Effectiveness of Workplace-Based Diet and Lifestyle Interventions on Risk Factors in Workers with Metabolic Syndrome: A Systematic Review, Meta-Analysis and Meta-Regression

Workplace health interventions are essential to improve the health and well-being of workers and promote healthy lifestyle behaviours. We carried out a systematic review, meta-analysis and meta-regression of articles measuring the association between workplace dietary interventions and MetS risk. We recovered potentially eligible studies by searching MEDLINE, the Cochrane Library, Embase, Scopus and Web of Science, using the terms “Metabolic syndrome” and “Occupational Health”. A total of 311 references were retrieved and 13 documents were selected after applying the inclusion and exclusion criteria. Dietary interventions were grouped into six main types: basic education/counselling; specific diet/changes in diet and food intake; behavioural change/coaching; physical exercise; stress management; and internet/social networks. Most programmes included several components. The interventions considered together are beneficial, but the clinical results reflect only a minimal impact on MetS risk. According to the metaregression, the interventions with the greatest impact were those that used coaching techniques and those that promoted physical activity, leading to increased HDL (effect size = 1.58, sig = 0.043; and 2.02, 0.015, respectively) and decreased BMI (effect size = −0.79, sig = −0.009; and −0.77, 0.034, respectively). In contrast, interventions offering information on healthy habits and lifestyle had the contrary effect, leading to increased BMI (effect size = 0.78, sig = 0.006), systolic blood pressure (effect size = 4.85, sig = 0.038) and diastolic blood pressure (effect size = 3.34, sig = 0.001). It is necessary to improve the efficiency of dietary interventions aimed at lowering MetS risk in workers.


Introduction
Metabolic syndrome (MetS) constitutes a major public health problem, not only because of its increasing prevalence-currently estimated at 25% worldwide [1]-but also because of its clinical, economic and humanistic impact.
The WHO defines MetS as a pathological condition characterised by abdominal obesity, insulin resistance, hypertension and hyperlipidaemia. Slightly different definitions have been provided for this syndrome. The harmonized definition of the International

Search Strategy
To define the search terms we consulted the Medical Subject Headings thesaurus developed by the U.S. National Library of Medicine. The search scheme was planned in three dimensions: The following filters were applied: "Humans", "Adult 18+ years", and "Comparative Study" or "Clinical Trial".
The search syntax was developed for MEDLINE, then adapted to each of the remaining databases. The search was performed from the first available date until the day of the last query of the databases (initial search in MEDLINE 2 May 2020). To reduce the possibility of publication bias, we reviewed the reference lists of the selected articles and of relevant guidelines. Furthermore, experts in the domain were contacted by mail to avoid issues regarding possible grey literature (materials and research produced by organisations outside the traditional commercial or academic publishing and distribution channels).

Final Selection of Articles
Registries that met the following inclusion criteria were accepted for review: articles with clinical trial or comparative study design that met the search criteria and that were published in peer-reviewed journals in English, Spanish or Portuguese.
Exclusion criteria were unavailability of the complete text, lack of causal relationship between metabolic syndrome and occupational health, and interventions not focused on diet in adults.
We preselected articles by carefully reading the titles and abstracts to verify suitability according to the established inclusion criteria. We then performed a full text review of all preselected studies, excluding duplicates and articles that did not provide the relevant data.
Eligible articles were independently selected by two authors (AGC and ELP). To validate study inclusion, we established that inter-rater agreement (Cohen's Kappa coefficient) should be greater than 60% [17]. Provided that this condition was met, any discrepancies were resolved by consulting a third author (CWB) and subsequently reaching a consensus among all authors.

Data Extraction
Double entry tables were used to ensure data accuracy. Where inconsistencies were detected between two entries, the original articles were consulted for verification.

Study Variables for Each Article
The studies were grouped by study variables to ease comprehension of the results. We extracted the following data from each article: • Author: first author. • Year: year of publication. Effect: causal relationship derived from the results. We recorded the clinical and anthropometric results derived from the interventions, as well as other outcomes of interest related to improvement of MetS and/or occupational health parameters.

Quality of Reporting of the Selected Documents
We used the CONSORT (Consolidated Standards of Reporting Trials) statements to assess quality of reporting [18]. This checklist contains 25 essential aspects that should be described in all studies. One point was assigned for each item present (if not applicable, it was not scored). When an item was composed of several points, the points were assessed independently and then averaged to give a final score. In this way, no item could receive a score above 1 point.

Obsolescence
We calculated Burton-Kebler half-life (median age) and Price Index (percentage of articles less than 5 years old) to determine whether clinical trials were up-to-date or obsolete.

Meta-Analysis and Meta-Regression
We analysed the global effect size through a meta-analysis of the included articles, considering the following variables: waist circumference (WC), body mass index (BMI), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-c), low-density lipoprotein cholesterol (LDL-c), triglycerides (TG), systolic blood pressure (SBP), diastolic blood pressure (DBP), and fasting blood glucose. For this analysis we adopted the random effects model. The effect sizes and 95% confidence intervals were presented in forest plots together with the coefficient of determination (R-squared), the Tau-squared value, and p value.
To examine the influence of each study on effect size, we used the leave-one-out method, which consists of omitting one study at a time and recalculating heterogeneity [19]. We also created the scatter plot introduced by Baujat et al. [20]. In this graph, the X axis shows the contribution of each study to heterogeneity, and the Y axis shows the influence of each study on the overall effect size, with the strongest influencers located in the upper right-hand corner.
Publication bias occurs because favourable results have a higher probability of being published compared with nonsignificant results. The absence of the latter may result in overestimations in meta-analyses. We analysed publication bias in our study using funnel plots, which show the effect measure of each study on the X axis and a measure of precision, such as the standard error, on the Y axis. A meta-analysis without publication bias shows a point cloud in the shape of an inverted funnel. Based on this assumption, we performed a nonparametric analysis called trim-and-fill, estimating the number of missing studies and recalculating the intervention effect after including these new "filled" studies [21]. We also used a less conservative method, proposed by Copas et al. [22], for estimating the number and results of the missing studies.
Meta-regression was used to establish whether the type or duration of interventions affected effect size. The included studies covered six intervention types: Basic education and general counselling on healthy living and diet (Int.1), specific diet/changes in diet and food intake (Int.2), behavioural changes/coaching (Int.3), physical exercise education and/or training (Int.4), stress and/or sleep management (Int.5) and internet/social networks (Int.6).
The authors of the selected articles presented their results in three different ways: mean (± standard deviation) before and after the intervention, difference in means (± standard deviation) before and after the intervention, and difference in medians (with interquartile range) before and after the interventions. For the meta-analysis we used the second method (mean difference ± standard deviation). To do this, we calculated the weighted difference in means and standard deviation for the first case, and for results presented as medians and interquartile ranges, we approximated the mean and standard deviation according to the methods of Luo et al. [23] and Wan et al. [24], respectively.

Systematic Review
Applying the study criteria listed above, we recovered 311 records: 16 in MEDLINE, 43 in the Cochrane Library, 32 in Embase, 33 in the Web of Science and 187 in Scopus. (Figure 1).
Of the 311 records identified, 50 were duplicates and 2 were redundant, leaving 259 records for screening. A further 56 records were discarded after screening, leaving 203 for full text review. Of the articles excluded during the screening stage, 16 were not related to MetS, 8 did not measure the relationship between occupational health and MetS, 14 did not include a dietary intervention, 7 had no available full text, and 11 were published in other languages (Chinese, Japanese, Korean) ( Figure 1).
In the second stage, 190 records were discarded, because 161 were not clinical trials or comparative studies, 19 did not include a population of workers, and data were not shown in 9. Only one study was carried out in young people.
Of the 311 records identified, 50 were duplicates and 2 were redund records for screening. A further 56 records were discarded after screening full text review. Of the articles excluded during the screening stage, 16 w to MetS, 8 did not measure the relationship between occupational health a not include a dietary intervention, 7 had no available full text, and 11 we other languages (Chinese, Japanese, Korean) ( Figure 1). In the second stage, 190 records were discarded, because 161 were n or comparative studies, 19 did not include a population of workers, and shown in 9. Only one study was carried out in young people.
When evaluating the quality of the 13 select articles using the CONS naire, the scores ranged from 9.5 to 20.5 with a median of 13.5 (Table 2). cording to Burton and Kebler was six years, and the Price Index was 30.8% that only four articles were less than 5 years old.
The inter-rater agreement in the selection of articles, measured by coefficient, was 62% (p < 0.001).
Regarding study design, all of the articles included in the review wer 12 randomised [27][28][29][30][31][32][33][35][36][37][38][39] and one nonrandomised [34]. All manuscrip in English, and the studies had been conducted in seven different countrie When evaluating the quality of the 13 select articles using the CONSORT questionnaire, the scores ranged from 9.5 to 20.5 with a median of 13.5 ( Table 2). The half-life according to Burton and Kebler was six years, and the Price Index was 30.8%, which implies that only four articles were less than 5 years old.
The inter-rater agreement in the selection of articles, measured by Cohen's Kappa coefficient, was 62% (p < 0.001).
Regarding study design, all of the articles included in the review were clinical trials, 12 randomised [27][28][29][30][31][32][33][35][36][37][38][39] and one nonrandomised [34]. All manuscripts were written in English, and the studies had been conducted in seven different countries. The USA [30][31][32]36,39] and Japan [34,37,38] were the biggest contributors. With respect to the clinical criteria used to evaluate metabolic syndrome, studies [31,32,36,39] (four of the five studies conducted in the USA) used the Third Report of the National Cholesterol Education Program Adult Treatment Panel (NCEP-ATP-III) [40]. Shrivastava et al. [29] and Inoue et al. [34] used these same criteria. The remaining studies used heterogeneous classifications. The criteria of Woo et al. [27] were based on but not defined by the guidelines for treating adult diseases (National Cholesterol Education Program 2002), published by the USA National Heart, Lung and Blood Institute [40]. Proeschold-Bell et al. [30] used the same classification to categorise BMI, and also referred to the MetS definition of the International Diabetes Federation [41], as did Puhkala et al. [33]. Chen et al. [35] and Nanri et al. [37] employed criteria specific to their countries, Taiwan and Japan, respectively [42,43]. Studies [28,38] did not specify the clinical indicators used to evaluate MetS.
The number of workers included in the different studies varied greatly, from 35 [34] to 1390 [31]. The male to female ratio also varied: three studies [34,37,38] included only men and one [35] included only women. Men predominated in studies [27][28][29], and women in [31,35,36,39]. The male to female ratio was not reported by Woo et al. [27] or by Puhkala et al. [33]. In most of the studies, the average age of participants was around 50 years.
The 13 included studies covered several places of work, though most were conducted with office or administrative workers. The participants of six studies worked either in a hospital [27,28,32,39] a medical company [31] or insurance company [38]. Shrivastava et al. [29], Inoue et al. [34] and Nanri et al. [37] included public or private sector office workers, without providing further details. Proeschold et al. [35] recruited United Methodist clergymen and Puhkala et al. [33] studied long-distance bus and truck drivers. Allen et al. [36] enrolled university employees and Chen et al. [35] do not specify beyond the term "career women" in their article.
All but two studies included individuals with overweight or obesity, either as an isolated MetS risk factor [28][29][30]32,33] or combined with other risk factors [27,31,[35][36][37][38]. Inoue at al [34] included workers who did not perform daily physical exercise and who usually travelled by train or bus. In the inclusion criteria of [34], workers' state of health was not taken into account.

Dietary Interventions
The included studies applied a variety of interventions led by experts who were lifestyle coaches or health managers [27,[30][31][32]35,36], nutritionists or dieticians [27,29,32,37,39], nurses [28,32,37,39], physical trainers [29,38,39] or a physiotherapist [33]. In the study by Inoue and colleagues [34], a cook in the staff cafeteria led the intervention. Table 3 summarises this information and the different components of each intervention. All but one study [34] used a multicomponent strategy, classified into four main groups: 1. Basic education and general counselling on healthy habits and diet; 2. Specific diet or changes in diet and food intake; 3. Motivational changes and/or coaching; 4. Physical exercise and stress and/or sleep management. Most interventions were fully or partially implemented through online platforms and/or social networks. Week 6: IG1 showed significant decrease in WC, BMI, TC, LDL-c, health promotion behaviours and self-efficacy, but not significantly greater than in IG2 or CG. Week 12: IG1 showed significant decrease in WC, BMI, TC, LDL-c. Self-efficacy and health promotion behaviours improved to a greater degree in IG1 than in IG2 and CG.      Both groups showed improvements in fitness, BP, HDL-c and LDL-c, and a slight reduction in weight, BMI and fat mass (greater reduction in IG). In the IG, the proportion of participants with the lowest Framingham Risk Score increased from 40% at baseline to 57% after 1 year. The prevalence of MetS reduced significantly in the IG from 38% to 25%, owing to improvements in HDL-c and BP. IG participants also increased fruit and vegetable intake from 4.7 servings at baseline to 7.8 servings at 6 months and 7.0 at 1 year (all p < 0.001); decreased consumption of saturated fat, fatty meals and fried foods (p < 0.001); and significantly increased total daily PA. CG showed significant but smaller improvements in fruit and vegetable intake, saturated fat intake and PA.

Basic Education and General Counselling on Healthy Habits and Diet
Six of the 13 included studies implemented an intervention focused on offering participants basic education and general counselling on healthy habits and diet [27,29,30,33,36,37]. In general, these interventions consisted of educative sessions aimed at improving participants' knowledge of the importance of leading a healthy lifestyle. The main nutrition topics included in the sessions were related to portion control; a healthy plate model; balancing fats, proteins and carbohydrates; increasing intake of fruits and vegetables; food types in terms of glycaemic index; reducing sugar or alcohol consumption; and sampling of healthy foods.
Woo et al. [27] and Allen et al. [36] also included sessions focused on explaining MetS risk factors or how to reduce risk of cardiovascular disease, diabetes or hypertension.

Specific Diet/Changes in Diet and Food Intake
Five studies included specific diets or dietary and/or nutritional changes, implemented through different interventions. Maruyama et al. [38] and Chen et al. [35] proposed a personalised nutritional plan based on previous analysis of participants' dietary and nutritional habits, setting specific goals and recording calorie intake.
Inoue et al. [34], meanwhile, evaluated the impact of a healthy Japanese diet on the prevention and reduction of metabolic syndrome. Puhkala et al. [33] used the healthy eating plate model developed in 2011 by Harvard University [44] to help change participants' dietary habits. Racette et al. [39] used the Weight Watchers© model, consisting of group sessions, incentives and a healthy snack cart.

Behavioural Changes or Motivational Coaching
Some interventions were based on behavioural change models that aim to improve workers' health behaviours through motivation, changing health beliefs and empowering self-efficacy. Woo et al. [27] used a programme based on Rosentock's 1990 health belief model [45], while Racette et al. [34] used the transtheoretical model of behaviour change developed by Prochaska and DiClemente in 1983 [46]. Proeschold-Bell et al. [30] delivered a stress management programme developed by Williams LifeSkills [47] to improve interpersonal relationship skills, and also offered theological content supporting healthy behaviour. Kramer et al. [32] and Nanri et al. [37] also used programmes based on behavioural change theory to modify participants' lifestyles.
In three studies [27,31,38], coaching techniques were included in the interventions to improve participants' commitment and help them to maintain behavioural changes [48].

Other Interventions: Physical Activity, Stress Management and Sleep Hygiene
Eleven of the 13 interventions included a component related to physical exercise. Some were centred on regular physical counselling and increasing daily steps [27,28,[35][36][37], while others included physical activity training sessions or personalized exercise plans [29,31,38]. Kempf et al. [28] included telemonitoring of participants' physical activity

Use of Internet and Social Networks
Although this aspect is not a type of intervention per se, the interventions in most studies were fully or partially implemented through web resources, platforms or social networks to facilitate follow-up and communication with participants. These digital tools included online health platforms and study-specific web portals [23,24,26,30,31,33,35], as well as apps or e-mail reminders [26,28]. Shrivastava et al. [29] tracked adherence to lifestyle changes through smartphone messages, a digital health platform, e-mail and phone calls. Regarding the use of social networks, Woo et al. [27] used an instant messaging app called KakaoTalk, and Steinberg et al. [31] used Skype. Most studies combined telematic resources (digital platforms, mobile applications, specific websites) with face-to-face activities in participants' workplaces [29,30,32,33,38,39].
Three studies used telematic means only [27,28,36], while the interventions of Inoue et al. [34] and Nanri et al. [37] included only the traditional face-to-face method.

Results of Dietary Interventions
The effect of the dietary interventions was measured through changes in the different variables proposed as health markers: WC, BMI, TC, HDL-c, LDL-c, TG, SBP, DBP and FBG. Table 1 summarises these results.

4.
Changes in Fasting Blood Glucose (FBG, mg/dL) Fasting blood glucose was measured to assess diabetes risk. Samples were taken between 7:00 and 8:00 in the morning. Most data were provided in mg/dL and the rest we transformed from mmol/L to mg/dL.

Improvements in Prevalence of MetS or in Number of Risk Factors for MetS
Four studies analysed the change in MetS prevalence after the intervention [30,33,37,39]. In the study by Proeschold et al. [30], for participants with at least one follow-up measurement, reductions in MetS prevalence were observed in all cohorts at 24 months, ranging from 3.7 to 6.6 percentage points. In Puhkala et al. [33], the between-group differences in the prevalence curves covering 24 months of intervention (evaluation at 0, 12 and 24 months) were not statistically significant (p = 0.11). In Nanri et al. [37], MetS prevalence decreased for both the intervention and control group, from 100% at the beginning to 65.3% and 62.3%, respectively, at six months. However, the authors found no statistically significant difference when comparing the two groups (p = 0.75). Racette et al. [39] found reductions in MetS prevalence in both groups, from 38% to 25% in the intervention group and from 29% to 18% in the control group. For participants who received the intervention, the authors attributed this improvement to a reduction in blood pressure and increase in HDL-c.
Some studies assessed the change in a cluster of risk factors for MetS [29,35,36]. Most studies stated which definition they had used for metabolic syndrome.
Shrivastava et al. [29] found that in their intervention group, the number of people with three, four or five risk factors reduced, while the number of people with one or two risk factors increased. Few participants in either group reduced their risk factors to zero. Specifically, the number of subjects with three risk factors lowered from 27% to 19% in the intervention group compared to an increase from 21% to 22% in the control group. For four risk factors, this change was 14% to 8% in the intervention group versus 18% to 15% in the control group, and for five risk factors, 7% to 3% versus 4% to 5%. In Chen et al. [35] the intervention group showed better results in terms of mean number of MetS components compared with the control group (−0.6 vs. 0.1, p < 0.05).

Health Beliefs, Health Promotion Behaviours and Self-Efficacy
The programme applied by Woo et al. [27] was based on Rosentock's 1990 health belief model [45], and health promotion behaviours of workers in the intervention group showed greater improvement compared with the control group. In Racette et al. [39], all components in the intervention were based on the transtheoretical model of behaviour change (Prochaska and DiClemente, 1983) [46], which centres on individuals' tendency to adopt a healthy or unhealthy behaviours. Many participants made behavioural changes such as increasing physical activity and improving eating habits, which contributed to clinically significant health improvements. The intervention programme with coaching adopted by Maruyama et al. [38] obtained improvements in dietary habits as well as in 14 of 17 clinical parameters.

7.
Changes in Food Group or Diet/Nutrition Intake Shrivastava et al. [29] found a significant change in dietary behaviour after the intervention, with participants choosing healthier options. Consequently, they observed a decrease in total daily caloric intake (from 1823 ± 353 to 1665 ± 367 kcal) and proportion of fat in total energy (from 35% to 32%).
In the study by Inoue et al. [34], the intervention group who consumed fewer than 50 of a potential 61 healthy Japanese-style lunches had reduced their total daily intake of energy and of carbohydrates at three months compared with baseline (energy: 2554 ± 392 kcal vs. 2104 ± 393 kcal, p = 0.042; carbohydrate: 359.6 ± 85.2 g versus 295.8 ± 45.3 g). Fibre intake increased significantly in the group who had 51 or more Japanese lunches (total dietary fibre: 15.3 ± 5.2 g vs. 30.4 ± 20.9 g, p = 0.047; total vegetables: 292.4 ± 146.6 g vs. 411.1 ± 155.9 g, p = 0.035).
After implementing a programme designed to promote healthy dietary habits and physical activity, Maruyama et al. [38] observed changes in typically consumed foods in the intervention group (p > 0.01). The magnitude of the intervention effect was 0.31 for food group A (fish, soya bean/soya products, green/deep-yellow vegetables, white vegetables and mushrooms/seaweed/konnyaku) and 0.35 for food group B (large portions of grains such as rice/bread/noodles, confectionery, sweet drinks, fatty meats, meat products, butter/margarine/dressing/mayonnaise, eggs/liver, fried foods, pickles, soup and alcoholic drinks).
Chen et al. [35] performed a group and time interaction analysis at 1.5 months, finding better exercise scores in the intervention group than in the control group. Allen et al. [36] suggested reaching a goal of 10,000 daily steps to maintain good health. The intervention group had achieved a 31% improvement after 12 months compared with baseline.

Meta-Analysis
In the meta-analysis we included 11 studies and 22 study groups who received interventions.

Effect Size
The effect sizes calculated in the meta-analysis are presented in Figure 2, together with the heterogeneity test results.

Heterogeneity of the Included Studies
The analysis of the nine parameters included showed that overall, the interventions have changed the baseline values, achieving reductions in all except HDL, which increased significantly (Figure 2d). It should be noted, however, that eight of the nine variables showed considerable heterogeneity. We therefore considered it appropriate to examine this heterogeneity between studies. Table 4 presents the results of the leave-oneout analysis.

Heterogeneity of the Included Studies
The analysis of the nine parameters included showed that overall, the interventions have changed the baseline values, achieving reductions in all except HDL, which increased significantly (Figure 2d). It should be noted, however, that eight of the nine variables showed considerable heterogeneity. We therefore considered it appropriate to examine this heterogeneity between studies. Table 4 presents the results of the leave-one-out analysis. The study groups with the greatest heterogeneity contributions are the two study groups of Shrivastava et al. [29] and one group from the Kempf et al. study [28], specifically control group 1 at 365 days. Omitting the Shrivastava et al. groups produces the greatest reduction in heterogeneity in the variables TC, TG, HDL-c and FBG, while the variables most affected by the Kempf et al. control group are SBP and DBP. However, the change in both SBP and DBP does not exceed 5% when this study is omitted (1.4% and 4.3%, respectively). Similarly, the leave-one-out analysis reveals subtle changes in the heterogeneity of the variables WC and BMI. Regarding LDL-c, one of the results collected by Woo et al. [27] at 84 days produces a change in heterogeneity exceeding 10%. However, these changes in heterogeneity should be accompanied by the influence on the final result. Figure 3 depicts Baujat plots, which plot the heterogeneity contribution of each study group against its influence on the pooled result. The scales are relative, meaning the graph only serves to identify studies with considerable influence on both heterogeneity and on the final result. The numbers shown in the figure correspond to the items in the ID column of Table 4.
groups of Shrivastava et al. [29] and one group from the Kempf et al. study [28], specifically control group 1 at 365 days. Omitting the Shrivastava et al. groups produces the greatest reduction in heterogeneity in the variables TC, TG, HDL-c and FBG, while the variables most affected by the Kempf et al. control group are SBP and DBP. However, the change in both SBP and DBP does not exceed 5% when this study is omitted (1.4% and 4.3%, respectively). Similarly, the leave-one-out analysis reveals subtle changes in the heterogeneity of the variables WC and BMI. Regarding LDL-c, one of the results collected by Woo et al. [27] at 84 days produces a change in heterogeneity exceeding 10%. However, these changes in heterogeneity should be accompanied by the influence on the final result. Figure 3 depicts Baujat plots, which plot the heterogeneity contribution of each study group against its influence on the pooled result. The scales are relative, meaning the graph only serves to identify studies with considerable influence on both heterogeneity and on the final result. The numbers shown in the figure correspond to the items in the ID column of Table 4. In Figure 3 we can see that for the four variables that may compromise heterogeneity (TC, HDL, LDL and TG), only the Shrivastava et al. study groups occupy the upper right-hand corner, reflecting the greatest contribution to heterogeneity and influence on the effect.

Heterogeneity Due to Missing Studies (Publication Bias)
Another possible source of heterogeneity is publication bias. For this reason, we created funnel plots and analysed their symmetry. These graphs are shown in Figure 4.  Table 5 shows the results of the most classic methods for estimating the number of missing studies and the influence they could have on the final result. The trim-and-fill method shows publication bias in seven of the nine variables. All seven effect sizes are reduced, but only those of HDL and TG are nonsignificant after this adjustment. With the Copas method, only TC and LDL show publication bias, and the adjusted effect sizes, while reduced, remain statistically significant (Table 5).

Moderator Analysis or Meta-Regression
Heterogeneity may also be influenced by covariates or moderators. Table 6 shows the influence of the six intervention types and intervention duration. The duration of treatment is associated with a reduction of DBP and an increase in FBG. The interventions with the greatest impact on effects are types 3 and 4, which reduce BMI and increase HDL. Intervention 3 also increases FBG. Type 1 interventions increase BMI and blood pressure (SBP and DBP).

Discussion
To the best of our knowledge, this is the first systematic review and meta-analysis to group and synthesise the available scientific literature and analyse the characteristics and effects of dietary interventions aimed at reducing MetS risk in the working population. The interventions with the greatest effect are those that include physical activity and that focus on health beliefs and behaviours and workers' motivations. Although many clinical results reflect only a modest impact on MetS risk, our results suggest that interventions are beneficial on the whole.
We restricted our search to clinical trials and comparative studies because we aimed to establish a cause-effect relationship [49]. In the end we included 13 studies. The low obsolescence of these studies reflects their validity and the timeliness of our review; the data obtained (Price Index and Burton-Kebler index) indicate lower obsolescence than typically found in nutrition science bibliometric results, showing that workplace dietary interventions for reducing MetS risk is an emerging topic of interest.
Although we found substantial methodological and clinical heterogeneity between the 13 studies, we were able to include 11 in the meta-analysis. Because the sample sizes were generally small (n < 125), this meta-analysis was needed to obtain more robust conclusions.
The study participants were aged between 30 and 60 years, as would be expected in a working population. Our review included several different work places, but most participants were office workers or had jobs that did not require great physical effort (longdistance drivers, clergy, health workers), and most had overweight, obesity and/or one or more other MetS risk factors. These workers perform sedentary activities, and this together with inadequate diet can lead to overweight and obesity, increasing MetS prevalence [10]. In addition, the high responsibility of these jobs and the complex tasks often involved can generate considerable pressure and lead to occupational stress, another risk factor for metabolic syndrome [50].
In the studies included in our review, the choice of MetS definition depended on the country where the study took place. In the absence of a single definition, several closely related but individual definitions have been proposed for MetS. Four studies used the criteria of the National Cholesterol Education Program Adult Treatment Panel III (NCEP-ATP-III) [40]. These criteria are straightforward and readily measurable, making them easy to apply clinically and epidemiologically [51]. For this reason, the NCEP-ATP-III definition is among the most widely used for MetS [51]. Not all studies stated which definition they applied; some simply evaluated the presence of risk factors such as insulin resistance, obesity, atherogenic dyslipidaemia and hypertension. In any case, all studies largely conformed to the harmonised classification proposed by Alberti et al. in 2009 [2]. The criteria listed in this classification constituted the main tools for measuring the effect of interventions, in some cases together with questionnaires on food intake, physical activity or stress.
Most follow up periods lasted one year or less and were not established according to any standard. Duration was a major limitation of many studies. Several authors recognised that longer follow-up times were needed to determine the sustainability of lifestyle changes and to be able to correct deviations accordingly. [29,37]. One strength was that most interventions were led by health professionals (doctors, nurses, dieticians), which added scientific rigour to the programmes.
The dietary interventions were grouped into six main types: 1. Basic education and general counselling on healthy habits and diet; 2. Specific diet/changes in diet and food intake; 3. Behavioural change and coaching; 4. Physical exercise education and training; 5. Stress management. 6. Internet and social networks.
All but one of the programmes included more than one intervention type. This made it difficult to isolate the effect of each type on the final result, justifying the meta-regression we carried out.
Our results show that most intervention groups achieved a significant improvement in the parameters assessed. Considered together, the interventions were beneficial. However, the clinical impact of these improvements on workers' health was moderate. For example, each 1 mg/dL increase in HDL-c is thought to reduce risk of coronary death by 6%, regardless of LDL-c values [52]. Regarding blood pressure, a recent meta-analysis including data from 48 clinical trials showed that a 5 mm Hg reduction in systolic blood pressure reduces the risk of major cardiovascular events by around 10%, regardless of cardiovascular disease history, and even at normal or high-normal blood pressure levels [53]. With respect to waist circumference (another diagnostic criteria of MetS), previous studies carried out in women show that a 5 cm reduction is related to at least 10% improvement in at least one cardiovascular risk factor [54]. When we consider all this evidence, we see that the differences recorded in the reviewed studies scarcely improve workers' risk profile. It therefore seems necessary to establish which interventions are most effective for reducing MetS risk and to analyse these interventions in depth with the aim of improving future clinical results.
In our review, the interventions with the greatest impact on effect size were those that used coaching techniques and/or that applied behavioural change theory to modify eating and lifestyle habits, and those that promoted physical exercise. This result reinforces the evidence that the most effective approach to MetS is achieved by targeting diet and physical activity [55], and suggests that to optimise the results of MetS-focused workplace programmes, dietary interventions should be combined with physical exercise. This finding was to be expected: physical exercise is a powerful tool in the fight to prevent and treat numerous chronic diseases [56]. Physical activity impacts on the components of MetS, such as cardiovascular risk, blood pressure, lipid profile and blood glucose levels. Previous studies have shown a link between physical exercise and improved MetS. Haufe et al. [57] conducted a randomized controlled trial in workers at a motor vehicle company, finding that a programme of regular and telemonitored physical activity reduced metabolic syndrome and improved autonomy in the workers who took part. Tsai et al. [58] assessed the effect of a 12-week physical exercise programme on MetS components in bank and insurance workers. Their results show that intense exercise helps to improve blood pressure levels and waist circumference.
In contrast, the most common interventions, which were generally group-based and focused on offering information to participants on healthy habits and lifestyle, led to an increase in BMI and blood pressure (both systolic and diastolic). This finding has been described by other authors, who state that subjects often seem to make wrong decisions when they receive information on health risk factors [53]. Interventions must go further than simply providing information and focus more on the concept of literacy, empowering participants and offering them tools to process the information relevant to their health.
Low health literary is associated with a lack of understanding of concepts, worse management of disease and self-care activities, lower use of preventive measures, errors in compliance with treatment, and difficulty understanding health advice [59]. Giving patients information is not enough; they need to understand the information, be able to identify information that is appropriate and true, then be able to interpret and apply it according to their circumstances and personal needs [60]. To ensure educational interventions are successful, or at least to prevent a contrary effect, planners should consider the varying levels of literacy that may exist among the workers of a single company.
Our results highlight the importance of motivating and empowering workers who receive a behavioural change intervention, as well as addressing their health beliefs. The interventions that used coaching techniques or that centred on behavioural change theory had the greatest impact on the effects. These cognitive-behavioural programmes aim to change beliefs and motivate workers to acquire and maintain preventive practices. They are personalised and take into account any obstacles to the application and maintenance of preventive practices, setting specific goals for each worker, and typically following up on their achievement [48]. Several studies vouch for the effectiveness of coaching in the clinical setting. A randomised controlled trial demonstrated that this approach, compared with traditional care, significantly improved HbA1c levels in patients with diabetes [61]. A recent meta-analysis of randomised controlled trials found that behavioural treatment strategies improved adherence to lifestyle intervention programmes in adults with obesity [62]. The authors concluded that these strategies should be routinely incorporated into lifestyle interventions and obesity management and weight loss programmes to improve commitment and adherence.
Among the studies included in our review, Shrivastava et al. [29] incorporated several successful components, which probably explains why this study contributed the most to heterogeneity in certain parameters, such as TC, TG, HDL and FBG. 1. The interventions were intensive, with considerable follow-up, and were led by a team of doctors, nutritionists and personal trainers. 2. They focused on raising awareness among workers and improving knowledge, attitudes and practices to achieve the desired results. 3. The programme included sessions every two weeks on healthy habits, diet and physical activity. These sessions covered eating out, cooking methods, reading food labels and meals during festive seasons. 4. The participants received personalised advice and reinforcement. 5. Practical physical activity sessions were provided and participants were encouraged to maintain their new habits. 6. Participants were offered occupational stress management sessions, as stress is another risk factor for MetS [63]. Adherence to lifestyle changes was monitored through individual interviews and digital tools such as smartphones, an online platform, emails and repeated phone calls. This type of follow-up is missing from all the other studies. The authors achieved not only a considerable reduction in the parameters of interest, but also a change in clustering of MetS profile status, which is the ultimate objective of this type of intervention. Theirs could serve as a reference for future worksite health promotion programmes.
Previous systematic reviews have assessed the effect of lifestyle interventions on MetS [64,65] in the general population with and without MetS. The results of these reviews indicate that lifestyle modification helps to reduce MetS prevalence and the severity of its individual components. For example, Van Namen et al. [5]. carried out a systematic review and meta-regression of 15 papers reporting data on 1160 participants from 10 randomised controlled trials, to investigate the effects of lifestyle interventions-including both dietary changes and supervised exercise on outcomes for people with MetS. Compared with usual care, lifestyle interventions achieved significant improvements in waist circumference (−4.9 cm, 95% CI −8.0 to −1.7), systolic blood pressure (−6.5 mmHg, 95% CI −10.7 to −2.3), diastolic blood pressure (−1.9 mmHg, 95% CI −3.6 to −0.2), triglycerides (SMD −0.46, 95% CI −0.88 to −0.04) and fasting glucose (SMD −0.68, 95% CI −1.20 to −0.15). The authors conclude that health services should consider implementing lifestyle intervention programs for people with metabolic syndrome to improve health outcomes and prevent progression to chronic disease.
In view of these findings, both health services and workplaces appear to constitute ideal settings for implementing lifestyle intervention programmes to improve health outcomes and prevent progression to chronic diseases. This conclusion is supported by international organisations such as the WHO, which has developed the Healthy Workplace Model and Framework [66], and proposes workplace programmes as one of the key strategies to improve population health. Workplaces have the infrastructure to provide workers with chronic disease prevention interventions at different levels [67]. Interventions in this setting could therefore make a significant contribution towards reducing chronic disease risk at the population level.

Limitations of the Review
Out study is not without its limitations. One important weakness is that we were unable to retrieve the full text of some articles because they were not published in the journal's website or did not feature in the main collections. Nor were we able to retrieve these texts through the university library network or by contacting the authors. Another limitation is that the articles did not report on intervention dose or cost-effectiveness of the interventions. This information might be useful for the development of an adequate evidence base to support practice.
In addition, the methodological differences between interventions and the varying profiles of the workers studied made it difficult to interpret the results. Nevertheless, the meta-analysis and meta-regression helped us to explain possible sources of heterogeneity and to analyse and synthesis the information obtained from the different studies.
Lastly, we found some publication bias, though not enough to cancel the effects.

Implications for Future Research
It is clear that workplaces are important settings for health promotion and disease prevention. In view of our results, it seems necessary to continue investigating which interventions are most effective for preventing MetS and to continue exploring new strategies. Almost all the interventions analysed in this review are individualistic, aimed at raising awareness among workers and educating them on healthy habits. Only Inoue et al. [34] applied what can be considered a mass catering intervention, in which the intervention groups received healthy Japanese-style lunches for three months in the staff cafeteria. Despite the short intervention time, the authors observed a reduction in the parameters assessed, and this reduction was more pronounced in the participants who consumed more healthy lunches. We believe this type of intervention should be studied in greater depth. Regardless of workers' awareness and motivation, it is often difficult for them to make healthy food choices at work because the food on offer in staff canteens lacks nutritional quality and/or variety. The WHO global strategy on diet, physical activity and health suggests that workplaces facilitate healthy food choices in order to reduce daily risk exposure [68]. The European FOOD programme (Fighting Obesity through Offer and Demand) is an initiative of the European Commission that aims to improve the nutritional quality of foods consumed during working hours as a complementary strategy to individual awareness and education on healthy habits [69]. It proposes that companies, workers and restaurants work together to ensure balanced nutrition at work. This is undoubtedly an interesting avenue to explore in future research.

Conclusions
It is necessary to improve the efficiency of the dietary interventions aimed at lowering MetS risk in workers. The totality of available evidence suggests work-based interventions have a positive yet modest effect on MetS risk. The interventions with the greatest effect are those that include physical activity and that focus on health beliefs and behaviours and on workers' motivations. Purely informative or educative interventions are common but have a contrary effect. Employers must take this into account. Our results may help to guide future health promotion programmes aimed at improving workers' health to reduce risks and possible productivity losses.

Funding:
The authors received no financial support for the research, authorship, and/or publication of this article. The protocol of this systematic review has not been registered.

Conflicts of Interest:
The authors declare no conflict of interest.