How effective are family-based and institutional nutrition interventions in improving children’s diet and health? A systematic review

The following electronic databases were searched from 1980 to December 2014: Medline, Central (Cochrane)/DARE, Embase and CINAHL, with a search strategy that incorporated terms including both Medical Subject headings and keywords for: 1. Food, 2. Family-based programs and Institutional nutrition programs, 3. Nutrition and health outcomes. In addition, filters for high income countries and study design were applied. A Medline search strategy was developed and adapted to the other databases as required. The complete search strategy is attached as a Additional file 1. In addition, the reference lists of included studies were assessed to identify other eligible studies. Only articles published in English were included in the review.

The search results were downloaded into an Endnote library and titles and/or abstracts assessed by one of two authors for eligibility. The full-text manuscripts of those deemed potentially eligible were assessed by either author and any discrepancies resolved by consensus with the other author. Data of eligible studies were entered into a standard template and data entry checked after data extraction completed. The primary health and nutrition outcomes, any adverse outcomes, together with details of the intervention, including length of follow-up, demographics of participants and theoretical basis were extracted. The study authors were contacted to try to obtain additional data if necessary.

The interpretation of the results was facilitated by converting outcomes presented as a mean with a measure of variance to a Cohen’s d effect size estimate using an online calculator [33]. The impact of the interventions was compared using effect size (Cohen’s d) as a standard measure, which was interpreted as shown: Small 0.1–0.2, moderate 0.3–0.5, large > or = 0.6. Due to the heterogeneity of the interventions and study outcomes, a narrative synthesis of the results is presented [29].

The included studies were assessed systematically for methodological quality and risk of bias using the Effective Public Health Practice Project (EPHPP) critical appraisal tool [34]. This tool assesses the risk of bias based on potential selection bias, study design, likelihood of confounding, blinding of outcome assessors and participants, appropriateness of data collection and completeness of follow-up. Studies were classified as high risk of potential bias if two or more of the above categories were assessed as weak (Weak), moderate risk of potential bias if one category was assessed as weak (Moderate) and low risk of potential bias if none of the above categories were assessed as weak (Strong).

The majority of the studies (82%) were undertaken in schools (n = 25) or preschools (n = 6) with the most common approach incorporating multiple components to promote healthy eating and/or other aspects of healthy lifestyle particularly increased physical activity (Table 1). These multiple component nutrition programs included combinations of class curriculum activities, school food service modifications, home activities, enhanced physical education/activities and strategies to engage parents/families (see Table 2). There were also four studies of school meal programs [40‐43], one school gardening study [44], a school fruit program study [45] and two school canteen studies [46, 47]. The duration of the studies ranged from 3 months to the ongoing STRIP studies [48, 49] which reported follow-up of children recruited as infants at 9 and 11 years. Twenty-three studies (59%) were between 6 months and 2 years in duration (Table 1).

Fourteen of the 25 school studies included efforts to involve families [50‐63], however, the uptake of this was limited in most studies where it was assessed [51‐53, 55, 60]. There was only one school study predominantly focussed on family activities, however, it was less intensive as it involved monthly nutrition or physical activity challenges [56]. Five of the six preschool studies involved parent-focussed activities [64‐68], with three of these including parents in nutrition education sessions [65‐67].

The participants in the primary school based programs were predominantly older children aged 8–12 with many of the studies at multiple schools involving hundreds of students. There were two smaller scale programs at single schools with 60–100 participants [62, 69]. There were three particularly large studies- a trial of universal free school breakfast program in the USA (compared with the existing means-tested school breakfast program) assessing 4358 students (153 schools) [42], an evaluation of a free school breakfast program in Wales involving 4350 students at 111 schools (1975 with follow-up at all time-points) [41], and the Child and Adolescent Trial for Cardiovascular Health (CATCH), a multi-component school program in the USA with 5106 students (96 schools) at baseline [60]. The majority of the school and preschool studies involved populations that were representative of the community and thus included children from families with varying socio-economic status (SES)/educational attainment. Six studies included predominantly disadvantaged children, either by selecting schools fulfilling criteria for low SES [41, 57, 70], an intervention for Native American children [53], or by undertaking the study in low income rural communities [54, 56].

The majority of the studies in schools and preschools involved a mixture of support for school staff, either teachers or cafeteria staff, and/or research staff to deliver enhanced nutrition education including opportunities for practical food experiences [44, 47, 50‐53, 57, 59, 61, 65‐72]. These were usually at one or more geographically clustered schools. In addition, two studies had a major focus on teachers and/or parents as role models of healthy eating [66, 72], while another study involved high school students as facilitators of the nutrition education [62]. There were six studies which instituted large scale programs- either developed directly with the input of the education or health department [58, 63, 73], or instituted with low-cost resources or requiring low intensity input by school staff that was consistent with the local school curriculum [55, 60, 64]. The three publicly funded programs included Food Dudes in United Kingdom primary schools [58], Project Energise in primary schools in Waikato, New Zealand [63] and Munch and Move in preschools in New South Wales, Australia [73]. In addition, there were studies evaluating publicly funded national school meals programs- breakfast programs in the USA [42] and Wales [41] and a school fruit program in Norway [45]. There were also smaller research trials of a school breakfast program in New Zealand [43] and a school lunch program in Denmark [40].

There were only eight family-based studies (Table 1) and these predominantly targeted the parents or parents together with their children [20, 49, 74‐79]. Four of the five family-based studies aimed at parents only involved monthly (or less frequent) nutrition education for individuals or groups together with individual assessment [49, 75, 76, 79]. However, the two studies involving parents with their children incorporated practical nutrition or physical activity sessions one to three times a week [77, 78]. The only study to directly target children involved a 4 week summer camp program in the USA with internet follow-up for the children and their parents [74]. These family-based studies had relatively small numbers of participants except for an Australian study involving parent groups for those with infants [75] and the Special Turku Coronary Risk Factor Intervention Project for Children (STRIP) [49]. The recruitment strategies used in these studies meant that unless a disadvantaged population was explicitly targeted [77, 78], then participants were predominantly not disadvantaged [74‐76, 79].

Of the 39 RCTs, 15 (38%) were assessed to be at low risk of bias, 15 were assessed to have a moderate risk of bias and 9 to have a high risk of bias (Table 1). Underlying theories and/or clear rationale for interventions were common among the included studies. The development of 26 of the studies were based on one or more specific theories of behaviour most commonly social cognitive theory (n = 13) and social learning theory (n = 7). In addition, the five school meals programs were developed with a clear rationale. There was no clear relationship between a theoretical basis or rationale and effectiveness of the intervention.

Nutritional intake was assessed using standardised self-reported (or parent-report) measures in 29 of the 38 studies which reported these data (76%) with two of these incorporating assistance by research staff and one using measured self-reporting. Five studies (13%) used measurement/observation by research staff and four (11%) used a combination of self-report and observation. In addition, most of the studies reported standard categories of macro- and micronutrient intake alongside other outcomes, which increases the likelihood of finding statistically significant differences by chance alone due to multiple comparisons [80]. Finally, only 12 studies (31%) reported on follow-up at a time after the intervention, and of these only three reported long-term follow-up greater than 1 year (Table 1).

Despite these potential methodological limitations, the consistent results from these RCTs and the observation that most appeared adequately powered to detect relevant improvements in health outcomes for individuals and populations provides support for the findings of this review.

The focus of this review and hence the most frequent outcome reported was dietary intake, with all but one study reporting on dietary intake through at least one measure. The most commonly reported outcomes were fruit and vegetable intake and fat intake. The intake of energy-dense nutrient-poor foods was another focus analysed in fewer studies. These are reported in detail below.

Many of these programs also aimed to promote healthy lifestyles. Hence in addition to dietary intake, physical activity levels and/or sedentary behaviour was reported by 15 studies (12 school studies and three family studies), screen-time in six studies and body mass index was reported by 19 studies (12 school studies and all seven family studies). Analysis of these outcomes was not undertaken. Other biomedical outcomes were reported in only four studies, which were generally the longer-term studies and/or more intensive studies aimed at reduction of fat intake and adiposity. The determinants of improved dietary intake were reported by a limited number of studies- attitudes and knowledge in five school studies, self-efficacy in two studies and F&V access in one study. Adverse effects were not reported in these studies.

There were 93 effect size estimates extracted or calculated in the 39 included studies. 31 (33%) of these effect sizes (Cohen’s d) were greater than 0.2 and had a 95% confidence interval that did not include the null value. However, there were also five studies which reported no impact on intake of energy-dense, nutrient-poor foods and three studies which reported no effect on at least one biomedical outcome, and no effect size estimates were calculated. The 31 positive effects were all increasing F&V intake or reducing fat intake; however there was no obvious relationship between study components and effectiveness. The impact of involving parents in school/preschool studies was only systematically assessed in the CATCH study [60], which did not find a difference in outcomes between the school-based intervention with or without parental involvement.

Twenty-five studies reported on F&V intake (Table 3). Of these studies, five reported a null effect on fruit and/or vegetable intake [44, 50, 52, 55, 75]. There were eleven studies which found a null to small effect on F&V intake [45, 47, 51, 54, 57, 64, 68, 70, 73, 79, 81]. Five studies reported a moderate effect on F&V intake [46, 58, 62, 65, 69] and four studies which found a large effect on F&/or V intake [20, 59, 72, 74]. In addition, two studies which reported a null to small effect on vegetable or overall F&V intake, also reported a moderate to large effect on fruit intake [50, 57]. Ten of the 25 studies were assessed to be at low risk of bias (Table 1) with eight of these reporting null to small effect on F&V intake.

Overall, there is evidence that both school-based studies and family-based studies, if designed and implemented well, have a positive impact on F&V intake, particularly fruit. These successful programs incorporated engaging and innovative strategies including multimedia programs to motivate the children. This impact was demonstrated for the duration of the intervention and for up to 12 months post-intervention follow-up. For example, the one study of a school garden program found no impact on F&V intake overall, but did report a significant increase in F&V intake in students at schools where a successful school garden was achieved [44]. However, the majority of the studies had a null or small effect on F&V intake.

There were two family-based studies and five school studies that reported a medium or large effect on F&V intake and were at low or moderate risk of bias. Five of these studies were based on social cognitive theory or social learning theory and involved high intensity interventions and/or innovative strategies to engage children. The family-based studies included the GEMS study involving 8 year-old girls at a 4 week summer camp in the USA with weekly internet sessions for girls and their parents afterwards for 8 weeks [74]. This study incorporated activities to promote F&V intake and physical activity into the regular camp program. The other family based study involved obese parents and their non-obese children in a comprehensive weight-control program for the parents comparing the impact of promoting F&V or low fat/sugar diets [20]. Both of these studies were delivered with high fidelity and had high follow-up albeit in small sample size. The five school studies included one which compared exposure to a teacher modelling eating fruit (or other healthy snacks) daily or healthy lifestyle curriculum with a control group [72]. Both of the intervention groups increased fruit intake at the end of the 1 year intervention. However, only the exposure group maintained this increased fruit intake 1 year later. There was also a lunch rewards program with observation 3 days/week [46]. Another school curriculum based program was of low intensity but involved high school students to deliver the program to younger children [62]. All three of these were delivered with high fidelity and achieved high follow-up. There were two other school programs, one involving daily videos and rewards together with F&V provision [58] and the other involving four physical activity and two nutrition education sessions weekly [69]. However, the fidelity of these two interventions and the follow-up achieved was not clear. Three of these five school studies [58, 62, 69] involved parents in some way while the other two had no parental involvement [46, 72], however there was no systematic assessment of the impact of involving parents in any of these studies (Table 3).

Fifteen studies reported fat outcomes (Table 4). Of these studies, six had a null effect [40, 44, 50, 54, 55, 59], four had a null to small effect [42, 53, 60, 69], three had a moderate effect [56, 78, 81] and two studies a large effect on dietary fat intake [48, 76]. Five of the 15 studies were assessed to have a low risk of bias (Table 1) with one of these reporting a null effect, three a small effect and one a moderate effect on fat intake. Overall, there is some evidence of benefit for studies that specifically target fat intake and are set in the home/parent-based intervention, although the majority of the studies had a null or very small effect.

The two studies with a large effect had reduction in fat as a main aim of the study, whereas the majority of the other studies focussed on fruit and vegetable intake. Hendrie et al [76] targeted parents of school-aged children to promote low fat dairy products. This study involved a low intensity intervention in a relatively small sample size; however it was delivered with high fidelity, based on social learning theory and a high follow-up was achieved. In addition to the reduction in fat intake, there was a small change in LDL cholesterol, but no change in BMI. The STRIP study [48, 49] is an ongoing, relatively low intensity program in which families with infants were recruited to attend twice yearly nutrition education and assessment promoting a low saturated fat diet, predominately via parents. This study also had high fidelity, but was not based on any explicit theory. These STRIP studies reported on the follow-up of a small sub-sample, aged 9–11 years at follow up. In addition to these two family-based studies, four of the eight school studies involving parents [53, 56, 60, 69] (all assessed to have a low risk of bias) reported smaller reductions in fat intake of children, although the only study which tested the impact of parental involvement systematically reported no additional benefit compared to no parental component [60] (Table 4).

There were nine studies which reported intake of EDNP foods. Of these, three found no effect [62, 66, 73], two studies reported a small effect for one outcome measure but not the other related to EDNP foods [64, 82], two showed a moderate effect (although findings not statistically significant) [20, 81]. and two studies had limited reporting of results [67, 70]. Overall, the studies demonstrated limited evidence for benefits in reducing EDNP food intake.

The studies reporting a moderate effect were family-based studies. Epstein et al. recruited obese parents, with non-obese children, to a weight control program, with a reduction in EDNP foods among the children (d = −0.51 95% CI -1.30, 0.27) for those in the arm that focused on reducing intake of high fat/sugar foods (compared with the F&V arm) [20]. This study had a small sample size, moderate intensity over 6 months and then 6 months post intervention follow-up and was at moderate risk of bias. The Bounce program involved healthy lifestyle activities after school for mother-daughter pairs, with reductions in high fat foods (Cohen’s d = 0.40 p = 0.26) and sweetened beverages (Cohen’s d = 0.36, p = 0.31) among girls in the intervention group [81]. This was also a small study with high intensity over 3 months and a moderate risk of bias. Given the nature of these studies, the results should be interpreted cautiously.

Five of the studies reported biomedical outcomes, although as the duration/follow-up of the studies was < 3 years, except for the STRIP study, the focus was on proxy measures. There was a null effect on blood pressure and lipids in the STRIP and CATCH studies [48, 60]. The STRIP studies found a small to moderate effect on reducing insulin resistance in 9 year old children [Cohen’s d = −0.16 (95% CI -0.58, 0.27) in girls and −0.58 (95% CI -1.02, −0.13) in boys] and null to small effects on lipids [48]. In a later follow-up, in 11 year olds, there was a small to moderate effect on improving endothelial function [Cohen’s d = 0.11 (95% CI -0.18, 0.39) in girls and 0.34 (95% CI 0.05,0.63) in boys] with reduced total and LDL cholesterol in boys only [49]. There was decreased blood pressure (BP) in the 10–12 year old children [systolic BP mean difference = −0.23 (95% CI -0.43,-0.02); diastolic BP mean difference = −0.14 (95% CI -0.30, 0.04)] but not the 5–7 year old children in Project Energise-a large school-based healthy lifestyle program in New Zealand [63]. Project Energise involved a change agent to champion the promotion of healthy nutrition and increased physical activity in curricular and extra-curricular activities in primary schools. The study also found small decreases in adiposity in 5–7 year old children only. In addition to the reductions in fat intake, Hendrie et al. found reduced LDL cholesterol mean difference = −0.15 mmol/L (95% CI -0.30, −0.01, Cohen’s d = 0.52) and total cholesterol mean difference = −0.12 (95% CI -0.28, 0.05) but no changes in other plasma lipids 3 months after the program [76]. In the school lunch study in Denmark, Damsgaard et al. found no change in the metabolic syndrome score with reductions in mean blood pressure, total cholesterol, TAG and insulin resistance offset by increased waist circumference and reduced HDL cholesterol [83]. Of these five studies, only the CATCH study was assessed to have a low risk of bias. Thus, the limited evidence of improvements in cardiovascular disease risk factors, should also be interpreted cautiously.

The five school-based studies and two family based studies which were targeted at disadvantaged populations (Table 1) did not clearly demonstrate that these programs were more or less effective in these populations compared to the overall assessments described above. Four of the studies reported on F&V intake with one reporting a large effect [57], one a moderate effect [81] and two a small effect [54, 70] (Table 3). Four of the five studies which reported on fat intake in these populations showed moderate effect [53, 56, 78, 81] and one showed no effect [54] (Table 4). The Bounce study, described above, involved disadvantaged African-American mother-daughter pairs in an urban setting and reported reduced fat intake, increased F&V and reductions in EDNP food, as noted previously [81]. This is consistent with the sub-group analysis reported in the 3 year follow-up CATCH study, which showed that overall dietary intake after this program in rural USA was the same in African-American and Hispanic- American children as in white children [84].

There were four studies of three or more years duration included in the review [42, 49, 53, 60]. Three were school-based programs, including two multi-component programs- the Pathways program which aimed to reduce adiposity in American Indian school children and the CATCH program which aimed to reduce risk factors for cardiovascular disease in primary school students (Table 1). Both reported small to moderate effect on reducing total fat and/or saturated fat intake after 3 years: Pathways Total fat intake mean difference-2.5% and CATCH Total fat intake mean difference-2% (Table 4). The other school-based program was the Universal School Breakfast Pilot Program in the USA [42]. This program operated successfully for 3 years; however, follow-up dietary intake data were only collected at 12 months. This showed that intake of a nutritionally substantive breakfast increased from 76% to 80% in intervention schools, but that breakfast skipping was 4% in both intervention and control schools. The other long-term study was the family-based STRIP study in Finland. The children participating in this program had lower intake of total and saturated fat at 9 years and saturated fat at 11 years than control children, although results varied by gender (Table 4).

Long-term follow-up of programs (> 1 year post-intervention) to assess the ongoing impact was reported by two studies in addition to the ongoing STRIP study described above. A 3 year follow-up of the CATCH study reported that these children still had significantly lower intake of total and saturated fat compared to controls although overall dietary intake was not significantly different between the two groups (based on the Healthy Eating Index) [84]. In a 3 year follow-up of the Free School Fruit program in Norway, increases in F&V intake, noted at the conclusion of the program and 1 year after the program, persisted although reduced from the conclusion of the program [85]. It was estimated that boys ate an extra 0.38 serves of F&V/day and girls 0.44 serves of F&V/day as a result of prior participation in this school fruit program compared to an extra 0.6 serves/day at the conclusion and 0.5 serves/day at 1 year [45]. Thus, there is consistent although limited evidence that ongoing multicomponent nutrition programs can improve dietary patterns sufficiently to impact on population health (if sustained) and that these effects persist, although attenuated for up to 3 years.