How effective are interventions at reducing socioeconomic inequalities in obesity among children and adults? Two systematic reviews

Overview

Eleven individual-level studies were identified that met the review inclusion criteria. 40–43,46–54 Because of heterogeneity in terms of study design and main outcomes, as well as the generally poor quality of data reporting in the studies (e.g. studies seldom reported means and standard deviations), it was not possible to conduct meta-analysis for this subset of interventions. The 11 studies are therefore synthesised narratively in terms of whether they followed a universal (n = 740,41,46,48–53) or a targeted (n = 442,43,47,54) approach. The results are also summarised in Table 37 (universal-approach studies) and Table 38 (targeted-approach studies) (see Appendix 5), with effect size data presented (when possible) in Tables 2–5 and implementation information provided in Tables 39 and 40 (see Appendix 5).

The majority of the studies (n = 840–43,46–50) were of weight-management diet and physical activity treatment programmes for childhood obesity conducted in medical/health-care or university settings. In addition, one study investigated a home-based obesity treatment programme51,52 and two studies, one treatment53 and one prevention,54 investigated interventions delivered primarily in participants’ homes. Four47,48,53,54 studies were conducted in the USA, two40,49,50 in Australia and one each in the UK,46 Germany,51,52 Belgium,41 New Zealand43 and Spain. 42 One study was published in a Spanish-language journal42 and one study was published in both German- and English-language journals. 117–119 The remaining studies were published in English-language journals.

The numbers of participants in the studies varied considerably, between n = 16 and n = 445, and the median follow-up time was 12 months (range 3–48 months). The studies were conducted among children aged from 0 to 18 years, with five42,43,46,48,53 including those of preschool age (from 0 to 5 years), 1040–43,46–54 including primary school-aged children (aged 6–12 years) and six40,41,43,46,47,49,54 including secondary school-aged children (aged 13–18 years). There were four48,50,53,54 experimental studies and seven40–43,46,47,49,51,52 observational studies. Seven40,42,47–50,53,54 of the studies were of moderate quality and the remaining studies were of low quality (using the EPHPP tool; see Appendix 3). All of the studies used measured primary outcomes. All of the studies reported some elements of how the intervention was implemented (see Appendix 5, Tables 39 and 40), particularly in terms of motivation, context and resources.

Universal interventions

Three48,50,53 experimental studies and four40,41,46,49,51,52 observational studies took a universal approach and measured outcomes between SES groups (see Appendix 5, Table 37). One experimental study evaluated a home-based intervention intended to reduce the sedentary behaviours (television viewing and computer use) of children (aged 4–7 years) who were either at risk of becoming overweight or obese or already overweight or obese. 53 The active control group received general information on parenting tips, activities and recipes. The study found more favourable intervention effects in terms of a reduced BMI for participants from a low-SES background than for those from a high-SES background. This study was of moderate quality although it had a relatively small final sample size (n = 67). Another moderate-quality experimental study of a health-care setting-based obesity treatment programme (compared with usual care) for preschool children (aged 2–6 years)48 found more favourable results in those of low-SES than in those of high-SES in terms of reduction in BMI (Table 2). The final experimental study (no-intervention control) of a health-care setting-based obesity treatment programme (moderate quality)50 found no relationship between SES and intervention effects on BMI, waist circumference or prevalence of obesity in children aged 5–10 years (see Table 2).

Four observational studies also followed a universal approach40,41,46,49,51,52 (Table 3). These observational studies found more beneficial effects in low-SES groups than in high-SES groups. Two studies (both of low quality) investigating the effects of obesity treatment programmes that targeted both diet and physical activity behaviours in children aged 2–18 years46 and 7–17 years41 found no association between SES indicators and outcomes, although the study by Sabin et al. 46 found a hospital obesity service to be effective in reducing BMI overall. One moderate-quality study found that a treatment programme targeting diet behaviours was more effective in high-SES children aged 10–17 years. 40,49 One low-quality study investigating a home-based diet and physical activity counselling intervention for overweight and obese children (mean age 6.5 years) found that the intervention was less effective in children of low SES than in those of high SES. 51,52 This study had a long duration (4 years); however, it had a very small final sample size (n = 16) and subgroup analysis should be treated with caution.

Targeted interventions

One experimental54 and three observational42,43,47 studies examined targeted individual-level interventions. The one experimental study (no-intervention control) was of moderate quality. 54 It found that a home-based mentor-based health promotion and obesity prevention intervention for children aged 11–16 years reduced the prevalence of obesity and reduced percentage body fat and increased the fat-free mass of the overweight and obese participants (Table 4). Of the three observational studies that followed a targeted approach, two investigated the effects of obesity treatment programmes based in a health-care setting, one in children aged 10–14 years47 and the other in those aged 2–13 years. 42 These moderate-quality studies both found improvements in at least one obesity-related outcome: a reduction in BMI47 and a reduction in the prevalence of obesity but not in BMI42 (Table 5). The other study investigated the effects of a more general nurse-led healthy lifestyle clinic that followed a holistic approach to health needs defined by each patient (not necessarily obesity). 43 This low-quality study found no intervention effect on BMI in patients aged 0–18 years.

Overview

Fifty-two community studies44,45,55–106 (from 54 papers) were identified that met the review inclusion criteria. Because of heterogeneity in terms of study design and main outcomes, as well as the generally poor quality of data reporting in the studies (e.g. studies seldom reported means and standard deviations), it was possible to conduct meta-analysis for only a small subset of interventions in this category (n = 13 targeted interventions). The fifty-two studies are therefore synthesised narratively in terms of whether they followed a universal (n = 1644,45,57,64,74–77,86,88,96–98,100–102,104,105) or a targeted (n = 3654,56,58–63,65–73,78–85,87,89–95,99,103,106) approach. The results are also summarised in Table 41 (universal-approach studies) and Table 42 (targeted-approach studies) (see Appendix 5). Effect size data (when possible) are displayed in Tables 6–9. The meta-analysis of the 13 suitable studies is reported separately at the end of this section with the raw data in Table 10. The implementation information for each study is contained in Tables 43 and 44 (see Appendix 5).

The majority of the studies (n = 3055–75,77–86) investigated interventions that were conducted in school (including preschool/kindergarten and after-school) settings. The setting of one of these studies was not clearly reported;60 however, the intervention was based around school semesters and therefore it is assumed that it was a school-based intervention. Seven studies investigated interventions conducted in community centres or community venues such as sports centres76,87–93 and two studies took place in Head Start centres (preschool centres similar to the UK’s Sure Start centres). 94,95 One study took place in both community centres and homes. 106 Ten studies investigated group-based childhood obesity treatment programmes. Seven of these were conducted in medical/health-care or university settings. 44,45,96–102 In the other three studies the settings were not clearly reported but each investigated group-based weight-management programmes and read as if they were held in medical or university settings. 103–105

The majority of the studies (n = 3255–58,61–83,85,86,89,93–95) were public health interventions that aimed to promote a healthy weight, either by preventing overweight and obesity in those of a healthy weight or reducing weight in those already overweight or obese until they reached a healthy weight. These interventions were targeted at populations of children regardless of their weight status, whereas the remaining 20 studies44,45,59,60,84,87,88,90–92,96–106 were treatment interventions for overweight and obese children. Sixteen44,45,57,64,74–77,86,88,96–98,100–102,104,105 followed a universal approach and either included subgroup analysis of different SES groups or explored associations between SES and intervention outcomes. The remaining 36 studies55,56,58–63,65–73,78–85,87,89–95,99,101,103,106 were targeted at low-SES or disadvantaged groups or areas.

The majority of the studies were from the USA (n = 31);55–62,70–71,76,80–85,90,93–97,99,101–106 in addition, four were conducted in Germany,44,45,72,100 three in Chile65–67 and two in each of Australia,69,78 Brazil,79,87 France64,86 and Denmark72,88 (the study reported by Nemer et al. 72 was conducted in both Germany and Denmark). Of the remaining studies there was one each from Israel,73 the Netherlands,63 Peru,89 Sweden,68 New Zealand,77 Finland98 and the UK. 91,92 Five studies were published in foreign-language journals,45,65,66,74,87 although two of the studies also had results published in English-language journals. 44,75

Most of the studies (n = 4244,45,55–59,61–63,65–67,70,72,74–88,90–93,96–103,105,106) were conducted among primary school-age children (aged 6–12 years), 10 studies61,64–66,73,77,82,87,89,94,95 were conducted among preschool groups (aged 0–5 years) and 1444,45,59,60,68,69,71,72,91,92,96,99–101 were conducted among secondary school-age children (aged 13–18 years) (14 studies spanned multiple age groups). Thirteen57,63,67,76,79,83,86,87,89,90,93,94,98 of the studies were of high quality, 1844,45,64–66,69,71,78,81,82,84,88,96,100–102 were of moderate quality and 2155,56,58–62,70,72–75,77,80,85,95,97,99,104–106 were of low quality (using the EPHPP tool; see Appendix 3). All of the studies included measured primary outcomes and all reported some elements of how the intervention was implemented (see Appendix 5, Tables 43 and 44), particularly in terms of motivation and context.

Universal interventions

Seven57,74–77,86,98,102 experimental studies followed the universal approach, with five57,74,75,77,98,102 evaluating diet and physical activity interventions and the other two76,86 evaluating physical activity-focused interventions.

Five studies,57,76,86,98,102 four of high quality57,76,86,98 and one of moderate quality,102 found no differences in intervention effects by SES. One school-based cardiovascular disease risk factor reduction intervention comprising nutrition and physical activity education and physical activity sessions was effective at reducing skinfold thickness in children aged 8–10 years (compared with usual care);57 one intervention aimed at increasing physical activity through education, extra physical education classes and activity events reduced the rate of increase in BMI in children aged 11–12 years for up to 3 years but this was not maintained at 4 years (compared with usual care);86 one intervention aimed at reducing sedentary behaviour (television viewing and video game use) in children aged 8–9 years reduced a number of obesity-related outcomes (BMI, triceps skinfold thickness, waist circumference and waist-to-hip ratio) (compared with usual care);76 and two smaller (n < 150) studies investigated health-care setting-based obesity treatment programmes in children aged 7–9 years (compared with usual care)98 and 8–12 years (compared with a low-intensity intervention) (Table 6). 102

One study of low quality did, however, find that a school-based obesity prevention intervention was effective at reducing the prevalence of overweight in high-SES children but not in low-SES children (mean age 6.3 years) (compared with no intervention),74,75 and another study of low quality found that another school-based obesity prevention intervention showed a trend (although not significant) towards more favourable intervention effects in higher-SES schools in terms of body fat increases over 2 years in younger children (5 years old at baseline) but not in older children (10 years old at baseline) (compared with no intervention). 77 Both of these studies had large sample sizes (n = 135277 and n = 176474,75) and long follow-up durations (277 and 474,75 years).

Nine44,45,64,88,96,97,100,101,104,105 observational studies also followed a universal approach. One of these studies followed both universal and targeted approaches as the study population was predominantly those of low SES but within this the results were also broken down by an indicator of SES (receiving Medicaid or not). 101 One study of moderate quality that included the evaluation of two levels of intervention found that a basic obesity information provision intervention was effective at reducing BMI of low-SES preschool children (aged 3–4 years) but not of children of higher SES. 64 This study also found that a reinforced intervention of obesity information provision along with a diet and physical activity education programme was effective at reducing BMI of children in both SES groups.

Seven studies44,45,88,96,97,100,101,105 of group- or community-based obesity treatment programmes (five of moderate quality44,45,88,96,100,101 and two of low quality97,105) found that the programmes led to reductions in BMI or percentage overweight and that the intervention effects were the same across different SES groups (Table 7). Each of these studies was conducted in children aged 7–12 years and/or adolescents aged 13–18 years. Jelalian et al. 104 also found that BMI or weight reductions were not associated with SES in a group-based weight control programme in children aged 13–18 years (low-quality study; overall results not reported). However, one low-quality study of a group-based weight loss intervention in a sample aged 8–12 years reduced the per cent overweight overall but larger reductions were observed in higher-SES children. 97

Targeted interventions

Twenty-three58,59,62,63,67–69,71,73,78–80,81,83,84,87–90,93–95,103,106 experimental studies followed a targeted approach (Table 8). Eleven58,59,63,67,73,81,83,89,93,94,103 examined interventions that targeted both diet and physical activity behaviours; 1068,69,71,78,80,84,88,90,95,106 investigated interventions that targeted physical activity and sedentary behaviours only; and two62,79 investigated interventions targeting diet behaviours only. Thirteen55,56,60,61,65,66,70,72,82,85,91,92,99,101 observational studies followed a targeted approach, with 1155,56,60,65,66,70,72,82,85,99,101 investigating interventions targeting diet and physical activity behaviours and two investigating interventions that targeted physical activity behaviour (Table 9).

Meta-analysis of community-level interventions

Effect estimates were pooled for the 11 experimental studies of physical activity/diet interventions for which there were sufficient data in terms of sample size and mean and standard deviation values for both the control group and the intervention group, both before and after the intervention (Table 10). 59,63,67,71,73,84,87,90,93,94,106 Two studies reported effects seperately by group: boys and girls67 and 6–9 years and 10–12 years;63 therefore, two sets of data are included in the analysis for each of these studies. The common outcome was BMI change. A random-effects model (in R statistics package ‘metafor’) was used to incoporate heterogeneity between studies, which may have been a result of differences in the interventions as well as in the samples (e.g. age). The level of heterogeneity means that the results of the meta-analysis should be treated with caution. Using Egger’s test (z = 0.0242, p = 0.9807), there is no indication of publication bias. Mean differences and 95% confidence intervals (CIs) are presented for the pooled BMI data.

Figure 3 shows the resulting forest plot. Only one of the 11 studies59 shows a positive effect for the intervention. The summary meta-analysis suggests that overall the interventions did not significantly reduce BMI among children (random-effects model pooled mean difference estimate of –0.45, 95% CI –1.20 to –0.30). There was evidence of subtantial heterogeneity between studies (I² = 71.74%, p = 0.0047). A sensitivity analysis, which adjusted for moderators (prevention compared with treatment, physical activity intervention compared with physical activity and diet intervention, and study quality), was conducted and this also found that there was no significant intervention effect [described further in Analysis of the robustness of the results (sensitivity analyses)].

FIGURE 3.

Random-effects meta-analysis of BMI change: child community-level intervention studies. a, 6–9 years; b, 10–12 years; c, boys; d, girls.

Overview

No studies of ‘healthy macro policies’ such as restrictions on advertising high-fat foods or agricultural subsidies were located that met the systematic review inclusion criteria (see Figure 1). However, in terms of improving living and working conditions, 10107–121 ‘societal-level’ studies were located that examined multicomponent school environment and education interventions intended to prevent increases in childhood obesity or overweight (summarised in Table 45). All of the studies evaluated interventions in disadvantaged areas (targeted approach). Eight studies examined the combined impact of nutritional education or physical activity alongside changing elements of the school food environment, particularly in terms of introducing nutritional standards for food sold in schools or introducing water fountains. 107–119 One study examined the effects of a school breakfast programme120 and one examined increased access to healthy meals and physical activity. 121

Most studies were conducted in the USA (n = 7107–111,113–116,121), with one each from Germany,117–119 Chile112 and Mexico. 120 There were no UK studies. Two studies112,120 were published in Spanish-language journals and one study117 was published in German, along with two English articles. 118,119 The other seven107–111,114,115 studies were all published in English-language journals. The studies were conducted among children aged from 3 to 17 years, with two studies conducted with children aged 0–5 years,114,121 six conducted with children aged 6–12 years107–111,115–119 and one conducted with children aged 13–18 years. 112 The age of the participants was not reported in the study by Perman et al. 113 but this study included schoolchildren. There were seven107–111,113–119 experimental studies and three112,120,121 observational studies. Seven107–109,112,114,116–119,121 of the studies were of high or moderate quality (using the EPHPP tool; see Appendix 3), with all including independently measured primary outcomes. Most of the studies reported some elements of how the intervention was implemented (see Appendix 5, Table 46), particularly in terms of motivation and delivery fidelity.

Because of heterogeneity in terms of study design and main outcomes, as well as the generally poor quality of data reporting in the studies (e.g. studies seldom reported means and standard deviations), it was possible to conduct meta-analysis for only a small subset of interventions in this category (n = 4 targeted interventions). The 10 studies are, therefore, synthesised narratively. The results are also summarised in Table 45 (see Appendix 5). Effect size data (when possible) are reported in Tables 11 and 12. The meta-analysis of the four107–110 suitable studies is reported separately at the end of this section, with the raw data used provided in Tables 13 and 14.

Targeted interventions

The seven107–111,113–119 experimental studies (five107–109,114,116–119 of moderate quality and two110,111,113,115 of low quality; all using no-intervention control groups) all examined targeted interventions among deprived populations, mostly of children from the 6–12 years age group, with one study carried out with preschool children. The interventions were of promising – albeit only limited and inconsistent – effectiveness as, although most did not reduce the prevalence of overweight and obesity or necessarily prevent new incidence of overweight and obesity (confirmed by the meta-analysis; see Meta-analysis of environmental-level interventions), they did tend to slow down the rate of incidence or weight or BMI gain among poorer children and thus decrease the size of the growth in the SES gap in prevalence (Table 11). In other words, they slowed the epidemic increase in risk of overweight or obesity among the low-income children under study.

Two observational studies112,120 (one120 of low quality and one112 of moderate quality) found no significant intervention effects, one in children from the 6–12 years age group and one in adolescents, whereas another, better-quality study121 found a decrease in obesity prevalence but not overweight in preschool children (Table 12).

Meta-analysis of environmental-level interventions

Effect estimates were pooled for four of the five experimental studies of nutritional interventions for which there were useable data107–110 in relation to two common outcomes: differences in prevalence of overweight and obesity (four studies107–110) and differences in prevalence of obesity (three studies108–110). Random-effects models were used in all cases to incorporate heterogeneity between studies. The heterogeneity between studies may have been a result of differences in the interventions as well as in the samples (e.g. age). Odds ratios (ORs) and 95% CIs are presented for the pooled prevalence data.

Prevalence of overweight and obesity

The four studies107–111 show a range of effectiveness, with the summary meta-analysis suggesting that overall the nutritional interventions did not significantly reduce the prevalence of overweight and obesity among children aged 4–11 years (random-effects model pooled OR estimate of 0.85, 95% CI 0.71 to 1.02) (Figure 4). There was evidence of subtantial heterogeneity between studies (I² = 71.74%, p = 0.0047) and this meta-analysis result should be treated with caution. Using Egger’s test (z = –0.6706, p = 0.5025), there is no indication of publication bias. Raw data for the studies included in the meta-analysis are presented in Table 13.

FIGURE 4.

Random-effects meta-analysis of the prevalence of overweight and obesity: societal-level intervention studies.

TABLE 13 - Raw data for the studies included in the meta-analysis of the prevalence of overweight and obesity: societal-level intervention studies

OR, odds ratio.

Study	Treatment	Overweight, n	Not overweight, n	OR (95% CI)
Coleman et al. 2005107	Intervention	147	253
	Control	152	192	0.73 (0.55 to 0.98)
Foster et al. 2008109	Intervention	204	275
	Control	164	201	0.90 (0.69 to 1.20)
Foster et al. 2010108	Intervention	1057	1250
	Control	1038	1258	1.02 (0.91 to 1.15)
Hollar et al. 2010110,111	Intervention	1207	1825
	Control	350	387	0.73 (0.63 to 0.86)

Prevalence of obesity

Three studies108–110 all found that nutritional interventions did not significantly reduce the prevalence of obesity among children aged 4–11 years and this is reinforced by the summary meta-analysis (random-effects model pooled OR estimate of 0.92, 95% CI 0.79 to 1.06) (Figure 5). There was no evidence of heterogeneity between studies (I² = 0%, Q = 3.46, p = 0.1765). Using Egger’s test (z = 1.1820, p = 0.2372), there was no indication of publication bias. Raw data for the studies included in the meta-analysis are presented in Table 14.

FIGURE 5.

Random-effects meta-analysis of the prevalence of obesity: societal-level intervention studies.

TABLE 14 - Raw data for the studies included in the meta-analysis of the prevalence of obesity: societal-level intervention studies

Study	Treatment	Obese, n	Not obese, n	OR (95% CI)
Foster et al. 2008109	Intervention	134	345
	Control	91	274	1.17 (0.86 to 1.60)
Foster et al. 2010108	Intervention	568	1739
	Control	611	1685	0.90 (0.79 to 1.03)
Hollar et al. 2010110,111	Intervention	667	2365
	Control	187	550	0.83 (0.69 to 1.00)

Overview

Three studies,122–124 although described as community level, contained elements that spanned each of the levels of interventions described in our framework (see Table 1) – individual, community and societal (environmental). Each of these studies investigated obesity prevention interventions that primarily took place in school settings but also involved the wider community through partnership (capacity-building) approaches. Two of the studies were conducted in the USA122,123 and one was conducted in Australia. 124 All of the studies were published in English-language journals. Because of heterogeneity in intervention types, meta-analysis was not conducted and the studies are therefore synthesised narratively in terms of whether they followed a universal (n = 2122,124) or a targeted (n = 1123) approach. The results are also summarised in Tables 47 and 48 (see Appendix 5). The effect size data of the studies (when possible) are reported in Tables 15 and 16. The implementation information is contained in Tables 49 and 50 (see Appendix 5).

Universal approach

Two of the studies, one experimental124 and one observational,122 followed a universal approach and explored differential effects by SES. The high-quality experimental study by Sanigorski et al. 124 found favourable effects (significantly lower increases) for waist circumference and BMI z-score in the intervention group compared with the no-intervention control group although BMI changes were no different (Table 15). There was no association between SES and intervention effects in the intervention schools; however, lower SES was associated with a greater gain in body fat and waist circumference in the control schools. The low-quality observational study by Chomitz et al. 122 also found non-stratified decreases in BMI z-score and prevalence of obesity (Table 16).

Targeted approach

One low-quality observational study by Hoelscher et al. 123 targeted a low-income school population. It observed larger decreases in the number of children who were overweight but not obese in the schools receiving an obesity prevention intervention with community involvement than in the schools receiving the obesity prevention intervention with no community involvement.

Individual-level interventions

Four high-quality experimental studies examined individual interventions, three universal48,50,53 and one targeted. 54 Two of the studies examined tailored weight loss programmes delivered through primary care for children of all SES. 48,50 One study investigated a screen time-reduction intervention that aimed to reduce television and computer use in overweight children of all SES. 53 The final study examined a mentor-based health promotion programme for black adolescents of low SES. 54 One study48 included children of preschool age (0–5 years), two studies50,53 included children of primary school age (6–12 years) and one54 study included children of secondary school age (13–18 years). Three of the studies were conducted in the USA48,53,54 and one50 in Australia. The studies of tailored weight loss programmes found either more beneficial effects in the lower-SES groups or no differential effect by SES after 1 year. The screen time-reduction intervention found beneficial effects in low-SES children but not in high-SES children, both in the short term and long term. The mentor-based health promotion intervention maintained (did not decrease or increase) obesity-related outcomes in all low-SES children and had beneficial effects for those who were overweight and obese in the long term.

Community-level interventions

In total, 13 high-quality experimental studies examined community-level interventions: four universal57,76,86,98 and nine63,67,79,83,87,89,90,93–95 targeted. Eight of the studies examined the effects on obesity-related outcomes of school-based health promotion interventions (three among children of all SES57,76,86 and five among children of low-SES63,67,79,83,94), three evaluated group-based weight loss programmes87,90,98 (one among children of all SES98 and two among low-SES children87,90) and two89,93 evaluated group-based weight gain prevention educational interventions in low-SES children (one targeted parents only89). Half of the studies (n = 657,76,83,90,93,94) were conducted in the USA with four conducted in South American countries (two from Brazil79,87 and one each from Chile67 and Peru89) and three conducted in Europe (the Netherlands,63 Finland98 and France86). The majority of the studies (n = 1057,63,67,76,79,83,86,87,90,93,98) included children of primary school age (6–12 years); three89,94,95 studies included children of preschool age (0–5 years). None of the studies included adolescents (13–18 years). All of the studies included boys and girls (usually an approximately 50/50 mix) with the exception of two studies90,93 that included girls only.

The evidence from school-based health promotion interventions suggests that nutrition and physical activity education combined with exercise sessions may be effective in school-aged children (6–12 years) after reasonably long follow-up times (≥ 6 months) both when targeted at low-SES populations and when delivered universally to children of all SES (and there were no differential effects by SES), but may not be effective in preschool-aged children in the short term. Education-only interventions (diet and/or physical activity) are not so consistently effective in low-SES school-aged children. Screen time-reduction interventions may also be effective in school-aged children (6–12 years) after 6 months, with no differential effects by SES.

The evidence from group-based weight loss programmes suggests that family-based educational and behavioural weight loss programmes may be beneficial in terms of short-term weight loss and long-term weight maintenance and work equally across the social class gradient in school-aged children (aged 6–12 years); and exercise-based weight loss programmes may result in short-term weight loss among low-SES school-aged children.

The evidence from group-based weight gain prevention educational interventions suggests that these interventions do not lead to beneficial effects after a relatively long follow-up period (1 year) in low-SES preschool and primary school-aged children.

Societal-level interventions

The ‘best-available’ evidence for environmental interventions comes from five moderate-quality experimental studies that were all conducted in low-SES schools (targeted approach). 107–109,114,116–119 All of the studies examined multifaceted school-based obesity prevention interventions. Four of the studies were from the USA107–109,114,116 and one was from Germany. 117–119 Four of the studies included children of primary school age (6–12 years)107–109,116–119 and one included preschool children (0–5 years). 114 The evidence from these relatively long-term (> 8 months) studies suggests that multifaceted school-based obesity prevention interventions are effective at reducing or preventing increases in obesity-related outcomes in low-SES primary school-aged children (6–12 years) but may not be effective among low-SES preschool children. No studies investigated macrolevel interventions.

A randomised cluster trial109 examined the effects of a 2-year School Nutrition Policy Initiative (a multifaceted educational and environmental intervention to increase nutritional knowledge and the availability of healthy food) in 844 children aged 11 years in a deprived area of the USA. Sodas, crisps and other high-calorie snacks were no longer sold in vending machines or cafeterias. At the 2-year follow-up, the incidence of overweight was 33% less in the intervention group (OR 0.67, 95% CI 0.47 to 0.96) and the prevalence of overweight was 35% less (adjusted OR 0.65, 95% CI 0.54 to 0.79). The reduction in prevalence was particularly effective for black pupils (adjusted OR 0.59, 95% CI 0.38 to 0.92).

Another randomised cluster trial108 examined the effects of a similar multifaceted educational and environmental intervention in 4603 children aged 11 years from low-SES schools in the USA. The 30-month study found non-significant differences in the prevalence of overweight and obesity in both the intervention group and the control group. However, the mean BMI z-score (p = 0.04) and waist-to-hip ratio were significantly lower in the intervention group (p = 0.04). There were significantly more cases of remission in the intervention group, with overweight or obese pupils at baseline having a 21% lower chance of being obese at follow-up than control group pupils (OR 0.79, 95% CI 0.63 to 0.98).

A non-randomised cluster trial by Williams et al. 114 examined the effects of an 8-month Healthy Start intervention to improve cardiovascular health in 676 low-SES preschool children. At post-intervention follow-up there were non-significant differences between the intervention group and the control group in the prevalence of obesity or overweight or in BMI z-scores.

A non-randomised cluster trial107,116 examined the effects of the 2-year preventative CATCH (Child and Adolescent Trial for Cardiovascular Health) initiative on 744 children aged 8 years from low-SES schools. At the 2-year follow-up, the percentage of children who were overweight and obese increased significantly for both girls and boys in the intervention group and the control group; however, the rate of increase was significantly lower in intervention schools (girls: intervention group increased by 2%, control group increased by 13%; boys: intervention group increased by 1%, control group increased by 9%; p < 0.05).

One randomised cluster trial117–119 examined the effects of a 10-month school-based educational and environmental intervention to increase water consumption in 2950 children aged 8 years from a deprived area in Germany. The intervention entailed the installation of water fountains in schools alongside education. At 10 months’ follow-up, the prevalence of overweight (BMI) was unchanged in the intervention group but increased in the control group. The risk of becoming overweight was significantly reduced in the intervention group compared with the control group (OR 0.69, 95% CI 0.48 to 0.98; p = 0.04).

Individual-, community- and societal-level interventions

The ‘best-available’ evidence for the multilevel individual, community and societal (environmental) interventions comes from one high-quality experimental study124 that examined the effects of a 3-year community capacity-building intervention (Be Active Eat Well) among 1807 children aged 4–12 years in Australia. The intervention was designed by a number of key organisations to build the community’s capacity to create its own solutions to promoting healthy eating, physical activity and a healthy weight and was delivered universally in all intervention schools. After 3 years, children in the intervention schools showed significantly lower increases in waist circumference (–3.14 cm) and BMI z-score (–0.11) than children in the control schools. There was no association between SES measures and intervention effects in the intervention schools; however, lower SES was associated with a greater gain in body fat and waist circumference in the control schools. Therefore, the intervention halted the widening of inequalities in obesity that would normally occur naturally over time.

Individual-level intervention

An uncontrolled prospective cohort (uncontrolled before-and-after) study conducted by Sabin et al. 46 investigated the effects of a primary care educational and behavioural weight loss programme among 61 children aged 2–18 years (approximately 50% girls). The study followed a universal approach in that the service was open to children of all SES and the study explored whether or not SES influenced a child’s level of success. After at least 1 year, 28% of participants achieved the target reduction in BMI [BMI standard deviation score (SDS) reduction of at least 0.5 or obtained normal BMI centiles for age]. There was no significant correlation between SES and BMI SDS reduction nor were there any differences in SES between achievers and non-achievers.

Community-level intervention

An uncontrolled prospective pilot cohort study91,92 explored the effects of a community-based counselling weight loss programme (WATCH IT) among 48 children aged 8–16 years (approximately 50% girls) living in deprived areas in the UK. There was no significant change in BMI SDS at 3 months; however, the BMI SDS was significantly reduced at 6 months’ follow-up (change –0.07; p < 0.01) (the intervention was particularly effective in girls and those aged ≥ 13 years).

Motivation

The majority of the studies clearly described the motivation behind the intervention investigated. The main motivation was to reduce or prevent obesity and/or overweight, or a combination of the two. In some cases this was in a particular population (e.g. low-SES, African American). In some studies the general motivation was the improvement of health and in some studies the focus was on the prevention of disease risk factors, including excessive weight/body fat. Only three studies did not clearly report a motivation behind the intervention. 62,80,110,111,115

Theory

Twenty-nine studies47,48,50,54–56,62,63,68–70,76,78,81,82,86,88,90–92,94,95,98,106,112,117–119,121,122,124 reported a theoretical underpinning of the intervention (or evaluation of the intervention). A number of studies reported using multiple theories, frameworks and/or approaches. The most commonly reported theory informing the interventions was social cognitive theory (reported by 11 studies), followed by social learning theory (three studies), self-efficacy theory (three studies), the theory of planned behaviour (two studies) and community-based participatory approaches (two studies). Other theories or frameworks reported included the chronic care model, the behavioural epidemiology framework, behavioural- and solution-oriented therapy, the social marketing framework, the transtheoretical model of behaviour change, self-care deficit nursing theory, behavioural choice theory, the bottom-up approach to health promotion and concept of empowerment (collective/community control over the design and implementation of interventions) and child quality theory.

Context

Twenty studies55–57,59,60,62,65,66,69,70,78–80,82,84,87,88,110,111,115,120 did not report the context in which the intervention was developed/delivered. For the remaining studies the most commonly reported context was social, usually led by the research team or a health-care or community group. Four studies reported a political context to their study: the intervention investigated by Hawthorne et al. 61 was developed in response to the Child Nutrition and Women, Infants and Children Reauthorisation Act in the USA; the intervention investigated by Hamad et al. 89 was guided by the WHO Integrated Management of Childhood Illness strategy; Ibarra and Alarcón113 evaluated an intervention developed as part of a university social responsibility agenda and inspired by the WHO Healthy Schools Initiative; and Frisvold and Lumeng121 studied the effects of changes to childcare settings that were influenced by national policies: the ‘War on Poverty’ in 1965 and welfare reform in the mid-1990s.

Experience

Fifty-eight of the studies40,42,43,46–52,54–59,62–68,71–75,77–85,87–94,96,98,100,101,103–119,122,124 reported some information regarding the experience of either those who developed the intervention or those who delivered it. In all of these cases interventions were delivered by those with appropriate experience or by those who were trained by others with appropriate experience. In some studies interventions were developed and delivered by multidisciplinary teams. In a number of cases, however, some details of experience were lacking, for example the experience of those delivering the intervention may have been reported but details of who developed the intervention were not given or were unclear, and vice versa.

Consultation/collaboration

Only 20 studies54,62,68,70,72,73,77,81,84,90,94,95,106,107,109,112,113,116,122–124 reported that some degree of consultation/collaboration took place and, within these studies, the level of detail provided varied as well as the level of consultation and/or collaboration. For example, in the study by Perman et al. ,113 a coalition between academic and community partners (including health departments, food retailers and banks) was formed and planning meetings were held with the teachers involved in the intervention, whereas in the study by Moore et al. 70 focus groups with children and teachers informed just a small part of the intervention. The majority of consultations appeared to take place during the planning stages; however, some studies did report extensive consultation at all stages of the research. 122,124

Delivery fidelity

Thirty-five studies40,47–50,55,56,61,62,65–67,71,73,76–78,81–83,86,87,89,90,91,92,95,97,103,106–112,115–119,123 reported details about whether the intervention was delivered as intended or about methods that were put in place to ensure delivery fidelity. This information included data on session attendance, completion rates of intervention components, observation of sessions, quality control audits, staff and researcher records/logs and additional evaluation such as interviews and focus groups with participants and intervention staff members. Methods used to ensure delivery fidelity included standardised training, the use of standardised manuals, practice ‘role-play’ sessions with feedback and regular meetings with trainers/supervisors/more experienced members of the intervention team, and observations of sessions.

Sustainability

Information regarding intervention sustainability was reported for 26 studies. 43,48,50,57,61,63,66,67,69,71–75,86,89,91,92,94,96,101,107,110–113,115–119,121,122 In a number of studies, interventions were integrated into existing health services and school curriculums and were delivered by existing or non-specialised staff. Some studies reported the continuation and expansion of interventions beyond the study. 63,72,107,116,122 However, problems affecting sustainability were reported in some studies66,89,101,121 and reliance on highly motivated staff, volunteers and/or in-kind contributions was reported in others. 61,86,91,92,113 One study relied on the intervention being unsustainable as intervention schools were later assigned as control schools. 74,75,125

Stakeholder support

Twenty studies50,59,61,62,67,70,72,73,81–84,86,89,98,107,112,113,116,123,124 provided information on stakeholder support. Stakeholders included general practitioners (GPs), community groups/individuals, health departments, volunteer groups/individuals, schools, teachers, parents, participants and funders. Stakeholder involvement and feedback were used to indicate stakeholder support but in most cases stakeholder support was implied by the authors without formal evaluation.

Resources

Information on resources was well documented, being reported by 57 studies;40,42,44,46–58,60–64,67–73,77–80,82–84,86–93,96–98,100–104,107,108,113,116–119,121,124 however, the information provided was mostly related to time, staff and equipment rather than the actual costs of the interventions or parts of the interventions, which were reported by just 11 studies. 50,72,77,80,90,93,107,108,113,116–119,121

Differential effects

Subgroup differential effects were also explored by the majority of the studies (n = 4740–42,44,46,48,50–57,61,63–67,73,76–79,86,88,90–92,94,96–98,100,103,106,107,109–112,114–116,121–124). As well as the 23 studies40,41,44,46,48–53,57,64,76,77,88,96–98,100,101,104,105,122,124 reporting differential effects by an indicator of SES (universal approach studies), other differential effects were explored for age, gender, ethnicity, weight status, geographical location (e.g. rural vs. urban), session attendance, parental marital status and parent variables (e.g. BMI, ethnicity).

Sensitivity analysis for the meta-analysis of community-level studies

The heterogeneity between the studies, as shown earlier, may be attributable to study-specific characteristics such as type of study [prevention (P) compared with treatment (T)], quality of the study (high, moderate or low) and intervention type (physical activity only or diet only or physical activity and diet). In the meta-analysis model we accounted for these mediators and the results are presented in Table 19. There is a statistically significant difference between the different types of study (P and T). Additionally, there is a significant difference between intervention types, with physical activity-only studies reporting a higher intervention effect than physical activity and diet studies. There is also a significant effect of quality score. The Q-statistic measure for heterogeneity between the studies is no longer significant (4.2188, p = 0.8369).

Given that the original heterogeneity between the studies is explained by the moderators (study type, intervention type and quality score), it would be desirable to perform the meta-analysis on a homogeneous set of studies. Table 20 shows the distribution of studies by study type, intervention type and quality score. Among all of the possible combinations, only the combination of preventative study, diet and physical activity intervention and high quality includes more than three homogeneous studies.

Figure 6 shows the forest plot for the preventative studies that have a quality score of high and include a diet and physical activity intervention. The individual-level studies do not show significant intervention effects and consequently the overall pooled effect is also insignificant. The Q-statistic (2.1074, I² = 0%, p = 0.8341) measure of heterogeneity is not significant, which implies that these studies are homogeneous. The result is still that there is no significant intervention effect (pooled effect size –0.04, 95% CI –0.29 to 0.20).

FIGURE 6.

Random-effects meta-analysis of BMI change for the high-quality, preventative studies of diet and physical activity interventions. a, 6–9 years; b, 10–12 years; c, boys; d, girls.

Sensitivity analysis for the meta-analysis of societal (environmental) studies

For both outcomes – the prevalence of overweight and obesity and the prevalence of obesity – the low-quality study by Hollar et al. 110,111 was excluded and the random-effects meta-analysis was re-run. The results were as follows. For the prevalence of obesity and overweight the pooled effect size based on the three better-quality studies107–109 was an OR of 0.91 (95% CI 0.75 to 1.1). There was also less heterogeneity (I² = 55.54%, Q = 4.52, p = 0.1040) between the remaining studies. 107–109 For the two better-quality studies of the prevalence of obesity,108,109 the pooled estimate of the OR was 0.99 (95% CI 0.77 to 1.26). As with the full meta-analysis, these sensitivity analyses suggest that, overall, the nutritional interventions did not significantly reduce the prevalence of overweight and obesity or obesity only among children.

Individual-level interventions

In total, we located 11 studies of individual-level interventions. 40–43,46–54 The ‘best-available’ international evidence comes from four48,50,53,54 moderate- or high-quality experimental studies and suggests that tailored weight loss programmes work equally well across the SES gradient and can even have more beneficial effects in the lower-SES groups; screen time-reduction interventions can have beneficial effects in low-SES children but not high-SES children, both in the short term and the long term; and mentor-based health promotion interventions can have beneficial long-term effects among disadvantaged children who are most at risk (overweight and obese). This evidence suggests that interventions of this type may help reduce SES inequalities in obesity. There were no studies that assessed the cost-effectiveness of interventions.

The UK evidence comes from one46 low-quality observational study of a primary care educational and behavioural weight loss programme that found positive results across the SES gradient.

Community-level interventions

In total, we located 5244,45,55–106 studies of community-level interventions. The ‘best-available’ international evidence comes from 1357,63,67,76,79,83,86,87,89,90,93,94,98 high-quality experimental studies which suggest that school-based nutrition and physical activity education combined with exercise sessions can be effective in low-SES school-aged children and when delivered universally to children of all SES groups after reasonably long follow-up times (≥ 6 months), but may not be effective in preschool children in the short term. School-based education-only interventions are not as consistently effective in low-SES children, and school-based screen time-reduction interventions can be equally effective across the SES gradient after 6 months. Family-based educational and behavioural group weight loss programmes can be beneficial in terms of short-term weight loss and long-term weight maintenance and work equally across the social class gradient. Group-based exercise-only weight loss programmes may result in short-term weight loss among low-SES school-aged children. Group-based weight gain prevention educational interventions have no effect in low-SES preschool and school-aged children. There were no studies that assessed the cost-effectiveness of interventions.

The UK evidence comes from one91,92 low-quality observational study of a community-based counselling weight loss programme, which found no effect initially but BMI reductions in low-SES children in the longer term (6 months).

Societal-level interventions

In total, we located 10107–121 studies of societal (environmental)-level interventions but no studies of societal (macro)-level interventions. The ‘best-available’ international evidence for the environmental interventions comes from five107–109,114,116–119 moderate-quality experimental studies and suggests that multifaceted school-based obesity prevention interventions are effective at reducing or preventing increases in obesity-related outcomes in low-SES children aged 6–12 years but may not be effective among low-SES preschool children.

There were no UK studies of societal-level interventions.

Individual-, community- and societal-level interventions

In total, we located three122–124 studies of multilevel interventions that spanned each of the individual, community and societal levels described in our framework. The ‘best-available’ international evidence comes from one124 high-quality experimental study, which found that a community capacity-building intervention halted the widening of inequalities in obesity that was observed in the control community.

There were no UK studies of multilevel interventions.

Table 21 illustrates where the included child studies fit within our framework described in Table 1.

What works to reduce inequalities in obesity, for whom and where?

The ‘best-available’ international evidence for the effectiveness of individual-level interventions (n = 4)48,50,53,54 in reducing inequalities in obesity among children suggests that tailored weight loss programmes may have more beneficial effects on obesity-related outcomes in lower-SES groups than in higher-SES groups in the longer term (1 year); that a mentor-based health promotion intervention also seemed effective in preventing an increase in obesity-related outcomes in SES children (but not reducing them) and that this particularly benefited those who were overweight and obese (1 year); and that, most notably, a screen time-reduction intervention (aimed at reducing television viewing and computer use in 67 children aged 4–7 years in the USA) found beneficial effects for up to 2 years in low-SES children but not high-SES children.

Similarly, the ‘best-available’ international evidence for the effectiveness of community-level interventions (n = 13)57,63,67,76,79,83,86,87,89,90,93,94,98 included evidence of some effective interventions in the short and long term among school-aged children (aged 6–12) but not among adolescents (no studies) or preschool children (not effective in the short or long term). There was evidence of longer-term effectiveness (> 6 months) in reducing obesity-related outcomes among school-aged children (6–12 years) of (1) school-based nutritional and physical activity education and exercise sessions and (2) school-delivered screen time-reduction interventions, with no differential effects by SES. There was evidence of short-term effectiveness (up to 6 months) in reducing obesity-related outcomes among school-aged children (6–12 years) of targeted (1) family-based educational and behavioural weight loss programmes and (2) exercise-based weight loss programmes.

We found only five107–109,114,116–119 moderate-quality experimental studies of the effects of more upstream environmental interventions and no studies of the effects of macro-level policy interventions on obesity-related outcomes among children. All of the studies examined multifaceted school-based obesity prevention interventions. The evidence suggests that multifaceted school-based obesity prevention interventions (educational and environmental interventions typically including nutritional education and/or physical activity and an environmental modification component such as increasing the availability of healthy food) are effective in the medium to longer term (> 8 months) at reducing or preventing increases in obesity-related outcomes in low-SES school-aged children (6–12 years) but may not be effective among low-SES preschool children. Similarly, a multilevel community capacity-building intervention was effective in preventing a widening of inequalities in obesity over the long term (up to 3 years) among schoolchildren aged 4–12 years.

There were just two46,91,92 studies conducted in the UK, both observational in design, with small sample sizes (< 75) and of low methodological quality. There was no evidence of effectiveness in terms of reducing inequalities in obesity-related outcomes of an individual-level primary care educational and behavioural weight loss programme for children aged 2–18 years. However, a community-based counselling weight loss programme targeted at children aged 8–16 years in a deprived area found reductions in BMI after 6 months and the intervention was particularly effective in girls and those aged ≥ 13 years.

It is important to reflect on ‘for whom’ and ‘where’ the interventions were – or more usually were not – effective. The ‘best-available’ international evidence was typically for interventions conducted in the USA or South America. However, caution should be applied in trying to extrapolate the effectiveness of the various individual-, community- and societal-level interventions beyond these countries. This is especially the case as the UK evidence base was extremely small (n = 2) and so provides little insight into how such interventions could work in the UK. However, the ‘best-available’ international evidence does suggest that interventions are universally much more effective among school-aged children (aged 6–12 years) than among preschool children. There were no studies of adolescents. This is such a consistent finding that it may also be applicable to the UK context and suggests that it is at the primary-school level that we need to intervene. Interventions appeared to be equally effective for boys and girls, although some studies did not distinguish their results by gender. Most of the studies were of interventions targeted at low-SES children/areas and often of ‘treatment’ interventions for those already overweight/obese. In terms of ‘where’ interventions appeared to be effective, the ‘best-available’ evidence was dominated by school-delivered interventions, which suggested that targeted school-based interventions appear, in general, to be effective. This supports the ‘whole school approach’ to tackling childhood obesity. The findings of effectiveness are therefore very much limited to the effectiveness of school-based interventions for low-income, primary school-aged children (6–12 years), particularly in the USA.

In terms of barriers to and facilitators of interventions, although most of the studies provided data for motivation, context and experience of the intervention team and resources, the type and level of information provided varied substantially for each of the domains, making comparisons between the studies difficult. There were no apparent differences between interventions that were successful in reducing inequalities in obesity and those that were not. For example, in terms of study motivation, it may be hypothesised that studies that focused primarily on reducing obesity would be more successful at reducing inequalities in obesity than those that aimed to improve health in general; however, both successful and unsuccessful studies reported both motivations. There appeared to be no differences in the experience of the intervention team between successful and unsuccessful interventions (e.g. trained or professional facilitators were reported for both). Additionally, the reporting of resources (incentives, supportive materials, contact time and training of facilitators) did not appear to be related to outcomes.

Implications for research

The evidence suggests that the direction of research and evaluation in this field would benefit from looking into how to implement effectively to scale interventions to manage childhood obesity, sustain the impacts over time and ensure equitable outcomes. We recommend larger, longer-term studies, powered to detect the small changes that are likely to be found and including an assessment of equity impacts, to enable translation of research findings into effective public health approaches for managing childhood obesity.

The nature of the evidence base has a number of implications for public health researchers. Most notably, although we found a very large international evidence base, the evidence found was largely observational and of moderate to low quality. It is worth noting that, for the same type of intervention, observational studies are more likely to show positive effects than experimental studies. It is reasonable to suggest, therefore, that the most useful information on the way in which obesity (preventative or treatment) interventions impact on health inequalities comes from moderate- to high-quality experimental studies of universal interventions. These were particularly lacking in the UK evidence base and in methodological terms the UK evidence did not compare well with studies from the USA. There were also very few studies of societal-level interventions, which might be expected to have more of an impact on the gradient in obesity. 29 We did search for reports of observational studies of societal interventions that we are aware of, and which might have met our inclusion criteria, for example EPODE (Ensemble Prévenons l’Obésité Des Enfants; see www.epode-international-network.com/; accessed 16 September 2014), Sure Start (see www.gov.uk/find-sure-start-childrens-centre; accessed 16 September 2014) and Healthy Towns (www.theguardian.com/society/2008/nov/10/obesity-healthy-towns1; accessed 16 September 2014); however, we were not able to find any relevant evidence.

The majority of interventions that we included in this review took a targeted approach to tackling obesity and were concerned with weight loss (‘treating’ existing obesity) rather than preventing weight gain (‘preventing’ obesity). These ‘treatment’ interventions are more likely to show positive effects than prevention interventions. 31 The targeted approach also has limitations because even when interventions are effective among low-income groups they are only able to reduce the health inequalities gap; they have little effect on the wider social gradient. Most studies were school based and aimed at primary school-aged children. We also found no studies that assessed the cost-effectiveness of interventions and meta-analysis could be conducted only on a minority of studies given their heterogeneity.

Our results suggest a need for more experimental studies of the effectiveness and cost-effectiveness of interventions to reduce inequalities in childhood obesity, particularly in (1) adolescents, (2) in the UK and (3) in terms of macro-level interventions that potentially address the entire gradient. There has been a real missed opportunity to evaluate the effects of such ‘real-world’ interventions, and future interventions (such as Fulfilling Lives: A Better Start, a Big Lottery-funded programme in Newcastle) could benefit from including such analysis.

Implications for public health

Our review has found a large international evidence base but only limited effectiveness of interventions with the potential to reduce SES inequalities in obesity. The body of evidence in this review provides some support for the hypothesis that obesity management interventions in children can be effective and that they do not increase health inequalities. Interventions need to be developed that can be embedded into ongoing practice and operating systems, rather than implementing interventions that are resource intensive and cannot be maintained long term. This review also highlights that, although we may now have a good understanding of the range of interventions that are feasible for use in reducing the risk of childhood obesity, we lack the knowledge of which specific intervention components are most effective to ensure that the equity gradient is reduced. Being able to answer this question is of critical importance to decision-makers.

The review provides evidence of significant positive outcomes for the more disadvantaged. There was no evidence of a widening of health inequalities as a result of obesity management interventions. In addition, the relatively large number of studies of interventions targeting disadvantaged population groups provides useful information about the implementation strategies needed for obesity prevention efforts targeting these high-risk groups. We advocate for an assessment of outcomes by measures of equity, such as those indicated by PROGRESS (Place of residence, Race/ethnicity, Occupation, Gender, Religion, Education, Socioeconomic status and Social capital), if a general population is targeted.

In relation to which interventions could now be implemented by the UK public health community, the findings of this review are mainly limited to non-UK evidence and we cannot assume that such interventions will be effective outside their country context. It is also difficult to distinguish which specific components of intervention programmes are necessary to achieve beneficial impacts on obesity in children across all SES groups. However, our review has found tentative evidence of some interventions with the potential to reduce SES inequalities in obesity. Most notably, school-delivered educational and combined educational and environmental interventions that are targeted at low-SES primary school-aged children appear to have some effectiveness in the long term in reducing obesity-related outcomes among such children. The evidence suggests that interventions of this type may therefore be worth commissioning in the UK by clinical commissioning groups or local authorities who wish to target services at low-income primary school children or children in deprived areas. However, these interventions could benefit from being piloted first and thoroughly evaluated using an experimental design.

Strengths and limitations

This review was very extensive as an extremely thorough search was conducted of the international literature, using very broad intervention inclusion and exclusion criteria, which has ensured that the entire relevant experimental and observational evidence base has been captured. However, we located few evaluations of societal-level interventions and this was probably because we did not include non-experimental study designs. The quality of the review is also high as double screening was applied and both data extraction and quality appraisal were independently checked. We also examined the implementation of the interventions and paid attention to the context within which interventions were carried out. However, the review is still subject to some methodological limitations as, for example, the quality assessment tool, although described as a tool for public health interventions, seemed to favour those that followed a more clinical model. We particularly found the blinding question unhelpful as it mostly resulted in moderate scores. The implementation tool was practical but enabled only a brief summary of implementation factors to be provided. The theoretical framework adapted from the health inequalities literature meant that most interventions were categorised as community-level interventions and we encountered difficulties in determining in which section of the framework particular interventions should sit. One final limitation that may be of particular relevance to the non-UK evidence base is our exclusion of studies that examined ethnic inequalities, which may have reduced the number of US studies identified, in which ethnicity is often used as a proxy for SES.

Overview

Thirty-three43,127–150,152,154–158 individual-level studies were identified that met the review inclusion criteria. Because of heterogeneity in terms of study design and main outcomes, it was possible to conduct meta-analysis for only a small subset of studies in this category (n = 4133,135,136,152 for weight change and n = 4132,135,149,152 for BMI change, all targeted interventions). The 33 studies are therefore synthesised narratively in terms of whether they followed a universal (n = 13127,129–131,138–146) or a targeted (n = 2043,128,132,134–137,140,141,143–148^-150,152,154–158) approach. The results are also summarised in Table 51 (universal-approach studies) and Table 52 (targeted-approach studies) (see Appendix 5). Effect size data (when possible) are presented in Tables 22–25 and the meta-analysis is reported separately at the end of this section, with the raw data included reported in Tables 26 and 27. Implementation information for each study is provided in Tables 53 and 54 (see Appendix 5).

Most of the studies (n = 2143,127–131,134,137,139,140,143–146,152,154,156–158) were prevention-type studies (they included participants regardless of their weight status). The other 12 studies132,133,135,136,138,141,142,147–150,155 were treatment studies in participants who were already overweight or obese. The greatest number of studies were conducted in health-care settings or study/university settings (n = 1943,131,133,134,136–138,141,143,144,147,150,152,155,157,158), with five129,130,135,148,149 studies conducted in participants’ homes (or in the home environment, e.g. shopping vouchers for fruit and vegetables), two128,142 delivered remotely by telephone or through the internet and seven127,132,139,140,145,146 being mass media or population-wide campaigns or prevention activities. Thirteen127,129–134,136,138,145,150,154,155 of the studies were conducted in the USA, with two127 of those also being conducted in Sweden. In addition, seven135,141,143,147,152 studies were conducted in the UK, three each in Australia,139,142,148,149 Israel137,157,158 and the Netherlands140,156 and one each in Finland,144 Germany,146 Spain128 and New Zealand. 43 One study was published in a foreign-language journal. 128 Seven129–136 studies targeted women only, with two133,136 targeting only African American women, two131,134 targeting pregnant women and one135 targeting mothers who were 6–18 weeks post partum. One137 study targeted men only and one128 recruited participants from industrial workplaces, resulting in a predominantly male sample.

The numbers of participants in the studies varied considerably, between n = 9 and n = 14,078 (median sample size 687), and the median follow-up time was 12 months (range 3–72 months). Eight of the studies were of high methodological quality (using the EPHPP tool; see Appendix 3), six were of moderate quality and the remaining studies were of low quality. All of the studies reported some elements of how the intervention was implemented (see Appendix 5, Tables 53 and 54), with 21 studies scoring ≥ 6.

Universal interventions

Three129,130,138,139 experimental and 10127,131,140–146 observational studies followed a universal approach (all diet and physical activity studies). One study145 was of moderate quality and three129–131,140 were of high quality, with the remaining studies of low quality. All of the experimental studies found no differential effects by SES. One found that a financial incentive weight loss programme (compared with a non-financial incentive weight loss programme) led to weight loss (low quality);138 one observed no intervention or differential effects of a weight prevention programme (no contact control; high quality);129,130 and one found that, although in the final (low-quality) study weight increased after a nurse-delivered counselling cardiovascular disease risk prevention intervention (usual-care control), there were no differential effects by occupation139 (Table 22).

One high-quality observational study found that an intervention to prevent excess gestational weight gain had more beneficial effects among low-income women than among high-income women131 (Table 23). A high-quality study observed beneficial intervention effects (BMI and waist circumference) after a population-wide/mass media cardiovascular disease prevention programme in both moderate to high and low-SES groups. 140 Three studies (all low quality) observed beneficial effects of weight-management programmes141,142 and a health trainer-led prevention programme,143 with no differential effects by SES. One low-quality study found beneficial effects of a national diabetes prevention programme in terms of weight, BMI and waist circumference reductions, which were similar across all SES groups144 (see Table 23). Two further studies of a mass media campaign (moderate quality)145 and a population-wide cardiovascular disease prevention programme (low quality)146 also found no differential effects by SES; however, no beneficial changes were observed in either study. One population-based health promotion intervention was evaluated using two different study designs (therefore this counted as two studies; both low quality) in two different countries (Sweden and the USA). 127 The evaluation using a controlled prospective cohort design found no differential effects by education but also no decreases in BMI in both countries, whereas the evaluation using serial cross-sectional surveys observed adverse effects (increases in BMI) both overall and in those with a low level of education in Sweden, but no intervention or differential effects in the USA or when data from the two countries were pooled.

Targeted interventions

Nine128,132–136,148–150,152 experimental and eleven43,137,140,141,143,147,154–158 observational studies followed a targeted approach (interventions targeted at low-SES adults or low-SES areas) (Tables 24 and 25 respectively). Four experimental studies (two of strong quality,133,135 one of moderate quality132 and one of weak quality148,149) observed beneficial intervention effects on obesity-related outcomes. Three of these studies were weight-management treatment programmes with diet and physical activity components133,135,136,147 (two usual-care controls133,136,147 and one lower-intensity intervention control135) and one was a physical activity and nutrition programme for sedentary older adults (no-intervention control). 148,149 A high-quality pilot study (using an enhanced standard care control) found no beneficial effects of a diabetes prevention programme for overweight adults, although there was a trend towards a reduction in body weight. 150 One high-quality study found a beneficial intervention effect of a telephone-based counselling intervention (participants were recruited from a manufacturing company but the intervention was home based) in terms of weight loss and BMI. 128 Two studies (both of moderate quality) found no beneficial effects of an intervention to prevent excessive gestational weight gain (standard care control)134,151 or a diet-only culturally appropriate fruit and vegetable promotion programme (with an active control; fruit and vegetable gift card only vs. gift card plus health education). 132 One moderate-quality study even observed an adverse effect on waist circumference of a lifestyle helper intervention using behavioural counselling (vs. a lifestyle helper intervention without behavioural counselling), although there was no effect on BMI. 152

Of the observational studies, three140,143,147 diet and physical activity interventions (two of weak quality143,147 and one of high quality140) led to favourable intervention effects in terms of weight loss and BMI: one weight-management treatment programme,147 one health trainer-led prevention programme143 and one population-wide/mass media cardiovascular disease prevention programme. 140 One physical activity-only intervention investigating a dog-walking programme also improved obesity-related outcomes (moderate-quality study). 154 A clinic-based nurse health promotion (diet and physical activity) intervention for overweight participants saw beneficial effects in those who were classified as full adherers, but this effect was lost when all participants were included in the analysis (low-quality study). 155 Marshall et al. 43 did not find any beneficial effects in terms of BMI or waist circumference in those attending a nurse-led healthy lifestyle clinic (low-quality study) and Verheijden et al. 156 saw no changes in BMI following a mass media campaign except for a small decrease in non-Dutch people following a second campaign (low-quality study).

Three studies,137,157,158 using different populations, evaluated the Community syndrome of Hypertension, Atherosclerosis and Diabetes (CHAD) programme in Israel. The preliminary study (moderate quality) in a sample of men found no beneficial effects in terms of body weight or prevalence of overweight. 137 The evaluation of the first 5 years of the programme observed reductions in the prevalence of overweight overall and BMI in women (high-quality study);157 however, these beneficial effects were not observed in the evaluation of the second 5 years of the programme, although it is worth noting that a different study design was used (moderate-quality study). 158

Meta-analysis of targeted individual-level interventions

Effect estimates were pooled for the five132,133,135,136,149,152 experimental studies of targeted physical activity/diet interventions for which there were sufficient data in terms of sample size, means and standard deviations for both the control group and the intervention group, both at baseline and at follow-up. Common outcomes identified for meta-analysis were weight change (n = 4133,135,136,152) and BMI change (n = 4132,135,149,152).

Random-effects models were used in all cases as the test for heterogeneity was significant (p < 0.20) and/or the I² statistic was moderate or high (> 50%). This heterogeneity may have been a result of differences in the interventions as well as in the samples (e.g. age). This level of heterogeneity means that the results of the meta-analysis should be treated with caution.

Individual-level interventions: weight change

Results for effects of interventions on weight were pooled for the experimental studies with usable data (n = 4133,135,136,152) and, using Egger’s test (–0.3452, p = 0.7300), there was no indication of publication bias. In Figure 9 the random-effects model shows a signficant pooled effect in favour of the intervention (mean difference –1.52, 95% CI –2.53 to –0.52). There was no evidence of heterogeneity (I² = 38.85%, p = 0.2565) between the studies. Table 26 shows the raw data included in the meta-analysis.

FIGURE 9.

Random-effects meta-analysis of weight change: adult individual-level studies.

Individual-level interventions: body mass index change

Results for BMI were pooled for the experimental studies with usable data (n = 4132,135,149,152). Using Egger’s test (0.2071, p = 0.8360) there was no indication of publication bias. In Figure 10 the random-effects model shows a significant, albeit small, pooled effect in favour of the intervention (mean difference –0.85, 95% CI –1.72 to 0.02). There was evidence of substantial heterogeneity (I² = 65.87%, p=0.0583) between the studies. Table 27 shows the raw data included in the meta-analysis.

FIGURE 10.

Random-effects meta-analysis of BMI change: adult individual-level studies.

Overview

Sixty community studies (from 62 papers)65,66,103,153,159–176,178–200,202–214,225,230,231 were identified that met the review inclusion criteria. Because of heterogeneity in terms of study design and main outcomes, it was possible to conduct meta-analysis only for a small subset of interventions in this category (n = 1365,103,168,179,182,183,198,211,212,230 for weight, n = 965,66,179,182,198,209,212,230 for BMI change and n = 365,66,179 for waist circumference change; all targeted interventions). The 60 studies are therefore synthesised narratively in terms of whether they followed a universal approach (n = 16161–163,172,173,175,186–196) or a targeted approach (n = 4465,66,103,153,159,160,164–171,174,176,178–185,197–200,202–214,225,230,231). The results are also summarised in Table 55 (universal-approach studies) and Table 56 (targeted-approach studies) (see Appendix 5). Effect size data (when possible) are presented in Tables 28–30. The meta-analysis of the 25 outcomes is reported separately at the end of this section along with the raw data in Tables 31–33. Implementation information for each study is provided in Tables 57 and 58 (see Appendix 5).

Twenty-seven65,66,103,161–163,167,171,172,174,180,181,185,198,190,194,197,199,201,202,205,206,209–211,213,214,231 of the studies were aimed at the prevention of obesity and 32153,159,160,164,165,168–170,173,175,176,178,179,182–184,186–189,191–193,195,196,200,201,203,204,207,208,212,225,230 were aimed at the treatment of obesity (these studies included participants who were overweight and/or obese only). One study166 used both prevention and treatment strategies. Twenty-five164,166–169,176,178,181–184,190,196,198,203,204,207,209,210,212–214,225,231 studies were conducted in community settings (e.g. community centres, churches and schools), 1765,66,159,162,163,170–174,187,191,193,194,199,200,205,206 were workplace interventions, seven153,180,185,192,195,211,230 were conducted in health-care settings, two165,188,202,208 were conducted in both health-care and community settings, two160,161 investigated health insurance training courses, one179 was conducted in a university, one189 investigated a diet club and one201 a weight loss training camp. The four103,175,186,197 remaining studies investigated group-based interventions but their settings were unclear.

The majority of the studies were conducted in the USA (n = 42164–172,174–176,178,179,181–186,190,191,193,195,196,198–200,202,203,205,206,210,213,214,230,231), one of which202 was conducted at the USA/Mexico border. Four studies153,189,192,208 were conducted in the UK. Three studies159–161 were carried out in Germany and two each in Australia,194,211 Brazil162,163,212 and Chile. 65,66 There was one study from each of the following countries: Denmark,173 Israel,188 Korea,197 Mexico201 and Turkey. 204 Five of the studies were published in foreign-language journals: three in German,159–161 one in Spanish (Chile)65 and one in Portuguese (Brazil). 162,163

In the majority of studies, more women participated than men. Twenty-three164–169,174,176,178–180,182–184,188,197,199,200,203,212–214,225,230 studies were exclusively targeted at women, with six targeting mothers in particular. 164–169 Six studies were conducted in manufacturing companies;159,170–174 with the exception of the study by Grandjean et al. ,174 which targeted blue-collar female employees, the participants in these studies were predominantly men. Two studies targeted men only. 153,175 Six studies explicitly targeted African Americans (five176,179,180,200,230 of which also included only women), two studies recruited Mexican American women only,182–184 one study recruited Filipinos only (predominantly women)185 and one study recruited Latino women only. 167

The study populations varied considerably, with between 19 and 10,368 individuals included (median sample size 99) and a median follow-up time of 6.5 months (range 3–48 months). Thirteen of the studies were of high methodological quality,167,171,176,183,186,189,197–200,210,211,225,230 21 were of moderate quality103,160,168,174,175,179,182,184,185,187,190–192,200,202–204,207,209,212 and the remaining 26 were of low quality65,66,153,159,161–166,169,170,172,173,178,180,188,193,195,196,201,205,206,208,213,214 (using the EPHPP tool; see Appendix 3). All of the studies reported some elements of how the intervention was implemented (see Appendix 5, Tables 57 and 58), with 43 studies scoring ≥ 6.

Universal studies

Four experimental studies (two diet and physical activity studies,186,187 one diet-only study189 and one physical activity-only study188) followed the universal approach (Table 28). In one of the diet and physical activity studies (high quality), no relationship was found between the intervention effect of a weight loss programme based on behavioural education and social support (vs. a behavioural weight loss programme without social support) and the education status of participants, although the intervention was ineffective overall for weight maintenance. 186 In the other diet and physical activity study (moderate quality), a workplace telephone- and internet-based weight loss intervention was effective at reducing body weight and waist circumference compared with an information provision-only control group and the intervention effect was not affected by the participants’ level of education. 187 A low-quality study investigating the effectiveness of adding an exercise session component to a weight loss programme found that both the intervention group and the active control group (weight loss programme without an exercise component) lost weight to the same extent and that there were no differences in weight loss between employed and unemployed women in the intervention group (although employed women did lose more weight than unemployed women in the control group). 188 In the diet-only experimental study two types of diet (low-carbohydrate vs. low-fat diet) were compared in a diet club in the UK for the treatment of overweight (high-quality study). 189 The low-carbohydrate diet appeared to be more beneficial in the short term (3 months), especially in those of low SES, but this effect was lost after 1 year. Neither diet was effective overall in either social group after 1 year.

Twelve observational studies (10 nutrition and physical activity studies,161,172,173,175,190–193,195,196 one diet-only study162,163 and one physical activity-only study194) followed the universal approach. Four moderate-quality studies175,190–192 and three low-quality studies161,173,193 found that the interventions investigated [four weight-management programmes, two health promotion (prevention) interventions and one workplace weight loss competition] led to reductions in BMI or weight overall and that SES indicators were not associated with these intervention effects. Another low-quality study investigating a workplace-based telephone coaching health promotion (prevention) intervention observed reductions in body weight that did not differ across income groups;172 there was even a tendency for more weight loss in those who were less educated (although this was not statistically significant). The moderate-quality physical activity study (a pedometer-based workplace health promotion intervention) observed reductions in waist circumference overall and no differential intervention effects by level of education. 194 A low-quality study found no differences in education level between those who were successful in a group-based weight loss programme and those who were unsuccessful,195 and another low-quality study found no association between employment status and weight gain during the maintenance period following a commercial weight loss programme. 196 Although these studies do not show any strong evidence of reducing SES inequalities in obesity, they do show promising results for the prevention of such inequalities. However, one low-quality study showed that the Workers’ Food Program in Brazil had adverse effects (weight gain, increase in overweight rates) in workers of low SES but not in those of high SES,162,163 suggesting that this intervention may contribute to the development of SES inequalities in obesity.

Targeted interventions

Twenty-four65,66,103,166–171,174,176,179,180,182–184,197–200,209–212,225,230 experimental studies followed the targeted approach (Table 29). Sixteen65,66,103,166–171,179,182,184,197–200,230 of these studies examined interventions with diet and physical activity components, five176,183,209–211,225 examined interventions with a diet component only and three174,180,212 were physical activity-only interventions. Twenty153,159,164,165,178,181,185,200–208,213,214,231 observational studies followed the targeted approach (Table 30). Sixteen153,159,164,178,185,200–208 of these studies targeted both diet and physical activity, one160 contained a diet component only and three181,213,214,231 were physical activity-only interventions.

Meta-analysis of targeted community-level interventions

Effect estimates were pooled for the targeted experimental studies for which there were sufficient data in terms of sample size, means and standard deviations for both the control group and the intervention group, both at baseline and at follow-up. Meta-analysis was conducted for a small subset of intervention outcomes (n = 1365,102,168,179,182,183,198,211,212,230 for weight, n = 965,66,179,182,198,209,212,230 for BMI and n = 365,66,179 for waist circumference).

Random-effects models were used in all cases as the test for heterogeneity was significant (p < 0.20) and/or the I² statistic was moderate or high (> 50%). This heterogeneity may have been a result of differences in the interventions as well as in the samples (e.g. age). This level of heterogeneity means that the results of the meta-analysis should be treated with caution.

Community-level interventions: weight change

Results were pooled for all of the experimental community-level studies with usable data (n = 1365,103,168,179,182,183,198,211,212,230 outcomes). Using Egger’s test (–0.4645, p = 0.6423) there is no indication of publication bias. In Figure 11 the random-effects model shows no significant pooled effect in favour of the intervention (mean difference –0.73, 95% CI –1.56 to 0.11). There was no evidence for significant heterogeneity (I² = 20.40%, p = 0.5420) between the studies. The raw data included in the meta-analysis are shown in Table 31.

FIGURE 11.

Random-effects meta-analysis of weight change: adult community-level studies. a, individual intervention group vs. control group; b, family intervention group vs. control group; c, white/Anglo women; d, African American women; e, Hispanic women.

TABLE 31 - Raw data included in the meta-analysis of weight change: adult community-level studies

SD, standard deviation.

Individual group vs. control.

Family group vs. control.

White/Anglo women.

African American women.

Hispanic women.

Study	Intervention			Control
	n	Mean	SD	n	Mean	SD
Janicke et al. 2011103	22	1.6	26.01	11	–0.6	18.56
Nichols 1995179	20	–2.39	16.03	17	0.66	14.67
Alves et al. 2009212	78	–1.3	8	78	0.4	11.15
Cousins et al. 1992182a	32	–2.1	14.05	27	–0.7	11.66
Cousins et al. 1992182b	27	–3.8	13.72	27	–0.7	11.66
Kain et al. 200965	38	–1.5	9.08	19	1.9	10.45
Ockene et al. 2012198	147	–1.13	2.78	142	0.28	3.79
Faucher and Mobley 2010183	7	–2.94	3.4	5	–1.27	4.62
Walker et al. 2012168c	8	–2.59	6.21	8	–1.17	1.81
Walker et al. 2012168d	9	1.5	2.85	11	–0.09	2.81
Walker et al. 2012168e	5	–1	1.85	9	–0.09	2.18
Befort et al. 2008230	14	–2.6	4.2	19	–3.2	5.7
Reid et al. 1995211	76	–0.4	4.44	73	–0.5	3.05

Community-level interventions: body mass index change

Results were pooled for all of the experimental community-level studies with usable data (n = 965,66,179,182,198,209,212,230 outcomes). Using Egger’s test (–0.7140, p = 0.4752) there is no indication of publication bias. In Figure 12 the random-effects model shows a significant pooled effect in favour of the intervention (mean difference –0.31, 95% CI –0.57 to –0.06). There was no evidence of substantial heterogeneity (I² = 21.73%, p = 0.7227) between the studies. The raw data included in the meta-analysis are shown in Table 32.

FIGURE 12.

Random-effects meta-analysis of BMI change: adult community-level studies. a, Individual group vs. control; b, family group vs. control.

TABLE 32 - Raw data included in the meta-analysis of BMI change: adult community-level studies

SD, standard deviation.

Individual group vs. control.

Family group vs. control.

Study	Intervention			Control
	n	Mean	SD	n	Mean	SD
Nichols 1995179	20	–0.95	8.4	17	0.32	9.12
Befort et al. 2008230	14	–1	1.5	19	–1.1	2
Cullen et al. 2009209	318	–0.3	0.42	240	–0.1	0.56
Alves et al. 2009212	78	–0.6	4.67	78	0.1	4.87
Ockene et al. 2012198	147	–0.4	0.92	142	0.11	1.64
Cousins et al. 1992182a	32	–0.8	7.28	27	–0.3	7.29
Cousins et al. 1992182b	27	–1.6	7.1	27	–0.3	7.29
Kain et al. 200965	38	–0.4	5.8	19	1	5.51
Kain et al. 201066	28	–0.3	5.23	19	–0.4	5.51

Community-level interventions: waist circumference change

Results were pooled for all of the experimental community-level studies with usable data (n = 365,66,179). Using Egger’s test (0.3589, p = 0.7197) there is no indication of publication bias. In Figure 13, the random-effects model shows a significant pooled effect in favour of the intervention (mean difference –4.26, 955 CI –4.83 to –3.69). Table 33 shows the raw data included in the meta-analysis.

FIGURE 13.

Random-effects meta-analysis of waist circumference change: adult community-level studies. a, Individual group vs. control; b, family group vs. control.

TABLE 33 - Raw data included in the meta-analysis of waist circumference change: adult community-level studies

SD, standard deviation.

Study	Intervention			Control
	n	Mean	SD	n	Mean	SD
Kain et al. 200965	38	–2	11.7	19	2.2	12.09
Kain et al. 201066	28	0.2	13.06	19	2.4	15.13
Nichols 1995179	20	–3.37	0.89	17	0.88	0.889

Overview

Eight215–222 environmental studies were identified that met the review inclusion criteria. All studies were aimed at prevention and combined some environmental-level changes with individual- and community-level interventions. Three studies conducted community-wide environmental interventions,215–217,222 one study incorporated both a community-wide and a workplace intervention218 and four studies conducted workplace interventions. 219–221 The results are summarised in Table 59 (universal-approach studies) and Table 60 (targeted-approach studies) (see Appendix 5), with implementation information provided in Tables 61 and 62 (see Appendix 5).

Most studies were conducted in the USA (n = 5219–222), with two215–217 from Norway and one218 from the UK. All of the studies were published in English and one was also published in Norwegian. 217 One of the studies was an experimental study219 and seven were observational studies. 215–218,222 Six of the studies were of moderate quality215–217,219,222 (using the EPHPP tool; see Appendix 3), with all using independently measured primary outcomes. Half of the studies215–219,221 reported most elements of how the intervention was implemented (see Appendix 5, Tables 61 and 62; high scores between 6 and 9), particularly in terms of motivation and delivery fidelity, but the other half218,220,222 reported only a few elements of implementation (scoring between 4 and 5). Two of the studies scored low on the quality appraisal because of the study design, having no control and using self-reported health measures. Because of heterogeneity in terms of study design and main outcomes, we could not conduct a meta-analysis of the eight environmental studies.

Universal interventions

The experimental study by Lemon et al. 219 (moderate quality; no-intervention control), which examined a universal intervention among a working-age population (aged 18–69 years) that combined environmental components (modification to stairways and canteens through the use of signs and improved street lighting and gritting) with other non-environmental components (including a social marketing campaign, farmers’ markets, walking groups, educational displays, newsletters and a website), showed adverse BMI intervention effects for lower-SES individuals. This study found that weight gain was most likely to be prevented in the groups with a higher income and a higher educational level, thus revealing that such environmental modifications may not be effective in narrowing socioeconomic inequalities in obesity but may increase them, although it is unclear whether this was solely the effect of the environmental modifications or a combination of these modifications and other aspects of the intervention (e.g. educational material).

Six observational studies followed the universal approach. Scoggins et al. 221 examined the effects of environmental modifications in the workplace (such as decorating stairways and replacement of unhealthy food in vending machines with healthy food) using two study designs: a 1-year controlled cohort study (moderate quality) and a 5-year uncontrolled cohort study (low quality). In both studies favourable weight loss effects for lower-SES adults were observed. This therefore suggests that such environmental modifications in the workplace may have favourable effects for lower-SES individuals. The observational study (moderate quality) by VanWormer et al. ,220 which also examined the effects of aesthetic stairwell enhancements and access to healthy food/beverages as well as other non-environmental interventions (such as the use of pedometers and website step-tracking to encourage physical activity, improved scale access for self-weighing, worksite advisory groups and site-wide publicity on nutrition and physical activity), found no difference in BMI change across SES groups. Therefore, these studies show contradictory results for the effect of environmental modifications in the workplace and community wide on lower-SES groups.

A controlled cohort study215–217 in those aged 30–67 years explored the intervention effects of environmental modifications (improved street lighting, pavement gritting and labelling of walking trails to increase the accessibility of areas for physical activity) and non-environmental modifications (including specifically designed leaflets, individual counselling, biannual fitness tests, organised walking groups and indoor group activity sessions) and found that the net proportion who increased their body mass was significantly lower in the intervention district than in the control district. This was found across all educational groups; however, it was more pronounced in low-SES men, thus showing potential favourable intervention effects on lower-SES groups as well as improvements across the social gradient.

One repeat cross-sectional study (moderate quality) conducted by Carleton et al. 222 explored the intervention effects of community-wide, environmental components (grocery store food labelling and healthy hearts menus) combined with non-environmental components (exercise courses and nutrition programmes at the public library) and found that BMI was lower in the intervention district than in the control district (similar to the results of the Jenum et al. study215–217). The other repeat cross-sectional study (low quality) by Tudor-Smith et al. ,218 which examined multiple interventions (both community wide and in the workplace) consisting of both environmental modifications (food labelling in a major grocery retailer, a restaurant and canteen scheme to increase the availability of healthy food choices, and smoke-free areas) and non-environmental modifications (smoking cessation television series), had significant biases that prevented the detection of intervention effects and differences between SES groups. These biases included possible contamination in the reference area because of other health promotion activities taking place there and the fact that the sample size for the reference area was too small to give it enough statistical power for the detection of likely net intervention effects.

Targeted interventions

One of the observational studies reported in the universal interventions section also followed a targeted approach as it was delivered in two low-SES districts. 215,216 The study explored the intervention effects of environmental modifications (improved street lighting, pavement gritting, and labelling of walking trails to increase accessibility of areas for physical activity) and non-environmental modifications (including specifically designed leaflets, individual counselling, biannual fitness tests, organised walking groups, and indoor group activity sessions) and found that the net proportion who increased their body mass was significantly lower in the intervention district than in the control district. Therefore, this intervention appeared to be effective in a low-SES population.

Targeted interventions

The two studies investigated the effects on obesity-related outcomes of the US Food Stamp Program (now called the Supplemental Nutrition Assistance Program), which is aimed at low-income families. The study by Kaushal224 was a natural experiment that took advantage of a change in the federal law that denied a subgroup of the population access to the programme. The results from this study suggest that the Food Stamp Program had no effect on BMI. The second study investigated the effects of the Food Stamp Program in women using longitudinal data from the Panel Study of Income Dynamics. 223 The results were presented by food security status. There was no change in body weight in those who were persistently food secure, or those who changed food security status, but increases in body weight were associated with Food Stamp Program participation in those who were persistently food insecure (the most deprived group). These results suggest that the Food Stamp Program may be associated with increases in weight in low-income, food-insecure women that could potentially increase SES inequalities in obesity. However, the authors concluded that participation in the Food Stamp Program does not necessarily cause weight gain.