A systematic review of the influences of food store product placement on dietary-related outcomes

Abstract Context Product placement strategies have been used to influence customers’ food purchases in food stores for some time; however, assessment of the evidence that these techniques can limit unhealthy, and promote healthy, food choices has not been completed. Objective This systematic review aimed to determine how product placement strategies, availability, and positioning, in physical retail food stores located in high-income countries, influence dietary-related behaviors. Data Sources From a search of 9 databases, 38 articles, 17 observational studies, and 22 intervention studies met the study inclusion criteria. Data Extraction Two reviewers independently extracted data relating to study design, study population, exposures, outcomes, and key results. Each study was also assessed for risk of bias in relation to the research question. Data Analysis Meta-analysis was not possible owing to heterogeneous study designs and outcomes. As recommended by Cochrane, results were synthesized in effect direction plots using a vote-counting technique which recorded the direction of effect and significance level according to the expected relationship for health improvement. Conclusions The majority of studies showed that greater availability and more prominent positioning of healthy foods, or reduced availability and less prominent positioning of unhealthy foods, related to better dietary-related behaviors. A large number of results, however, were nonsignificant, which likely reflects the methodological difficulties inherent in this research field. Adequately powered intervention studies that test both the independent and additive effects of availability and positioning strategies are needed. Systematic Review Registration PROSPERO registration no. 42016048826


INTRODUCTION
The current food environment is obesogenic and encourages individuals to habitually overconsume foods in a way that is inconsistent with dietary recommendations. 1 In the late 1990s, it was identified that modern food environments heavily promote the sale and intake of energy-dense, nutrient-poor foods and beverages. 2 It took until 2007 for the first significant government document, the Foresight report, to highlight the key role of food environments in fueling obesity. 3,4 Although published in the United Kingdom, this report has had international impact. Yet now, more than a decade later, food environments remain obesogenic and obesity levels continue to rise worldwide. 5 Human behavior is often responsive to environmental stimuli in settings frequently visited. 6 Food stores, such as supermarkets and corner stores, are the main sources of food for many people living in highincome countries; they are likely having a significant influence on the food choices of their consumers. 7 Marketing strategies are used extensively in food stores and commonly comprise the 4 Ps of marketing: product, price, promotion, and placement. 8 Product placement strategies have been used to influence customers' purchases in these stores for some time, and their successful effects have been documented in the marketing literature. 9,10 Assessment of the evidence that these techniques can be successfully used to limit unhealthy, and promote healthy, food choices has not yet been completed. Gray literature suggests that two-thirds of placement strategies are used to promote unhealthy food and beverages in supermarkets. 11 Comprehensive assessment of academic research examining the healthrelated effects of placement strategies on store-level food sales, household-level food purchasing, or individual-level dietary outcomes would help guide future government intervention across the world. Some governments are already taking, or considering taking, legislative actions against food and beverage placement promotions. For example, Chapter 2 of the UK government's Childhood Obesity Plan, released in 2018, included a population-level proposal to ban marketing strategies used in food outlets that promote the overconsumption of unhealthy foods and beverages. 12 A number of systematic reviews have narratively summarized the influence of supermarket interventions on diet-related outcomes. Existing reviews have largely examined the evidence for intervention strategies related to product price and healthier product promotion, including product swaps, product signage, and product labeling. [13][14][15][16][17][18] Only a very small number of studies included in these reviews assessed the role of product placement on dietary and food purchasing behaviors, and no quantitative evidence synthesis has been conducted to date. Reviews of observational research investigating the association between in-store food retail environments and dietary-related outcomes have not exhaustively examined product placement either -primarily because the literature in this area has grown rapidly since 2012, when 2 critical reviews on this topic were published. 19,20 Policy makers would benefit from a systematic review of recent observational and intervention research investigating the role of product placement strategies in retail food stores on outcomes related to health, such as food sales, purchasing, dietary intake, or BMI.
According to the Typology of Interventions in Proximal Physical Micro-Environments (TIPPME), product placement consists of 2 distinct intervention types: availability and position. 21 Availability describes the addition or removal of products to increase or decrease their variety, number, or range. Position refers to altering the position, proximity, or accessibility of products, rendering them easier or harder to engage with.
There is some evidence to indicate that public health strategies that alter environmental influences on health behaviors may be more equal in their effectiveness across socioeconomic groups than those requiring conscious or reflective engagement, which appear most beneficial for more advantaged groups. 22 Assessing whether product placement strategies in retail food stores has a differential effect on dietary-related behaviors could provide important evidence to help address dietary inequalities. Thus, this systematic review aims to adopt a quantitative approach to answer the following questions: 1. Does an association exist between the availability of healthier and/or unhealthy food products in retail food stores and BMI, dietary behaviors, purchasing, and sales of these foods? 2. Does an association exist between the prominent positioning of healthier and/or unhealthy food products in retail food stores and BMI, dietary behaviors, purchasing, and sales of these foods? 3. Do these associations differ according to socioeconomic position?

Recommendations made by the Preferred Reporting
Items for Systematic Reviews and Meta-Analyses (PRISMA) group were followed throughout this review. 23 Table S1 (please see the Supporting Information online) shows the PRISMA checklist for the review. This systematic review was registered with the Prospective Register for Systematic Reviews (PROSPERO) CRD: 42016048826.

Data sources
Nine electronic databases were searched (Medline, DARE, Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, EBSCOhost, PsycINFO, Science Direct, EconLit, and Scopus). A combination of medical subject headings (MeSH) and free-text terms relating to "diet," "feeding behavior," "food," "beverages," "food supply," and "food industry" were used to identify studies that described the association between in-store food environments and diet, sales, purchasing, and BMI outcomes in adults. All studies were published in English between January 2005 and February 2019. A landmark article describing different types of food environments in relation to diet and health was published in 2005, prior to this date little research was published in this area. 24 The inclusion of this time frame captures the most recent literature in the research field, reflects current food store layouts and provides useful and applicable evidence for policy makers. The complete search strategy and list of search terms can be found in Table S2 (please see the Supporting Information online). All titles and abstracts were screened by one author (S.C.S.) against the study PI(E)COS (population, intervention/exposure, comparison, outcome, and study design) criteria to ensure eligibility for inclusion (Table 1). Observational and intervention studies were included if they involved adult participants (aged 18 years and older), were conducted in high-income countries, included an exposure/intervention that investigated either the positioning or availability of food items in physical food stores and had an outcome relating to food sales, purchasing, dietary intake, or BMI. If it was unclear from the abstract alone whether an article was eligible for inclusion, the full text was reviewed. The bibliographies of the included studies were also screened for additional articles.

Data extraction and risk-of-bias assessment
Data were extracted to capture the relevant information for the research questions. Separate data extraction forms were created for observational studies and intervention studies. The full text for each article was assessed independently by 2 reviewers (S.C.S. and C.A.V.). Details regarding the study characteristics (study design, setting, participant details, exposures, outcomes, results, and funding sources) were extracted. Concurrent with the data extraction, risk of bias was assessed for each eligible article to determine the risk of bias in relation to the research questions. This process was conducted using predefined assessment criteria based on those described by the NHS Centre for Reviews and Dissemination. 25 Separate risk-of-bias assessment criteria were used for observational and intervention studies. Thirteen domains in observational studies and 18 domains in intervention studies assessed the elements of study design, participant selection, attrition, assessment methodologies, statistical analyses, and handling of confounding effects (Tables S3 and S4; please see the Supporting Information online). A risk-of-bias score of þ1 (low risk of bias), 0 (medium risk of bias), or À1 (high risk of bias) was allocated for each domain. If, for any reason, an element of the assessment criteria was not applicable, a score of 0 was applied for that domain. For example, a 0 rating was applied for the "cohort follow-up percentage" domain if the study was cross-sectional. The reviewers (S.C.S. and C.A.V.) compared the risk-of-bias assessment ratings for consistency. Any discrepancies were discussed in depth until a quality score was agreed. An overall risk-of-bias score was allocated to each study based on the number of "À1" ratings a study received. Intervention studies with 6 or more À1 ratings, and observational studies with 5 or more À1 ratings, were classed as having a high risk of bias overall. If the number of À1 scores was 2 for intervention studies or 1 for observational studies, the overall risk-of-bias score was classified as low. Intermediate ratings were allocated a moderate overall risk-of-bias score.

Data synthesis
Separate summary tables were created for observational and intervention studies (Tables S5 and S6; please see the Supporting Information online). Each study was categorized according to the placement strategy (availability or positioning) of the exposure or intervention. Availability and positioning were defined according to the TIPPME recommendations. 21 Studies were further classified to reflect whether the exposure/intervention focused on the "placement of healthy foods" or "placement of unhealthy foods." These categorizations were based on the Eatwell Guide (Public Health England). 26 Foods that were inadequately described or did not clearly align with this guide were excluded from this review. For example, popcorn, vegetarian pepperoni, and wheat-square cereal -covered in the study by Holmes et al 27 -were not included in the quantitative assessment as an expected direction for health outcomes could not be determined. Exposures/interventions that considered both healthy and unhealthy food items together were categorized under "placement of healthy and unhealthy food." Meta-analysis was not possible owing to the heterogeneity of the study designs, exposures/interventions, and outcomes. A vote-counting method was therefore used to summarize the findings of this review. Cochrane's advice for accurate vote counting was followed throughout this review and requires that each study's effect estimates are categorized according to their direction in terms of showing a health benefit or harm, in order to produce a standardized binary metric. 28 This systematic review hypothesized that greater availability and/or more prominent positioning of healthy foods resulted in a benefit for health through greater sales/purchasing/consumption of the healthy food, reduced sales/purchasing/consumption of unhealthy food items, or lower BMI. In addition, this review also hypothesized that reduced availability and/or no prominent positioning of unhealthy foods resulted in a benefit for health through decreased sales/purchasing/consumption of the unhealthy foods, increased sales/purchasing/consumption of healthy foods, or lower BMI. Each outcome result from an article was classified as either positive (supports hypothesis) or negative (rejects hypothesis). In cases where the direction of the outcome result could not be determined, results were categorized as inconclusive. Each article's results were further classified according to the significance level (significant P 0.05 or nonsignificant P > 0.05). Only results that were deemed to be relevant to the research question were extracted during the data synthesis process. The vote-counting results were summarized visually using bar charts and in detail using effect direction plots, as recommended by Cochrane. 28 Effect direction plots indicate studies that report on more than one similar outcome (diet, sales, purchasing, BMI) in a way that is not captured by a bar chart.
Arrows were used in effect direction plots to represent the combined direction and significance level of outcomes for each study. The method of combining results was based on previous criteria for variation in effect and significance 29 : If !70% of outcomes report similar direction use an arrow (~[positive] or ! [negative]) to represent the direction.
• If <70% of outcomes report a similar direction, use a diamond (᭛) to represent inconsistent results. • If effect direction similar AND >60% outcomes are statistically significant, use a solid arrow (~) to represent a significant result. • If effect direction similar AND <60% of outcomes are statistically significant, use a hollow arrow (~) to represent a nonsignificant result. Figure 1 is a PRISMA diagram representing the literature search process. After removal of duplicates, 16 342 references were identified from the 9 databases searched. A further 2 articles were included from bibliographic review. All titles and abstracts were screened and 69 full-text articles were reviewed for eligibility. Thirty-one articles were excluded because of insufficient detail or inappropriate population, exposure/intervention, or outcome. Overall, 38 articles were deemed appropriate for inclusion. These articles described 17 observational studies and 22 intervention studies. Two of the intervention studies were reported in the same article but used different data sources and addressed different research questions. This article was therefore treated as 2 separate studies in this review (these studies are presented throughout as 'Ejilerskov et al 2018a1' and 'Ejilerskov et al 2018a2'). 30

Study characteristics
Observational studies. Publication dates of the included observational studies ranged from 2008 to 2017. In total, 13 769 participants and over 1487 food stores were studied in the included observational literature. All but one of the observational studies had a cross-sectional design (n ¼ 16; 94%).  Table S6 (please see the Supporting Information online) provides a detailed summary of the study design, study setting, participant demographics, key findings, and quality for all intervention studies.
Research question 1: Does an association exist between the availability of healthier and/or unhealthy food products in retail food stores and BMI, dietary behaviors, purchasing, and sales of these foods?
Observational studies. As shown by the effect direction plot in Table 2, 14 observational studies (82%) assessed food availability, 50-55,57-66 of which over half (n ¼ 8/14; 57%) 50,52,59-64 found that product availability in retail food outlets was associated with outcomes that supported the review hypotheses and showed health benefits (3 positive significant [þs] outcomes, 5 positive nonsignificant [þns] outcomes). Two studies reported Of the 13 studies (n ¼ 13/14; 93%) that assessed availability of healthy food products, 50,52-54,58-66 57% (n ¼ 8/14) showed results in the expected direction for health (5 þns, 3 þs). 50,52,[59][60][61][62][63][64] In addition, one study (n ¼ 1/14; 7%) assessed the availability of healthy food items separately to unhealthy items. The results showed a nonsignificant positive relationship between unhealthy food availability and BMI but inconsistent findings for the availability of healthy foods and BMI. 66 One study showed that having a greater proportion of shelf space dedicated to fruit and vegetables, compared to unhealthy drinks and snacks, was associated with healthier purchases (þns). 60 One study assessed the availability of chocolate and confectionery, finding an overall inconsistent relationship with the consumption of these items -specifically, a nonsignificant positive association for confectionery exposure and confectionery consumption but no clear trend between chocolate exposure and chocolate consumption. 57 The 4 studies that considered purchasing outcomes demonstrated the most consistent support of the review hypotheses; 80% (n ¼ 3/4) found significant positive associations 62-64 and 20% (n ¼ 1/4) nonsignificant positive associations. 60 Half of the studies (n ¼ 4/8; 50%) with diet outcomes indicated a relationship with food availability in the expected direction for health benefit; however, none were statistically significant. 50,52,59,61 BMI showed no clear relationship with food availability; 2 of the 3 studies showed inconsistent results, 54,66 and one identified a nonsignificant relationship between greater healthy food availability and higher BMI. 53 Reported effect direction and significance for multiple outcomes: -All outcomes report effect in same direction and with same level of statistical significance OR -Where direction of effect varies across multiple outcomes: Overall result direction determined if !70% of outcomes report similar direction; Overall result significance level determined if !70% of outcomes report similar statistical significance Inconsistent findings rated as inconsistent if <70% of outcomes report consistent direction of effect (᭛).
Intervention studies. Table 3 shows the effect direction plot for intervention studies. Four intervention studies (n ¼ 4/22; 18%) described manipulation of food availability. 36,37,43,44 Of these, none showed results in the expected direction for health benefit. One study had results in the unexpected direction (negative significant, Àns), 36 and 3 showed inconsistent results. 37,44,61 All 4 of these studies targeted the availability of healthy foods, with none reducing the availability of unhealthy food items; additionally, all 4 were implemented as part of multicomponent interventions. 36,37,43,44 For the studies assessing sales/purchasing (n ¼ 3/4, 75%), inconsistent results were observed in 2 studies, 37,44 and one study showed that increasing the availability of healthy food items resulted in reduced sales of these items (Àns). 43 In the 2 studies assessing dietary outcomes (n ¼ 2/4; 50%), one found results in the unexpected direction (Àns), 36 and the other showed inconsistent results. 43 The one study (n ¼ 1/4; 25%) that assessed BMI found that improving the availability of fruit, vegetables, low-fat milk, and whole-grain products in convenience stores resulted in a nonsignificant Sales represents sales/purchasing. d~P ositive result (P < 0.05);~positive result (P > 0.05); ! negative result (P < 0.05); ! negative result (P > 0.05); ᭛ inconsistent results. Number of outcomes within each category is 1 unless indicated in subscript beside effect direction. Reported effect direction and significance for multiple outcomes: -All outcomes report effect in same direction and with same level of statistical significance OR -Where direction of effect varies across multiple outcomes: Overall result direction determined if !70% of outcomes report similar direction; Overall significant level determined if !70% of outcomes report similarstatistical significance Inconsistant findings rated as inconsistent if <70% of outcomes report consistent direction of effect (᭛).
increase in BMI among intervention customers compared with control customers. 43 Research question 2: Does an association exist between the prominent positioning of healthier and/ or unhealthy food products in retail food stores and BMI, dietary behaviors, purchasing, and sales of these foods?
Observational studies. Seven observational studies (41%) provided no description of the socioeconomic backgrounds of the study area or study participants. Of the 10 (59%) that reported socioeconomic data, 5 were conducted in study areas with varying levels of socioeconomic position (SEP) 55,57,58,63,65 and 5 were conducted with participants of lower SEP or in areas of lower SEP. 51,53,59,62,64 Only one study had an inclusion criterion that specifically targeted low-income participants. 53 No observational studies explicitly examined the interaction between SEP and food placement strategies. However, from the studies conducted amongst predominantly disadvantaged groups (ie, low income, high prevalence of government assistance, or deprived area), findings showed consistently that healthier placement strategies were associated with better diet and sales outcomes, but the association with BMI outcomes was inconsistent.
Intervention studies. Fifteen intervention studies (71%) described the socioeconomic backgrounds of the study area or study participants. Twelve of the studies (n ¼ 12/15; 80%) reporting socioeconomic data were conducted in deprived neighborhoods, and 3 (n ¼ 3/15; 20%) among populations of varying SEP. Only one study specifically analyzed the differential intervention effects according to household social class (occupation of highest earner) and found no clear trend across social class quintiles. 42 In the 12 studies that focused on populations of lower SEP, half showed results indicating that the intervention was beneficial for health. The other half of the studies, however, showed inconsistent results.

Risk of bias
Tables S5 and S6 (please see the Supporting Information online) present the risk-of-bias assessment results for each article.

Summary of findings
This systematic review is the first to consider the overall direction of effect for food placement strategies on healthy eating behaviors. Considering the need for action on the complex public health concerns of poor diet and obesity, and the difficulties in conducting randomized controlled trials in this research field, this systematic review finds moderate evidence using a practicebased evidence perspective, 67 from both observational and intervention studies, for food placement strategies in food retail stores positively influencing healthy eating behaviors. This review indicates weaker, but still meaningful, evidence of an effect when adopting the more traditional evidence-based practice approach. Although the majority of findings showed that greater availability and more prominent positioning of healthy foods, or reduced availability and less prominent positioning of unhealthy foods, related to better dietary behaviors, many were not statistically significant. The small sample size and lack of power demonstrated in many of the studies, particularly the intervention studies, may have contributed to the high number of nonsignificant results, and likely indicate the difficulties in conducting these types of field studies with a high-quality scientific design.
Analyzing the results with greater granularity, according to placement type the literature reveals moderate observational evidence for an association between product availability and dietary-related outcomes in the expected direction for health; evidence from intervention research was more limited and equivocal. A large proportion of both observational and intervention literature focused on improving the availability of healthy foods; hence, drawing conclusions on the effectiveness of solely limiting the availability of unhealthy foods is not yet possible. Both observational and intervention literature indicated moderate evidence for product positioning strategies in food stores affecting dietary-related health outcomes; most intervention studies indicated that more prominent positioning of healthy foods, and less prominent positioning of unhealthy foods, results in better dietary, or healthier sales, behaviors. Good evidence from the intervention studies included in this review exists to support strategies that combine the availability and positioning of both healthy and unhealthy foods to provide benefit for health. A number of interventions were multicomponent, which does somewhat weaken the conclusions we can draw about the effect of availability and positioning measures alone; however, the majority (80%) of the multicomponent interventions in this study showed findings in the expected direction for health. Sales outcomes, which were assessed in the majority of intervention studies and in almost half of the observational studies, provided the most consistent results in the expected direction for health benefit. The least abundant and least consistent evidence was found for BMI outcomes; this is perhaps not surprising given the multiple determinants of body weight and the challenges of assessing BMI in largescale studies.
Observational studies suggested that strategies to improve the placement of healthy foods, and limit the placement of unhealthy foods, could have a positive impact on diet and sales in populations of low socioeconomic status. However, these results were not replicated among the intervention literature, which showed inconsistent findings in populations of low socioeconomic position.

Policy implications
The results from this review provide policy makers with evidence to justify the implementation of populationlevel policies incorporating placement strategies in food retail stores to improve dietary-related behaviors. Although more research is needed to quantify the magnitude of effect of availability and positioning strategies, this review's findings suggest that placement strategies combining both availability and positioning have the greatest potential to improve the healthfulness of sales and dietary patterns.
The evidence that is currently available -summarized in this review and the 2016 systematic review by Bucher et al, 68 which found that manipulating the order and proximity of food in eateries and food service outlets influenced food choice -supports the UK government's intention to ban the positioning of unhealthy food items in prominent locations in food retail stores and food service outlets. 69 Other governments could also consider introducing similar policies to improve dietary quality across high-income countries. Even though the current research findings do not meet the "gold-standard" level of evidence that is usually required for the scientific community to provide certainty of effect, these reviews provide a sufficient body of evidence to recommend government action. The introduction of government policies to promote healthy food retail establishments would lead to a "level playing field" between retailers. If such placement strategies are only implemented on a voluntary basis, the high level of competition within this setting may result in some businesses not implementing such strategies, and that would likely limit the positive impact on public health. 70 Next steps for the research field Although this review indicates moderate evidence for food placement strategies in retail food stores influencing purchasing and dietary behaviors, there are a number of ways in which the body of evidence can be strengthened. The lack of power calculations described in intervention research in particular is an issue that needs addressing to optimize external validity of the evidence. No intervention articles in this review described their power calculations or justified the study's sample size. Placement studies require power calculations that take account of clustering at the store level because the intervention is store-based. In a cluster-designed study, it is the number of clusters, rather than the number of individuals within each cluster, that is most potent in determining statistical power. 71 The need for a large number of stores and the opportunistic nature of many interventions studies are key reasons why no highquality intervention research has been conducted in this field. Considerable commitment is required from commercial collaborators to allow for the required number of stores; however, mounting societal and political pressure for food retailers to engage in healthy eating strategies could enhance future prospects of adequately powered studies being conducted.
Improving the design of future food placement research, particularly considering novel trial designs and longitudinal observational studies, would further improve the evidence base. Less than two-thirds of the intervention articles in this review included a comparison group, and only 4 were randomized controlled trials. This finding indicates that store-based placement interventions do not easily conform to scientific gold standards. Researching in real-world settings, however, provides valuable knowledge to policy makers about intervention effectiveness in complex social contexts, particularly when studies are rigorously designed. Parallel designs with control groups matched on area characteristics and store sales (plus adjustment for confounders) offer a robust design and were used in approximately one-third of the current intervention evidence, all published in the last 5 years. Alternative designs including natural experiments, stepped-wedge designs, synthetic controls, and propensity scores could be further explored for use in future food placement intervention or policy evaluation studies. [72][73][74][75] There is a gap in the evidence to describe how reducing the availability of unhealthy foods in food retail stores affects diet-related outcomes as most of the literature investigating placement strategies has focused on healthy foods. Although more challenging commercially than the "win-win" of targeting healthy foods in placement interventions, 45 future research should focus on limiting the availability and prominent positioning of unhealthy foods -perhaps by replacing them with nonfood items in an attempt to reduce overall food intake. Compensation to food retail chains for any loss in revenue may need to be considered. As highlighted in a recent scoping review, business outcomes of food retail strategies to improve health should be consistently reported in academic literature. 76 There is also a need for greater harmonization of in-store assessment measures that act as exposures in observational research or fidelity assessments in intervention studies. Currently, 3 categories of availability measures are used: shelf space, variety, and composite scores. Considerable within-category variation exists, and the composite scores typically measure in-store factors such as price and quality in addition to availability. Positioning measures have focused almost entirely on prominent store locations such as checkouts, ends of aisles, and front of store; only one observational article measured shelf placement in this review. An in-store assessment tool that expands the existing validated GroPromo tool 55 to include measures of availability and shelf placement would help to harmonize data in this field. Intervention research should include measures of both availability and positioning in their fidelity assessments because these two placement strategies are often intertwined. Positioning products at the checkouts or ends of aisles typically extends the availability of those products as they are located both in an aisle and in a prominent location. Only one intervention article in this review specifically considered both these aspects of placement in its evaluation. 47 Consistent with the findings of previous reviews of supermarket interventions, 15,16 many of the studies in this review contained multiple intervention components. Food placement strategies were a core component of these interventions. It was not possible, however, to decipher the isolated effects of changing product availability or positioning because additional strategies such as signage or staff training were employed at the same time. Future research that aims to test multiple intervention components, such as the 4 Ps of marketing, should consider the study by Wensel et al, 34 which tested both the independent and additive effects of 4 intervention elements. This type of research, combined with studies that test single-component placement interventions, will be scientifically advantageous and ensure efficient and cost-effective packaging of interventions.
While sales outcomes were most frequently used and showed greatest consistency of effect in this review, loyalty card data were not used in any of the observational or intervention studies. Loyalty card data are a form of "big data" that offer a potentially economical method of analyzing how in-store determinants affect household purchasing. 16,77 Little is known from the available literature about the populations who are most affected by placement strategies in food stores. This gap could be addressed by measuring intervention effects at a household or individual level rather than at a store level. Despite existing evidence suggesting that those from disadvantaged backgrounds are more susceptible, in dietary terms, to unhealthy in-store environments than those of more affluent groups, 78 it is currently unclear whether placement strategies exacerbate or reduce dietary inequalities. Further evidence from adequately powered studies is needed to determine differential effects by SEP. Moreover, assessments of dietary outcomes, nutritional biomarkers or metabolites, and food waste are needed alongside loyalty card data to provide intelligence on the accuracy of this big data source, and the correlation between purchasing and intake patterns. It would be particularly useful if future research included dietary assessments from more than one household member to understand more clearly which household members are being affected by placement strategies. Finally, as outcome measures with frequent time points (weekly sales or purchasing data) become commonplace in this research field, more advanced statistical methods such as the time-series analyses used by Ejlerskov et al, 30 and appropriate adjustment for clustering and confounding, should be more consistently applied.

Strengths and limitations
This review is strengthened by the adherence to PRISMA guidelines throughout. Two reviewers independently conducted a risk-of-bias assessment and data extraction from each of the included studies to ensure consistency and rigor. In addition, the inclusion of both observational and intervention studies is a strength of this study as this allowed for a more thorough assessment of the overall relationship between placement strategies in food-store settings and diet-related outcomes. Product availability has been researched most extensively in the observational literature, while product positioning has been the focus of many intervention studies. However, including both types of literature presents some challenges with interpreting the quality of the evidence and summarizing overall results. Separate risk-of-bias assessment criteria were used for observational and intervention studies, so the final quality scores could not be compared. Intervention research is considered a higher grade of evidence and should be treated as such when drawing conclusions. This review also only included studies that assessed physical in-store environments, excluding virtual and online settings. This approach allows for the assessment of strategies that have greater external validity and the ability to be implemented in real-life settings. The findings of this review, however, are not applicable to the growing online grocery sales market.
The search strategy for this review did not include literature published prior to 2005 or forward searching for identified articles through citations. It is therefore possible that some articles of interest may have been excluded. However, 2005 marked the year a landmark article in the field of food environment research was published. 24 This article, along with the Foresight obesity report, 3 was among the first to highlight the importance of understanding the role of food environments on population health. Another limitation of this systematic review was the exclusion of gray literature or unpublished data in this topic area. Consequently, it is possible that the results of this review are subject to publication bias. Studies showing positive and significant effects may have an increased likelihood of publication, and our findings were potentially skewed as a result of this bias. In addition, owing to the heterogeneous nature of the study exposures, interventions, and outcomes, it is difficult to draw definitive conclusions from the available evidence. Though meta-analysis was not possible, a quantitative summary of the evidence was achieved by using a direction-based vote-counting technique or effect direction plot, as is recommended by Cochrane when meta-analysis is not feasible. 28 This technique, however, is limited by its lack of consideration of the magnitude of effects and differences in study size. 28

CONCLUSIONS
Drawing on recent evidence from observational and intervention research across high-income countries, this review suggests that more prominent placement strategies are associated with higher sales and consumption of both healthy and unhealthy foods, but not weight status. Even though further high-quality research is required in this area, the balance of evidence suggests that the introduction of government interventions may be beneficial by providing a "level playing field" between retailers and to increase the availability and prominence of healthy foods and reduce the availability and prominent positioning of unhealthy foods. Future research priorities should focus on designing adequately powered intervention studies that test both the independent and additive effects of reducing the availability, and limiting the prominent positioning, of unhealthy foods. A greater understanding of who is most affected by placement strategies is required; this could be achieved through the use of loyalty card data as the primary outcome holds potential, alongside dietary assessments from more than one household member. Declaration of interest. S.C.S. and G.N. have no conflicts of interests to declare. J.B. and C.A.V. have a nonfinancial research collaboration with a UK supermarket chain. J.B. has received grant research support from Danone Early Life Nutrition. The study described in this manuscript is not related to any of these relationships.

Supporting Information
The following Supporting Information is available through the online version of this article at the publisher's website.