Dietary intakes of young children are suboptimal [1]-[3]. Low intakes of fruits and vegetables and high intakes of energy-dense, nutrient-poor (non-core) foods in early childhood are likely contributing to high rates of overweight, obesity and chronic disease across the lifespan [4]-[6]. Interventions to improve these aspects of young children’s diets are therefore vital.

Health promotion interventions for young children are most likely to be effective if they involve parents [7],[8], and target factors likely to be important influences on children’s diets. Maternal nutrition knowledge has shown associations with better child diet quality in preschoolers [9],[10]. Child feeding practices which have shown associations with healthier diets of young children include lower parental pressure [11], lower parental control of feeding [12], higher parental covert and overt control [13], higher child control [14], and lower use of foods as rewards [14]-[16]. Maternal self-efficacy in child feeding has shown associations with higher children’s vegetable intakes [17], and lower child intake of non-core foods [17],[18]. Furthermore, maternal role modeling of dietary intakes, reflected by maternal diet, is associated with young children’s diets, for example, for fruits and vegetables [12],[16],[19],[20], and sweets and soft drinks [16]. When providing a child health promotion intervention to parents, it is therefore likely to be important to target these constructs. Improvements in these maternal constructs should promote improved child diets, however, these hypotheses have been infrequently assessed, particularly with parents of children under five years of age.

Mediation analysis facilitates the assessment of how an intervention effect was achieved, and an understanding of which aspects of an intervention were most effective or important and which may need improvement. In statistical terms, mediation analysis seeks to identify an intermediate variable in the causal sequence relating an independent variable to an outcome variable [21]. By testing the theories on which an intervention is based, mediation analysis can assist in the interpretation of trial results and can inform the development of more parsimonious future interventions [21]. The importance of mediation analyses to assess intervention and prevention trials has been previously highlighted [22]-[25], however, relatively few studies have undertaken such analyses of dietary interventions in children [26].

The Melbourne Infant Feeding Activity and Nutrition Trial (InFANT) Program was a health promotion intervention which significantly improved diet quality of children at 18 months of age [27]. The intervention aimed to achieve this via improving mothers’ knowledge of child feeding and nutrition, child feeding practices, self-efficacy for promoting healthy eating and modeling. Therefore the hypothesis tested in this study is that these maternal correlates would act as mediators of the intervention effect on children’s diet quality, as outlined in Figure 1. This study aims to assess which components of maternal behavior were impacted by the intervention and acted as mediators of the intervention effect on child diet quality.

The Melbourne InFANT Program was a cluster-randomized controlled trial to promote obesity-protective behaviors in the early childhood years, conducted from 2008-2010 [28],[29]. The intervention targeted nutrition, physical activity and sedentary behavior, but only nutrition and feeding-related outcomes will be discussed in this paper. The study was conducted through the existing social setting of first-time parents’ groups, which are run within local government areas (LGAs) in Victoria, Australia, by Maternal and Child Health nurses (about two thirds of new mothers attend [30]). Fourteen randomly selected LGAs were recruited, then parents’ groups within those LGAs were randomly selected and approached. Groups with at least eight eligible families (English-speaking, first-time parents) who provided written, informed consent, were randomized to either the intervention or control arm. An independent statistician undertook all stages of randomization using a random number schedule, to ensure there were no biases in allocation. Mothers and fathers were both permitted to participate, but most attendees were mothers, who are the focus of this paper.

The intervention involved six interactive sessions over 15 months, from when children were approximately four months of age. Sessions included educational activities, a DVD, promotion of peer discussion, and written materials. The intervention had a major focus on improving children s diet quality (specifically promoting fruits and vegetables and discouraging non-core foods), via improving maternal knowledge, feeding practices, self-efficacy and role modeling. Key principles of the intervention included offering anticipatory guidance [31], promoting division of responsibility in child feeding [32] and emphasizing parenting skills [28]. Participants in the control arm received quarterly newsletters on topics unrelated to the intervention, in addition to the usual care available to them in their local area from their Maternal and Child Health nurse. Ethics approval was granted by the Deakin University Ethics Committee and the Victorian Office for Children.

Post-intervention, children were 18.0 ± 1.5 months of age (at first 24-hour recall). Of the 480 participants (91% of baseline sample) who completed the trial, complete data for child diet recalls and at least one maternal mediator was provided by 375 mothers (71% of baseline sample). Those participants who provided complete data were more likely to have a higher education level than those who discontinued the study or provided insufficient data (data not shown). However, there were no differences in completion rates by pre-pregnancy BMI or maternal age at childbirth.

The relationship between the maternal mediators and child OPDI score (β pathways) was significant for nine of the 18 pathways, and three further pathways showed trends towards significance. The β values in Table 3 indicate the difference in ODPI score between the highest and lowest category of the mediator. All trends and associations were in the hypothesized direction.

Joint significance was therefore demonstrated for three mediators, as shown in Table 3. The difference of coefficients test was then conducted for these three mediators. The direct intervention effect (c) for the sample with complete data for all three mediators was 1.50 (p =0.01, n =357). The indirect effects (c-c’) for each mediator are shown in Table 3. Higher maternal knowledge of child feeding intervention messages and lower maternal use of food as a reward in the intervention compared to control arm mediated the intervention effect on child OPDI score, explaining 33% and 22% of the intervention effect respectively. Additionally, use of pressure in feeding, intentional modelling, and self-efficacy for limiting unhealthy foods could be considered to show trends towards mediation of the intervention effect, with trends towards significant α and β pathways.

These analyses are novel and valuable because they inform the targets and content of future interventions by highlighting maternal domains likely to be most important to achieving an impact on young children’s diets. The results indicate that higher maternal knowledge of child feeding and nutrition may contribute to small improvements in child diet quality, even with minimal concurrent improvements in maternal behaviors and self-efficacy. The finding of knowledge as a mediator of the intervention effect is consistent with other health promotion studies in older children and adults [62]-[64]. An important distinction and strength of this study, however, was that knowledge was measured by a set of 12 items, rather than a single item. Measurement using a factor comprised of multiple items is likely to be more robust than individual item analysis. The cumulative results of these studies highlight the value and importance of improvements in nutrition knowledge, and suggest that maternal knowledge can directly influence children’s diets, but that the effect of knowledge alone may be quite small. In this study, higher knowledge in the intervention arm explained 33% of the modest intervention effect on child diet quality.

The finding that lower use of food as a reward mediated the intervention effect is novel and informative. While one other study reports that improvements in parental feeding strategies mediated the effects of a nutrition promotion intervention on young children’s non-core foods intakes, it assessed a composite of feeding practices as the mediator rather than distinguishing between practices [18]. Two further studies in a similar age group have also reported intervention effects on parental use of food as a reward [65],[66], though those did not assess mediation. Considered together, these studies highlight the benefits of including parental feeding practices as targets of an intervention, and a need for further investigation of whether and why some feeding practices may be more amenable to interventions than others.

It is not known why reduced use of food as a reward acted as a mediator while other feeding practices did not. It was not emphasized more strongly in the intervention, and it did not show more need for improvement than other factors. Qualitative assessments and process evaluation of the intervention have also not identified any reason for this difference (findings not yet published). It is possible that this practice is particularly amenable to change. Following the Melbourne InFANT Program, in general it was the less healthy practices, or those not consistent with feeding recommendations, which differed between trial arms. Intervention participants used less food as a reward and less pressure, and trended towards being more confident to limit unhealthy foods, with minimal impact on improving healthy behaviors. Intervention effects on child TV and non-core sweet snacks, and mothers’ unhealthy dietary patterns, have previously been reported, but with no impact on child or maternal physical activity or fruit and vegetable intakes [29],[67]. It is possible that participants found it easier to reduce their unhealthy behaviors than to increase their healthy behaviors.

Important findings from these mediation analyses were the significant associations between many of the potential mediators and child diet quality (as shown by significant β pathways). Such associations support the conceptual theory behind the choice of maternal targets of the Melbourne InFANT Program, as maternal factors were related to child diet quality in the expected direction, even if the maternal factors were not influenced by the intervention. These findings highlight that these are worthy intervention targets for future programs because they are likely to influence child diets. However, the action theory and strategies used to influence these maternal mediators may benefit from refinement in future interventions. For example, many aspects of maternal diet were associated with child diet, so if intervention impact on maternal diet could be improved, and this act as a mediator, the effect on child diet may be greater. These findings are similar to those of Fletcher et al., who found that parental self-efficacy for providing healthy foods was associated with young children’s intake of non-core foods, but was not improved by a nutrition promotion intervention [18].

Only one previous study is known to have assessed parental mediators of young children’s diets following a nutrition promotion intervention [18]. That study offered a one month telephone-based intervention for parents with children aged 3-5 years. It found that child access to non-core foods and child feeding strategies mediated reduced child non-core food intake at 2 months (assessed by FFQ), but there was no intervention effect at 6 months. The current study therefore extends this small evidence base by reporting a wider selection of parental mediators, assessing child diet quality as the primary outcome, evaluating a 15 month group-based intervention, and focusing on parents of children under two years. Other studies which have assessed mediation of child nutrition promotion interventions have primarily involved older children, and assessed children’s own knowledge, behaviors and beliefs as potential mediators [57],[58],[62],[68]-[70]. While three of those papers did report parental diet and home fruit and vegetable availability [62],[68],[70], they were not found to be significant mediators of the intervention effect on child diets, likely due to the school-based nature of the studies and consequent minimal parental involvement.

A limitation of this study was the modest sample size, as this can limit the analytic techniques appropriate for mediation, and make it difficult to detect statistically significant mediation effects, particularly if the effect size is also small [71]. Additionally, those participants who provided adequate data for these analyses were more likely to have a higher education level than those who discontinued the study or provided insufficient data, which limits generalizability. It is also acknowledged that this single time-point data does not account for possible bi-directional relationships between maternal mediators and child dietary patterns [72]. However, longitudinal assessment was not applicable in this study, given that child feeding practices would not yet have been established at baseline when children were aged three months.

The unacceptable internal reliability of the child control subscale (Cronbach’s α 0.52) meant that this key aspect promoted within the intervention was not assessed as a mediator. In comparison, across the three validation studies in the United States, Cronbach’s α of this factor was variable: 0.49, 0.69 and 0.70 [33]. According to the principles of division of feeding responsibility [32], parents letting their child choose their foods from those served could be considered appropriate, and parents preparing alternatives if their child disliked what was served could be considered inappropriate. However, items assessing these behaviors load in the same direction on the factor, suggesting the factor may measure greater allowance of child control, not necessarily more appropriate child control. Further refinement and validation of tools to measure parental feeding practices in young children is required in future.

Strengths of this study include the high quality dietary assessment, modelled on the United States Department of Agriculture 5-pass system [73]. This method is used in national nutrition surveys [73],[74], and was strengthened in this study by multiple days of recalls [75]. Additionally, use of a diet quality index is a novel strength of this study, allowing assessment of relevant dietary outcomes in combination, which is important for interventions which target multiple dietary behaviors [76],[77]. Furthermore, the assessment of a variety of potential mediators, and the use of factors and scores rather than individual questionnaire items as mediators, distinguishes this study. This is also the first study known to have assessed maternal mediators of a health promotion intervention aimed at improving the diet of children under two years of age.

Higher maternal knowledge of child feeding intervention messages, and lower maternal use of food as a reward in the intervention arm, mediated the effect of the Melbourne InFANT Program on child diet quality. This highlights the importance of these elements in public health nutrition interventions for parents. A number of other potential maternal mediators were shown to be appropriate intervention targets, but require further research to consider how they may be more effectively impacted in future interventions to promote behavior change.