Many indicators of health, including disability-free life years and life expectancy, are positively associated with socioeconomic status (SES) [1]. While socioeconomic inequalities in health cannot be fully explained by behavioural factors [2], behaviours such as smoking and diet play a role [3, 4]. Inequalities in health behaviours emerge during childhood [5], while recent international evidence indicates that these inequalities are widening, in line with growing economic inequality [6].

Attempts to improve young people’s health behaviours commonly involve school-based interventions, often targeting lower SES pupils or schools [7]. While intuitively appealing, targeting usually ignores the graded nature of associations between socioeconomic status and health [1] and may produce stigma [8], while imprecise targeting methods often fail to reach at-risk individuals. Targeting deprived schools for example may fail to reach poorer children in more affluent schools, or address inequalities within such schools [9]. Hence, universal interventions which disproportionately benefit lower SES groups may have greater potential to improve population health, while reducing inequality [1]. However, depending on the nature of intervention, universal approaches may also worsen inequalities [10, 11].

McLaren and colleagues [11] argue that the likelihood of universal interventions worsening or improving inequalities depends on whether intervention involves ‘superficial’ (i.e. focusing upon individual agency) or ‘radical’ change (i.e. targeting structural factors). Some empirical evidence lends support to notions that ‘downstream’ interventions are more likely to worsen inequality [12]. Interventions targeting individual factors such as knowledge may work where ignorance is the sole reason for not ‘choosing’ healthier options, though may do little for individuals facing greater structural constraints [13]. However, social and environmental interventions may be resisted where incongruent with local norms and reducing inequality may require intervention at multiple levels [14, 15]. Some reviews indicate that school-based interventions based on education are often ineffective [16], while multi-level interventions have greater effects [17]. One would perhaps hypothesise that school-based interventions based purely upon education may be more likely to worsen inequalities, while interventions targeting factors at multiple levels simultaneously, may be more likely to reduce inequality.

However, evaluations have traditionally paid little attention to effects on inequalities; in a 2013 Cochrane review of school-based interventions to prevent adolescent smoking, socioeconomic inequalities are not mentioned [16]. In a 2013 review of physical activity interventions, none of the 44 trials reviewed examined effects upon inequalities [18]. In a 2014 review of the WHO Health promoting schools framework [19], 2 of 67 included studies reported effects by SES. A recent review of equity impacts of tobacco control policies, identified 5 childhood studies which reported effects by SES [20], including 2 in a school setting. One study reanalysing European trials of ‘effective’ interventions in four behavioural domains (smoking, diet, physical activity and alcohol use) reported that effects often varied by SES, but drew few conclusions regarding why due to the small number of studies [21–24].

Inattention to sub-group effects may in part reflect tensions between needs of policymakers and practitioners, and a desire for scientific purity. Sub-group analysis, while critical for policy and practice, is often viewed as statistical malpractice, with studies typically powered to detect a main effect [25]. It may also reflect assumptions that universal delivery ensures movement of the risk distribution to the left, with no effect on inequality [26]. Schools are commonly seen as useful venues for health improvement precisely because they may reach almost all children. However, interventions do not ‘work’ upon passive recipients, but are interacted with in ways which are shaped by pre-existing contextual conditions [27, 28]. Inattention to inequality may also reflect utilitarian approaches, focused on achieving the greatest gain for the greatest number. Recent discourses surrounding inequalities have attempted to highlight economic consequences for society as a whole [29], while others maintain that addressing health inequalities remains a moral issue. Intervention developers’ and evaluators’ views on the moral and economic importance of inequality, or dominant discourses within their countries may shape the nature of interventions developed or evaluated, and decisions on whether effects on SES should be tested.

In 2007, Whitehead [30] called for all interventions which aim to improve population health to be evaluated in terms of effects on inequality. This review examines the extent to which, and the ways in which effects of universal school-based interventions on socioeconomic inequality have been evaluated in peer-reviewed evaluations published since 2008. Authors such as McIntyre have highlighted the fallibility of assuming all measures of SES to be equal [31], with socioeconomic trends differing by means of measurement (for example, whether assessing income or education level). Hence, the types of SES markers collected will also be summarised. A secondary aim is to synthesise existing evidence on the types of interventions which are more likely to worsen or improve inequality, although conclusions will be drawn tentatively due to the shortage of available studies. No previous reviews to date have examined how socioeconomic inequalities are discussed within evaluation studies, in order to gain insights into why an inequality focus features so infrequently. Hence, we report a content analysis of discussion of socioeconomic inequality within the rationale for interventions and interpretation of findings within published articles of school-based interventions.

Effects on inequality have been tested and reported in only one in five peer-reviewed evaluations of universal school-based interventions published since 2008. These analyses indicate that universal school-based interventions have the potential to improve or worsen inequalities, though conclusions regarding how or why are hampered both by a lack of routine testing of effect on inequality, and by biases in consideration of inequality. For example, evaluators of interventions focused on education alone appear least likely to test effects on inequality. Given the often held assumption that educational interventions may make inequality worse, researchers who see reducing SES inequality as a goal of universal intervention perhaps avoid developing or testing interventions based solely on education. Inequality is also rarely considered in studies from North America, though more commonly in European studies.

Within studies which report differential effects, there is inconsistency in how SES is measured, with studies typically providing little justification for their choice of SES measure (e.g. education, SES), or its validity. Furthermore, there is inconsistency in how testing is conducted and reported; some report subgroup effects, some interaction terms and others both. For the most part, where studies do not demonstrate a significant differential effect, test statistics are unreported and it is simply concluded that the intervention was equally effective across SES subgroups. This is despite the fact that studies were usually not powered to detect this interaction, and failure to reach significance is inevitable unless interactions are large [25]. Hence, it is difficult to ascertain whether certain types of intervention are having consistent small to medium effects on inequality which cannot be detected in individual studies. Outcomes are sometimes analysed selectively with, for example, differential effectiveness examined only for variables with a significant main effect, despite the fact that a significant main effect is not a pre-requisite for an effect on inequality.

Nevertheless, the review provides some tentative insights into which types of intervention might affect inequality. No interventions based solely on education reduced inequality, while all interventions which worsened inequality included educational components. By contrast, interventions which resulted in a narrowing of inequality included environmental change components. This offers tentative support for arguments of Mclaren and colleagues [11] and Whitehead [30], that interventions based on structural change may be more likely to narrow inequality. However, effects of interventions which combined education with environmental change were inconsistent. This may indicate differences in the emphases on components at each level, or varying overall effectiveness of components. For example, it is plausible that if an effective education component is combined with an ineffective change in social or physical environments, this may widen inequality.

There was some patterning by behavioural domain. Five studies with a negative social gradient focused on diet and physical activity, with 4 reporting bigger effects on obesity outcomes for the higher SES group, or in one case, a negative effect on obesity in the lower SES group. Notably, alongside smoking interventions, diet and physical activity interventions were the intervention type for which effect on inequality was most commonly tested, perhaps reflecting widespread acknowledgement of the contribution of smoking and obesity to health inequalities. Such studies were more likely to use anthropometric measures of change, rather than relying upon self-reports of behavioural change, which may lack the sensitivity to detect SES differences in effect due to less valid completion in some subgroups [66].

Content analysis of the 98 published evaluations provided key insights into reasons why effects on inequality receive so little attention. Notably, only in a minority of studies was socioeconomic patterning discussed in describing the problem the intervention aimed to address. Hence, SES inequality in health behaviours, despite consistent evidence of its presence throughout adolescence [5], is simply not seen as part of the problem intervention aims to address in most cases. North American studies were least likely to emphasise SES patterning in adolescent health and health behaviours within the rationale for the intervention. This is perhaps linked to high levels of health inequality, and limited political will to address inequalities, in the US compared to many other developed countries [67]. Some European funders such as the NIHR in the UK have begun to mandate analysis of effects on inequality. One study funded by this body stated that they conducted analyses by SES due to research funders’ pressure, though did not report findings in their peer reviewed outcomes paper, due to a lack of statistical power, highlighting a perceived tension between policy need for information on what works for whom vs the desire for scientific purity [25]. Consistent reporting of sub-group effects across intervention studies, even where individual studies are underpowered, would allow systematic reviewers and meta-analysts to pool effects across studies to identify whether there are consistent interaction across studies.

School-based interventions were sometimes portrayed as inevitably reaching all SES groups and having universal effect; an assumption challenged by the finding that many universal interventions worsened or narrowed inequality. A more minority assumption, contradicted by the finding of variable effects of school-based intervention on inequality was that interventions would inevitably work better for groups who have the most to gain, due to ceiling effects in more affluent groups. Theory or empirical evidence on effects on inequality were rarely drawn upon throughout evaluations. A minority of studies included process evaluations, some of which emphasised SES differences in perceptions of the intervention, none of which translated into greater effects for lower SES groups. None provided more in-depth insights into how the contexts in which lower or higher SES pupils experienced the intervention impacted its effectiveness.

A strength of this review is the fact that the search strategy incorporated seven databases, ensuring a high level of coverage. However, this could have been extended to grey literature and research published in languages other than English. The review, in part for reasons of resource, also covers a relatively narrow time period, and earlier evaluations may have contributed different understandings of intervention effects on inequality.

Nevertheless, the review indicates that universal school-based interventions may narrow inequalities, or make them worse. Peer-reviewed reports of evaluation studies indicate that SES inequalities are rarely considered as a part of the problem intervention stakeholders are aiming to address, while assessment of impact on inequality remains a minority activity. The review provides support for notions that interventions based on education may be more likely to widen inequality, with those based on environmental change more likely to narrow it. However, conclusions are highly tentative due to the fact that the vast majority of relevant studies provide no such analysis, while those that do tend to be linked to specific locations and types of intervention. At present, such analyses are almost exclusive to European studies; much could be learned through more routine assessment of inequalities impact in other contexts, including North America, which currently generates a large proportion of intervention studies, though pays least attention to effect on inequality. There is also a need for guidance on the most appropriate methods for analysis and reporting of sub-group effects in order to allow for comparison across studies. Authors of evaluations need to clearly justify the measure of SES used. Different dimensions of SES may function differently in shaping health across contexts – in the US, for example, the racialization of inequality may mean that ethnicity in itself can be considered a valid proxy measure of SES in that context. Hence, while it may not be appropriate to standardise SES measures across all studies, the measure selected should be justified with reference to its role in shaping the health outcomes under investigation in the context where the intervention is delivered. It is vital that authors present interaction effects and sub-group analyses fully, regardless of whether the individual study was sufficiently powered to detect a moderating effect. Accumulation of consistently applied methods to assess intervention effects on inequality could allow future research to use methods such as meta-regression or qualitative comparative analysis to examine what features of interventions were associated with positive, negative or neutral effects on health inequalities, in what contexts. To facilitate this, guidance for the conduct, analysis and reporting of evaluations of complex interventions should include recommendations on measuring, analysing and reporting effects on inequality. Greater use of process evaluation to facilitate understandings of how interventions are experienced by young people with varying socioeconomic backgrounds may help in building theory and providing deeper insights into why some school-based interventions worsen inequalities while others reduce it.

Not applicable.