Interventions using behavioural insights to influence children's diet‐related outcomes: A systematic review

The global prevalence of children with overweight and obesity continues to rise. Obesity in childhood has dire long‐term consequences on health, social and economic outcomes. Promising interventions using behavioural insights to address obesity in childhood have emerged. This systematic review examines the effectiveness and health equity implications of interventions using behavioural insights to improve children's diet‐related outcomes. The search strategy included searches on six electronic databases, reference lists of previous systematic reviews and backward searching of all included studies. One‐hundred and eight papers describing 137 interventions were included. Interventions using behavioural insights were effective at modifying children's diet‐related outcomes in 74% of all included interventions. The most promising approaches involved using incentives, changing defaults and modifying the physical environment. Information provision alone was the least effective approach. Health equity implications were rarely analysed or discussed. There was limited evidence of the sustainability of interventions—both in relation to their overall effectiveness and cost‐effectiveness. The limited evidence on health equity, long‐term effectiveness and the cost‐effectiveness of these interventions limit what can be inferred for policymakers. This review synthesises the use of behavioural insights to improve children's diet‐related outcomes, which can be used to inform future interventions.


| INTRODUCTION
The number of children aged five through 17 with obesity is projected to increase globally from 77.8 million in 2013 to 91.2 million by 2025 without substantial intervention. 1 Obesity in childhood has negative long-term consequences on health, social and economic outcomes. 2 The primary driver of obesity is energy imbalance, which is often caused by a poor diet consisting of an excess of energy-dense foods. 2 Many children live in obesogenic environments, in which the consumption of energy-dense foods is encouraged through their increased availability, affordability and promotion. 2 To address the obesogenic environment and reduce the global burden of obesity in childhood, multiple policy levers are required. 2 Policies informed by 'behavioural insights' (BIs) have demonstrated potential for improving children's diets. BIs is a broad term used to encapsulate the results of research on human behaviour undertaken in disciplines such as economics, psychology, sociology and neuroscience, providing an understanding of how human beings make choices (behave) given individual endowments, experiences and external influences. 3 BIs draw on a range of theoretical frameworks from behavioural economics and psychology and other social sciences, including Kahneman's Dual Process Theory, 4 Thaler's and Sunstein's Nudge Theory 5 and Bandura's Social Cognitive Theory. By drawing on a wide range of concepts and findings, BIs provide a more comprehensive understanding of the factors that influence human behaviour than individual theories of disciplines.
Typically, interventions using BIs influence an individual's behaviour through subtle changes to the social and physical environment without actively restricting available options. 6 These changes attempt to reduce the cognitive biases in the decision-making process that prevent individuals from adopting self-interested behaviours, 7 such as healthy eating. Previous research has shown incentives, social norms and environmental cues are important drivers of children's dietary choices. 8 Thus, children's reliance on social norms (adult or peer influence) environment cues (meal size/composition indicators) to dictate appropriate eating behaviour suggests that children may be particularly amenable to interventions using BIs. Interventions using BIs can also be implemented at a relatively low cost and preserve free choice, thus are often easier to implement than major mandatory regulatory changes-such as taxation. 9 Previous systematic reviews on diet-related topics have focused on a single BIs precept. For example, many reviews have focused on nudge interventions, 7,10-14 which centre on modifying the choice architecture without significantly changing financial incentives or restricting options. 15 Other reviews have focused on singular behavioural economics precepts [16][17][18][19][20][21] such as incentives or anchoring, which are derived from economic theory and social psychology. 15 This review extends beyond previous reviews as we focus on a broader range of insights from the behavioural and social sciences, which are reflected in Table 1. This current review used the framework proposed by Bauer and Reisch, 3 who classified interventions into one of five categories: (1) provision of information; (2) use of salience and social norms; (3) changes in the defaults; (4) changes to the physical environment; and (5) incentives and preplanning (Table 1).
Interest in BIs has grown in many governments, which has led governments in countries such as the United Kingdom 22 and the United States 23 to establish dedicated teams tasked with applying BIs to public policy. Additionally, the Organisation for Economic

Co-operation and Development, World Bank and European
Commission have produced policy reports on BIs in recent years, [24][25][26] which cover the policy efficacy of BIs in areas including energy, health and finance.
Major health inequities are created via the unequal distribution of obesity burden by sociodemographic characteristics. 2 Health equity is a commitment to reduce, and ultimately eliminate, disparities in health. 27 In this review, we focus on 'health equity elements', which are the sociodemographic characteristics (e.g., sex, socioeconomic status and ethnicity) by which obesity is unfairly distributed, leading to major health inequities. The World Health Organization (WHO) has highlighted the centrality of equity in the design and implementation of obesity policies to protect against worsening inequities and actively reduce them. 28 Evidence of the health equity potential of policies using BIs is unclear, with some evidenced of exacerbating 3,29 and others reducing existing inequities. 30,31 To date, there appears to be only one review of systematic review that focused on and classified BIs interventions. 3 Previous systematic reviews of diet-related interventions using BIs have focused on education, 13,21,32 home 7 and food retail 14 settings but have not compared the effectiveness by setting. The populations within these reviews also vary, as some exclusively focus on adults, 12,14,17,32 or children, 7,19,21 or both. 3,10,11,16,18,20,33,34 In addition, previous reviews have not prioritised health equity in their evaluations. We aim to assess the effectiveness and equity potential of interventions using BIs to improve children's diet-related outcomes in all settings.

| Data extraction and risk of bias assessment
Data extraction and risk of bias (RoB) assessment procedures were conducted by two independent reviewers. One reviewer conducted all procedures in full while a second reviewer conducting a crosscheck of 10% of the studies at each stage of the data extraction and RoB assessment procedures. [35][36][37] Abstract and title screening were conducted using the reference managing software, Rayyan. 38 39 An adapted Newcastle-Ottawa RoB tool was used for nonrandomised studies. 40 We opted for the Newcastle-Ottawa RoB tool as our review contained a large amount of beforeafter studies that have not been validated using tools such as ROBINS-I. Our approach is consistent with recent reviews on obesity containing before-after studies and nonrandomised studies.

| Data analysis
The main analysis was conducted at the intervention level. Unique

| Search strategy
The review was conducted and reported in adherence to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses. Figure 1 shows the Preferred Reporting Items for Systematic Reviews and

| Overall effectiveness
The characteristics and effectiveness of the 137 unique BIs interventions are displayed in Table 3. In total, 102 (74%) of interventions had a statistically significant and positive effect on the targeted dietrelated outcome.

| Study design
Over half of the interventions were RCTs (n = 79, 58%), followed by before-after studies (n = 29, 21%) and controlled before-after studies (CBA) (n = 18, 13%). Before-after studies produced statistically significant results 97% of the time, compared with CBA studies and non-RCT studies that had statistically significant results only 50% and 55% of the time, respectively (p = 0.001).

| Settings
Education settings were the most common setting for interventions, covering 78% (n = 107) of all included interventions. The remaining studies were equally conducted in different settings. In terms of effectiveness within settings, lab-based interventions (100%, n = 7) and primary schools (80%, n = 80) showed the most consistent effectiveness compared with the food retail (25%, n = 8) and home settings (57%, n = 7) (p = 0.018).

| Follow-up
Follow up refers to the time between the outcome measurement at the end of the intervention and any subsequent outcome measurements. Evidence of sustained effectiveness was limited; only 21 (15%) interventions had long-term follow-ups (greater than 1 month) and 16 (12%) interventions had short-term follow-ups (less than 1 month).
Evidence suggests intervention effects are time sensitive as only 56% and 52% interventions showed sustained effectiveness in the short term and long term, respectively. Of the 10 long-term follow-ups without sustained effectiveness, three had initially showed effectiveness post intervention but not at follow-up. Likewise, for interventions with short-term follow-ups, five of the seven interventions evidenced effectiveness post intervention, but it was not sustained at follow-up.

| Food type
The most commonly reported intervention outcome was F&V con-

| Behavioural insights type
Panel A of Table 4 shows the effectiveness of interventions and the number of interventions that evaluated health equity elements. We Whereas there were no statistically significant differences between BIs intervention categories in relation to their health equity implications (p = 0.065), interventions using incentives, which tested for differential effects, reported statistically significant differences 58% of the time compared with only 14% for salience and 20% for changes to the physical environment.
Panel B: BIs, no multiple category a n (%) n = significant/total (%) n with any equity element/total (%) n = significant/total (%) Note the percentage of studies will cumulatively be higher than 100% because each study could utilise multiple behavioural insights. salience (n = 7, 30%) and physical environment (n = 5, 17%

| Information provision
In total, 16 interventions utilised information provision (n = 16, 56% effective); seven of these interventions focused on information provision exclusively (n = 7, 29% effective), and nine used it as part of a multiple BIs intervention (n = 9, 78% effective). The most common type of intervention included the provision of nutritional facts such as "carrots contain vitamin C" (n = 11, 64% effective), followed by manipulations of serving size pictures on packaging (n = 3, 67% effective) and provision of caloric information (n = 2, 0% effective).
Health equity elements were evaluated in 10 of the 16 interventions, with two finding significant differences based on sociodemographic characteristics. In both cases, the interventions used multiple BIs so these differential effects cannot necessarily be entirely attributed to information provision. Health equity elements were evaluated in 23 of the 62 interventions, with seven interventions finding statistically significant differences by sociodemographic characteristics differences. Three of these came from a single intervention that showed that an on-screen prompt during preordering meals was more effective with non-White students, low SES and male students compared with their White, high SES and female counterparts. 41 Three interventions showed differential effects for age where tangible rewards had greater effectiveness for younger children (aged 5-8) than older children (9)(10)(11)(12). 42,43 One study showed differential effects by the child's BMI and demonstrated that children with healthy weight were more sensitive to social modelling of a peer's eating compared with children with overweight or obesity. 44 Seven interventions had short-term (less than 1 month) and

| Salience and social norms
13 long-term (greater than 1 month) follow-ups. Six of the seven short-term follow-up showed sustained effectiveness, while only seven of the 13 long-term follow-ups showed sustained effectiveness.

| Defaults
In total, seven interventions used defaults (n = 7, 71% effective). Five of these interventions focused on defaults exclusively (n = 5, 80% effective), whereas another two used defaults as part of a multiple BIs intervention (n = 2, 50% effective). The defaults usually involved changing the default side option from an unhealthy side option (e.g., fries) to a healthier option (e.g., apple slices) (n = 5). Other examples include changing the default meal to a healthier option with a less healthier option available upon request (n = 2). None of these interventions evaluated the health equity implications or the sustained effectiveness.

| Physical environment
In total, 51 interventions made changes to the physical environment  [45][46][47] In one other intervention, sliced fruit increased consumption for younger children (aged 6-9 y/o) more than for older children (10-12 y/o). 48 In one intervention, increased dishware size increased the amount of food self-served and consumed by children of low SES. 49 One of these interventions had a short-term follow-up and two had a longterm follow-up; all showed sustained effectiveness.

| Incentives and precommitments
In total, 46 interventions used incentives to change children's diet-

| Randomized control trial
In total, 37 studies had only one to three out of eight RoB categories graded as low risk, 46 had four or five RoB categories graded as low risk and only six had six to eight, which suggests most studies had many sources of potential bias. In general, the greatest RoB came from the randomisation procedure (87% of studies graded high or unclear risk), attrition bias (59% graded high or unclear risk) and knowledge allocation (100% graded high or unclear risk). Studies generally accounted for differences in baseline characteristics (18% high risk) and outcomes (31% high risk).

| Nonrandomised studies
Overall, 17 studies were graded as low RoB. The remaining 41 studies were graded as high RoB, with 29 of these being before-after studies.

| Secondary outcomes-cost
When the cost of the intervention was reported, there was evidence that behaviour change could be cost-effective. However, only eight of the interventions reported either implementation costs or a costeffectiveness analysis. 6

| Study characteristics
Interventions that used before-after study designs consistently reported positive statistically significant findings compared with interventions that used RCTs and CBA study designs. Before-after studies tend to overestimate intervention effects due to the uncontrolled biases inherent in such designs. 58 In contrast, the findings in highly controlled lab-based RCTs (n = 7) may not translate into real-world settings. 10 These results reinforce researchers' calls for BIs interventions to implement more robust study designs, with a higher degree of external validity. 3,17 Expectedly, 78% of the interventions in this review took place within educational settings. Schools provide greater accessibility to large numbers of children and control over some of the environmental conditions that reduce potential sources of confounding. The overemphasis on school settings has left a sizable gap in the literature, particularly in the home and food retail environments, which are settings where a substantial amount of children's food choices are made and calories are consumed. 59 The lack of evidence in the food retail environment is disappointing given the extensive research into the physical and social elements within such environments that drive consumer behaviour conducted by private companies. 60 Fruit and vegetable consumption was the most common outcome investigated. When F&V and healthy MSSB outcomes were combined and compared against unhealthy MSSB, we found that interventions targeting healthier options were typically more effective than those targeting unhealthy options. This result contrasts to a previous systematic review, which included adult populations, that showed that reducing unhealthy eating was easier than increasing healthy eating. 10 However, children often struggle to evaluate the long-term health consequences of their decisions against short-term rewards and have been consistently shown to have higher discount rates (presentoriented preferences) compared with adults. 61 Therefore, it is less surprising that children may not be able to appropriately weigh-up the short-term reward of eating tasty and convenient unhealthy foods against the intangible and delayed reward of long-term health.

| Behavioural insights type
The lack of evidence of information provision interventions for children may be due to existing evidence in adults that show limited effects of these types of interventions. 10 Our results also showed the limited effectiveness of interventions using information provision alone (29% effective). A previous review showed that cognitively oriented interventions, like information provision, were the least effective largely because they require the greatest cognitive input and thus are most susceptible to cognitive biases. 10 Interventions using information provision in combination with other BIs were more effective (67%), suggesting that information provision can contribute to behaviour change but is insufficient by itself.
We also found interventions that used health facts rather than calorie information were more effective. Health facts may be more effective as they often elicit an emotional response, whereas, for many, caloric information is an abstract or foreign concept. 17 People generally find interpreting numeric information difficult, which is a major health communication problem, particularly for children. 62 Interventions most commonly used salience. Social modelling was both the most effective and prevalent subtype of interventions using salience. Social modelling creates social norms that signal a code for appropriate behaviour and is particularly powerful if presented at the time of decision making and by actors that are important to children. 63 Visual cues were equally as effective as social modelling, and implementation generally involved minimal time and less monetary investment. Visual cues go beyond simple information provision by providing relevant cues that appeal to the emotional state of individuals. 17 For children, simple cues such as smiley face emoticons or attractive names of healthy products were enough to improve dietrelated outcomes. 64,65 Few studies directly examined the impact of defaults alone, but those that did demonstrated high effectiveness. Defaults influence behaviour largely through status quo bias, by which, cognitive inertia guides people to stay with the current option (e.g., having to evaluate alternative options) or the default option communicates the socially expected choices (e.g., 'regular size' signals). 3 Both mechanisms are likely to influence children disproportionately compared with adults as children often defer to people of authority for decisions or look to others for social norms. Defaults are typically low-cost as they change the choice architecture with no redistribution of underlining resources. 9 The behaviour change potential of defaults makes their scarcity in this review even more surprising and highlights an area that deserves further investigation.
Changes to the physical environment was a common and effective intervention in this review. Our results support previous findings with adults that show the efficacy of modifying portion size, increasing accessibility and presentation of healthy foods. 3,11,12 Increased portion sizes were generally associated with increased consumption but seemed to have less effect for younger children (under 5 y/o) than for older children, which suggests portion size may become a more important environmental cue as children get older. Portion size control may be particularly important for unhealthy, usually ultraprocessed, foods that may promote overconsumption through their high energy density, high palatability and disruption of gut-brain signalling. 66 Accessibility was frequently used in these interventions to either make healthy foods more accessible or unhealthy food less accessible.
Accessibility relates to the time or effort required to mentally and physically access an option. 12 Food retailers have contributed substantial resources to research and experimentation of how accessibility impacts consumer behaviour. 60 Unfortunately, these insights are often not publicly available. Instead, they are commonly used in retail environments to increase profits by driving demand for products with higher margins. Other times, they are used to charge companies extra for premium shelf space. In both, these profit mechanisms tend to favour the promotion of unhealthy foods. In environments where the incentives are shifted from profits to child well-being, such as the school environment, manipulating the accessibility of healthy foods is a particularly low-cost and effective approach.
Incentives were among the most effective BIs interventions.
Incentives varied, but their effectiveness did not differ between tangible and social incentives. Incentive-based interventions are regularly viewed as costly or labour intensive; however, children were motivated by very small tangible rewards (e.g., pencils, toys and glow sticks). The efficacy of social rewards may highlight cost-effective and sustainable insights into deploying health interventions with limited resources. For example, two interventions used competition with fellow students or students from fictional schools to increase F&V consumption. 67,68 Likewise, simple social rewards such as verbal praise from school staff or ringing a bell during school lunch were effective incentives for children. 51,69 Preordering or precommitment devices were rarely used in this review but highlight a key area for further exploration given the proposed mechanisms behind their efficacy. First, preordering removes hunger-based decisions typical of mealtime and removes the environmental cues associated with selecting less healthy options. 70 Second, preordering enables the modification of the choice architecture to promote healthy default options 71 or opportunities for targeted healthy nudges. 41

| Health equity
The WHO recommends equity be a central component in the design and implementation of health policies to ensure inequities do not worsen and are actively reduced. 28 Over half of the interventions in this review failed to explore even one equity element. Sex and age were most commonly tested for differential effects, as these characteristics are typically readily available during data collection. More evidence of the long-term effects of these interventions is required to make conclusions about their sustainability for policymakers.
Another issue for assessing the sustainability of interventions using BIs was the relatively short duration of studies. Forty percent of studies lasted less than one week and 92% lasted less than 6 months.
It is entirely likely that changes in diet during a

| Strengths and limitations
This review has several methodological strengths. First, the search strategy included a sensitivity analysis, which is particularly pertinent given the majority of papers contained no reference to BIs terms. Second, the search strategy also included six databases, references from 10 systematic reviews and backward reference searching of included studies. Third, a second reviewer ensured that the data extraction and RoB assessment procedures were applied consistently.
Although this review provides a comprehensive examination of

| Future research and policy implications
Future research on BIs interventions should focus on providing evidence of their sustainability. To better inform policy, future interventions require longer study durations and follow-ups to access the sustainability of behaviour change. There is also a need for better documentation and analysis of the costs of interventions to determine their cost-effectiveness. There is also an urgent need for evidence of the health equity implications of BIs interventions, particularly for analyses of SES and ethnicity. Future studies should attempt to obtain sample sizes with sufficient statistical power to detect meaningful differences, including enough data to test for differential effects by sociodemographic characteristics.
There is currently an overemphasis on interventions in education settings, thus additional studies in food retail and home settings would provide a substantial contribution to the existing evidence. Finally, there is a need for additional interventions utilising a non-lab based RCT study design to overcome the reliance on studies with a high RoB, such as before-after studies, which may misrepresent the effectiveness of these interventions.
This review highlights areas where BIs may be useful for developing and implementing obesity-related policy. Importantly, this review has reinforced previous findings suggesting that information provision alone is insufficient for changing behaviour. It is clear that more effective BIs, such as changing defaults or the physical environment, are required to induce and sustain behaviour change.
The large amount of studies in the school environment with significant findings, coupled with the governments' ability to regulate most school settings, suggest this is an area where governments can most easily affect behaviour change in children. Secondary findings in this review demonstrate that such interventions can be implemented at a very low cost. However, in sum, the current evidence does not provide strong evidence for policymakers as there is a lack of evidence of sustained effectiveness, impact on obesity-related outcomes or implications for health equity.
Additionally, only seven interventions reported either their implementation costs or a cost-effectiveness analysis, which is a major barrier to providing policy recommendations. Commission is not liable for any use that may be made of the information it contains.

CONFLICT OF INTEREST
No conflict of interest was declared.

AUTHOR CONTRIBUTIONS
TC conceptualised the review, developed and implemented the search strategy, abstract screening, full text review, data extraction and data analysis. ABS conducted the double reviewing and contributed to the development of the RoB assessment. All authors contributed to the manuscript. FS is the PI on the STOP project and provided oversight on all aspects of the review.