Childhood overweight and obesity are associated with an increased risk of diabetes, CVD, adult overweight and consumption of ultra-processed foods including sugar-sweetened beverages (SSB)(Reference Must and Strauss1–Reference Rauber, Steele and da Louzada3). Drinking water during meals has been shown to reduce hunger and promote satiety but may not impact calories consumed(Reference Lappalainen, Mennen and van Weert4,Reference DellaValle, Roe and Rolls5) . Substitution of water for SSB has been associated with reduced energy intake, increased energy expenditure and increased fat oxidation in studies of obese adults and children(Reference Wang, Ludwig and Sonneville6,Reference Stookey7) .
Food insecurity, a chronic lack of ‘access to enough food to support an active, healthy life’, is a risk factor for childhood overweight and obesity(8–Reference St. Pierre, Ver Ploeg and Dietz13). People experiencing an unpredictable food supply may be more prone to weight gain to buffer for times of food scarcity(Reference Dhurandhar14,Reference Bateson and Pepper15) . The stress of an unreliable food supply may impact self-regulation in the presence of food, decreasing satiety and increasing emotional overeating(Reference Wardle, Guthrie and Sanderson16–Reference Eagleton, Na and Savage18). Childhood food insecurity is associated with higher consumption of total calories, fat, sugar and fibre(Reference Fram, Ritchie and Rosen11,Reference Sharkey, Nalty and Johnson19) . Mothers, infants and toddlers with food insecurity are more likely to consume SSB and consume them more frequently than those who were food-secure(Reference Cunningham, Barradas and Rosenberg20,Reference Fernández, Chen and Cheng21) . Low-income households are commonly found in areas where there is a high concentration of unhealthy food outlets, many of which sell SSB(Reference Elbel, Tamura and McDermott22). The relative affordability of SSB, and the ubiquity of SSB, and SSB advertisements in these communities promote the sale of SSB over healthier beverage options(Reference Blecher23).
School-based drinking water interventions that promote the substitution of water for SSB have been shown to increase water consumption, reduce SSB consumption, decrease flavoured milk purchases and reduce the prevalence of childhood overweight(Reference Muckelbauer, Libuda and Clausen24–Reference Kenney, Gortmaker and Carter26). The Water First drinking water access and promotion intervention increased the frequency of water consumption and reduced overweight prevalence among low-income, ethnically diverse, fourth-grade students in the San Francisco Bay Area(Reference Patel, Schmidt and McCulloch27). Studies found that adults with food insecurity experienced reduced benefits from nutrition interventions, but little is known about the impact of food insecurity on children’s responses to nutrition interventions(Reference Grilo28,Reference Brimblecombe, Ferguson and Barzi29) . Informed by this research, we hypothesised that students experiencing food insecurity would be less likely to benefit from the intervention. This would in turn reduce the impact of the Water First intervention in preventing unhealthy weight gain among students.
Methods
The Water First cluster-randomised controlled trial was a drinking water promotion and access intervention conducted with predominantly low-income and ethnically diverse fourth-grade students(Reference Moreno, Schmidt and Ritchie30). Enrolled schools served low-income households (≥ 50 % of students eligible for free and reduced-priced meals) and were not already promoting drinking water by offering appealing water stations and/or providing cups or reusable water bottles. A total of twenty-six elementary schools (cohorts of 6–8 schools per year) in four school districts in the San Francisco Bay Area, California, were enrolled from August 2016 to March 2020. Half of the schools within each district cohort participated in the intervention, while half served as controls. Data from eight schools enrolled in the 2019–2020 cohort were incomplete due to COVID-related school closures and therefore omitted from this analysis(Reference Patel, Schmidt and McCulloch27,Reference Moreno, Schmidt and Ritchie30) .
Intervention
In each intervention school, a tap water dispenser with disposable cups was installed in the cafeteria and two reusable water bottle filling stations were installed in additional high-traffic locations, including areas where students had physical education classes or recess. Students in schools randomised to the intervention were given reusable water bottles and engaged with Water First staff in eight 15-min classroom activities highlighting the health, financial and environmental benefits of drinking water. Schoolwide activities included assemblies and awarding of small prizes to students drinking water. Details of the study protocol are published elsewhere(Reference Moreno, Schmidt and Ritchie30).
Data collection
At three time points, baseline (at the start of the school year), and 7 and 15 months later, Water First staff measured students’ height and weight using methodology consistent with National Health and Nutrition Examination Survey Anthropometry Procedures Manual(31), and students completed surveys reporting frequency of beverage consumption. Diary-assisted 24-h dietary recalls were conducted at baseline and 7 months. Surveys at 15 months captured students’ self-reported child food insecurity (CFI) status(Reference Moreno, Schmidt and Ritchie30).
Outcome variables
The primary outcome for the Water First study was prevalence of overweight (BMI for age and sex: ≥85th percentile). Secondary weight status outcomes included prevalence of obesity (BMI for age and sex: ≥95th percentile), BMI, BMI percentile and BMI z-score(Reference Moreno, Schmidt and Ritchie30,32) . Dietary intake, also a secondary outcome, was assessed in two ways. Diary-assisted 24-h dietary intake recalls conducted by trained researchers using the multiple-pass method were used to evaluate water, food and beverage intake over the previous 24 h(Reference Thompson, Subar, Coulston, Boushey and Ferruzzi33). Food and beverage calories were estimated using the US Department of Agriculture’s Food and Nutrient Database for Dietary Studies(34). An adapted instrument for students, used in prior studies(Reference Moreno, Schmidt and Ritchie30,Reference Neuhouser, Lilley and Lund35) , was used to assess the frequency of student intake of plain water, SSB, juice, flavoured milk and plain milk.
Food insecurity
Food insecurity was quantified using five of the nine statements from the Child Food Security Assessment (CFSA)(Reference Fram, Ritchie and Rosen11,Reference Fram, Frongillo and Jones36,Reference Fram, Frongillo and Draper37) . The US Department of Agriculture Food Security Survey Module for Youth was not used as it includes questions only on food quality and quantity and is designed only for children 12 years and older(38). In contrast, the CFSA was developed based on interviews with children as young as 7 years old and taps into children’s cognitive, emotional and physical awareness of food insecurity(Reference Fram, Frongillo and Jones36,Reference Fram, Frongillo and Draper37,Reference Connell, Nord and Lofton39) . To achieve a reasonable student survey length, five items from the CFSA were selected as the most accurate for assessing student awareness of food insecurity and reliably measuring food insecurity in children aged 7 years and up(Reference Fram, Ritchie and Rosen11,Reference Fram, Frongillo and Draper37) . Students were asked how often in the previous 12 months did they experience the following:
-
1. We can’t get the food we want because there is not enough money.
-
2. I worry about how hard it is for my parents to get enough food for us.
-
3. I worry about not having enough to eat.
-
4. I feel hungry, because there is not enough to eat.
-
5. I get really tired, because there is not enough to eat.
In accordance with the assessment guidelines, responses were coded as 0 (never), 1 (1 or 2 times) and 2 (many times) and summed across all statements for a relative CFI score (0–10)(Reference Fram, Ritchie and Rosen11). The CFI score was categorised into three subgroups: score=0 (no CFI), score=1 or 2 (medium CFI) and score>2 (high CFI)(Reference Ezennia, Schmidt and Ritchie40). These cut-offs were selected based on distribution to establish categories with similar sample sizes and to provide meaningful interpretation of results. The distribution of CFI scores is presented in Table 1.
Table 1. Baseline characteristics of Water First food insecurity study participants
* Student participants per school assessed using linear regression clustering on school.
† Differences in age, hours of screen time yesterday and food insecurity score by intervention status were assessed using mixed-effects linear regression models accounting for school and class effects.
‡ Percentage of female students, race/ethnicity and frequency of physical activity/week were assessed using mixed-effects logistic regression models accounting for school and class effects.
§ Overweight/obesity is defined as BMI for age and sex: ≥85th percentile.
|| Obesity is defined as BMI for age and sex: ≥95th percentile.
Covariates
Covariates were prespecified in the Water First study protocol to adjust for potential imbalance that is more common in cluster-randomised controlled trials than trials that randomise at the individual level. Covariates assessed via student self-report at baseline included age, gender, race/ethnicity, physical activity and screen time. Physical activity was assessed using questions from the Physical Activity Questionnaire for Older Children and Adolescents(Reference Benítez-Porres, López-Fernández and Raya41). Students were asked how many times in the previous 7 d did they spend their free time doing things that involved physical effort that made them breathe hard or sweat. Reporting categories were never (0 times), sometimes (1–2 times), often (3–4 times), quite often (5–6 times) and very often (7 or more times) in the previous 7 d. Screen time was reported as a continuous variable summed over three categories: playing video or computer games, watching movies or programmes on TV or computer, or doing other things on a computer or phone such as searching the internet, social media, email or texting. Students reported during the previous day how much time they had spent for each category: no time at all, less than an hour, 1–2 h, 2–3 h, 3–4 h, 4–5 h, or 5 or more hours.
Data analysis
Using Stata version 17, mixed-effects logistic regression models including a three-way interaction between food insecurity, the intervention and time were employed to predict differential changes in weight status, number of times per week different beverages were consumed, and food and beverage energy intake. Models controlled for covariates listed above. Potential clustering of students in classes and schools was addressed in the models through inclusion of random effects for the school, class and student. To achieve convergence of obesity regression models, the covariates were limited to race/ethnicity and potential clustering was addressed through inclusion of random effects for students.
Dietary intake data were log-transformed prior to regression analysis to account for the skew of variable distributions. Regression coefficients were subtracted from baseline estimates for each time period and exponentiated to estimate percent changes in median predicted values for dietary intake outcome variables.
Differences in predicted estimates resulting from interaction of the intervention with food insecurity over time were evaluated for statistical significance. Because of the poor precision in estimating interaction terms, it has been suggested to raise the P-value to declare statistical significance of an interaction term to as high as P < 0·20(Reference Durand42). To balance multiple testing concerns with this poor precision, we elected to declare interactions with P < 0·05 as statistically significant. Outcomes found to be significantly modified by food insecurity and the intervention over time were further evaluated by food insecurity subgroups.
Results
The study sample included 1056 students for whom food insecurity status was reported (84 % of the total sample – 206 students [16 %] were lost to follow-up or did not respond to the food insecurity questions) (Fig. 1). At baseline, study students had a mean age of 9·6 years (sd = 0·41), 47·4 % were female and 63·6 % were of Mexican American/Latino/Hispanic race/ethnicity. There were significantly fewer Asian/Native Hawaiian and Other Pacific Islanders in the intervention group (n 62) compared with the control group (n 85) (P = 0·005). No other significant differences were found in demographic variables between the intervention and control groups. The food insecurity score distribution was skewed (mean 1·8, sd 2·1) with 65 % of students reporting some level of food insecurity in the previous 12 months (Table 1).
Figure 1. CONSORT flow diagram for the impact of the Water First promotion and access intervention on elementary school students
CFI status significantly interacted with the prevalence of obesity (P = 0·04) and the volume of water consumed (P = 0·04). No significant interaction was observed for the Water First primary outcome, overweight. There were also no significant interactions for other secondary outcomes including: BMI z-score, BMI percentile, overall calories, food calories, beverage calories, SSB calories, and frequency of milk, flavoured milk, SSB, 100 % juice and water intake.
Subgroup analyses of significant interactions were conducted to understand the predicted outcomes associated with the prevalence of obesity and the volume of water consumed within each CFI category. Among students with no CFI, those exposed to the intervention had a reduced prevalence of obesity between baseline and 7 months (–0·04, CI –0·08, 0·01) compared with an increase among no CFI controls (0·01, CI –0·01, 0·04) (P = 0·04) (Table 2). Among students with high CFI, the intervention group had significant increases in volume of water consumed between baseline and 7 months (86·2 %, CI 21·7, 185·0) compared with a decrease (–13·6 %, CI –45·3, 36·6) observed in the high CFI control group (P = 0·02). There was no evidence of significant interaction between the intervention and CFI relative to other outcomes of interest.
Table 2 Changes in prevalence of obesity and water intake from pre- to post-intervention by child food insecurity categories
* Child food insecurity score based on student survey responses to five questions from the Child Food Security Assessment. Responses were coded as 0 (never), 1 (1 or 2 times) and 2 (many times) and summed across all statements for a relative food insecurity score (0–10)(Reference Fram, Ritchie and Rosen11). The child food insecurity score was categorised into three categories: score = 0 (no child food insecurity), score = 1 and 2 (medium child food insecurity) and score > 2 (high child food insecurity)(Reference Ezennia, Schmidt and Ritchie40).
† P-value calculated from analysis of regression model for interaction of food insecurity, intervention and time.
‡ Multilevel mixed-effects logistic regression models used to examine intervention impacts on changes in outcomes adjusting for intervention status, time point and race/ethnicity. Models included random effects for students.
§ Multilevel mixed-effects logistic regression models used to examine intervention impacts on changes in outcomes, adjusting for intervention status, time point, age, race/ethnicity, gender, screen time, physical activity and time. Models included random effects for school, class and student change over time.
Discussion
Consistent with our hypothesis, the Water First intervention did not reduce the prevalence of obesity among children with food insecurity even though it was effective for others(Reference Patel, Schmidt and McCulloch27). The change in the prevalence of obesity was significantly lower among students with no CFI in the intervention group (–13·8 %) compared with the control group (4·5 %). There was no significant difference between the change in the prevalence of obesity over time among students with CFI in the intervention group compared with those in the control group, suggesting that students with no CFI may have benefitted more from the intervention.
Among students with high CFI, those in the intervention group significantly increased their volume of water intake during the trial. Increased water intake is likely to be attributable to the intervention, which focused on water promotion. Concomitant reductions in SSB consumption were not observed in students with CFI exposed to the intervention. This study was not equipped to investigate the intervention’s mechanisms of action, but the literature suggests a range of possible mechanisms. Prior studies find that adults and children with food insecurity frequently eat beyond satiety and experience emotional overeating(Reference Eagleton, Na and Savage18). Moreover, the SSB industry selectively price and market their products to low-income consumers(Reference Blecher23), and SSB intake may be habitual in households with food insecurity as a low-cost source of calories(Reference Fernández, Chen and Cheng21). High-calorie, low-nutrient diets have been identified as a potential link between food insecurity and poor health outcomes that may be impacted by interventions(Reference Cunningham, Barradas and Rosenberg20,Reference Fernández, Chen and Cheng21) .
As a result of these possible mechanisms, children with food insecurity could experience higher barriers to drinking water over SSB that were not overcome by the Water First intervention. Results suggest that water promotion efforts should be designed in ways that enhance the benefit to children with food insecurity. Like other studies(Reference Muckelbauer, Libuda and Clausen24–Reference Kenney, Gortmaker and Carter26), Water First focused on making changes to the school environment and did not evaluate or impact the availability of water outside of school. Nor did it alter the widespread availability of ultra-processed foods, including SSB, in the child’s food environment at home or in the community(Reference da Louzada, Costa and Souza43). Future studies should attempt to maximise the benefit of water promotion by engaging parents in the intervention with the intent of impacting the food environment both inside and outside school, especially in low-income communities.
A strength of this study is the use of student self-reporting to define food insecurity. Self-reporting has been identified as a more accurate assessment of CFI compared with parent reporting. Even when parents reported shielding children from household food insecurity, children reported household food insecurity(Reference Fram, Frongillo and Jones36,Reference Frongillo, Fram and Escobar-Alegría44) .
This study has several limitations. The control group included significantly more Asian/Native American and other Pacific Islanders than the intervention group; analyses controlled for within person changes so any potential bias should be minimal. Estimation of interaction terms coupled with multiple testing issues introduce poor precision to regression models used in this evaluation; selecting a significance level of P < 0·05 is conservative, but these precision concerns may limit the relevance of the results presented in this study. The Water First cluster-randomised controlled trial was not powered to evaluate subgroup interaction effects on the intervention. CFI was only measured at the 15-month follow-up and therefore may not accurately reflect changes in the level of food insecurity in the study population throughout the study period.
Conclusion
School-based drinking water interventions may be impacted by the presence of CFI among students. Future, adequately powered studies may enhance the understanding of the interaction between nutrition interventions and food insecurity. Consideration of food insecurity in the design of nutrition interventions may maximise the benefits to all populations.
Authorship
LG conceived and implemented this evaluation, performed data analysis and interpretation, drafted the initial manuscript, and reviewed and revised the manuscript critically for important intellectual content. LB participated in data collection, helped lead the intervention, performed data cleaning and data analysis, and provided input in data interpretation, and reviewed and revised the manuscript critically for important intellectual content. CM contributed to study conceptualisation, study design, data collection, provided specific expertise in data analysis and interpretation, and reviewed and revised the manuscript critically for important intellectual content. LR contributed to study conceptualisation, study design, led the collection, cleaning, and analysis of food and beverage diary data, provided input on interpretation of study findings, and reviewed and revised the manuscript critically for important intellectual content. VO conducted data collection, data entry and cleaning and provided details on study implementation. LS contributed to the study conceptualisation and interpretation and reviewed and revised the manuscript critically for important intellectual content. AP conceptualised, designed, and implemented the study, led the development of the intervention, and reviewed and revised the manuscript critically for important intellectual content. All authors conceptualised ideas and study design, interpreted findings, reviewed drafts of the manuscript, approved the final manuscript as submitted and agreed to be accountable for all aspects of the work.
Financial support
Research reported in this publication was supported by the National Heart, Lung, and Blood Institute of the National Institutes of Health under award numbers R01HL129288 and K24HL169841. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Competing interests
There are no conflicts of interest.
Ethics of human subject participation
This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving research study participants were approved by the Internal Review Board of Stanford University (Internal Review Board Number 42210). Written informed consent was obtained from the primary caregiver of all subjects/patients. Written assent was obtained from all student participants.
Open access
