Systematic review of the effect of policies to restrict the marketing of foods and non‐alcoholic beverages to which children are exposed

Summary This systematic review examined the effectiveness of policies restricting the marketing of foods and/or non‐alcoholic beverages to children to inform updated World Health Organization (WHO) guidelines. Databases were searched to March 2020. Inclusion criteria were primary studies of any design assessing implemented policies to restrict food marketing to children (0–19 years). Critical outcomes were exposure to and power of marketing, dietary intake, choice, preference, and purchasing. Important outcomes were purchase requests, dental caries, body weight, diet‐related noncommunicable diseases, product change, and unintended consequences. Forty‐four observational studies met inclusion criteria; most were moderate quality. Pooling was conducted using vote counting by direction of effect, and GRADE was used to judge evidence certainty. Evidence suggests food marketing policies may result in reduced purchases of unhealthy foods and in unintended consequences favorable for public health. Desirable or potentially desirable (for public health) effects of policies on food marketing exposure and power were also found. Evidence on diet and product change was very limited. The certainty of evidence was very low for four outcomes (exposure, power, dietary intake, and product change) and low for two (purchasing and unintended consequences). Policies can effectively limit food marketing to children; policymakers should prioritize mandatory approaches aligned with WHO recommendations.

content (e.g., branding, promotional characters, emotional appeals, and animation) in food marketing that produces such compelling commercial messages so as to influence children's behavior and health outcomes. 11 In other words, the impact of food marketing is a function of both exposure to the marketing message and its persuasive power. 12 Given this evidence of impact, and with diet-related noncommunicable disease (NCD) risk and obesity prevention in children being public health priorities in many countries internationally, bestpractice recommendations have been issued by the World Health Organization (WHO) and other authoritative bodies for governments and industry to restrict HFSS food marketing to children. In May 2010, the World Health Assembly unanimously adopted the WHO Set of Recommendations on the Marketing of Foods and Nonalcoholic Beverages to Children through resolution WHA63.14. 13 The primary purpose of these recommendations was to guide Member States in the optimal design of new policies, or in strengthening existing policies, to maximize the achievement of public health goals. Also in response to the mandate of that resolution, WHO published a framework for policymakers to support the implementation of recommendations in individual territories, 12 and WHO have led on the development of region-specific nutrient-profiling models to support policymakers in identifying products that should be restricted in marketing to children. 11 Implementation of the WHO recommendations so far has been limited, with a lack of comprehensive approaches. 14 Numerous food industry groups have established self-regulatory programs that refer to encouraging more "responsible advertising" while a small but growing number of countries have enacted mandatory policies. 15 To date, focused evaluations have suggested that self-regulation has not meaningfully reduced children's exposure to unhealthy food marketing 16 or sales of unhealthy foods. 17 Similarly, the few existing assessments of mandatory policies have reported mixed findings as to whether not the policies resulted in reductions in unhealthy food advertising in affected media 15 although effects on unhealthy food sales have been reported. 17 In some studies, decreases in HFSS advertising covered by the policy were accompanied by increases in HFSS advertising not covered by the policy such that overall exposure did not substantially change. 15 There is an urgent need to comprehensively evaluate the effectiveness of existing policies against a range of relevant indicators 12 (behavior and health but also market responses and consequences for wider society) and specifically to identify which policy design elements are most effective at achieving meaningful improvements. 15 Therefore, WHO commissioned the current review to inform the development of updated recommendations regarding policies to restrict food marketing to children.

| METHODS
We conducted a systematic review following Cochrane methods 18 reported as per Preferred Reporting Items for Systematic Reviews and Meta-Analyses. 19 The WHO Nutrition Guidance Expert Advisory Group Subgroup (NUGAG) on Policy Actions determined the research question, policy types, and outcomes to be captured by the review and ranked all outcomes for priority (see Supporting Information). Key terms were used as defined by WHO, namely, "marketing" as a commercial communication, "exposure" as the reach or frequency of the marketing message, and "power" as the creative content of marketing. 12 Policies were defined as either mandatory (legally enforceable measures including statutory approaches, regulations, legislation, or any "order" used by a jurisdiction's legal system) or nonmandatory (including self-regulatory measures, pledges, or codes). The protocol was pre-registered with Prospero in May 2019 (CRD42019132506, available from https://www.crd.york.ac.uk/prospero/display_record.php?RecordID= 132506).

| Search strategy and selection criteria
Primary studies (of any design) were considered for inclusion if they assessed implemented policies that aim to restrict (i.e., to reduce exposure and/or power of) food marketing to children aged 0-19 years compared with no policy (e.g., before the policy was implemented) or a weaker policy (e.g., partially implemented) and reported on one or more outcome of interest. Exclusion criteria were reviews of studies (narrative or systematic) and studies assessing action plans, strategies, programs, initiatives, or potential impact of policies yet to be implemented. Critical outcomes (critical for decision making 20 ) comprised exposure to food marketing, power of food marketing, and food intake, choice, preferences, or purchasing (by children or on behalf of children). Important outcomes (important but not critical for decision making 20 ) were purchase requests (by children to a caregiver), dental caries/erosion, body weight/body mass index (BMI)/obesity, diet-related NCDs (including validated surrogate indicators), product change (e.g., portion size and product reformulation), and unintended consequences to wider society (e.g., revenue and jobs). Database searches were supplemented with (i) hand searching reference lists of retrieved systematic reviews and eligible studies, (ii) contact with topic experts, (iii) forward and backward citation searching of included studies, and (iv) a WHO evidence call for data. 21 No language or date restrictions were applied.

| Quality assessment
There is no established tool for the assessment of quality for observational studies evaluating policy effectiveness, so an adapted version of the Newcastle-Ottawa Scale (NOS) was applied. Modifications were the removal of nonapplicable characteristics (e.g., selection of the non-exposed cohort) or components of characteristics (e.g., sample size calculations). Point allocations for outcome measures were not altered. Bias assessments were conducted by one reviewer and independently checked by a second. Discrepancies were discussed until a consensus was reached.

| Data extraction
Two reviewers (EB, LM) independently extracted data using prepiloted forms, and again, discrepancies were discussed until a consensus was reached. The reviewers extracted the following information: study information (e.g., authors, year, study country, funding, and conflicts of interest); study design (e.g., description of study design and media assessed [if relevant]); population (where relevant, e.g., number of participants in intervention and control groups); intervention (e.g., policy type, scope, definitions, and level of implementation); outcome measures (e.g., volume of marketing).

| Data synthesis and analysis
It was not possible to conduct formal quantitative analyses for any outcome because of the diverse range of effect measures used and the limited reporting of p values or the data required for the computation of effect sizes. 22 Therefore, vote counting based on direction of effect was adopted. 22 This necessitated the selection of one effect per outcome per study, and decision rules were used to determine the most appropriate effect (namely, the most comprehensive measure, e.g., overall unhealthy food marketing instead of marketing of individual food groups). 3 | RESULTS

| Description of included studies
A total of 31,063 titles were assessed for eligibility, and 28,682 were ineligible ( Figure 1). Of 2381 full-text articles assessed, 44 studies were included in the systematic review. Table 1 summarizes the main characteristics of the included studies; extracted outcome data are shown in Tables S1-S6). Information about the policies evaluated by included studies is provided in Table S7.
With respect to important outcomes, those reported were unintended consequences (3 studies) and product change (2 studies).
None of the included studies reported on the important outcomes of product requests, dental caries/erosion, BMI/obesity, or dietrelated NCDs.

| Interventions
All interventions were within a single category: policies to restrict children's exposure to food marketing and its persuasive power. To support the WHO guideline development process, additional comparisons were conducted ( Table 2).
The overall effect of the intervention (i.e., policy to restrict food marketing) on all available outcomes is synthesized in the GRADE Evidence Profile (Table 3) using the approach set out by Murad et al., 68 for rating certainty in the absence of a single estimate of effect (see Supporting Information) and reported as per GRADE guidelines. 69 In this section, we present the results of the synthesis of the effects of the intervention on all critical and important outcomes for  Table 4 for details), and the results are reported in full in the supplement with a brief synopsis of subgroup results for Comparison 1 provided below.  The direction of effect varied considerably across the included studies: 3 reported a clear effect favoring the intervention, 2 reported an unclear effect potentially favoring the intervention, 1 reported no effect, 6 reported an unclear effect potentially favoring the control, and 6 reported a clear effect favoring the control. We therefore judged the evidence for this outcome to have very serious inconsistency and downgraded the certainty of evidence once for inconsistency.

|
c Based on one study of only moderate quality due to methodological limitations (comparability of samples, outcome assessment). d The effect varied across the two studies: 1 reported a clear effect favoring the control, 1 reported no effect. e One of two studies used an indirect measure of marketing policy impact (cereal price).
f Based on just two studies, but one study included data on 17 brands in 6 provinces and the other included 66 cereal brands (so substantial number of data points overall), therefore, deemed "serious" rather than "very serious" imprecision.
regulations. 39 The U.K. Office of Communication (Ofcom)'s repeated cross-sectional surveys reported reduced impacts (a measure of advertisement viewing) of TV advertisements for HFSS foods in and around programming dedicated to children or "of particular appeal" to children post-U.K. Government policy compared with pre-policy. 35,36 In other studies, the mean number of TV food advertisements viewed by children per day reduced post-CFBAI implementation 38  reported that a repeated cross-sectional survey found that RCMI signatory companies were responsible for a greater proportion of noncore food advertisements as a percentage of all food advertisements on the main free to air TV channels compared with non-signatories.
Others reported a greater frequency of non-core food advertisements per hour per channel on thematic channels for children following the implementation of public health laws on food marketing using a repeated cross-sectional content analysis design. 26 In three studies all using cross-sectional survey designs, it was reported that CFBAI participating companies were responsible for a greater increase (percentage change over time) in number of confectionery advertisements viewed by children 55  Advertising Initiative (CAI) signatory companies compared with nonsignatory companies. 60 A repeated cross-sectional content analysis found a greater frequency of TV food advertisements during likely child viewing hours post-CFBAI implementation compared with preimplementation. 40 One cross-sectional survey found a greater number and proportion of food advertisements for less healthy items around TV programs with a child audience share ≥35% being from CAI signatory versus non-signatory companies, 61 and another cross-sectional survey found a greater proportion of food ads that are "high in nutrients to limit" during children's programming by CFBAI signatories versus non-signatories. 58 Three studies (all assessing the effect of voluntary policy vs. no policy) clearly favored the control, reporting undesirable effects (i.e., increased exposure to food marketing) including a repeated cross-sectional content analysis that found significantly increased proportions of non-core food advertisements on TV channels popular with children and adolescents following implementation of the EU Pledge. 29 Two Canadian studies evaluated the marketing by CAI signatory companies versus non-signatories. One, using cross-sectional content analysis and survey, found a significantly greater number and proportion of less healthy food advertisements by signatory companies during children's preferred television, 49  of these, one clearly favored the intervention, and five potentially favored the intervention (see Figure 2). Certainty of evidence was deemed very low (Table 3).

| Critical outcome: Power of food marketing
Sixteen studies within this comparison (any policy vs. no policy) reported on power outcomes.
Two studies reported a clear effect favoring the intervention, with desirable effects on power of food marketing (i.e., reductions) as a result of the policy. Both studies used repeated cross-sectional content analysis designs. Kunkel et al. 66 reported that a significantly lower proportion of TV advertisements for unhealthy foods ("whoa" foods, as a proportion of all food advertisements) featured a licensed character following implementation of the CFBAI (Table S7) (Table S7)  reported an 84% decrease in child impacts for TV food advertisements featuring a licensed character post-policy (full implementation) compared with pre-policy.
One study found no effect. reported on a repeated cross-sectional content analysis that found a significant increase in the propensity for non-core food TV advertisements to contain a promotional character following implementation of the EU Pledge compared with pre-implementation, and Neyens and Smits 57 similarly reported a significantly greater presence of spokes characters on the websites of EU Pledge signatory companies compared with non-signatories using a cross-sectional content analysis design. Galloway and Calvert 52 reported on a repeated cross-sectional content analysis that found that compared with non-signatories, CFBAI companies marketed more products with media characters on "Whoa" (nutritionally deficient) than "Go" (highest nutritional content) products and on "Whoa" than "Slow" (nutritionally beneficial but limit consumption) products. Potvin Kent et al.'s cross-sectional content analysis and survey 49 found that the number (and proportion) of TV food advertisements featuring media characters that were for less healthy foods was significantly higher for CAI signatory than non-signatories, and Warren et al. 40 found that the proportion of TV food advertisements featuring the production technique of animation was 10% higher following CFBAI implementation compared with preimplementation using a repeated cross-sectional content analysis design.
Three of 16 studies were judged to be high quality, and of these, none clearly favored the intervention and two potentially favored the intervention (see Figure 2). Certainty of evidence was deemed very low (Table 3).

Subgroup analyses for power (see Supporting Information for details)
For policies where children sought for protection were 12 years or under, one study (of 12) clearly or potentially favored the intervention. Where policies also sought to protect children over 12 years, three studies (of four) clearly or potentially favored the intervention.
For policy effects on power of unhealthy food advertising on television, three studies (of 10) clearly or potentially favored the intervention; for digital marketing, no study (of three) clearly or potentially favored the intervention; and for product packaging, one study (of three) clearly favored the intervention.
For studies evaluating policy effects on use of promotional characters in unhealthy food marketing, three studies (of 10) clearly or potentially favored the intervention. For child-appealing persuasive strategies, one study (of five) clearly or potentially favored the intervention, and for the production technique of animation, the single study identified did not clearly or potentially favor the intervention.

| Critical outcomes: Food preferences and choice
None of the included studies reported these outcomes.

| Critical outcome: Food purchasing
Four studies within this comparison reported on purchasing outcomes.
A repeated cross-sectional survey 31

| Critical outcome: Dietary intake
One study within this comparison reported on dietary intake outcomes.
This repeated cross-sectional content analysis and survey reported that the self-reported consumption score of children (9-16 years) for potato chips was significantly lower postimplementation of the Singapore Code of Advertising Practice compared with pre-implementation. An effect reported to be driven by the 13-16 year old's in the sample as no significant difference was found for those aged 9-12 years. 33 In summary, for the dietary intake outcome, one moderate quality study clearly favored the intervention 33 (100.0% [95% CI 5.5% to 100%], p = 1.00). Certainty of evidence was deemed very low (Table 3).
3.3.6 | Important outcomes: Product requests, dental caries/erosion, body weight/BMI/obesity, dietrelated NCDs None of the included studies reported these outcomes.

| Important outcome: Product change
Two studies within this comparison reported on product change outcomes.
One cross-sectional survey 62 evaluated the sugar content of all ready-to-eat breakfast cereals (excluding granolas) available from two U.S. grocery stores between signatory and non-signatory companies of the CFBAI, finding that mean sugar content per ounce was significantly higher for signatories.
Another cross-sectional survey reported that there was no significant difference in the average price of children's brand breakfast cereals per 100 g between Canadian provinces with no regulation (analysis conducted pre-CAI implementation) and Quebec, subject to the Quebec Consumer Protection Act (see Table S7). 47 In summary, for the product change outcome, no studies clearly favored or potentially favored the intervention. One study was deemed to be high quality; this found no effect. 47 Certainty of evidence was deemed very low (Table 3).

| Important outcome: Unintended consequences
Three studies within this comparison reported on unintended consequences.
One repeated cross-sectional survey reported that there was a statistically significant reduction of £15.2 million in TV HFSS advertising expenditure following implementation of the U.K. regulations, compared with pre-implementation. 44 One repeated cross-sectional content analysis of the mandatory policy from the Republic of Korea narratively reported that the total amount of money invested in TV advertising for energy-dense nutrient poor food promotion during regulated hours fell pre-post policy 32 and one repeated cross-sectional survey narratively reported that there was a 26% reduction in net food and drink advertising revenue on children's channels pre-to post-U.K. Government policy implementation. 35 In summary, for the unintended consequences outcome, one study clearly favored the intervention 44  to 100%], p = 0.248); all were high quality. Certainty of evidence was deemed low (Table 3).

| DISCUSSION
This review identified and synthesized evidence from 44 observational studies evaluating policies to restrict the food marketing to which children are exposed. Evidence suggests food marketing policies may result in reduced purchases of unhealthy foods and in unintended consequences favorable for public health. Desirable or potentially desirable (for public health) effects of policies on food marketing exposure and power were also found. Evidence on diet and product change was very limited. Overall, the certainty of the evidence was very low for four of the six outcomes (exposure, power, dietary intake, and product change) for which data were available and low for the two remaining outcomes (purchasing and unintended consequences).
The use of GRADE in this context has the potential to underestimate the certainty of evidence. The relatively low ratings reflect not just the inconsistency of effects (study heterogeneity) and, for some Powell et al. 58 was a cross-sectional survey comparing marketing activity between signatory and non-signatory companies of the CFBAI and reported potentially undesirable effects of the policy on exposure.
In the case of voluntary measures, there may be other differences between signatory and non-signatory companies (e.g., size of business and product portfolio) that can influence marketing practices and may therefore confound any observed effects 72,73 such that studies comparing marketing activity pre-post policy implementation may be more informative.
The observed heterogeneity also likely stems from the sampling approach used in studies, which can also reflect fundamental disagreements between different actors as to what the aims of food marketing regulation are or should be. For example, Ofcom,35 Ofcom, 36 Adams et al., 24  viduals so may be considered to be a better measure of exposure than count data, 24 whereas five "impacts" could be five children seeing the same single advert once or one child seeing it five times-each of which may have differential effects on behavioral outcomes. The concept of "unhealthy food" is also defined differently across studies including core/non-core (e.g., Whalen et al. and Hebden et al. 41,63 ), energy-dense, nutrient poor (e.g., Kim et al. 32 ), "high in" (e.g., Dillman Carpentier et al. 28 ), HFSS (e.g., Ofcom 35,36 ), which is sometimes not the same as the way in which foods are classified as unhealthy in the policy (e.g., Whalen et al. 41 use the core/non-core classification in the evaluation of a policy that uses a nutrient profile model to define HFSS foods). More generally, it is important to note that the synthesized effect measures vary in terms of the extent to which they capture differences in exposure by the specific population targeted by the policies, for example, advertising GRPs for "children's advertising" during regulated timeslots (e.g., Kim et al. 32 ), syndicated data for certain age groups (e.g., Dembek et al. 27 ), or advertising on popular channels or during peak child viewing times, such that child exposure is inferred rather than measured directly (e.g., King et al. 65 ).
According to GRADE, the evidence in this review is very uncertain about the effect of food marketing policies on power of food marketing. This heterogeneity also reflects the differences in study design and sampling discussed above for the exposure outcome and the selection of individual effect measures per outcome for synthesis in this review. Similar to the evidence discussed above for exposure, specifically the migration of advertising to adult airtime (which may be considered a "spill over" effect of the policy), Ofcom35 and Ofcom36 both reported that while the use of promotional characters in TV food advertising reduced post-policy compared with prepolicy, the use of celebrities increased in parallel (by more than 100%). Therefore, the specific "power" component(s) selected for inclusion in individual studies and syntheses (such as the current review) could affect whether a policy was analyzed as effective or not. There was also some limited evidence that the level of implementation of policies can affect policy efficacy. Ofcom35 was conducted when the U.K. policy was partially implemented, and Ofcom36 was conducted upon full implementation of the same policy. Both studies reported unclear effects on exposure and power potentially favoring the intervention, but, as would be expected, the magnitude of the effect was greater at full implementation (52% decrease in exposure effect, 84% decrease in power effect) than at partial implementation (39% decrease in exposure effect, 56% decrease in power effect).
In terms of policy design elements, subgroup analyses indicated that studies were more likely to report effects favoring the intervention when evaluating mandatory policies, policies designed to restrict food marketing to children including those >12 years, policies addressing television advertising, and policies using a nutrient profile model as the basis with which to classify foods to be restricted. Studies were also more likely to report desirable effects of policies on the use of promotional characters than on other marketing techniques.
There is some overlap between these policy components-of the eight studies evaluating policies in which the age of a child included those over 12 years old, seven were reporting on mandatory policies, 24,26,28,32,35,36,39 and three reported on a mandatory policy using a nutrient profile model. 24 shown to yield a reasonably accurate estimate of overall diet quality but not dietary intake of specific nutrients. 76 According to GRADE, the evidence in this review is very uncertain about the effect of food marketing policies on product change given the inconsistency, indirectness, and imprecision of the data.
Only two studies reported on this outcome; one reported a clear effect favoring the control, as mean sugar content was significantly higher in breakfast cereals by signatory versus non-signatory companies of the CFBAI 62 and one study reported no effect of the Quebec Consumer Protection Act on the average price of children's breakfast cereals. 47 There is evidence that other forms of public health policy (e.g., a soft drinks industry levy) drive product reformulation, 77  However, caution is to be used when interpreting changes in advertising expenditure as trends observed can result from reduced advertising costs (e.g., due to increased numbers of commercial television channels and therefore greater availability of advertising spots) rather than a change in the amount of advertising activity taking place. 35,36 None of the included studies reported on food preferences, food choice, product requests, dental caries/erosion, body weight/BMI/ obesity, or diet-related NCD outcomes. As with dietary intake, it is challenging for studies to disentangle effects of policies on these outcomes from effects of secular trends. Modeling, or data simulation, studies were not eligible for inclusion in the current review, but recent studies using this approach have indicated that food marketing policies may significantly reduce obesity in children (via reduced consumption of unhealthy food products). 78, 79 We could not conduct formal subgroup analyses by socioeconomic status, age, gender, or rural/urban residential status of participants or country income because enough studies did not (i) separate data by these variables in similar ways and (ii) provide/conduct analysis based on these variables.
Although focused comparisons and subgroup analyses (including those reported in the supplement) have allowed us to draw some conclusions about which policy components appear important for policy efficacy, more research will be needed to enable an in-depth analysis.
However, there are several key policy design components that may affect the effectiveness of policies to meet regulatory goals, namely, the age range of children sought for protection and how marketing to this age group is defined, which media are restricted, which forms of marketing are restricted, and how the specific foods and beverages to be restricted are classified. These issues have been discussed extensively in previous reviews of policies and their implementation (e.g., WHO Europe and Taillie et al. 14,15 ), but it is important to note that the evidence presented in the current review suggests that if a Government was to fully implement and enforce a comprehensive mandatory policy fully aligned with the spirit of the WHO recommendations, then it is likely that the policy would be effective as assessed against key indicators. 12

| Strengths and limitations
The strengths of this work are that it updates and builds upon the (as is reflected in this review), and therefore, it is difficult to separate the specific effect of a policy from that of an ongoing secular trend. Some studies (e.g., evaluations of the Quebec Consumer Protection Act) took place years after the implementation of the policy (in 1980) so were cross-sectional comparisons (i.e., comparing groups more or less likely to be exposed to the effects of the regulation, e.g., English-speaking and French-speaking households in Quebec) rather than pre-post implementation. Study designs may not be ideal in their ability to provide the evidence that policymakers are looking for because researchers are responding to real-life events, often with limited time to collect pre-policy data and without the facility to randomize to intervention conditions.
The use of vote counting, though necessitated by the data available, is a limitation, as the selection of single effects per outcome per study limits the ability of a synthesis to address the complexity of the evidence. Also, this method allows a conclusion to be drawn as to whether there is any evidence of an effect but provides no information on the magnitude of effects and does not account for differences in the relative sizes of the studies.

| CONCLUSION
Restrictions in food marketing may result in reduced purchases of unhealthy foods by or on behalf of children 31,33,44,48 and in unintended consequences favorable to public health. 32,35,44 Several studies in this review also report desirable (or potentially desirable) effects of food marketing policies on child food marketing exposure 27,28,[30][31][32][33][35][36][37][38][39]41,46,63,64 and/or power. 35,36,39,45,66 The certainty of evidence was very low for four outcomes (exposure, power, dietary intake, and product change) and low for two (purchasing, unintended consequences), largely due to inconsistency of effects (study heterogeneity). Mandatory policies, those seeking to protect children beyond 12 years of age, restricting unhealthy food advertising on television and using a nutrient profile model to classify foods were more likely to be evaluated as effective. Given that the impact of food marketing is a function of both exposure and power, 12 and there is substantial evidence to demonstrate that food marketing is extensive, powerful, and impactful upon behavior, 8,82 it is positive for public health that these findings show it is possible for policies to effectively reduce the food marketing to which children are, or are likely to be, exposed and its persuasive power. The challenge for policymakers lies in designing and implementing policies with maximum potential for efficacy.

ACKNOWLEDGMENTS
This article was supported by funding from the World Health Organization. The WHO NUGAG Subgroup on Policy Actions specified the PICO criteria (including exposure and outcome measures) and confirmed or modified the certainty judgments but had no role in data collection, data analysis, other data interpretation, or writing of the article.

CONFLICT OF INTEREST
We declare no competing interests.

AUTHOR CONTRIBUTIONS
EB was responsible for the systematic review, wrote the manuscript, and was involved in the interpretation of results. AJ was responsible for the statistical analyses and was involved in interpretation of results. LM, MM, JH, and AB were involved with the systematic review and the interpretation of results. EB and LM accessed and verified the data. All authors were involved in devising and agreeing the final protocol for this work, had full access to all the data in the study, had final responsibility for the decision to submit for publication, reviewed and commented on the draft manuscript, and approved the submission of the final manuscript.