Effect of Early Life Income and Sugars Intake on Child Oral Health: Marginal Structural Modelling Using a Birth Cohort Study Open Access

The early years of a child’s life are a time when the foundation for future health and development is formed [1, 2]. Unhealthy children are less able to learn and develop and are more likely to grow into unhealthy adults. In many cases, unhealthy children will be more likely to have unhealthy offspring themselves, leading to intergenerational disadvantages, for instance, poor maternal oral health is associated with poor child oral health [3]. Recent research has suggested that the roots of chronic conditions need to be traced back to as early in life as possible [4‒6]. Variations in health and development exist in the early years of life [7]; although of lower magnitude than those observed among adults, they can lead to substantial divergence in health trajectories later in life [8]. Targeting the sources of variations in health and development in early life is important for health and economic benefits for both individuals and society, as well as addressing socioeconomic inequalities in health.

There are many reports from the developed world which have shown that socioeconomic status (SES) is a strong determinant of oral health [9, 10]. In Australia, socioeconomic inequalities in children’s caries experience have been reported [11, 12]. Children from low SES households had already accumulated considerably greater caries experience than children from high SES groups, even by 5 years of age [13, 14]. There is also evidence that the socioeconomic gradient in child oral health has widened during the past decades [15]. It is, therefore, imperative to identify early life determinants of socioeconomic inequalities in oral health.

Oral health is influenced by a multitude of factors [16] at different levels – societal, community, social network, family, as well as individual. These multilevel influences are operative from early in life [17]. It is speculated that they interact with each other to impact child oral health as well as through key oral health determinants, such as free sugars intake, an individual determinant which was found to impact child oral health both immediately and later in life [18, 19]. However, much of the available evidence has not been confirmed using a causal inference approach.

Evidence of potential causal pathways and the magnitude and direction of effects is important to inform appropriate actions [20]. Longitudinal research is capable of such a task [21]. Given the complexity of the relationships among factors, properly developed causal questions and analytical methods are required to estimate the effects consistently [22, 23]. Application of causal inference approaches in dental research has been discussed and promoted recently [24, 25].

This paper aimed to investigate pathways through which SES early in life impacts dental caries. A causal question was to be answered: does household income at childbirth cause dental caries by age 5 years, not mediated by the free sugars intake at age 2 years? To do so, we applied a causal inference approach and generated a marginal structural model to consistently estimate the controlled direct effect of household income at children’s birth and the direct effect of intake of free sugars at age 2 years on dental caries experience at age 5 years.

The Study of Mothers’ and Infants’ Life Events Affecting Oral Health (SMILE) is a population-based birth cohort study established in Adelaide, Australia, in 2013. SMILE applies an observational study design to follow prospectively a cohort of socioeconomically diverse newborns and their mothers from birth of the children [26].

Parents of the children completed multiple age-specific questionnaires to collect data on a wide range of factors in six waves. Data from three of those waves (baseline, age 2 years and age 5 years) were used in this study (see online suppl. Table 1; for all online suppl. material, see https://doi.org/10.1159/000546215).

At baseline, data on parental SES was collected, together with other factors. The first exposure (E1), annual household income at baseline (i.e., childbirth), was categorised as: low (<AUD 60,000), medium (AUD 60,000–AUD 100,000), and high (>AUD 100,000). The second exposure (E2), intake of free sugars at age 2 years, was estimated as grams of free sugars/day from a customised 98-item food frequency questionnaire (FFQ) used to collect detailed dietary data [27]. The intake of free sugars in grams was used to group the sample into having a low (<12 g free sugars/day, equal to <5% estimated energy requirement [EER]); medium (12 to <25/g free sugars/day, equal to 5 to <10% EER); or high (≥25 g free sugars/day, equal to ≥10% EER) intake, based on WHO guidelines [28]. EER was used as a proxy for total energy intake that could not be estimated from the FFQ.

At 5 years of age, the children were examined for dental caries by a small team of trained and calibrated oral epidemiological examiners [26]. Repeated examinations were not conducted, instead, the examiners conducted joint examinations frequently to maintain consistency. The primary outcomes were (a) the prevalence of dental caries in the primary dentition by age five (the proportion of children with dmfs >0), and (b) their accumulated primary dentition dental caries experience (the dmfs score).

A confounding variable, area-level SES at baseline, was estimated using postcode-based Index of Relative Socioeconomic Advantage/Disadvantage (IRSAD) [29]. Other confounding variables were collected at baseline and included parental education attainment, parental occupation, parental work status, maternal country of birth and household composition. Those factors were expected to confound the relationship between the exposure and the outcome (see online suppl. materials for details of those variables).

A causal inference framework was needed to guide the analytical process to investigate causal relationship using observational data [21]. Understanding those causal pathways provides evidence of early life influences on socioeconomic inequalities in child oral health. A causal directed acyclic graph (DAG) was developed based on existing knowledge (shown in Fig. 1) [21]. Household income at childbirth influenced the intake of free sugars at age 2 years [30]. This relationship could be influenced by a set of confounding factors, which also influenced dental caries experience at age 5 years. This specified DAG informed estimation of total effect and controlled direct effect of household income, not mediated by the intake of free sugars at age 2 years (path 1: E1–Y), and direct effect of free sugars on dental caries experience (path 2: E2–Y). Those effects could be estimated consistently using a marginal structural model [21, 31].

It was necessary to control for the confounding effects of baseline factors on the exposure and the outcome [31]. Two sets of inverse probability of treatment weights (IPTWs) were estimated: three levels of household income given the confounding factors; and three levels of intake of free sugars, given the confounding factors and household income. A stabilised IPTW was estimated as the product of the first two IPTWs (see online suppl. methods). When investigating the IPTW values, 1 case was found to have an extreme value (35.0); it was excluded from the analysis. There were no cases with a weight of 0. The mean of the stabilised IPTW was 0.99 (SE 0.03).

The stabilised IPTW was used in the analysis to control for the confounding effect. It was important to demonstrate that the use of the stabilised IPTW resulted in a balance of baseline confounding factors, mimicking randomisation. We estimated distributions of the exposure by confounding factors with and without the stabilised IPTW. Differences between levels of confounding factors were examined to determine their balance by the exposure with the stabilised IPTW.

A multinomial regression model was used to investigate the link between the three levels of household income and the three levels of intake of free sugars. The confounding effect of baseline SES was controlled for using the stabilised IPTW.

Three multivariable regression models were generated for each primary outcome. Log binomial regression was used for the prevalence to estimate adjusted prevalence ratios (PRs). Log negative binomial regression was used for the experience of dental caries to estimate adjusted mean ratios (MRs) because dmfs scores are a count variable. Model 1 estimated the total effect of household income on dental caries experience, adjusted for baseline confounding factors. Model 2 was model 1 with intake of free sugars added to estimate effect of household income and intake of free sugars using conventional approaches. Model 3, using marginal structural models with the stabilised IPTW, estimated the controlled direct effect of household income on dental caries (path 1) and the direct effect of intake of free sugars on dental caries (path 2). All models were adjusted for child sex and age at examination. Estimates are reported with their 95% confidence intervals (95% CIs).

Natural indirect effects (NIEs) were estimated as the proportions of effects of medium- or high household income that were mediated by the intake of free sugars. NIE was adjusted for confounding effects and child age and sex.

As a sensitivity analysis, IPTW was estimated as above using imputed data of the confounders. The MSM was again generated as above (see online suppl. Table 2).

Analyses were conducted with SAS version 9.4 (SAS Institute Inc., North Carolina, USA). The study followed the STROBE Guidelines for reporting cohort research [32].

A total of 879 children had completed data collection at age 5 years and were included (Table 1). Over 23% of the sample had dental caries, with a mean dmfs score of 1.4. Among those with dental caries, median of dmfs score was 3 (IQR: 1–7). A quarter of the sample had low household income while 37% had high income. Over 42% of the sample had a high intake of free sugars while a quarter had low intake of free sugars. Over half (59%) had at least one parent with tertiary education, and most (81%) had at least one parent working full-time. Those in the low-income group were more likely to have dental caries and more tooth surfaces with caries experience than those in the high-income group. Similarly, the association between free sugars intake and the prevalence and experience of dental caries were marked.

Distributions of low-income households and of children with high intake of free sugars were estimated with and without the stabilised IPTW (Table 2). The proportion of low-income households was strongly associated with all confounding factors without the stabilised IPTW. Households with parental education at school level were twice as likely to have low income than households with parental education at tertiary level (40.0% vs. 20.9%). Similarly, those residing in the most disadvantaged areas were twice as likely to have low income than those in the most advantaged areas. The largest difference was observed between households with at least one full-time working parent and the other groups. After having been weighted by the stabilised IPTW, the distributions with low income became mostly balanced between levels of the confounding factors. Distributions of high intake of free sugars were also strongly associated with confounding factors. The largest difference was observed between the most disadvantaged areas and the most advantaged areas. Those distributions have also become balanced between levels of the confounding factors after having been weighted by the stabilised IPTW (Table 2; online suppl. Fig. 1).

There was a direct effect of household income at children’s birth on the intake of free sugars at the age of 2 years (Table 3). The income-related gradient in the proportions of children with high sugar intake was clear. Those in the low-income group had 2.78 times the odds of having higher levels of intake than those in the high-income group.

Three sets of models were generated sequentially for the prevalence of dental caries at age 5 years (Table 4). Model 1 showed that household income at birth was associated with the prevalence of dental caries. Children from low-income families had 1.6 times the prevalence and 3.5 times the cumulative experience of dental caries than those from high-income families. The effect of household income was attenuated in model 2 when the effect of intake of free sugars at age 2 years was also estimated. Both household income and free sugars intake were associated with the prevalence of dental caries in this model. The marginal structural model (model 3) showed further attenuation of the effects of household income on the prevalence of dental caries. Household income at birth was associated with negligible controlled direct effects on the prevalence of dental caries (path 1). The direct effect of intake of free sugars (path 2) on the prevalence of dental caries was sizable: children among the high-intake group were 1.55 times more likely to have dental caries than the low-intake group.

For the experience of dental caries, model 1 showed a sizable total effect of baseline household income, with a marked gradient in effect (Table 3). Children from low-income households had 3.5 times the dmfs score of those from high-income households. The effects attenuated in model 2 when free sugars intake was added. Both household income and free sugars intake showed marked gradients with cumulative dental caries experience. The marginal structural model (model 3) showed that both household income and free sugars intake had measurable effects on cumulative dental caries experience by age 5 years. Household income at childbirth had sizable controlled direct effects on dental caries experience (path 1): children from low-income families had 1.65 times the dmfs score than their counterparts from high-income families. The direct effects of intake of free sugars at age 2 years on dental caries experience (path 2) were of a relatively larger magnitude (MR: 2.64).

Estimated NIE values show that relatively larger proportions of effects of medium income were mediated by the intake of free sugars. For the prevalence of dental caries, the NIE values were 12.8% and 6.9%. For the experience of dental caries, those values were 11.6% and 8.0%, respectively.

The study has conceptually identified and empirically evaluated pathways through which socioeconomic determinants impact child oral health. Under the counterfactual framework, household income at childbirth had a measurable controlled direct effect on dental caries experience, and intake of free sugars at age 2 years had large direct effects on both the prevalence and cumulative experience of dental caries in the primary dentition. It provides consistent evidence that household income (a key measure of SES) and free sugars intake (a key indicator of health-related behaviours) are primary determinants of child oral health. Importantly, those effects were observed in early childhood.

This study has demonstrated that variations in determinants of oral health can be observed early in life. Using the sophisticated causal inference approach, the study demonstrates that early life socioeconomic factors, in this case household income, can both directly impact child oral health as well as through key determinants of child oral health, such as intake of free sugars. Those impacts can result in socioeconomic variations in child oral health later in life. Such variations are the result of complex interactions between a multitude of environmental factors surrounding young children. The observed association among early life factors, key immediate determinants of oral health and child dental caries experience can be explained by the dependence of young children on their immediate environment, which, in turn, can be impacted by complex interactions with other socioecological factors.

A major strength of the study lies in the prospective nature of its data collection, allowing sequential development of a causal inference framework from birth, age 2 and age 5 years. The study sample was representative of the general population [26], allowing for generalisation of the study findings to the source population. The prevalence and experience of dental caries reported in our SMILE sample were similar to those data reported in the National Child Oral Health Study [14]. Intake of free sugars was collected in detail using a 98-item food frequency questionnaire that measured free sugars intake in grams and estimated percent of intake of sugars relative to energy requirement. Such detailed measurements allow for direct policy and practice recommendations. A limitation of the study was loss to follow-up, which is a problem common to all longitudinal studies. There was evidence that the attrition rates among those from lower SES backgrounds were relatively higher than those from higher SES backgrounds [26]. This was anticipated when designing the study and oversampling of participants from lower SES backgrounds resulted in the analysis sample still being representative of the target population [33]. There were potential issues with social desirability in responses. However, collecting free sugars intake using 98 items of multiple foods and drinks limited such issue. The nature of the study meant that dental examinations for most children were conducted in their own home, so it was difficult to organize repeat examinations for a reliability assessment. However, there were only three examiners for this clinical examination round who have worked together in oral epidemiological assessment since 2006. The examiners frequently worked together as a team, alternating roles as examiner and data recorder to maintain consistency during the fieldwork.

This study is one of the first few internationally to investigate causal relationships between key determinants of oral health and child dental caries by applying the causal inference approach to observational data. This approach is a new methodological advance in investigating causation [21, 34], and it has been described as the way forward for epidemiology [20]. We have successfully developed a causal framework to guide the analysis and estimated and applied IPTWs to control for confounding effects. It has allowed the consistent estimation of controlled direct effects of household income at childbirth and direct effects of free sugars intake at 2 years of age on child dental caries at a later age. The advantages of our approach over conventional multivariable regression have been empirically demonstrated. The study findings provide evidence with a lower level of bias through confounding effects than conventional regression. This approach offers new opportunity for dental research and should be embraced and promoted [24].

Household income was used as a primary exposure in this study because it is easily defined and often used as a main criterion for government healthcare policies. Its use as a primary exposure allows direct application of the study findings. For example, the only public policy for child dental care in Australia, the Child Dental Benefit Schedule (CDBS), provides subsidised dental care for children aged 2 to 17 years of low-income households [35]. Our study provides evidence that expansion of CDBS to childbirth may bring larger benefits in providing timely information for those in need. Our recent research based on the SMILE cohort has developed a short screening tool for measuring free sugars intake at ages 2 and 5 years [36]. Such a tool allows for quick assessment by healthcare professionals to advise parents of young children in timely controlling intake of free sugars.

This study’s findings lend support for the integration of oral and general public health measures targeting the first 1,000 days. Most preventive programs target children through educating about oral hygiene practices such as tooth-brushing with fluoridated toothpaste from an early age and reducing child consumption of foods and drinks high in free sugars. However, those programs do not reach far enough into the early life of children. Little has been done to support people with various socioeconomic backgrounds to care for their offspring’s oral health. It is noted that health education alone does not sufficiently lead to sustained behavioural change in individuals. Oral health behaviours and oral health problems are enmeshed in more complex daily habits, which are largely determined by a broad set of psychosocial, economic, and environmental factors [37, 38].

In addition to focusing on children, we suggest strategically improving the environment in which children are born and develop. It is crucial to recognize that maternal health attitudes and practices are likely to influence behaviours of children through either (i) the direct provision of foods and drinks or (ii) role-modelling, and so they play an important role in the initiation and continuation of favourable oral health practices [39‒41]. As has been highlighted, maintenance of child oral health must commence as early as possible [42]. Health promotion activities should be targeted at parents (particularly those in low socioeconomic background) to give them the opportunity to embark on healthy dietary practices, not only for themselves but also for their offspring. There should be concerted efforts to enhance a healthy environment at early childhood facilities, childcare, and preschool facilities that directly influence young children.

The study has provided consistent evidence that socioeconomic variations at birth directly influence oral health inequality later in life. Large proportions of those effects were not mediated by immediate determinants such as intake of free sugars, which also had direct significant effects on oral health. Oral health preventive measures need to commence as early as possible, targeting both socioeconomic and immediate determinants. Timely and appropriate addressing of these determinants may help limit inequities in oral health.

The SMILE research team and study participants are gratefully acknowledged.

Ethical approval was obtained: Women and Children Health Network Human Research Ethics Committee (HREC#13/WCHN/69, 07/08/2013); the University of Adelaide Human Research Ethics Committee (HREC#H-2018-017, 16/10/2018). Parents of SMILE children provided informed written consent at the baseline, and prior to all clinical assessment.

Prof. Dr. David J. Manton was a member of the Journal’s Editorial Board at the time of submission. The authors have no other conflicts of interest to declare.

SMILE is supported by NHMRC Project Grants # APP1046219 2013-17 and APP1144595 2018-22.

D.H. Ha, L.G. Do contributed to conception and design, data acquisition, analysis, and interpretation, drafted manuscript, and critically revised manuscript. L. Bell, G. Devenish-Colman contributed to conception, data acquisition and analysis, critically revised manuscript. A.J. Spencer, W.M. Thomson, J.A. Scott, and D. Manton contributed to conception, data acquisition and interpretation, and critically revised manuscript. S. Leary contributed to conception, data analysis and interpretation, critically revised manuscript. All authors gave final approval and agree to be accountable for all aspects of the work.

The data that support the findings of this study are not publicly available due to their containing information that could compromise the privacy of research participants but are available from the corresponding author [L.G.D.] upon reasonable request.