Interventions to increase the consumption of water among children: A systematic review and meta‐analysis

Summary The aim of this study was to conduct a systematic review and meta‐analysis on the effectiveness of interventions to increase children's water consumption. A systematic literature search was conducted in seven electronic databases. Studies published in English before 18 February 2019 that evaluated any type of intervention that measured change in water consumption among children aged 2 to 12 years by applying any type of design were included. Of the 47 interventions included in the systematic review, 24 reported a statistically significant increase in water consumption. Twenty‐four interventions (17 randomized controlled trials and seven studies with other controlled designs) were included in the meta‐analysis. On average, children in intervention groups consumed 29 mL/d (confidence interval [CI] = 13–46 mL/d) more water than did children in control groups. This effect was larger in eight interventions focused specifically on diet (MD = 73 mL/d, CI = 20–126 mL/d) than in 16 interventions focused also on other lifestyle factors (MD = 15 mL/d, CI = 1–29 mL/d). Significant subgroup differences were also found by study setting and socioecological level targeted but not by children's age group, intervention strategy, or study design. In conclusion, there is evidence that, on average, lifestyle interventions can lead to small increases in children's daily water consumption. More research is needed to further understand the specific intervention elements that have the greatest effect.


Summary
The aim of this study was to conduct a systematic review and meta-analysis on the Significant subgroup differences were also found by study setting and socioecological level targeted but not by children's age group, intervention strategy, or study design. In conclusion, there is evidence that, on average, lifestyle interventions can lead to small increases in children's daily water consumption. More research is needed to further understand the specific intervention elements that have the greatest effect.

| INTRODUCTION
Water is a healthy alternative to sugar-sweetened beverages (SSBs), of which high consumption has been associated with weight gain [1][2][3] and tooth decay 4,5 in both children and adults. Guidelines therefore recommend introducing plain water when children are 6 months old and that it should be the principal source of hydration for children older than 1 year. [4][5][6] In addition, the consumption of cow's milk for children older than 1 year is also recommended, because milk can contribute nutrients to childrens diet. 6 Evidence from longitudinal studies suggest a weight-reducing effect when consuming water instead of SSBs in children and adolescents 7 as well as in adults. 8 Some controlled trials have also found that promoting water consumption among children reduces weight gain. 9,10 Different mechanisms might underlie these findings. The total amount of calories consumed may be reduced as water contains no calories, whereas SSBs do. 11,12 Another mechanism supported by Varsamis et al may be that consuming SSBs is linked to elevated glucose responses and sustained elevation in plasma insulin during a day of prolonged sitting, 13 which could lead to higher calorie intake in subsequent meals. A review by Daniels and Popkin suggested that the consumption of water instead of SSBs during or before meal times might reduce the energy intake during the meal. 14 Choosing to drink water as the main beverage is a habit that is likely formed in childhood. 15,16 The family environment is viewed as the principal place where dietary habits are shaped, especially during early childhood. 17 Parents create the food environment in the home and often act as the role models and gatekeepers for the dietary behaviours of their children. [18][19][20] When children become older, the preschool and school environment can also influence the consumption behaviours of children. 21,22 Most interventions that target dietary and obesogenic behaviours have therefore been conducted in either the home or school environments; the latter, in particular, have received a lot of attention. 23,24 Some of these lifestyle interventions focus specifically on changing children's diet and the consumption of specific foods or beverages 25,26 or multiple types of foods or beverages. 27 Other interventions focus on changing both dietary behaviour and other obesogenic lifestyle behaviours such as physical activity and sedentary behavior. 28,29 In a previous systematic review by our team, we identified potentially modifiable factors that were associated with children's water consumption; these factors were the child's self-efficacy, parental self-efficacy, and parental restrictive and encouraging feeding practices. 30 By targeting such factors, lifestyle interventions may be able to promote the water intake among children. A positive effect of such lifestyle interventions has also been found for related outcomes such as the reduction of SSB consumption 21,22,31 and reduction of weight gain. 22,32,33 Although limited, recent evidence provides some indication that water consumption among children may indeed be promoted by interventions. 24,34 A systematic review by Cradock et al on interventions to increase drinking water access and consumption in children younger than 5 years found that 12 of the 18 studies that measured children's water consumption reported positive effects on water intake. 34 A meta-analysis on the effectiveness of lifestyle (diet with/without other obesogenic lifestyle behaviours) interventions on SSB and water consumption among children and adults by Vargas-Garcia et al only included seven studies that targeted children and found an increase of 67 mL/d in children's water consumption. 24 However, to date, no comprehensive and rigorous evidence exists on the effectiveness of interventions to promote water consumption among children of preschool and primary-school age. The aim of this study was therefore to conduct a systematic review and meta-analysis on the effectiveness of interventions to increase the consumption of water among children aged 2 to 12 years. We focused on children older than 2 years because patterns of and recommendations for beverage intake among children aged 0 to 2 years change substantially for water, breastmilk, cow's milk, and juice. 35 2 | METHODS  30 This search was updated in all seven electronic databases on 18 February 2019. A combination of key words was used in the search: (water or beverage* or drink* or related key words) and (child* or infant* or toddler* or related key words) and (intervention* or trial* or strateg* or effect* or promot* or related key words).

| Search strategy
The search strategy was adapted to each database. The full search strategies are presented in the Supporting Information. In addition to the database search, references of relevant articles were screened for other studies. The systematic review protocol for this study was registered in the PROSPERO registry under registration number CRD42019124808 on 18 April 2019.

| Selection process
The duplicates of records retrieved in the search were removed. Subsequently, two independent reviewers (C.F., and L.W. or M.B.) performed title and abstract screening of the remaining records in order to identify studies that met inclusion criteria. Then, copies of full-text articles were ordered for all remaining studies, and full-text screening was performed by two independent reviewers (C.F., and L.W. or M.B.). At both stages, disagreements that arose were discussed between them and, if necessary, resolved by consultation with a senior reviewer (H.R.).

| Inclusion and exclusion criteria
We applied specific inclusion criteria in the selection process. (a) We included participants with mean age between 2 and 12 years at baseline. (b) For the systematic review, we included any type of study design that allowed us to study the effectiveness of an intervention to increase the consumption of water among children, such as randomized controlled trials (RCTs), non-RCTs, and other controlled and noncontrolled quasi-experimental designs. For the meta-analysis, we studied the mean difference between control group and intervention group in millilitres of water per day. Therefore, we could only include controlled studies that measured, within a specified time period, water consumption amount (millilitres, litres, grams, ounces, cups, glasses, and servings) and/or frequency (consumption occasions, consumption frequency, and consumption times). (c) We included any type of intervention strategy that aimed to promote water consumption among children. For studies that had a control group, the control group was defined as children who were not exposed to the intervention designed to promote water consumption. The main exclusion criteria that were applied during the selection process were as follows: (a) studies that only included participants from clinical populations (eg, obesity, malnutrition, and gastroenteritis), as we focused on the general population; (b) studies with data of less than 10 participants; and (3) studies that did not use human subjects. When more than one article was published on the same data set, the article with the longest follow-up period was used. Pilot studies, five in total, [36][37][38][39][40] were included when a full trial of the intervention was not available.

| Data extraction
After discussion and consensus among the study team, a standardized data extraction form was developed. This form was used to extract data from the studies by one researcher (C.F.). The information that was extracted included author, year and country of study, study design and name, intervention content (setting, strategy, socioecological level targeted, focus, frequency, and theory used), control condition, length of intervention and follow-up time, population age and characteristics, number of participants in intervention and control groups and number of clusters (if applicable), how water consumption was measured, participation and retention rate, and outcome data: effect of the intervention on water consumption among children. Key data (intervention content and outcome data) were checked by a second researcher (M.B.). If available, published protocol papers were obtained and used during data extraction.
For the purpose of the meta-analysis, continuous data were extracted either as mean with standard deviation or as adjusted mean difference with standard error. When studies had multiple follow-up time points, the time point with the longest follow-up time was chosen as suggested in the Cochrane Handbook. 41 For two studies that had multiple intervention arms with different intervention elements, 42,43 we used the intervention arm with all intervention elements combined. For one study that had two slightly different intervention arms, 44 the average of the two intervention arms was used as recommended in the Cochrane Handbook. 41 For specific choices for each paper, see Table S1. When data were missing, the authors were contacted to obtain the missing data.

| Risk of bias assessment
The risk of bias was assessed independently by two reviewers (C.F. and M.B.). For RCTs, the Cochrane Collaboration's tool for assessing risk of bias was used. 45 This tool assesses bias in random sequence generation; in allocation concealment; in blinding of participants, personnel, and outcome assessors; because of incomplete outcome data; because of selective reporting; and because of other reasons. For each domain of bias, the study was categorized as having "low" or "high" risk of bias. When it was not possible to determine the risk of bias for a certain bias domain because of missing information in the article, the domain was coded as "unclear." The most serious rating across these bias domains determined the overall risk of bias; eg, if a study was categorized as having a "low" risk of bias in five domains but a "high" risk of bias in one domain, the overall risk of bias was high. For other designs, the Risk Of Bias In Nonrandomized Studies of Interventions (ROBINS-I) was used. 46 The ROBINS-I tool assesses bias because of confounding, in the selection of participants into study, in classification of exposures, because of departures from intended exposures, because of missing data, in measurement of outcomes, and in selection of the reported result. For each domain, the study was categorized as having "low," "moderate," "serious," or "critical" risk of bias. For example, for the "bias due to confounding" domain, we assessed whether the study corrected for confounding variables, such as the child's sex and age. When it was not possible to define the risk of bias for a specific bias domain because of missing information, the domain was coded as "no information." Again, the most serious rating across bias domains defined the overall risk of bias. When there were discrepancies in the judgement of bias between the two reviewers, these were resolved through discussion.

| Analysis
For the qualitative synthesis, we calculated the number of interventions that found a (statistically significant) positive effect on water consumption among children out of the total number of interventions included in the systematic review. We conducted a meta-analysis only with the subset of interventions with a controlled study design and appropriate outcome data available. A random-effects meta-analysis was conducted to account for the between-study variance using the mean difference in millilitres of water consumption at follow-up between the intervention group and control group and the standard error of this difference. The overall mean difference and 95% confidence interval (CI) across all studies were estimated, and forest plots were created that graphically display these results. The meta-analysis was conducted in Review Manager (version 5.3, Cochrane Library).
The Cochrane Handbook was used for guidance regarding missing data and combining of data, 41 and the Cochrane calculator was used for making calculations. Results from the most adjusted models were used, wherever available. If the mean difference in water consumption between the control group and intervention group at follow-up was not reported, the mean water consumption at follow-up was extracted separately for the intervention group and control group, and the mean difference between intervention and control group was calculated. Using follow-up scores instead of change from baseline to follow-up scores is suggested to generate more conservative results in meta-analyses. 33,47 If in the paper water consumption was reported in a different quantity than in millilitres or within a different time period than 1 day, consumption was recalculated to millilitres of water per day. If the size of the portions was not reported in the paper, we used a portion size of 225 mL per drink or consumption occasion. This was chosen because among the included papers that reported portion sizes, these varied between 200 and 250 mL. 9,48-51 If CIs were presented instead of standard errors or standard deviations, these were calculated with the Cochrane calculator. When data were presented stratified by subgroups, such as by sex, subgroups were combined. If standard deviation at follow-up was missing but standard deviation at baseline was available, this was used. For specific calculations for each paper, see Table S1.
The I 2 test was used to assess heterogeneity across studies; above 25% is considered low variance between studies, above 50% is considered moderate variance, and above 75% is considered high variance. 52 As specified in our protocol, subgroup analyses were performed with (a) potential moderators: (1)

| Study selection
The inclusion and exclusion of articles are described using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses

| Study characteristics
The overall characteristics of the studies that were included in the systematic review are shown in Table 1; specific details of each study are shown in

| Effectiveness of interventions on water consumption in children
Of the 47 studies included in our review, 24 reported statistically significant effects on children's water consumption (Table S2)

| Effect by intervention strategy
Eight interventions that only used education as an intervention strategy had an overall smaller mean difference between intervention and control groups ( Table 2  T A B L E 2 Subgroup analyses using random-effects models of the mean difference in childrens water consumption between intervention and control groups (n = 24)

| Effect by intervention setting
There were 13 interventions that were in a school setting, five interventions that were based in a nonschool setting (community or home), and six interventions that were based in both a school and nonschool setting. Interventions based in both a school and nonschool setting had the smallest mean difference between intervention and control groups ( Table 2 and Figure S3: MD = 5 mL/d, CI = −9; 20 mL/d, P = .96, I 2 = 0%), followed by interventions that were based in a school setting (MD = 33 mL/d, CI = 6; 60 mL/d, P = .02; I 2 = 77%) and interventions that were based in a nonschool setting (MD = 65 mL/d, CI = 34; 96 mL/d, P < .001; I 2 = 0%). Differences between these three groups were statistically significant (P = .002).

| Effect by mean age of children
Ten interventions that targeted children with a mean age at baseline of between 2 and 5 years had an overall smaller mean difference between intervention and control group (Table 2 and Figure S5:

| Sensitivity analyses
The 17 RCTs had overall smaller mean difference between intervention and control groups ( Table 2 and Figure  there were fewer small studies that had a negative effect than what would be expected ( Figure S8). For two studies, 9,40 the mean difference between intervention and control groups was larger (outside of the 95% CI) than those of the other studies ( Figure S8). When excluding these two studies, the average mean difference between intervention and control groups and heteroge- These interventions, sometimes called "combined" or "holistic" lifestyle interventions, have been found to be particularly effective in reducing weight among both general populations of children and children with obesity. 23,32 We, however, found that interventions that specifically focused on diet or beverage consumption on average had a larger effect on water consumption among children than had these combined lifestyle interventions. A reason for this could be that within these broader interventions, the message to drink water receives less attention or gets lost within a multitude of other themes such as physical activity and active play. The design of the intervention strategy itself influences the uptake of messages related to water intake. A combined lifestyle RCT by Contento et al did not have an effect on water consumption but did decrease SSB intake. 48 This finding illustrates that children may not necessarily replace SSBs by water. Contento et al reported that with regard to the intervention, more time was spent on behaviours related to energy balance and diabetes and that the activities children engaged in were more "memorable" than were behaviours related to water. 48 When wanting to increase water and decrease SSB consumption, messages that promote water consumption may need to be prioritized alongside messages that limit SSB consumption. Additionally, intervention fidelity might also be lower for combined interventions due to having to divide time and resources over multiple interventions goals. 89 Intervention goals that are easier to implement might then be prioritized over goals that are more difficult to implement. 61,80 Siega Riz et al noted that replacing SSBs by water in vending machines was not possible in all schools involved in their combined lifestyle intervention. 80 De Coen et al found that most schools did not meet their suggested snacks and beverage policy guidelines, which included the installation of water fountains. 61 Of note is that many countries promote both water and cow's milk as healthy beverages, and these may be competing for messaging space. Especially in settings where malnutrition is a major public health concern, (fortified) cow's milk can contribute important nutrients to a child's diet. 90 24 The greater involvement of parents in home-based interventions compared with school interventions was suggested to be an important factor in the greater success of these interventions. 24 Indeed, all nonschool-based interventions in our meta-analysis either were based at home or involved parents directly in a community setting. Targeting only the child may not be the best intervention target, as we found that interventions that only targeted the child had a smaller effect than had interventions that also or only targeted parents, peers, and/or the environment. For future interventions, this emphasizes the importance of viewing childhood consumption behaviours within a socioecological framework, as children may be particularly receptive to their social and structural environments. In addition, interventions for young children may be more effective if the caregiver is targeted rather than the child, since the caregiver selects and provides most meals and drinks. So far, research has mainly focused on the association between child-related factors and water consumption and especially environmental factors have been understudied. 30 Environmental interventions such as choice architecture interventions may be a promising approach to promote healthy dietary behaviors. 99,100 Which specific parental and environmental factors need to be targeted in order to improve water consumption among children and which specific components of interventions are most effective in doing so need to be studied in more detail.

| Strengths and limitations
To the best of our knowledge, this was the first systematic review and meta-analysis that focused solely on the effectiveness of interventions in promoting water consumption among children aged 2 to 12 years. The literature search was performed in seven databases, and a rigorous procedure was followed for the inclusion of studies in our review. 101 However, our review also has some limitations that must be acknowledged. We included RCTs, non-RCTs, and other quasiexperimental designs in our systematic review and meta-analysis.
Non-RCTs and other quasi-experimental designs are considered to provide lower-quality evidence and more often show significant results than are RCTs when there are none. We found a lower effect in RCTs compared with other controlled designs in our meta-analysis, although this difference was not significant. Heterogeneity was moderate to high across the studies included in our review, and subgroup analyses were only partly able to explain this variation. Other differences between studies that were not explored may have explained this variation. Whether or not interventions are theory based may be an important factor in the effectiveness of interventions; however, only two studies included in our meta-analysis were not theory based.
Further, Cochrane advises to have at least 10 studies in each subgroup, 41 which was not the case for some subgroup analyses in our review. However, findings from subgroup analyses in the metaanalysis were confirmed in the qualitative subgroup analyses performed with a larger number of studies, which strengthened our findings. Risk of bias was high in most studies, which was, to a large part, due to measurement of outcomes. Retrospective report, which was most commonly used in the studies included in this review, is considered to be imprecise because of poor recall and (parents of) children with low levels of water consumption reporting higher amounts than actual amounts consumed. [102][103][104][105] For our meta-analysis, we estimated water consumption in millilitres per day for studies that did not report water consumption in volume per day by using a serving size of 225 mL. This may appear more imprecise compared with volume of water consumed per day, although it is uncertain to what level of precision children and parents can estimate their water consumption. 105 For younger children aged 2 to 5 years, a serving size of 225 mL may be relatively large; however, because national standardized serving sizes are between 200 and 250 mL, other studies have used similar serving sizes in this age group. 24,49 Changing serving size to 150 mL for studies with young children that did not report serving size 73,74 did not change our overall estimate. Finally, we did not include studies published in non-English languages and studies that were not published in a peer-reviewed journal; this may have an impact on the generalizability of our results and may have introduced publication bias. 106

| Conclusions
In conclusion, our systematic review and meta-analysis indicate that interventions can on average lead to a small increase in daily water consumption among children. Dietary interventions and interventions that focus on beverage intake specifically appear to have greater effects on improving children's water consumption than have interventions that focus on both diet and other lifestyle factors. Effects also appeared to vary by study setting and socioecological level targeted. However, more research is needed to further understand the specific intervention elements that have the greatest impact on the water consumption of children. Future research is also needed to determine the effectiveness of these interventions on improving weight and other health-related outcomes in children.