The effectiveness of interventions on changing caregivers' feeding practices with preschool children: A systematic review and meta‐analysis

Caregivers' feeding practices are critical in shaping preschool children's eating habits and preventing childhood obesity. We conducted a systematic review and meta‐analysis to evaluate the effectiveness of existing interventions targeting caregivers of preschool children, which aimed to promote child healthy eating and/or manage child weight and/or prevent child nutrition‐related problems and included feeding practices as one of the outcomes. Eighteen studies with 18 intervention programs and 3887 respondents that completed baseline evaluations were eligible for data synthesis. Behavior change techniques (BCTs) frequently used included the following: instruction on how to perform the behavior and demonstration of the behavior. The pooled effects of randomized controlled trials (RCTs) on pressure to eat (pooled standardized mean difference [SMD] = 0.61; 95%CI: −1.16, −0.06), use of food as a reward (pooled SMD = −0.31; 95%CI: −0.61, −0.01), and emotional feeding (pooled SMD = −0.36; 95%CI: −0.66, −0.06) were found statistically significant compared with control groups at post‐intervention. However, there were no pooled effects on restrictive feeding and pressure to eat at other follow‐ups or on other feeding practices at post‐intervention. Interventions may have short‐term effects on decreasing the adoption of coercive control. Future interventions should directly and adequately optimize feeding practices, include components of individual support, and contribute to the maintenance of the effects over the long term.


| INTRODUCTION
Childhood overweight and obesity is a major public health issue that affected 39 million children under the age of 5 and over 340 million children and adolescents aged 5-19 in 2020 across the world. 1 It is well known that eating behaviors and physical activities have substantial impacts on the risk of obesity across the lifespan. 2Child eating behaviors and obesity can be understood within an ecological framework in which children's characteristics interact with the environment and influence their health consequences. 3,46][7] Therefore, more and more research focused on studying feeding practices to identify ways to help caregivers provide their children with more healthy dietary options, and thus aid in obesity prevention.
Feeding practices refer to specific behaviors or strategies that caregivers employ to manage what, when and how much their children eat and shape their children's eating patterns. 8,9Based on the conceptual models proposed by Vaughn et al. 10 and O'Connor et al., 11 feeding practices can be classified into three food parenting constructs: coercive control (e.g., pressure to eat and restrictive feeding), structure (e.g., monitoring and modeling), and autonomy support (e.g., nutrition education and encouragement).8][19] Therefore, caregivers' feeding practices are critical in shaping preschool children's eating habits and a key target in efforts to prevent childhood overweight and obesity. 20,21Although more interventions are focusing on preschool children's eating behaviors and weight management, especially in the developed countries, [22][23][24][25][26] limited number of studies included their caregivers' feeding practices with preschool children as one of the intervention components or one of the outcomes. 22,23us, there is a need to summarize the existing interventions that included caregivers' feeding practices with preschool children as one of the outcomes, which could provide guidance on developing or adapting an intervention to improve feeding practices in the specific context, particularly in the low-to-medium economic districts.Some existing studies tested the effects of interventions that included caregivers' feeding practices with preschool children as one of the outcomes in the context of child healthy eating and nutrition promotion and weight management.The results showed that their effects on feeding practices were mixed.8][29][30] For example, a randomized controlled trial (RCT) in Australia (n = 132) adopted a 6-week intervention to promote healthy body image and eating patterns in preschool children and parents that included learning resources and workshop and reported significant decreases in parental negative feeding practices (e.g., pressure to eat). 22Leung et al. 27 also conducted a cluster RCT with 191 parents and their children from 24 preschools in Hong Kong, China, consisting of 20 sessions covering strategies to promote child development in physical health, cognitive and psychosocial areas; the result demonstrated improvement in parental healthy feeding practices in the intervention group.In addition, a randomized pilot study consisting of 8 weekly sessions, examined the feasibility of a tablet computer-based intervention on childhood obesity prevention among 32 low-income Chinese mother-child dyads in the US and found small effect size for parental restriction of food. 28However, some studies did not find such statistically significant changes of feeding practices after conducting the interventions. 31For example, an RCT (n = 146) conducted in Australia applied DVD modules and resources related to child feeding and parenting style (e.g., dietary fats, fussy eaters, food budgeting, and reading food labels) and reported the absence of an intervention effect on any parental feeding domain (i.e., restriction, pressure to eat, and monitoring). 31Similarly, a systematic review and meta-analysis with 3,162 parents of 0-12-year-old children studied twelve web-based intervention programs in the context of healthy eating promotion that included feeding practices as one of the outcomes; the results showed that the pooled effects for all included feeding domains were not statistically significant. 32In addition, some studies found mixed effects of interventions on changing caregivers' feeding practices. 23,33For instance, Hammersley et al. 23 conducted an 11-week program focusing on childhood obesity prevention in Australia with six modules about four obesity-related behaviors of two-to five-year-old children (i.e., nutrition, physical activity, screen time, and sleep) and found significant decrease in parental pressure to eat, while no differences in the other feeding practices (e.g., restriction and modeling).Overall, the effects of existing interventions on changing caregivers' feeding practices with preschool children have been inconsistent and there is a need to identify and test the effectiveness of the relevant interventions, which could thereby make the available evidence better informed for program designers in the future.
In addition, effective features of the relevant intervention programs should be identified to support the development of future evidence-based professional interventions.Behavior change techniques (BCTs) offer a means of breaking down variable training programs into observable, replicable, and irreducible features, which adequately describe intervention features. 34Some BCTs, for example, instructions on how to perform the behavior, social support, and goal setting, are frequently adopted by interventions to change parental feeding practices. 23,27,35,36Some systematic reviews also applied BCTs to specify behavioral intervention programs that enable rigorous evidence synthesis and comparison with the wider professional development of the intervention. 32,37Therefore, it is necessary to use BCTs to specifically analyze the components of the included intervention programs in this review.This systematic review and meta-analysis of the existing interventions on changing caregivers' feeding practices with preschool children in the context of child healthy eating promotion and weight management addresses the following questions: 1) Which interventions adopted caregivers' feeding practices with preschool children as one of the outcomes?2) Which BCTs are used to promote changes in caregivers' feeding practices?3) What is the effectiveness of these interventions on changing/improving the specific feeding practices?
To the best of our knowledge, it is the first systematic review and meta-analysis to summarize the existing interventions on changing caregivers' feeding practices with preschool children using BCTs and test their effectiveness.Findings from this review will enhance our understanding of the existing interventions and inform future interventions to optimize feeding practices.

| Data sources and search strategy
This systematic review and meta-analysis complied with the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) 38 and was registered in PROSPERO (registration number: CRD42021292366).
A systematic literature search was carried out on MEDLINE, Embase, PsycINFO, Web of Science core collection, CINAHL and the Cochrane Library from inception to June 2022.To minimize publication bias, we also searched for studies in grey literature sources including Grey literature report (http://greylit.org/),and Open grey EU (http://opengrey.eu/).The search was limited to publications published in English.The free text and Medical Subject Headings (MeSH) terms based on the PICO (participant, intervention, comparison, outcomes) were used for the search, including child, preschool, pediatric, toddler, caregiver, parent, grandparent, mother, father, family, feeding practice, food parenting, food control, pressure to eat, food availability, intervention, randomized controlled clinical trial, and non-randomized controlled trial.The detailed search strategy is provided in Table S1.A manual search of the reference lists of the included studies was performed to identify additional studies.

| Inclusion and exclusion criteria
All studies were included if they met the following criteria: 1. Study design was RCT or non-randomized controlled trial.
3. Studies aimed to improve caregivers' feeding practices and/or promote child healthy eating and/or manage child weight and/or prevent child nutrition-related problems.
5. Caregivers' feeding practices were one of the outcomes.
Studies were excluded if they: 1. Were reviews, editorials, commentaries, letters, methodological papers, or abstracts only.
3. Were only directed to preschool children.
4. Sampled a clinical (e.g., patients with physical and mental illness) or non-general population (e.g., pregnant women only).
5. Focused on caregivers of children with diseases that might influence their eating and nutrition.

| Study screening and data extraction
The PRISMA flow chart was followed during the screening stage. 38e investigator (JW) screened the title and abstract for initial inclusion.We excluded the articles from initial screening if they did not report related endpoints or exposures.The articles were not retrieved for full text if they were non-human based studies, or they were reviews, editorials, commentaries, letters, and abstracts only.A second reviewer (XW) randomly verified 10% of initial screening (n = 1,290, 96% agreement).Full texts were reviewed independently by the two investigators (JW and XW) for further screening (n = 332, 98% agreement).The agreement percentage at initial screening and at full texts both reached 95%.Data extracted was compared and summarized to have one final document on which analysis was conducted.The information extracted included: name of the first author, year of publication, country that the study was conducted in, name of the intervention program, study design, participants, theoretical framework, mode of intervention delivery, measurement of the related outcomes, follow-up time point, focused intervention topics, description of intervention and control condition, and main results.A small number of the included studies conducted three to four groups (e.g., completed intervention group, mini-intervention group [i.e., a part of completed intervention], and control group).We extracted the information from the completed intervention group (i.e., most intensive intervention group) and control group.Quantitative results that may represent the changes of caregivers' feeding practices, including mean and standard deviations (SD)/standard error (SE), or mean changes and SD/SE/95% confidence intervals (CI) were extracted.We contacted the corresponding authors if a study did not provide necessary numerical results.For any disagreements that occurred during the screening and data extraction stages between the two investigators, the reviewer team was consulted (KW, Y-SC, and YC).

| Behavior change technique codification
All intervention content extracted from peer-reviewed publications was compiled for coding.Two reviewers (JW and XW) completed certified training (available at www.bct-taxonomy.com) and independently coded the intervention with Michie's et al. taxonomy (BCTT v1), 34 according to the six coding principles.This taxonomy includes 93 BCTs, organized in 16 BCT clusters.Disagreements were resolved through discussion with the reviewer team (KW, Y-SC, and YC).

| Outcomes and BCTs codification
We included caregivers' feeding practices outcomes based on the conceptual model proposed by Vaughn et al. 10 and O'Connor et al., 11 which were classified into three food parenting constructs: coercive control, structure, and autonomy support.These three categories of feeding practices were commonly recognized by the existing studies. 32,39,40ercive control also known as non-responsive feeding refers to caregiver-centered feeding strategies including coercion and psychological control. 10,11They serve caregivers' goals and desires and may not take the children's emotional or psychological needs into account. 10e coercive control was classified into five categories 10,11 including restriction, pressure to eat, use of food as a reward, emotional feeding, and intrusive control.Four specific non-responsive feeding practices were included in our meta-analysis as follows.
1. Restriction means that the caregivers enforce strict limitations on children's access to food or opportunities to consume specific food. 10Typically, restrictive feeding is used to control children's intake of unhealthy food [41][42][43][44] and children's weight. 42 Pressure to eat means that caregivers insist, demand, or physically struggle with the child to have the child eat enough or specific food.10,[42][43][44] 3. Use of food as a reward is also called instrumental feeding, which bribes and threatens children to eat food.10,45 4. Emotional feeding means that caregivers use food to manage or calm children when they are upset, fussy, angry, hurt, or bored, 10,46 for example, using food to soothe, intrigue, and/or relieve.
Structure refers to caregivers' consistent enforcement of rules and boundaries about eating, strategies they use to help their children learn and maintain certain dietary behaviors, and the organization of their children's food environment. 10Structure includes practices such as rules and limits, guided choices, monitoring, role modeling, home food availability and accessibility, and food preparation. 10,11Monitoring was one of the most commonly used feeding practices in the previous studies and was also included in our meta-analysis, which means that caregivers track what and how much the children are eating so that he/she can make sure the child eats enough healthy foods and avoids overeating unhealthy food. 10,43,44tonomy support means that caregivers provide sufficient structure for their children to be involved in making food choices at a developmentally appropriate level; engage in conversations with children about reasons for rules and boundaries about food; and create an emotional atmosphere during these caregiver-child food interactions in which children unconditionally loved, valued, and accepted. 10,47This food parenting construct included practices such as nutrition education, child involvement, encouragement of healthy eating and praise. 10,11Encouragement of healthy eating was included in our meta-analysis, which means that caregivers suggest or offer healthy food to their children as a prompt and encourage them to eat more healthy food. 10,43

| Study quality assessment
The risk of bias for the included RCTs was assessed using the RoB2, a revised Cochrane tool that provides a framework to evaluate the risk of bias in RCTs. 48The five domains of this tool (i.e., randomization process; deviations from intended interventions; missing outcome data; measurement of the outcome; and selection of the reported result) correspond to the main types of bias that can influence the study results.The quality appraisal of the included non-randomized controlled trials (non-RCTs) was assessed using the Joanna Briggs Institute (JBI) Critical Appraisal Checklist for Quasi-Experimental Studies. 49This tool evaluates the methodological quality of a study and determines the extent to which a study has addressed the possibility of bias in its design, conduct, and analysis.Two reviewers (JW and XW) independently performed the assessment, checking for possible sources of bias, attrition, and validity of survey instruments.The final assessment was achieved upon discussion (see Tables S2 and S3).
Given the risk of bias may affect the internal validity of their findings, including the results of meta-analyses across studies, 50 the results of quality assessment were discussed by two reviewers and no studies were identified for exclusion.

| Statistical analysis
Quantitative approaches were any method that used a numerical method to summarize across at least two studies and were further divided into semi-quantitative and quantitative approaches (metaanalysis).We used a semi-quantitative approach to summarize the findings of all the included articles (i.e., study characteristics, outcomes, and intervention features) narratively, as adopted by recent reviews [51][52][53][54] because not all included studies had necessary numerical results for meta-analysis.Information was retrieved by one reviewer (JW) and checked by a second reviewer (XW).Mean and SD/SE or mean changes and SD/SE/95% CI from intervention and control groups were used to calculate standardized mean differences (SMDs) and 95%CI in the meta-analysis.Interpretation of SMDs was conducted: 0.2-0.49indicated small effect, 0.5-0.79indicated moderate effect, and > 0.80 indicated large effect. 55We conducted several meta-analyses based on the follow-up duration (i.e., post-intervention, 3-month, 6-month and 12-month follow-up).The meta-analyses were separated for RCTs and non-RCTs because of the risk of bias of the study.Results from different study designs should be expected to differ systematically, resulting in increased heterogeneity. 50If the same category of outcome (e.g., restriction for weight, restriction for health) was multiply evaluated in one study, the results were first synthesized within the study, and the summarized data were then used for the meta-analysis.When required, 95% CI or SE values were converted to SDs by using the following formula: The heterogeneity of the included studies was investigated using the I 2 statistics and interpretation of I 2 was defined as follows: I 2 ≤ 25% was defined as low heterogeneity, I 2 > 25% and ≤ 50% was defined as moderate heterogeneity, I 2 > 50% and ≤ 75% was defined as high heterogeneity, and I 2 > 75% was defined as considerable heterogeneity. 56The random-effect meta-analysis was chosen because it estimates the mean of a distribution of effects rather than a single effect assumed to be common to all studies. 50This approach fitted with the considerable heterogeneity expected from including studies with differences in population, study design, and intervention and control groups.The potential publication bias was assessed by the combination of Egger's test and visual inspection of the funnel plot. 57e leave-one-out (LOO) analysis was also performed as the sensitivity analysis to investigate the influence of a single study on the pooled effect. 58The subgroup analysis and meta-regression were used to evaluate effect heterogeneity.The exploratory subgroup analysis was based on some potential factors 7,53,59,60

| Search results
A total of 21,679 articles were identified.After removing duplicates, 12,894 articles remained for the initial screening, from which 332 articles were retrieved.After full text screening, 18 studies were retained for analyses.The PRISMA flow diagram is shown in Figure 1.

| Characteristics of the interventions
Table 2 showed the classification of caregivers' feeding practices used as outcomes in the included studies.Some specific feeding practices (i.e., desirable feeding practices, 27 inductive feeding strategy, 62 ageinappropriate feeding, 63 persuasive feeding, 66 distrust in appetite 66 and non-coercive child feeding practices 67 ) were excluded because they did not belong to three categories of feeding practices listed.Four studies considered feeding practices as primary outcomes, 25,29,31,33 three studies classified them as secondary outcomes, 23,35,64 one study considered pressure to eat as primary outcome and restriction as secondary outcome, 30 and the remaining studies included feeding practices as outcomes but did not specify if they were primary or secondary outcomes.
The detailed information of included intervention programs is reported in Table S4.The intervention programs lasted from 1 33 to 36 31 weeks and most programs lasted 6 to 8 weeks (n = 5). 22,25,28,62,65The focused topics of the included intervention programs can be categorized into child obesity prevention/weight management, 23,25,28,35,63,64,66 caregivers' feeding practices, 31,33,61 child development (i.e., physical and psychosocial health), 27 child healthy body image, 22 child eating patterns/behaviors, 22,29,30,36,62,67 and nutrition and parenting education. 65Some interventions also included individual support/personal guidance (e.g., counselling with a social worker, motivational interviewing from a dietitian and individual support through phone call) as one of components in the program. 23,27,35,36,61,66,67Two non-RCTs had no control group. 61,65Most programs offered some sort of intervention in the control condition while three programs 22,25,62 included a wait-listed control group or did not provide any intervention for control group.The number of BCT clusters used in each study varied, ranging from 1 31 to 9. 23,25,29 Three studies 23,25,29 used more than eight BCT clusters.Each intervention program reported different specific BCTs, of which Mobley et al 25 Tabak et al 36 and Sobko et al 29 used the most.The number of specific BCTs ranged from 1 31 to 15. 25,29,36 The example for identified BCT clusters and specific BCTs was shown in Table S5.

| Study quality assessment
Figure 2 presented the quality assessment of RCTs using the Cochrane Risk of Bias 2 tool.Three studies 31,35,64 were assessed to have "low risk" of bias.Seven studies 22,23,27,28,30,33,63 were evaluated as "some concern" of bias and the other three studies 29,36,67 were rated "high risk" of bias (Table S2).One study 67 reported that the participants were randomized but they did not adopt allocation concealment and there were significant differences of characteristics between intervention and control groups; therefore, it was classified as "high risk" of bias.Three studies 22,36,67 reported deviations from intended interventions and this domain was thus rated as having "some concerns".Two studies 36,67 were assessed "high risk" of bias regarding the domain "Missing outcome data" because they excluded the missing data without explanation and information.One study 29 was rated as "high risk" of bias in the domain "selection of the reported result" because there were inconsistences of outcome measurements between protocol and the published article.
The quality assessment of the included non-RCTs using JBI Critical Appraisal Checklist for Quasi-Experimental Studies is presented in Table S3.All studies clarified the aim and objectives.Two studies did not include control group. 61,65All studies used valid measurements to assess caregivers' feeding practices and used appropriate statistical analysis to test the effects of the interventions.The overall evidence should be considered moderate (two non-RCTs) to high quality (three non-RCTs).
T A B L E 2 Classification of caregivers' feeding practices in the included interventions.

| The results of meta-analysis
A meta-analysis was conducted on pooled data from eleven studies, 22,23,25,[28][29][30][31]33,63,64,66 which compared the intervention groups with the control groups. evaluated te effects of the interventions on restrictive feeding at post-intervention appears symmetric (Figure 4).No statistically significant publication bias was detected (Egger's test P = 0.437).The LOO sensitivity analysis indicated that the results were robust, and only one estimate was statistically significant (Figure 5).As there were no more than three studies summarizing the pooled effects of the interventions on restrictive feeding at post-intervention (non-RCTs),

| Restrictive feeding
3 months and 6 months follow-up, the funnel plots were not provided.No statistically significant publication bias was detected (Egger's test P = 0.159).The LOO sensitivity analysis indicated that the results were robust, and only one estimate was statistically significant (Figure 8).As there were no more than three studies summarizing the pooled effects of the interventions on pressure to eat at 3 months, 6 months and 12 months follow-up, the funnel plots were not provided.

| Subgroup analysis
The subgroup analysis and meta-regression aimed to identify the potential sources of heterogeneity and examine the stability of the pooled effects of the interventions from RCTs on restrictive feeding and pressure to eat at post-intervention across different categories (Table S6).Subgroup analysis by the country's level of development showed a significant decrease of restrictive feeding at postintervention in the developed countries (pooled SMD = À0.291;95% CI: À0.521, À0.061; n = 4).Restrictive feeding was found to be less likely to be used in the subgroup with fewer BCT clusters (pooled SMD = À0.291;95%CI: À0.521, À0.061; n = 4).The result of subgroup analysis also showed statistically significant decrease of pressure to eat in the subgroup with control groups that applied usual care Funnel plot with pseudo 95% confidence limit for randomized controlled trials on restrictive feeding at postintervention.
F I G U R E 5 Pooled SMDs of the effects of randomized controlled trials on restrictive feeding at post-intervention in leave-one-out analysis.
or active control (pooled SMD = À0.301;95%CI: À0.544, À0.059; n = 3).However, the results of meta-regression indicated that the associations were not statistically significant change in all subgroup comparisons.F I G U R E 7 Funnel plot with pseudo 95% confidence limit for randomized controlled trials on pressure to eat at post-intervention.

| DISCUSSION
F I G U R E 8 Pooled SMDs of the effects of randomized controlled trials on pressure to eat at post-intervention in leave-one-out analysis.
feeding practices with preschool children and evaluated evidence for their effectiveness.Eighteen articles included eighteen intervention programs and 3,887 participants at the baseline assessment.Only four studies considered caregivers' feeding practices as primary outcomes.
Fifteen studies considered coercive control as an outcome of which restrictive feeding and pressure to eat were the most frequently studied feeding.Twelve studies included structured feeding as an outcome and the majority of which focused on monitoring and modeling.
Nevertheless, only four studies included outcomes regarding autonomy support.[15][16] Our findings from meta-analysis suggested that the pooled interventions may have short-term effects on most coercive control (i.e., pressure to eat, use of food as a reward and emotional feeding) and had no statistically significant long-term effects.Specifically, the results of meta-analysis showed that the pooled effects of RCTs on caregivers' pressure to eat, use of food as a reward and emotional feeding were found the significant decrease compared with control groups at post-intervention.Similarly, the interventions on restrictive feeding from the included non-RCTs in the meta-analysis found a mean increase at post-intervention.However, there were no statistically significant pooled effects of RCTs on restrictive feeding and pressure to eat at other follow-ups or on other feeding practices at post-intervention.
The statistically significant short-term effects in the meta-analysis may be attributed to several elements.First, some interventions focused on improving feeding practices, 31,33,61 as shown in Table S4, including the components directly changed the extent to which parents believed coercive control (e.g., restriction and pressure to eat) were beneficial.These direct intervention components may be more effective in helping participants reduce the adoption of inappropriate feeding practices 33,61 compared with the indirect components on feeding practices (e.g., child obesity prevention) in some other interventions. 27,28,63For example, some included studies discouraged coercive control by featuring them with undesirable outcomes (e.g., child food fussiness and unhealthy weight status), 33 which may improve caregivers' awareness of their inappropriate feeding.
Although some interventions focused on promoting child healthy eating, 29,30 they included some components regarding improving feeding practices, such as maintaining a regular schedule of meals and snacks, providing and serving food, which might contribute to positive changes in their feeding practices (e.g., the reduction of pressure to eat). 30In contrast, a healthy weight intervention embedded in a home-visiting program 63  potential indirect effects of this intervention were thus not sufficiently substantial to manifest as differences between groups in participants' feeding practices.Second, the component of individual support/personal guidance in the interventions may contribute to positive changes of feeding practices, 36,61,66 as shown in Table S4.
For instance, a non-randomized controlled trial reported that showing segments of the evening meal video recording to the mothers generated a personalized detail-rich discussion related to their feeding practices, increasing participants' readiness and motivation to improve feeding practices. 61Some interventions also adopted motivational interviewing to address parents' feeding problems and assist them in identifying primary targets for promotion; the results showed improvements in feeding practices. 36,61It may be attributed to emotional social support (one of the BCTs) for individuals. 34These findings indicated that adding personalized components in the whole intervention may be an effective strategy for improving caregivers' feeding practices.On the other hand, evidence showed that facilitating learning played an important role in the related interventions that successfully optimized caregivers' feeding practices.Some studies showed that use of animations and short videos, which were classified as instruction on how to perform a behavior and demonstration of the behavior in BCTs, 34 can enhance the learning of health information in individuals with low health literacy and income levels and thus promote health behavior changes. 72,73Therefore, it is crucial to provide caregivers with support (e.g., videos) to facilitate their learning even after the intervention. 61In addition, the mode of the intervention delivery may play significant roles on the effects of intervention content. 22For instance, Hart et al 22 conducted an RCT with four groups; the results showed that parents who had received workshop and printed resources showed more improvements of feeding practices over time than those receiving resources only.These findings suggested that providing face-to-face training, which can be classified as instruction on how to perform a behavior in BCTs, 34 may be a superior method of delivering the intervention than providing the written resources alone.Simultaneously, it also reflected higher dose of the intervention may increase the likelihood of positive changes. 62For instance, Duncanson et al 31 adopted the least BCTs (i.e., instruction on how to perform a behavior), which was self-directed technologybased education resources; the result showed non-significant changes of any feeding practices in the intervention group.However, the included studies used several BCTs, which included different intervention components (e.g., sessions, homework activities, and group discussion), mostly reported positive changes of feeding practices. 23,25,29,36,66It seems further studies should deliver the sufficient dose of the intervention that may contribute to significant effects on feeding practices.
However, no statistically significant effects were shown in other feeding practices (i.e., monitoring and encouragement of healthy eating).Likewise, there were no differences of restrictive feeding and pressure to eat at follow-ups except post-intervention between groups.There may be several reasons.First, some studies recruited relatively well-educated participants who have already realized the importance of desirable feeding practices. 22,31In this case, the intervention effect may be absent even when parents are provided with appropriate, evidence-based resources.For example, Duncanson et al 31 conducted an RCT and the result showed that all included feeding practices were not significantly different between baseline and follow-ups.It might be attributed to the high mean scores of desirable feeding practices that highly educated parents already adopted at the baseline.Thus, further studies require a more targeted intervention in the specific sample with less advantaged backgrounds.
Second, caregivers may find difficult to focus on changing more than one of the specific feeding practices simultaneously, 23,28 which may result in non-significant effects on other included feeding practices.
Thus, it is possible that the result of meta-analysis showed no statistically significant effects on these specific feeding practices because of most non-significant effects of the interventions.Third, the changes in motivation and learning behavior might take place more gradually and the changes might not be obvious within short length of followup. 276][67] Therefore, follow-up data need to be collected to ascertain long-term maintenance effects of the intervention.Another explanation for the absence of the intervention effect is the high quality of the program that control group received, 63 resulting in diminishing the differences between groups.Lastly, small sample size may result in minimizing differences between groups.Some studies also reported low attendance rates, 25,61,64 which may also decrease the effectiveness of the intervention.Future studies require larger sample sizes which have enough power to assess the impacts of the intervention on improving caregivers' feeding practices.Meanwhile, it is critical to increase participants' interest in the intervention program, which may encourage caregivers to attend the entire intervention.For example, previous reports have indicated that convenient program timing is an important factor for participants. 74

| Strengths and limitations of the studies included in the review
To the best of our knowledge, this review was among the first that comprehensively synthesized data on the interventions to improve caregivers' feeding practices with preschool children.We used a comprehensive and rigorous methodology in this review, including a broad search strategy and a range of databases.Our study also adopted a wide range of search terms to retrieve all potential articles published in English, including grey literature.In addition, we included all trials where feeding practices were not considered to be primary outcomes.
However, there are several limitations to this systematic review and meta-analysis.First, some studies did not report the estimates representing the changes of caregivers' feeding practices, which precluded us from pooling all the extracted data and conducting subgroup analysis based on other important factors (e.g., child weight status, specific BCTs) because of the small number of articles included in the meta-analysis.Thus, the results from the meta-analysis should be interpreted with caution.Second, all included studies employed self-reported questionnaires to assess our outcomes of interest, which may be subject to recall bias.Third, limited studies focused on improving caregivers' feeding practices, which may result in non-significant effects and thus potentially diminish the estimates in the metaanalysis.On the other hand, many included studies were feasibility pilot designs and only assessed effects at post-intervention, which did not have the opportunity to report whether these changes have been sustained in the longer term.Simultaneously, the sample size in some included studies was small, which may influence the effectiveness of the interventions.Moreover, the results of meta-analysis across the included studies may be affected by the risk of bias 50 because the results were absent from some included studies that should have been included in the meta-analysis and some included studies evaluated as "some concern" and "high risk" might over-estimate or underestimate the true intervention effects.Furthermore, all included articles were embedded with different intervention programs, which may not tell the effects of one type of intervention in this review.Thus, the substantial heterogeneity of the included interventions makes it difficult to make recommendations for best practices.It should also be acknowledged that while we described the individual components using BCTs, many of the interventions included in our study were multi-component; as a result, we were unable to separate out the contribution of different intervention features.Finally, the included studies were mainly conducted in Western countries such as the US and Australia.Findings from this systematic review might not be extrapolated to other populations (e.g., Asian).

| Implications for practice
We found the effects of the existing interventions on improving caregivers' feeding practices with preschool children were inconsistent.
Limited studies considered feeding practices as primary outcomes and included explicit content around responsive feeding practices; further studies should deliver the sufficient dose of the intervention and provide straightforward content to comprehensively improve feeding Writing-review and editing: Jian Wang, Kirsty Winkley, Yan-Shing Chang and Yang Cao.
(i.e., country's level of development, sample size, caregivers' education, family income, study quality, mode of intervention delivery, number of BCT clusters and control condition).All analyses were performed in Stata 17.0 (StataCorp, College Station, TX, USA).All tests were two-sided, and the statistical significance was set as a P-value < 0.05.
(e.g., protocol).The result showed that 11 of 16 BCT clusters were used, of which shaping knowledge (n = 18), comparison of behavior (n = 16) and goals and planning (n = 12) were mostly adopted.Other clusters were also coded in more than 50% of the included interventions: social support (n = 10), natural consequences (n = 11) and repetition and substitution (n = 11).The included intervention programs reported 30 different BCTs of the 96 BCTs classification in the taxonomy, ranging from 1 to 18.The BCTs instruction on how to perform the behaviour (n = 18), demonstration of the behavior (n = 16), problem solving (n = 12), and behavioral practice/ rehearsal (n = 11) were frequently used in the included intervention programs.

8 . 6
Generalization of a target behavior 8.7 Graded tasks T A B L E 3 (Continued)

a
Identification of self as role model ✓ 13.2 Framing/reframing 13.4 Valued self-identity Sum (BCT clusters) Feeding practices noted with the primary outcomes.b Feeding practices noted with the secondary outcomes.a1

a
Feeding practices noted with the primary outcomes.b Feeding practices noted with the secondary outcomes.a1

Figure 3
Figure 3 showed that the pooled effects of the interventions on caregivers' restrictive feeding at post-intervention (RCTs
This systematic review and meta-analysis summarized the available interventions (including RCTs and non-RCTs) on changing caregivers' F I G U R E 6 Forest plot of the effects of interventions on pressure to eat at post-intervention, 3 months, 6 months, and 12 months follow-up.
primarily aimed to change child body mass index (BMI) targeting caregivers' own obesity-related behaviors and did not include components to directly change feeding practices.The F I G U R E 9 Forest plot of the effects of interventions on monitoring, food as a reward, emotional feeding and encouragement at postintervention.
practices.On the other hand, most studies focused on feasibility evaluation of the intervention and only reported the effects at postintervention.Next steps should extend the feasibility trial to a fullscale randomized trial include using an attention control group instead of a wait-listed comparison group.Future research aimed at developing new interventions to promote feeding practices should also investigate how to sustain the long-term effects of the intervention.Because of a small number of articles included in the meta-analysis, results should be interpreted with caution and considered as a preliminary test of the effectiveness of interventions on caregivers' feeding practices with preschool children provided by these programs.More relevant studies are needed with necessary numerical estimates (e.g., mean and SD/SE or mean changes and SD/SE/95% CI from intervention and control groups) in the future.Moreover, some included studies showed low attendance of participants, resulting in small effects of the intervention.Further studies should highlight the needs to optimize the strategies adopted (e.g., focusing on key program constructs, convenient program timing, reducing the number of assessment instruments) to raise participants' interest in learning to improve their practices and eventually augment the success of the program.In addition, future research may benefit from using factorial designs to isolate effects of certain intervention strategies enabling reviews to examine aspects of interventions most likely to contribute to positive intervention effects.Lastly, because of the recall bias generated by self-reported data, future research could strengthen these findings by incorporating direct observation of parent-child interactions during mealtimes.5| CONCLUSIONIn summary, the effects of the existing interventions on improving caregivers' feeding practices with preschool children were inconsistent.Some interventions were proven to be effective to optimize feeding practices, while other interventions only reported small or non-significant effects.The result of the meta-analysis also showed that the effectiveness of the included interventions was mixed.It seems that existing interventions may have short-term effects on reducing the adoption of coercive control (e.g., pressure to eat, emotional feeding), but no statistically significant long-term effects were found in our results.The findings suggest that further trials should include explicit content around all types of feeding practices and focus on key constructs.Given most included trials only conducted feasibility test, it is essential to test the efficacy of such an intervention with well powered sample.The limitation of a small number of studies included in the meta-analysis indicated that necessary numerical results related to the effects of intervention programs should be thoroughly reported in the future.Further research should also highlight the need to optimize the strategies adopted (e.g., regarding session contents and methodologies, flexible program timing and makeup sessions, and the dosage of the intervention) to augment the success of the program.In addition, the development of the related intervention should understand how the improved parent outcomes change or are maintained over the long term with the rigorous methodology.AUTHOR CONTRIBUTIONS Conceptualization: Jian Wang, Kirsty Winkley, Yan-Shing Chang and Yang Cao.Data curation: Jian Wang, Xiaoxue Wei and Yan-Shing Chang.Formal analysis: Jian Wang, Kirsty Winkley, Yan-Shing Chang and Xiaoxue Wei.Investigation: Jian Wang.Methodology: Jian Wang and Kirsty Winkley.Project administration: Jian Wang, Kirsty Winkley, Yan-Shing Chang and Yang Cao.Supervision: Kirsty Winkley, Yan-Shing Chang and Yang Cao.Validation: Jian Wang.Software: Jian Wang.Visualization: Jian Wang.Writing-original draft: Jian Wang.

Table 3 presented
BCTs identified in the intervention description available in the selected articles and other published articles BCTs implemented in the included intervention programs.
Abbreviation: fat restriction; WR: weight restriction; RFB: reward for behavior; RFE: reward for eating; * Coercive feeding strategy.aFeedingpractices noted with the primary outcomes.bFeeding practices noted with the secondary outcomes.a1Feeding practices noted with the primary and secondary outcomes.cTwo specific feeding practices recategorized into one feeding practice: emotional and instrumental feeding practices.dStructured meal setting.eStructured meal timing.fFamily meal setting.gFood availability (vegetables).hFood availability (non-potato vegetables).T A B L E 3