steps do not rule out MNAR.

Under MCAR, complete-case analysis (CCA) is valid. CCA is also valid when the analytical model includes baseline variables that predict outcome and missingness (i.e. CD-MCAR). However, to approach the analysis as intention to treat, individuals with missing baseline variables would need to be included.

The approach for identifying the missing data mechanism followed that suggested by Faria et al. 115 First, the proportions of missingness and of missing cost and QALY data are reported to observe whether or not there are differential rates of missing data across fluoridated and non-fluoridated groups. Second, logistic regressions of missing cost and QALY indicators are performed with a fluoridation group indicator, baseline covariates and baseline QALY as explanatory variables. If rates of missing data are similar and no baseline covariate or fluoridation indicator is a significant predictor of missing data, then the assumed missing data mechanism is MCAR and the CCA is the preferred model specification. If rates of missing data are similar and baseline covariates are significant predictors of missing data, then the assumed missing data mechanism is CD-MCAR and the CCA is the preferred model specification. If rates of missing data are similar and baseline QALYs are significant predictors of missing data, then the assumed missing data mechanism is MAR and imputation of missing data is the preferred methodological approach.

The study does not have multiple time points of follow-up, meaning that assessments of patterns of missingness and assessments of associations between missingness and observed outcomes are not applicable.

Missing data imputation

Missing at random was approached by multiple imputation, whereby missing data were replaced with predicted values. The chained equations approach was used to impute the data. Here, imputed values for variables were used to predict missing values in other variables. 118 Imputed values were obtained via predictive mean matching. The imputation was performed m times, where m was determined by the proportion of missing data (identified in the assessment of missing data above). The multiply imputed data sets account for uncertainty in the imputation. Each imputed data set was estimated and then combined to obtain a mean estimate and standard error (SE). 116

Imputation was performed by fluoridation group, with all baseline variables (i.e. age, sex and IMD quintile), costs (i.e. fluoridation costs, dental costs and DGA costs) and CHU9D domains included in the imputation model.

The multiply imputed costs and QALYs, domains of resource (costs) and CHU9D domains were compared with the CCA costs and QALYs to assess whether or not the imputation appeared to be valid. To obtain incremental cost and QALY estimates, seemingly unrelated regression models of the multiply imputed data sets were estimated.

Sensitivity analysis

Sensitivity analyses were used to explore the impact of changing the methods used on the estimates of whether or not WF was cost-effective, and this included re-running the analyses for the following:

• analysis where multiple imputation of missing observations gives estimates for missing costs and outcomes

• allocations of WF costs on a per capita of the population (all ages)

• analysis of cost-effectiveness with –

  • the birth cohort:

    • the count of dmft is the outcome measure

    • presence of decayed teeth as an outcome measure.

  • the older school cohort:

    • the volume of DMFT (including crowns) for teeth erupted post baseline

    • presence of decayed teeth for teeth erupted post baseline.

Subject recruitment and retention

Figure 8 provides details of the subject flow through the study, demonstrating the recruitment at the two principal hospital sites. Further information on recruitment is presented in Appendix 1, Figure 16.

FIGURE 8.

A STROBE flow diagram for the birth cohort. a, Invited to participate is based on area of recruitment; b, a child could be removed from the study if the study team was advised to remove them, the study team were informed they had moved out of the area or if they had died. Two children died who were part of the study, both were in the control area. This information was not treated as missing and not included within analysis as it was not connected to the study.

Figure 8 shows that the consent rate was higher in the non-fluoridated areas than in the fluoridated areas, despite an identical consent process, with most parents approached face to face when attending hospital for their scan or for the birth of their child. The numbers for ‘clinical examination 1’, which occurred when children were approximately 3 years old, were significantly smaller than expected. Examinations at age 3 years were particularly challenging, as examinations could occur within dental practices only, meaning that parents/guardians had to take their children to this appointment. Despite multiple sites and times being offered, with reminders over a month, as well as a being voucher offered as compensation, only 21% of children who had their final examination took part at ‘clinical examination 1’.

Examination status at the final clinical examination (from individuals who originally consented to take part) is presented in Table 13, according to fluoridated or non-fluoridated water status. There were fewer participants in the fluoridated area than in the non-fluoridated area at baseline (owing to the geographic restrictions of the fluoridated area). When comparing the percentage of participants who had a final examination, 67% of participants in the non-fluoridated group (of those originally recruited) completed their final examinations, compared with 78% of participants in the fluoridated group, with lack of access to schools in the non-fluoridated area (on account of COVID-19) being the principal source of the differential attrition. Further analysis on examination status is presented in Appendix 2, Table 61.

Demographics and behaviours were explored to assess if there were any differences between the fluoridated and non-fluoridated groups, and to look at differences between participants who had an examination and participants who did not (from participants who were consented). Table 14 shows that the relative proportions of males and females were similar in both groups. We observed that the proportion of people living in a most deprived area was higher in the WF group (30.2%) than in the no WF group (19.8%); however, the distributions across the quintiles of deprivation of people both examined and not examined were similar in the WF and no WF group. (See Appendix 2, Tables 62–111, for all other tables, descriptive statistics and analysis relating to potential confounders and bias.) When looking at potential confounders around behaviour that could affect dental health, there were only two variables that resulted in a statistically significant difference over time, and these were consumption of sweets/chocolate (i.e. sweets/chocolate consumed more than three times per week) and whether or not the child had ever visited the dentist. In the case of sweets consumption, the difference, of approximately a 10%, occurred after 12 months, with the prevalence of frequent sweets consumption (more than three times per week) being higher in the fluoridated area. These data were not included in the regression, as multiple (n = 11) behaviours were investigated, with the majority indicating no difference between groups over time. Therefore, we had to consider the risk of there being a type I error. In addition, the low response rate would have meant a reduced sample size and would have underpowered the final results.

Secondary outcome: decayed, missing or filled teeth (primary)

Table 16 details the count of dmft (generated from clinical scores, where the decayed component is into dentine) for each participant. The chi-squared with trend test demonstrated a statistically significant difference between the fluoridated and non-fluoridated areas when looking at count of dmft, with participants in a fluoridated area showing an, on average, a lower dmft score.

These data were further explored in a negative binomial regression with the count of dmft as the dependent variable, deprivation (quintile), age and sex entered as covariates and an offset for the number of erupted teeth. The regression shows that the incidence rate of dmft for children in a fluoridated region was 0.61 (95% CI 0.44 to 0.86) that of children living in a non-fluoridated region, holding other variables constant in the model. The rate of dmft was significantly lower for children in WF groups when controlling for these variables (see Appendix 2, Tables 94 and 95, and Figure 20).

Secondary outcome: dental general anaesthetic for extractions

There was no evidence of a difference in the DGA experience by exposure group, when controlling for deprivation (quintile) and sex as covariates (OR 0.759, 95% CI 0.397 to 1.468) (see Appendix 2, Tables 92 and 93 for further details).

Secondary outcome: questionnaire data – dentally related problem

Generalised estimating equations (GEEs) were used to determine if there was a difference between groups for a child experiencing a self-reported dentally related problem (e.g. pain, trouble eating, impact on sleep) over time (i.e. 5 years of questionnaire data). No significant difference was detected by groups over time (see Appendix 2, Tables 96–99, for further details).

Additional analysis from NHS Business Services Authority data

Analysis was performed on data collected by the NHS BSA. These data contain the charge to complete a course of treatment and the dmft/DMFT recorded following an examination at the dentist. Children were included if they were part of the CATFISH study and aged either 3 or 4 years at the time of dental examination. A dmft score was calculated from data collected by NHS BSA, which detailed the number of filled, missing or decayed teeth. The largest number of decayed, missing or filled teeth was taken from any examination from when a child was 3 years old. Data were available for 860 participants.

There was no evidence of an effect on presence/absence of dental decay (based on a dmft score of 0 or a dmft score of ≥ 1) from WF according to a logistic regression when adjusting for deprivation (quintile) and sex as covariates (OR 1.04, 95% CI 0.705 to 1.525) (see Appendix 2, Tables 101 and 102). The breakdown of deprivation for children included in these data is also presented in Appendix 2, Table 100, and is comparable to the breakdown for the overall CATFISH population.

Health inequalities: primary outcome

A logistic regression model, including an exposure by deprivation interaction term, was undertaken to determine the effects of WF on the presence or absence of decay at each level of the IMD. This analysis was supplemented by calculating the resulting marginal predicted probabilities for interpretability. A multidegree of freedom test of the interaction term following the logistic regression indicated that there was no evidence of an exposure by deprivation interaction (F = 4.78; p = 0.3109) (see Appendix 2, Tables 103–105 for further details). From our analysis, there is no evidence that deprivation influences the relationship between water exposure and the presence or absence of decay.

Figure 9 shows the resulting marginal predictive probabilities for decay (into dentine) or no decay in primary teeth based on a model that included age, sex and an interaction term for exposure by deprivation quintiles for the clinical assessment at the median age of 5 years. The predictive probabilities are displayed separately for males and females in Appendix 2, Figure 23.

FIGURE 9.

Adjusted predictions of decay or no decay including an interaction term for area across deprivation quintiles with 95% CIs at 5 years old. Decay is at tooth level, with the decay component being any decay into dentine.

Health inequalities: secondary outcomes

A negative binomial regression, including an exposure by deprivation interaction term, was undertaken to determine the extent to which WF reduced health inequalities. This was achieved by examining the count of decay and the interaction between WF status and deprivation, and including the covariates age, sex and offset by the log of number of erupted teeth. This analysis was supplemented by calculating the resulting marginal predicted probabilities for interpretability. A multidegree of freedom test of the interaction term following the negative binomial regression indicated that there was insufficient evidence of an exposure by deprivation interaction (F = 1.29; p = 0.8639). From our analysis, there is no evidence that deprivation influences the relationship between water exposure and the severity of dental caries, as indicated by dmft. Full results from the model are reported in Appendix 2, Tables 106–108, and Figure 26.

Subject recruitment and retention

Figure 10 provides details of the subject flow through the study, demonstrating the recruitment in fluoridated and non-fluoridated areas.

FIGURE 10.

A STROBE flow diagram for the older school cohort. Potential participants included those attending reception in Cumbria schools in September 2013–August 2014. Invited to participate/identification was based on area of recruitment (school).

Examination status at the final clinical examination is presented in Table 20, according to fluoridated or non-fluoridated water status. There were fewer participants in the fluoridated area than in the non-fluoridated area at baseline (owing to the geographic restrictions of the fluoridated area). When comparing the percentage of participants who had a final examination, the non-fluoridated group had 66% of participants completing their final examinations (of those originally recruited) compared with 79% of participants in a fluoridated area, with lack of access to schools in the non-fluoridated area (on account of COVID-19) being the principal source of the differential attrition.

Demographics and behaviours were explored to assess if there were any significant differences between the two groups (i.e. intervention and control). Table 21 details the demographics and deprivation for the fluoridated and non-fluoridated groups across each wave/clinical examination. Only children who had an examination at that time are included in Table 21. Ages were similar between groups at each wave. Deprivation followed the same pattern at each wave, with children in the fluoridated group having an, on average, higher (i.e. more deprived) deprivation score. In addition, we explored oral hygiene and related oral health behaviours periodically throughout the study (i.e. at baseline and then at each clinical examination). GEEs were carried out to determine if significant differences occurred between groups over this time period and whether or not they needed to be included in a regression analysis to adjust for these variables (see Appendix 3, Tables 112–149 for all tables, descriptive statistics and analysis relating to potential confounders and bias). Variables included within regression analysis were deprivation (quintiles), sex, dmft at baseline and age.

Secondary outcome: dental general anaesthetic for extractions

In the intervention group, 2.1% of children experienced a DGA. In the control group, 2.8% of children experienced a DGA. There was no evidence of a difference in the DGA experience by exposure group when controlling for deprivation (quintile) and sex as covariates (OR 0.81, 95% CI 0.51 to 1.26) (see Appendix 3, Tables 136 and 137, for further details).

Secondary outcome: decayed, missing or filled teeth (permanent)

Table 23 details the count of DMFT (generated from clinical examination scores) for each participant.

These data were further explored in a negative binomial regression in which deprivation quintile (reference category 1, least deprived quintile), age, dmft at baseline and sex (reference category male) were entered as covariates with an offset of number of erupted permanent teeth. The regression shows that the incidence rate of DMFT for children in a fluoridated region was 0.69 (95% CI 0.52 to 0.93) that of children living in a non-fluoridated region, when holding other variables constant in the model. The rate of DMFT is significantly lower for children in the WF group than for children in the no WF group when controlling for these variables (see Appendix 4, Tables 139 and 140 and Figure 32).

Health inequalities: primary outcome

A generalised linear model with logistic link function and including an exposure by deprivation interaction term was undertaken to determine the effects of WF on the presence or absence of permanent decay at different levels of the IMD. This analysis was supplemented by the resulting marginal predicted probabilities for interpretability. A multidegree of freedom test of the interaction term following the logistic regression indicated that there was insufficient evidence of an exposure by deprivation interaction (F = 2.68; p = 0.613). From our analysis, there is no evidence that deprivation influences the relationship between water exposure and the presence or absence of decay. Full results from the model are reported in Appendix 3, Tables 141–144.

Figure 11 shows the resulting predictive probabilities for decay or no decay in permanent teeth based on a model that included age, sex and an interaction term for exposure by deprivation quintiles for the median age of clinical assessment at 10 years.

FIGURE 11.

Adjusted predictions of decay or no decay including an interaction term for area across deprivation quintiles with 95% CI at 10 years old. Decay is at tooth level, with the decay component being any decay into dentine.

Health inequalities: secondary outcome

A negative binomial regression was undertaken to assess if WF contributes to reductions in health inequalities by examining the count of decay in permanent teeth and the interaction between WF status and deprivation, including age, dmft at baseline and sex as covariates, offset by erupted permanent teeth. A multidegree of freedom test of the interaction term following the negative binomial regression indicated that there was insufficient evidence of an exposure by deprivation interaction (F = 2.85; p = 0.5833). From our analysis, there is no evidence that deprivation influences the relationship between water exposure and the severity of dental caries as indicated by DMFT. Full results from the model are reported in Appendix 3, Tables 144–146 and Figure 35.

Sample

A total of 2035 children were consented into the study and 1444 children had an examination at school; however, three children had no identifier to link to the questionnaire data, giving a sample of 1441 children in the study. A total of 514 children completed the questionnaire at 5 years (927 children did not complete the questionnaire). Two children had no valid deprivation quintile score (one child with questionnaire data and one child with no questionnaire data), giving 513 children with and 926 children without questionnaire data on deprivation. Data were assessed to determine whether or not IMD and/or fluoridation status was associated with completing a questionnaire (see Appendix 4, Table 150). There was no evidence that questionnaire completion was associated with children residing in a fluoridated area (p = 0.291), but completing the questionnaire was associated with living in less deprived areas. Interacting fluoridation with deprivation quintile found that the relationship between deprivation and completion did not vary across fluoridation groups, implying that the potential bias caused by deprivation is a similar bias across fluoridation groups.

Of the 513 children with questionnaire data, eight had no follow-up dates (seven children because no birth date was provided and one child because no date of questionnaire completion was provided). In addition, one child had a follow-up date of zero, which was deemed a coding error. Subsequently, this gives a sample of 504 children in the birth cohort for the health economic analyses.

Child Health Utility 9-Dimensions health status, utility and quality-adjusted life-years

The responses to the individual domains of the CHU9D are presented in Appendix 4, Tables 151 and 152, for children with complete CHU9D data and children with complete CHU9D data and cost data (i.e. the CCA sample). Table 27 includes the utility scores estimated from the CHU9D domains and published population weights. The utility scores are anchored at 0 (dead) and 1 (full health), with some states also valued as worse than death. There were no clear differences between the fluoridated and non-fluoridated groups at follow-up.

Service use and costs

Cost and quality-adjusted life-year differences and cost-effectiveness of water fluoridation

Table 33 reports the net costs and QALYs after adjusting for differences in covariates. The data suggest a net lower cost for WF and a small net gain in QALYs. It is important to note that both the net cost and QALY data are associated with wide CIs that include zero and, therefore, differences are not statistically significant.

Table 34 reports the bootstrap simulations of cost and QALY differences, as well as the results of the cost-effectiveness acceptability analyses. The cost-effectiveness plane (Figure 12) illustrates the scatter of the 10,000 simulated cost and QALY differences and Figure 13 shows the cost-effectiveness acceptability curve. Overall, there is no clear difference in costs or QALYs, with a mean net benefit of £246 (95% CI –£416 to £905). In this scenario, at a WTPT of £20,000, the probability that WF is cost-effective is 0.7698.

FIGURE 12.

Cost-effectiveness plane of cost and QALY differences, primary analysis.

FIGURE 13.

Cost-effectiveness acceptability curve: probability that WF is cost-effective, primary analysis.

Sensitivity analyses

A range of sensitivity analyses were performed, including assuming that the missing data mechanism is MAR, attributing costs of fluoridation to the population aged 0–12 years, and an assessment with clinical measures of the volume of decayed, missing and filled teeth avoided and the presence of no decay.

Sample

There were 1662 children in the study, with 1192 children having an assessment at age 11 years. Nineteen children had missing data at follow-up, five children because follow-up was > 7 years (a clear coding issue) and 14 children because of incomplete follow-up or baseline assessment dates. Of the 1173 children with dates at follow-up, nine had no age recorded and a further 19 had no deprivation recorded and, therefore, this gave a sample of 1145 children. Of the 1145 children, 392 had CHU9D data at baseline (incomplete, n = 2) and 388 had CHU9D at follow-up (incomplete, n = 4). In total, 388 (33.89%) children had complete CHU9D data at both time points, and this gives a sample of 392 children in the older school cohort for the health economic analyses.

In the case of the 1145 children with complete dates, age and deprivation quintile score related to their postcode, we assessed whether or not IMD, age and/or fluoridation was associated with questionnaire completion (see Appendix 4, Table 155). There was some evidence that questionnaire completion was positively associated with children residing in a fluoridated area (p = 0.050) and/or in less deprived areas. Interacting fluoridation with deprivation quintile found that the relationship between deprivation and completion did not vary across fluoridation groups, implying that the potential bias caused by deprivation is an equivalent bias across fluoridation groups.

Child Health Utility 9-Dimensions health status, utility and quality-adjusted life-years

The responses to the individual domains of the CHU9D are presented in Appendix 4, Tables 156 and 157, for those with complete CHU9D data and Appendix 4, Tables 158 and 159, for those with complete CHU9D data and cost data (i.e. the CCA sample). Table 45 includes the utility scores estimated from the CHU9D domains and published population weights. The utility scores are anchored at 0 (dead) and 1 (full health), with some states also valued as worse than death. There were no clear differences between the fluoridated and non-fluoridated groups at baseline or follow-up.

Service use and costs

Cost and quality-adjusted life-year differences and cost-effectiveness of water fluoridation

Table 51 reports the net costs and QALYs after adjusting for differences in covariates. The data suggest a net lower cost for WF and a small net gain in QALYs.

Table 52 reports the bootstrap simulations of cost and QALY differences, as well as the results of the cost-effectiveness acceptability analyses. The results show similar cost and QALY differences to the analysis adjusted for covariates. The cost-effectiveness plane (Figure 14) illustrates the scatter of the 10,000 simulated cost and QALY differences and Figure 15 shows the cost-effectiveness acceptability curve. Overall, there is no clear difference in costs or QALYs, with a mean net benefit of £255 (95% CI –£779 to £1250). In this simulation, with a WTPT of £20,000, the probability that WF is cost-effective is 0.6831.

FIGURE 14.

Cost-effectiveness plane of net costs and QALYs, primary analysis.

FIGURE 15.

Cost-effectiveness acceptability curve, probability that WF is cost-effective, primary analysis.

Sensitivity analyses

A range of sensitivity analyses were performed, including assuming that the missing data mechanism is MAR, attributing costs of fluoridation to the population aged 0–12 years, and an assessment with the clinical measures of the volume of decayed, missing and filled teeth avoided and the presence of no decay.

Timing and settings of the studies

Improvements in oral health over recent decades is a substantial confounder when assessing the appropriateness of existing models, and these improvements will vary over settings too. The setting of the studies is important, given that oral health and the provision of dental care will vary across settings. Two119,120 of the 24 studies were UK studies. The UK studies119,120 identified relate to data pre 1990 and, as such, are more than 30 years old.

Outcome measure being investigated in studies

Our prime focus for the study is in cost–utility analysis, using the QALYs outcome obtained via utility measures sourced from the CHU9D. Studies identified in the scoping review comprised 15 cost–benefit studies,121–135 nine cost-effectiveness studies120,132,133,136–141 and four cost–utility studies. 119,142–144 Of the four economic evaluations identified as cost–utility analyses, two142,143 measured disability-adjusted life-years, one119 measured quality-adjusted tooth-years and one144 measured QALYs. The two UK studies related to cost-effectiveness analysis (i.e. DMFT and DMFT averted)120 and cost–utility analysis (i.e. quality-adjusted tooth years). 119

Perspectives taken in studies

Two UK studies119,120 took the payer perspective. This aligns with the within-study evaluation in this evaluation, which took a payer perspective, assessing NHS and WF costs.

Time horizon in studies

The time horizon should be sufficient to capture all impacts of the intervention. Given that decay and DMFT are absorbing states (one cannot revert back to no decay of DMFT), a lifetime horizon is appropriate. The time horizon differed over the two UK studies,119,120 that is, one took a lifetime approach and the other adopted a time horizon of 14 years.

Conclusion

These two UK studies, given their setting, perspective and time horizon, appear most appropriate to inform an economic model of WF in Cumbria. However, when the UK studies were carried out, oral health was somewhat different from how it was during the time frame of this study, and beyond. Furthermore, the studies rely on the same literature as the York and Cochrane reviews,16,71 which were criticised for being of poor quality and at high risk of bias. The studies also have a different health outcome. The single study assessing QALYs was based in New Zealand144 and this suggests that there is insufficient data in existing studies to inform the long term-effects of WF.

Mariño and Zaror84 concur with this, noting limitations of the existing evidence base with regard to the current context of declining dental caries incidence, greater tooth retention and an ageing population. In addition, adverse effects, such as fluorosis, were considered to be negligible and the study did not include costs associated with this, prompting the authors to recommend future models to consider these effects.

The long-term impacts of WF need to be examined. Although the within-study evaluation finds evidence that WF is likely to be cost-effective, the study does not inform potential impacts on the adult population, nor does it incorporate potential adverse effects, such as fluorosis. Evaluations of the impacts of WF on adults in Cumbria are needed, as too are assessments of fluorosis both in adults and in the child cohorts of the within-study evaluation.

The birth cohort

As expected, the prevalence of caries in the primary dentition was lower than in previous studies (i.e. 21% compared with levels almost double this in studies included in the recent Cochrane review), demonstrating and emphasising the need for the study. The difference in the primary outcome measure (i.e. the proportion of children who developed dental caries) was statistically significant, with an adjusted OR of 0.74 (95% CI 0.56 to 0.98). Despite this, the difference in the proportion of children with dental caries was modest, with 21.4% of children developing dental caries in the non-WF group and 17.4% of children developing dental caries in the WF group and an absolute difference of 4% might be judged by some stakeholders to be not sufficiently large or of clinical or public health significance. Nevertheless, the absolute difference is in line with the latest results from Public Health England’s Water Fluoridation Health Monitoring Report for England 2022,76 which demonstrated an absolute reduction of 4.7% in the prevalence of dental caries experience of 5-year-olds between a WF area (above 0.7 ppm) and a non-WF area.

When levels of deprivation were considered, the analysis involving an interaction term within the regression model looked to see whether or not the interpretation of the effect of WF depended on the value of deprivation (and vice versa), with the null hypothesis that the intervention effect is equal across subgroups. The performance of WF on dental caries experience was seen to be similar across deprivation quintiles. It is important to note that the overall effect size for WF on dental caries experience was smaller than expected. Therefore, it is not surprising that differences were not observed across the deprivation quintiles. There was no clear pattern when looking at deprivation, WF and dental caries experience, with no evidence of a shift in benefit for either those most or least deprived.

Secondary outcomes in this cohort demonstrated a significant decrease in the count of dental caries according to dmft, but no significant difference in the number of children requiring extractions of teeth under DGA or reporting issues relating to poor oral health, which included pain, problems with talking, eating or sleeping and being upset.

The use of blinded remote scoring of dental caries by an independent examiner suggested no underlying systematic bias in the clinical scores used for the main analysis.

Questionnaires to capture modifying factors were poorly completed, thereby reducing their value. Eleven different factors were assessed, including, diet, weaning history, toothbrushing and toothpaste use. Of these, only two differed between the control and intervention groups [i.e. consuming sweets on ≥ 3 days in each week (which was higher in the WF group) and having ever attended a dentist (which was lower in the WF group)]. Given the direction of the statistical difference seen in the primary outcome and the possibility of type I error, and given that the low response rate would have resulted in a reduced sample size, these two factors were discounted in the analysis.

The older school cohort

The low caries levels seen in the primary dentition of the birth cohort at 5 years old were also seen in the permanent dentition of the older school cohort when children were approximately 11 years old, with caries in the permanent teeth of 21.9% prevalence in control group and 19.1% in the intervention group. For the primary outcome measure (i.e. the difference in the proportion of children who were caries free in each group), there was insufficient evidence of an effect on the presence or absence of decay in the permanent dentition according to WF, with an adjusted OR of 0.8 (95% CI 0.58 to 1.09). However, it should be noted that dental caries is a chronic, long-term condition and develops throughout life and, hence, it is plausible that, as the disease progresses, the small difference (of 3%) recorded between the two groups may widen. Again, there was no observed impact on health inequalities.

When the difference in DMFT between groups was assessed (which included only teeth that had erupted after the intervention), there was evidence to suggest that WF was likely to have a beneficial effect on DMFT count, but there was uncertainty surrounding the estimate, which ranged from a substantial benefit to a less clinically important benefit. No difference was seen in the numbers of children requiring extractions under DGA.

When considering the health questionnaire data, across all the modifying factors measured, the only significant difference was seen in post-brushing behaviour, with children in the intervention group more likely to rinse their mouth after brushing than to spit residual toothpaste out.

Stability of intervention

Community WF is defined as the artificial supplementation of fluoride to drinking water supplies at the required concentration in England of 1 ppm. This dosing concentration has usually been assumed to be consistent and optimal during studies assessing effectiveness of WF. However, we know from the work of Moore et al. 89 that this is not always the case, as supplies are often interrupted and dosing may not be provided consistently at the optimal level.

During the current study, there was substantial interruption to the dosing of water supplies caused, in part, by a series of major flooding events that hit Cumbria at the end of 2015 and start of 2016, as well as the innate fragility of the plants themselves.

Figure 7 (see Chapter 7) displays the dosing from each of the two treatment plants during the duration of the study and shows both the impact of the 2016 floods and the general picture of suboptimal dosing activity, even from these newly commissioned plants. The Cornhow supply was more stable and more resilient to the impact of the flooding event than Ennerdale. It is important to note the profile of dosing seen in these plants is by no means unusual,89 as most plants in England fail to dose consistently or optimally.

The analyses undertaken for both the primary and secondary outcomes followed an ‘intention-to-treat’ model in that the actual dosing was not considered within the models. This is a legitimate treatment of the study data, given the public health nature of the intervention and the wide variation in water treatment plant performance seen across fluoridation schemes in the UK.

However, for completeness, a further analysis was undertaken, splitting the participants by their water treatment supply and, therefore, creating the three principal groups of Ennerdale (interrupted dosing), Cornhow (more stable dosing) and a non-fluoridated (control) supply.

The results of this analysis are shown in Table 19 for the birth cohort and in Appendix 3, Tables 147–149, for the older school cohort. There was insufficient evidence of a difference in caries experience in those children who received their drinking water from Ennerdale (interrupted dosing), whereas a significant difference was seen in those served by Cornhow (more stable dosing) for the birth cohort. No significant difference was observed for the older school cohort between either of the water treatment plants and the control. For the birth cohort, this could indicate that, if dosing remains stable at the optimal level, then the results from the CATFISH study could slightly underestimate the effectiveness of a perfectly delivered dosing system. However, this is a post hoc ancillary analysis, comparing optimal and non-optimal intervention delivery with a control, and so the results should be treated with caution. From our previous work,89 we know that optimal performance in fluoride delivery is rare and unlikely to become commonplace.

Health economic findings

The economic evaluation was undertaken from an NHS and local authority perspective. Costs included the capital and running costs of WF, as well as NHS dental activity. The measure of health benefit was the QALY. The cost and QALYs were combined to estimate the ICER (i.e. cost per QALY gained). For the within-study analyses, the time horizon was 5 years for the birth cohort (i.e. ages 0–5 years) and 6 years for the older school cohort (i.e. ages 5–11 years). QALYs gained from baseline to end of follow-up were estimated as the number of days survived multiplied by utility values for health-related quality of life reported on the CHU9D questionnaire. There were a large number of missing data due to questionnaire non-response (31.76% in the birth cohort and 33.89% in the older school cohort). The estimates of net costs and outcomes were bootstrapped (10,000 bootstraps) to generate cost-effectiveness acceptability curves of the probability that fluoridation was cost-effective at a range of threshold values for the willingness to pay for a QALY.

Sensitivity analyses included imputation of missing data, where costs of WF are apportioned to only children aged 0–12 years, and the use of alternative clinical outcome measures, that is volume of decayed, missing and filled teeth avoided and presence of no decay. The sensitivity analyses provided a consistent picture of a high probability of WF being cost-effective. Our approach to apportion WF costs has been conservative, as we assume a capital life of 6 years and this would overstate the costs for the WF group. Additional sensitivity analyses could explore allocating a longer lifetime for capital, although this would increase the probability of cost-effectiveness only by decreasing costs for children in the WF group.

Water fluoridation represented a small proportion of costs, at £14.14 per capita (£105.63 when apportioned to each child aged 0–12 years), in comparison with NHS dental services, which were over 10 times this amount for the birth cohort and over three times this amount for the older school cohort. Changing the study perspective to one of the NHS only would only further the dominance found for WF.

For the birth cohort, there is a 77.0% probability of WF being cost-effective at a willingness to pay £20,000 per QALY. For the older cohort there is a 68.3% probability of WF being cost-effective. The figure of £20,000 was chosen because this is the standard threshold used to determine whether or not interventions constitute good value for money for the NHS. All of the economic analyses give a high probability (p > 0.62) that WF is cost-effective based on willingness-to-pay thresholds of £20,000 per QALY/DMFT avoided/decay avoided.

For research

For policy

Over the last 30 years, population caries levels have diminished very significantly so that even in this deprived population only 20% of children were at risk of developing caries into dentine. Although this figure is still high when we consider disease burden and demonstrates that further intervention is necessary to continue to reduce the number of children experiencing dental decay, it becomes much harder to make the case for a whole population intervention given these declining prevalence levels, especially when its benefits are small at best and could be outweighed by risks of fluorosis. The importance of continuing to follow up the birth cohort to determine the impact of both fluorosis and caries in the permanent teeth has been highlighted in this report. Prevalence levels are an important factor to take into account when considering new WF schemes and the applicability of the results of this study to areas with widely different caries levels. We suggest that, where prevalence is low, consideration should be given to targeted approaches that may be as or more effective, as this may attract less criticism from those who believe that WF is an inappropriate intervention. Such targeted approaches may include modification of known risk factors, such as sugar consumption, rather than relying on the biopharmaceutical effect of fluoride alone.

The ‘silver bullet’ approach that was effective in the 1950s is failing to address dental caries in the most deprived populations and, hence, a hybrid approach with behavioural elements should be considered, alongside considering the social determinants model described the Marmot review,49 to address what is a complex and multifactorial disease. Although the attractiveness of what appears to be a simple intervention to universally prevent disease is clear, it is also apparent that the changing epidemiological nature of dental decay requires further consideration in the future.

The inability of the public health authorities to introduce a new scheme in the UK in the last 30 years demonstrates the logistical and political problems of delivering this intervention. The history of WF over the last 30 years has resulted in resources being ploughed into trying to implement new WF schemes where resources may have produced greater benefit if they had been deployed in other areas. Into the future, assuming the continued use and availability of fluoride toothpaste, policy-makers will have to trade-off the costs of the introduction of new schemes against other public health interventions.

Previous members

• Paul Wilson (University of Manchester, formally University of York) left after moving to Manchester.

• Brenda Gannon (University of Manchester) moved to Queensland, Australia, and could no longer be part of the committee.