Socio‐economic inequalities in the effectiveness of workplace health promotion programmes on body mass index: An individual participant data meta‐analysis

Summary This individual participant data meta‐analysis assessed the effectiveness of workplace health promotion programmes on body mass index (BMI) across socio‐economic groups and whether study and intervention characteristics explained inequalities in effectiveness. Studies were eligible if they assessed the effect of a workplace health promotion programme on BMI in the Netherlands, included workers of at least two different socio‐economic positions (SEPs) and had a study design with premeasurement and postmeasurement and control condition. Data of 13 studies presenting 16 interventions (5183 participants) were harmonized. In a two‐stage meta‐analysis, the interaction between intervention and SEP on BMI was tested with linear mixed models for each study. Subsequently, the interaction terms were pooled. The influence of study and intervention characteristics on the effectiveness of workplace health promotion programmes was evaluated using meta‐regression analyses. Compared with control conditions, workplace health promotion programmes overall showed a statistically non‐significant 0.12 kg/m2 (95% CI: −0.01, 0.25) decrease in BMI, which did not differ across SEP. Interventions evaluated within randomized controlled trials, agentic interventions, those that focused on high‐risk groups, included a counselling component, consisted of more than five sessions, or were offered at the individual level did statistically significantly reduce BMI. No evidence was found for intervention‐generated SEP inequalities.

However, there are concerns that health promotion programmes might increase, rather than reduce, inequalities due to a higher reach and/or effectiveness among individuals with a high socio-economic position (SEP) compared with those with a low SEP. 4 Only limited information is available about the differential effectiveness of public health interventions across socio-economic groups. 5,6 In several reviews, the majority of the included studies did not find differential effects of public health interventions targeting health behaviour across socio-economic groups. [5][6][7][8][9][10] According to a framework for the likely impact of obesity prevention strategies on socio-economic inequalities in body weight, interventions can be categorized based on the degree to which an intervention involves the capacity of individuals to make independent, purposive choices (i.e., individual agency). 11 This framework distinguishes agentic interventions (high individual agency), agento-structural interventions (some individual agency) and structural interventions (no individual agency). It is hypothesized that higher socio-economic groups will benefit more from interventions with a higher level of individual agency. 11 Indeed, some studies showed that agentic interventions, such as health education or counselling programmes or mass media campaigns, could widen socio-economic inequalities in health behaviour or health. 5,6,9,11 This is in contrast with structural interventions, such as removing unhealthy food options or fiscal interventions, which facilitate healthier choices and may contribute to reducing inequalities. 6,9,11 Agento-structural interventions do consider the environment, but individual agency is still important, for example, in interventions providing healthier food options in canteens. 11 Health promotion activities can be implemented in different settings. The workplace has been identified as a promising setting for health promotion due to the substantial time adults spent at work and the ability to reach large groups of participants in a natural social network. A recent review has shown positive, but small, effects of workplace health promotion programmes on body mass index (BMI). 12 Because workplace health promotion programmes can vary in the degree of agentic involvement, ranging from entirely agentic (such as health education or counselling programmes) to structural interventions (such as removal of vending machines containing unhealthy food and drink options), understanding the equity impact is highly relevant.
However, information on intervention-generated inequalities for workplace health promotion programmes is scarce. Overall, the majority of studies included in previous reviews on workplace health promotion programmes did not find differential effectiveness across socio-economic groups. 8,[13][14][15] Yet most reviews compared workplace health promotion programmes provided to the general working population with those targeted to blue collar workers or workers with a lower SEP only. 13,14 Equity-specific subgroup analysis within specific interventions can contribute to understanding the differential effectiveness of interventions, which is possible in an individual participant data (IPD) meta-analysis. An IPD meta-analysis furthermore provides the opportunity to investigate which type of studies and interventions could contribute to reducing socio-economic inequalities in BMI. As described above, it is hypothesized that agentic interventions will increase socio-economic inequalities in BMI.
The current IPD meta-analysis enables analyses that go beyond those that have been performed in the original studies and conventional meta-analyses. The current study contributes to the existing literature in two ways. First, more in-depth insight into the differential effects within and across workplace health promotion programmes can be assessed with equity-specific subgroup analyses. Second, the influence of study design and intervention characteristics on these differential effects can be studied, which contributes to the understanding of the influence of specific study and intervention components on the effectiveness of workplace health promotion programmes. This IPD meta-analysis will be performed in the Dutch context, which enables to study the effectiveness of the interventions in heterogeneous populations in a rather homogeneous occupational health and social security context. The study aims to investigate the differential effects of workplace health promotion programmes on BMI between socio-economic groups and the extent to which study and intervention characteristics explain possible differences in effectiveness across groups.

| Search strategy and selection of studies
This IPD meta-analysis was performed according to the earlier published protocol, 16 which was also registered with PROSPERO (CRD42018099878). The PRISMA-IPD guidelines were used for reporting our findings. As described in detail in the protocol paper, a systematic search was performed to identify relevant studies aimed at promoting healthy behaviour or preventing obesity among workers.
The current paper evaluates differential effectiveness of workplace health promotion programmes on BMI. Another paper, using the same generic data set, evaluates differential effectiveness on health behav-

| Methodological quality
Methodological quality was also assessed by one author (S.R., was used. 18 This consisted of nine criteria regarding randomization, blinding of participants, similarity of groups, compliance, loss to follow-up, intention-to-treat, adjustment for confounders, data collection methods and follow-up duration. 17 On each item, a study could score positive if the quality criterion was met (1 point), negative if the criterion was not met (0 points) or unclear if the publication and/or an additional information request by authors provided insufficient information to make a judgment (0 points). Summary scores were categorised as poor (0-2 points), fair (3-4 points), good (5-7 points) or excellent (8-9 points). Both the data extraction form and methodological quality scale were sent to the corresponding author of the original study for verification.

| Harmonization
Data from all studies were harmonized. If a study contained more than one intervention arms, these arms were all considered as separate interventions. In case of more than one control arms, these arms were combined into one control group. All information from the included studies, both the harmonized IPD and the data extracted from the original articles, were merged into a single dataset.

| Body mass index
A continuous measure of BMI (kg/m 2 ), obtained from self-reports or objective measures, was used from pre-intervention and postintervention measurements. The measurements could be assessed directly after the intervention (immediate effects) or after a longer follow-up period (sustained effects). The timing of these measurements differed between studies.

| Socio-economic position
Most interventions included education as indicator of SEP, which was divided into low (pre-primary, primary and lower secondary education), intermediate (upper secondary education) and high (postsecondary education), based on the 1997 International Standard Classification of Education (ISCED-97). In one study, where information on educational level was lacking, occupational class was used to define SEP. 19 Here, among workers from a construction company, the construction workers were categorised as low SEP and the office workers as intermediate SEP.

| Covariates
As in the original studies, age was used as a continuous variable and gender was dichotomized into male and female.

| Statistical analysis
A two-stage meta-analysis approach was performed. In the first stage, IPD data of each study were analysed separately using multilevel linear mixed models. In the second stage, the results per study were pooled in a meta-analysis. In the first stage, a random intercept for participant was used, and, for studies with a clustered design, a random intercept for cluster was added to take into account the clustering of participants. Overall effects and interaction effects with SEP (intervention * SEP) were analysed, and all models were stratified by SEP. In case a SEP group in an included study consisted of less than 10 participants, no subgroup analysis or interaction analysis was performed for that specific SEP group in that particular study. For the two studies without any workers with a high SEP, 19,20 the effects among workers with a low SEP were compared with workers with an intermediate SEP. As no statistically significant intervention * time interaction effects were found, both immediate and sustained effects were added jointly in the mixed model. All models were adjusted for baseline BMI, age and gender. Two studies evaluated more than one intervention arm, 21 were agentic interventions, five had a combination of agentic and agento-structural intervention elements, and there were no structural interventions. Twelve interventions included a counselling component, [19][20][21][22][23][24][26][27][28][29] whereas six interventions included (also) a change in the work environment (such as free or healthy food options at work, or signs to promote stair use) 21,30,31 (Table 1). Ten interventions consisted of universal prevention strategies, 20,21,23,24,26,28,30,31 whereas six interventions were offered to high-risk workers, that is, workers with an unhealthy behaviour, high BMI or high cardiovascular risk. 19,22,25,27,29 There was overlap between study and intervention characteristics: interventions focused on a high-risk group were all individual-level interventions, while this was the case for only three out of the 10 universal interventions. Moreover, interventions focused on a high-risk group more often had more than five sessions compared with universal prevention strategies.
In seven studies, sufficient participants of all three socioeconomic groups were represented to estimate the effectiveness of the intervention stratified by SEP. 22

| Overall effects
As shown in Table 2 3.2 | Differential effects and effects within socio-economic groups CI: −0.26, 0.08; Table 2). As shown in Figure 2, in the low socioeconomic group, only one out of 10 interventions showed a statistically significant reduction in BMI. 19 In the intermediate 19,25 and high 22 socio-economic group, two out of 16 and 14, respectively, interventions showed a statistically significant reduction in BMI.  Table 3). The association between study design and intervention effectiveness did not differ across socioeconomic groups. Overall, the effectiveness of interventions in which BMI was measured through self-report did not differ from those inter-  Table 3). The influence of these intervention characteristics on the effectiveness did not differ across socio-economic groups.

| DISCUSSION
No differential effects of workplace health promotion across SEP on BMI were found. In all socio-economic groups, a small, but statistically non-significant, decrease in BMI was found. This IPD meta-analysis showed that interventions evaluated within an RCT, agentic interventions, intervention that focused on a high-risk group, included a counselling component, consisted of more than five sessions, or offered at the individual level did reduce BMI. However, the reduction in BMI was 0.32 kg/m 2 or lower.

| No differential effectiveness of workplace health promotion programmes on BMI
Theoretically, public health interventions could generate socioeconomic health inequalities in different ways, for example, by differences in delivery, reach and compliance, or by having greater effects among individuals with a high compared with a low SEP.
Concerning the delivery, it is remarkable that most studies focused on intermediate and high educated workers (79% of the IPD sample). The interventions were either more often provided to workers in high SEP or these workers were more likely to participate in offered interventions. Only two studies targeted workers in the construction industry, 19,20 the majority of these participants had a low SEP. Offering effective interventions mainly to workers in high SEP would lead to intervention-generated inequalities. In this IPD metaanalysis, information on reach (initial participation) was lacking or not well defined in the individual studies. However, a systematic review investigating reach of workplace health promotion programmes did not find clear inequalities in reach across socio-economic groups. 32 We hypothesized that workplace health promotion programmes would be less effective among workers in low SEP compared with workers in higher SEP. Following the framework for evaluating the impact of obesity prevention strategies on socio-economic inequalities in body weight, this was in particular expected for agentic interventions as those interventions focus on cognitive-behavioural strategies to support making independent choices, for example, health education interventions. Our IPD meta-analysis, however, showed no differential effects on BMI across socio-economic groups for workplace health promotion programmes. This is in line with several reviews on public health interventions that showed that the majority of the included studies on the prevention of unhealthy behaviour or obesity did not have differential effects across socio-economic groups. [5][6][7]9,10,14,15,33 However, as unhealthy behaviours and obesity are more prevalent in workers in low SEP, the need for effective interventions for these workers remains of eminent importance. According to Hillier-Brown et al., 34 implementing effective interventions targeted specifically to individuals in low SEP, for example, blue collar workers, might be effective in reducing the socio-economic gradient in obesity.

| Associations with study design and intervention characteristics
Overall, the reduction in BMI was small and not statistically signifi- health promotion interventions. 18 The seven RCTs in our meta-

| Need for effective interventions among workers in low SEP
As the majority of the included interventions were not more effective in reducing BMI than control conditions, regardless of the SEP of the participants, this raises the question of which interventions are needed to reduce BMI in low socio-economic groups or to reduce socio-economic inequalities in BMI. The studies in this IPD metaanalysis consisted of agentic or agento-structural interventions, often counselling in combination with health education. All included interventions required individuals to make independent choices (e.g., free fruit at the workplace, healthier food options at the canteen or food steps to promote stair use). It was hypothesized that in particular, workers in higher socio-economic groups would benefit from these kinds of interventions. None of the included studies were considered to evaluate structural interventions, while it is expected that such interventions are more likely to be effective among persons in low SEP. The current study showed, however, no evidence for an increase in inequalities after agentic or agento-structural interventions.
Although some interventions contained, to some extent, an environmental change, they could not be considered as structural because individuals still needed to make their own choices. According to the framework for the likely impact of obesity prevention strategies on socio-economic inequalities in population body weight, structural interventions have more potential to reduce inequalities, because the individual choice is largely removed, such as providing only healthy food options at the canteen. 10,11 Such interventions would provide a context for healthy behaviour and could be combined with counselling interventions addressing high-risk groups. However, a first step could be to make a comprehensive analysis of the determinants of the inequalities in health behaviour and BMI and design integrated interventions targeted to workers in low SEP.

| Strengths and limitations
After combining original data from 16 interventions of 13 studies into one dataset, and analysing the results across socio-economic groups, this IPD made it possible to assess socio-economic inequalities in the effectiveness of workplace health promotion programmes and to provide insight into the association of study design and intervention characteristics with this effectiveness. Most workplace health promotion programmes did measure an indicator of SEP, but analysing differential effects across sociodemographic groups was mostly not performed.
A limitation is that-in contrast to what has been described in the protocol-the influence of reach and work-related characteristics on the effectiveness of the studied worksite health promotion programmes could not be investigated. This information was not available in most of the included studies or was too heterogeneous to be harmonized, as a result of which these analyses could not be conducted. We recommend to include relevant process information, such as reach and uptake, and information on work-related characteristics in publications on the effectiveness of workplace health promotion programmes. Work-related characteristics have been found to be associated with BMI, for example, a higher BMI among workers with an imbalance between perceived high efforts and low rewards at work, 36,37 and among workers with high physical work demands. 38,39 The focus of the manuscript is on differential effects of workplace health promotion programmes on BMI across SEP groups. Therefore, only studies with at least multiple SEP groups were included. However, we found one other study that concerned only a single SEP group (high SEP) but met all inclusion criteria. 40 The effectiveness for this study was comparable with the included interventions. Although in the selection process no studies were identified that were restricted to low SEP workers only, providing tailored and effective interventions to workers in low SEP only could reduce socio-economic inequalities in BMI. For six studies, no data were available. Five of these six studies were more than 10 years old. Although it is a strength that all studies were performed in the Netherlands within a homogeneous occupational health context, generalization to other contexts should be done with caution.
It would be relevant to perform a similar analysis in different countries and compare the results across countries.

| CONCLUSION
In conclusion, small statistically non-significant intervention effects of workplace health promotion programmes on BMI were found. No evidence was shown for intervention-generated inequalities in BMI for workplace health promotion programmes. These findings are in line with previous studies showing no differential effectiveness on BMI across socio-economic groups. Interventions evaluated within an RCT, agentic interventions, interventions focusing on high-risk groups, with counselling components, more than five sessions or being offered at the individual level did statistically significantly reduce BMI. No evidence was found for intervention-generated SEP inequalities in BMI.