An umbrella review of the effectiveness of fiscal and pricing policies on food and non‐alcoholic beverages to improve health

Poor quality diets represent major risk factors for the global burden of disease. Modeling studies indicate a potential for diet‐related fiscal and pricing policies (FPs) to improve health. There is real‐world evidence (RWE) that such policies can change behavior; however, the evidence regarding health is less clear. We conducted an umbrella review of the effectiveness of FPs on food and non‐alcoholic beverages in influencing health or intermediate outcomes like consumption. We considered FPs applied to an entire population within a jurisdiction and included four systematic reviews in our final sample. Quality appraisal, an examination of excluded reviews, and a literature review of recent primary studies assessed the robustness of our results. Taxes and, to some extent, subsidies are effective in changing consumption of taxed/subsidized items; however, substitution is likely to occur. There is a lack of RWE supporting the effectiveness of FPs in improving health but this does not mean that they are ineffective. FPs may be important for improving health but their design is critical. Poorly designed FPs may fail to improve health and could reduce support for such policies or be used to support their repeal. More high‐quality RWE on the impact of FPs on health is needed.

changefor example, through attention to changes in a product's nutritional quality, label, or priceis lower in lower income groups so diet-related FPs may be ineffective in reducing health disparities or risk exacerbating disparities. 11,12 This underscores the need to consider the socioeconomic context into which an FP is introduced. In concert with international agencies such as the World Health Organization, some governments have prioritized actions in a coordinated response that involves state and non-state actors to address current dietary intake. 4 This has sparked interest in a range of measures that policymakers may wish to consider to improve public health.
FPs, such as taxes or subsidies, are seen as an important tool to change consumer behaviorby affecting the price of goods and by encouraging manufacturers to reformulate their productsand ultimately improve population health. 5,13,14 Upstream interventions rely less on an individual's ability to engage with the policy, which may be linked to socioeconomic factors (e.g., health literacy or education more broadly), but without removing individual agency. [15][16][17] The distributional impacts of FPs can be both positive and negative; a tax on sugar-sweetened beverages (SSB) might impose a larger proportionate tax burden on low-income families but also confer a larger proportionate health benefit. 8,[18][19][20] Much of the evidence on the distributional impact of FPs as well as their overall effect on health outcomes come from modeling studies rather than RWE. 20 As the number of jurisdictions implementing FPs has grown, so too has the number of studies examining their effectiveness, as well as reviews synthesizing these studies. Umbrella reviews provide an important tool for policymakers by summarizing the highest level of evidence, namely systematic reviews and meta-analyses, in relation to a research topic/question. 21 We present an umbrella review of the effectiveness of FPs in reducing overweight, obesity, and diet-related NCD. We also consider intermediate outcomes that may contribute to improved health such as consumer behavior or product reformulation. 22,23 As disparities in diet-related health outcomes are evident across socioeconomic and minority groups, 24 they are a critical consideration in examining the success of such policies; where possible, we examine the distributional impact of policies according to PROGRESS-Plus categories (Place of residence; Race/ethnicity/culture/language; Occupation; Gender/sex; Religion; Education; Socioeconomic status; and Social capital). 25 2 | METHODS

| Search strategy
The protocol for the umbrella review was registered at the Interna- We used a range of search terms (Table S2) classified under three themes: study type, intervention, and outcome. Search terms for "study type" focused on systematic reviews or meta-analyses. Terms for "intervention" and "outcomes" focused on policy variables, behavior, and health categories. "Intervention" terms included those related to government FPs, and "outcome" terms included intermediate outcomes, such as consumer behavior or product reformulation, and final outcomes, namely bodyweight or diet-related NCDs. Within each theme "or" operators were used to combine terms and across each theme, "and" operators were used. Searches were tailored according to the functionality of each database. Where possible, MeSH terms were used (MEDLINE, EMBASE, PsychInfo) that corresponded to the thematic areas (Table S2). Where a database provided tools to further limit the search strategy, we restricted to studies of "humans" (MEDLINE, EMBASE, PsychInfo) and applied a database tool designed to achieve best balance between sensitivity and specificity of searches for systematic reviews (MEDLINE). 26 No restrictions were placed on language.
To further ensure specificity of our search, four systematic reviews were identified a priori, 18,19,27,28 which we expected to include based on the eligibility criteria used. These were used to test the search strategy but not to design it. We validated our search strategy across search engines based on whether or not they returned these articles. We followed the Peer-Review of Electronic Search Strategies (PRESS) guidelines in designing our search strategy though it was not peer-reviewed. 29

| Screening
One reviewer (LB) executed the search strategy, collated the results, and removed the duplicates in EndNote 20 software. 30 Two reviewers (CON & LB) independently screened article titles first to remove redundancies and compared results before finalizing a list of articles for abstract screening. Disagreements were discussed until consensus was reached and provision made for input from a third reviewer (FK), but this was not required. Next, abstracts were screened using the same process followed by full-text screening. We searched reference lists, contacted authors, and sought expert opinion to identify relevant studies and/or acquire full-texts. For any articles that required translation into English, initially, an online translation software (Google Translate) was used because this has shown reasonable accuracy in translating articles for systematic reviews. 31 To avoid risk of error from poor translation, this was used to identify any clear reasons for exclusion and otherwise a professional translation service was used.

| Inclusion/exclusion criteria
We included all systematic reviews examining the effectiveness of FPs implemented by governments to improve population diet.
Reviews eligible for inclusion were those that: i. Conducted a systematic review with or without meta-analysis, ii. Examined at least one government-enacted FP, iii. Included RWE, that is, not reviews that included only simulated models of expected effectiveness, iv. Included policies that targeted the consumption of food and nonalcoholic beverages, that is, not agricultural policies with unintended impacts upon consumption, v. Included impact upon final outcomes, for example, overweight or obesity, or intermediate outcomes, for example, consumer behavior or product reformulation, and vi. Included policies that applied to the entire population of its jurisdiction, that is, reviews that included only experiments of price discounts in supermarkets or targeted food programs were excluded.
Criteria for qualification as a systematic review were taken from the Cochrane Handbook for Systematic Reviews of Interventions. 32 Reviews were therefore excluded if they did not provide (i) a clearly stated set of objectives with pre-defined eligibility criteria for inclusion; (ii) an explicit, reproducible methodology; (iii) a systematic search that sought to identify as many studies as possible that would meet the study's eligibility criteria; (iv) an assessment of the validity of the findings of the included studies; and (v) a systematic presentation, and synthesis, of the characteristics and findings of the included studies.
We excluded reviews of only modeling/simulation studies (i.e., those that simulate a result) or theoretical studies. If a review included studies that satisfied our inclusion criteria as well as studies that did not (e.g., a modeling study of the same intervention), we included the review and reported results only relating to the relevant studies provided these were presented separately. If the original reviewers had combined the findings of modeling or theoretical studies with those examining RWE, we used the combined results while noting reviews that included a mix of results in our synthesis. The same process was followed where a review examined price changes in relation to a FP as well as variations in prices for other reasons, for example, supermarket promotions. Reviews evaluating interventions applied only within a limited setting such as supermarkets, airports, or schools that only involved price variation unrelated to an implemented government policy or that targeted only at a specific group, such as low-income groups, children, or pregnant women, were excluded.
Where a review reported on FPs targeted at a specific group of the population as well as FPs applied to an entire population, we reported on both. Patient/population, intervention, comparison, outcomes, and study design (PICOS) criteria are described in Table 1 below.

| Data extraction
Data extraction was carried out using an online form that included review aims; methods; eligibility criteria and search strategy; funding sources; setting; participants; intervention (and comparator where available); outcomes measured; research design; any sub-group analyses related to specific groups of participants; any distributional impacts on outcomes or the tax burden; as well as criteria relevant to our quality appraisal, for example, whether any conflicts of interest were reported. Distributional impacts of the tax burden or health outcomes were examined using axes of differentiation according to PROGRESS-Plus, 25 given the potential for inequalities to arise across groups differentiated in ways other than socioeconomic status (SES). 33,34 The extraction form was trialed by two reviewers (LB & CON), each using three systematic reviews before being finalized. 35 As part of the original protocol, it was intended that data extraction and quality appraisal would be conducted in duplicate; however, because of time constraints, extraction was performed in duplicate (LB & CON) on 25% of reviews. These results were compared and guidelines developed on how best to extract information from the remaining reviews in a consistent manner and this was completed by one reviewer (LB). For example, it was decided that "research design" should be documented according to whether the systematic review's included studies were: Randomized Controlled Trials (RCTs), Non-Random Studies of Interventions (NRSI) (e.g., controlled before and after, cohort, longitudinal, or cross-sectional studies), Modeling Studies (which simulated results), or other reviews. 36

| Quality appraisal
We assessed the methodological quality of 25% of the systematic reviews included for final review in duplicate (LB & CON) using the AMSTAR 2 (A MeaSurement Tool to Assess systematic Reviews) tool. 37 The AMSTAR tool provides a comprehensive critical appraisal instrument to evaluate systematic reviews of health interventions.
Updates to this instrument, in the form of AMSTAR 2, include criteria for the appraisal of non-random interventionsmaking it well-suited for our appraisaland also encourage a more disaggregated examination of quality according to critical and non-critical domains. 37 Disagreements were resolved by discussion, and although provision had been made where necessary to seek input from a third reviewer (FK), T A B L E 1 PICOS criteria used to define the research question. Conducted as a systematic review with or without meta-analysis and included real-world evidence this was not required. The rest of the reviews were appraised by one reviewer (LB) and results discussed with an independent expert advisory panel (EAP) established to advise on the conduct of the umbrella review. The EAP comprised researchers with international reputations in the areas of public health and economics for work in this area.

Participants
Details of the EAP are contained in Table S3. The AMSTAR 2 tool allows a broad indication of whether the quality of a review is high, moderate, low, or critically low. Reviews that received a critically low score were excluded from our final review but were listed for transparency. The quality rating applied to reviews in this umbrella review does not necessarily reflect the overall quality of the included review but rather reflects its quality in addressing the aim of our umbrella review. For example, a review may have included information that merited inclusion as part of our eligibility criteria but this may not have been the focus of the original review.

| Synthesis
Given the relatively few reviews that met our eligibility criteria (of which only one included a meta-analysis) and the heterogeneity across them, a narrative synthesis of the included reviews was conducted following exclusions based on our quality appraisal. We include only reviews in our synthesis rather than the primary research from the identified reviews. A summary of each review is presented in Table 2. Where quantitative results were provided as part of a meta-analysis, these are reported in Table 3 as were any distributional/subgroup impacts.

| Robustness check
Reviews that have more than one critical flaw, according to the AMSTAR 2 rating, receive a critically low score. As a robustness check, characteristics and results were extracted for reviews that had only two critical flaws, that is, those studies that give one change in score would have been included in our final review. Results from our main synthesis were compared with the results from these reviews to examine whether different conclusions would have been reached with their inclusion.
While umbrella reviews summarize high-level evidence from systematic reviews, they are restricted in their timeliness because they can only use evidence from primary studies that had been published before the most recent included systematic review search date; while our search covered up to June 2021, the most recent systematic review search date was October 2019. 38,39 As countries have continued to implement FPs, in order to examine the robustness of our umbrella review results against more recent evidence, we conducted a literature review of primary studies published between January 2020 and November 2021 (see Figure S1 for flow diagram).
This allows us to examine whether recent primary evidence supported or conflicted with our umbrella review results. We used the search strategy from the umbrella review applied only to EMBASE (given its high yield of eligible reviews, see below) and focused only on English language journal articles referencing tax or subsidies in the title and excluding references to alcohol or tobacco in the title (Table S4).
Although this search was more restricted (by database and language), was conducted at a high-level by only one reviewer (LB), and did not involve any quality assessment, it provides a useful overview of primary studies after 2019 that examine the intermediate and final (including distributional) impacts of taxes and subsidies to compare with our umbrella review results.

| Screening
We identified 16,883 records through database searches, resulting in 9,996 unique records once duplicates were removed ( Figure 1). After title and abstract screening, this was reduced to 75 records, with an additional four records identified based on recommendations from experts in the field and searches of reference lists but these were excluded in subsequent screening. Following the Google Scholar search, no new records were included that had not been identified in the original database search. One article in Russian was translated online for full-text screening, and all other articles were in English.
Following full-text screening, a further 54 records were excluded (the Russian paper was excluded as it did not qualify as a systematic review), leaving 25 potentially relevant records for data extraction.
The list of all 25 potentially relevant reviews is in Table S5, along with a summary of the databases in which they were indexed: 100%

| Quality appraisal
Following quality appraisal, only four systematic reviews/metaanalyses were included as part of our synthesistwo reviews were rated as "High" 38,39 and two as "Low" 40,41the other 21 reviews received a "Critically Low" AMSTAR 2 rating. Six reviews received a marginal "Critically Low" rating (two critical AMSTAR 2 domain flaws) and were therefore included in sensitivity analysis (below). Not including a list of potentially relevant but excluded studies (n = 6/6) and not justifying publication restrictions (n = 4/6) were the main reasons for these ratings (Table S6). Where a review restricted inclusion to studies of humans only, no justification was necessary, and thus, no penalty was incurred in the AMSTAR 2 rating. All the included reviews T A B L E 2 Characteristics of reviews included in the narrative synthesis.
Author Note: We have described the population as they were described by the review authors. Where the population inclusion criteria were not described in the review we have assumed that 'any' population, age, setting or country had the potential to be included. Abbreviations: FP, fiscal and pricing policy; NRSI, non-random studies of interventions.

T A B L E 3 Effects of intervention strategies synthesized by included systematic reviews on intermediate and final outcomes.
Author -"… the interventions assessed here fail to achieve an effect on consumption that could plausibly be considered as clinically significant, that is, as having an effect on individuals' nutritional intake to the extent that it would reduce the incidence of overweight, obesity, and related chronic diseases." Abbreviations: CI, confidence interval; FPs, fiscal and pricing policies; SMD, standardized mean difference; SSB, sugar-sweetened beverage.

T A B L E 3 (Continued)
Author/year reported either no conflicts of interest or, where conflicts of interest were reported, they described how these were managed so as not to influence results. Table 2 provides an overview of each of the four included reviews. labeling, 41 two reviews focused more specifically on taxes of sugar and sugary food and beverages, 38,40 and one on fat taxes. 39 None of the four reviews included results from modeling studies; they all examined NRSIs using RWE on the implementation of FPs. Two reviews were conducted as Cochrane reviews, 38,39 which were also the reviews that received a "High" quality rating; they included a combined total of four studies suggesting the Cochrane review evidence base for the effectiveness of taxes on the sugar, fat, or other nutrient content of foods in changing behavior or health is meagre.

| Intermediate outcomesdirect effects
All four included reviews provided information on the effectiveness of FPs, namely taxes, in changing consumption or purchasing of the targeted food or non-alcoholic beverage(s). The reviews covered 55 primary studies with some overlap of included studies between reviews and also included research into non-FPs, for example, menu labeling. 41 The evidence across reviews consistently supported the effectiveness of taxes on unhealthy foods or non-alcoholic beveragesin particular SSBsin reducing their consumption, often proxied by purchasing behavior. Those reviews with the highest quality rating included the fewest studies 38,39 but also focused on more specific nutrients in foods; see Table 2.
Estimates of own-price elasticity were only identified for SSBs, (1 month to 6 years) observation periods varied widely. 40 In their meta-analysis, they presented evidence that taxes on SSBs are associated with reduced purchases, and used a two-step approach to adjust for both jurisdiction (n = 6) and study-outcome level heterogeneity (n = 17) while also conducting a range of sensitivity analyses. They estimated an own-price elasticity of À1.00 (95% Confidence Interval

| Lower quality systematic reviews
Review characteristics and results were extracted for the six reviews that had only two critical flaws in the AMSTAR 2 quality rating 18,45-49 (Tables S7 & S8). When considering the results of these reviews, the key messages from our narrative synthesis remained unchanged: 1) there is evidence to support the use of taxes and subsidies in changing consumption/purchasing of taxed/subsidized goods; 2) there is some evidence of substitution effects towards consumption/ purchasing of untaxed goods or in untaxed jurisdictions; and 3) there is a paucity of RWE supporting the effectiveness of FPs in changing health outcomes. Many of these additional reviews included modeling studies that were more likely to support the use of FPs in affecting final outcomes (e.g., body mass index [BMI] or diet-related NCDs).
Three of the reviews examined the distributional impacts of FPs. 18,46,49 They found some evidence of FPs that are applied to entire populations as having a neutral or positive impact on inequalities in outcomes according to SEP and may, only to a small degree, be regressive (i.e., resulting in a significantly greater proportionate tax burden on lower socioeconomic groups). However, the majority of this evidence came from modeling studies, and the selection of primary studies using RWE found that taxes had a neutral impact on the socioeconomic distribution of anthropometric outcomes.

| Recent primary studies
Our additional search identified 24 primary studies published between January 2020 and November 2021 that presented RWE on the effectiveness of taxes and subsidies on food and non-alcoholic beverages in improving diet and health ( Figure S1 & Table S9). These covered FPs implemented in India, 50 Spain, 51 France and Hungary, 52 Denmark, 53 South Africa, 54,55 Cook Islands, 56 Tonga, 57 Saudi Arabia, [58][59][60] Thailand, 61 Mexico, 62,63 and various jurisdictions within the US. [64][65][66][67][68][69][70][71][72][73] Almost all of these (n = 23) examined taxes on non-alcoholic beverages, namely SSBs. The other study, conducted on the Navajo Nation, examined the effect of subsidies on healthy food and found that the availability of F&V increased relative to neighboring unsubsidized jurisdictions. 64 Twenty studies found evidence to support the effectiveness of taxes in reducing consumption of taxed items. [51][52][53][54][55][56][57][59][60][61][62][63][65][66][67][68][69][70][71][72][73] Two of these showed that reformulation was part of the mechanism underlying reduced sugar intake. 55,66 Three studies showed that when a tax was reduced or repealed, there was a subsequent increase in consumption of taxed or previously taxed beverages. 53,56,67 Many studies found evidence of substitution whereby individuals switched to untaxed alternatives within stores, shopped in untaxed jurisdictions, or in some cases, substitution was implied because total sugar intake was unchanged after the tax, in spite of decreased purchases of taxed items or effects were negligible at a population level. 51,52,54,55,57,62,66,[68][69][70][71] Only one study examined final outcomes and it found no evidence that a SSB tax in South Africa (known as the Health Promotion Levy) reduced BMI at the levels implemented. 54 Finally, five of the studies examined the distributional impacts of the taxes. These focused on inequalities in outcomes rather than on the tax burden with all but one consistently finding evidence that the tax was most effective in reducing consumption among lower income, lower educated or unemployed individuals/households, and among those with high baseline intake. 54,57,61,65 The study that found greater reduction in SSB consumption among higher educated individuals was from a distinct sample of employees from a healthcare provider. 63 One study conducted among a sample of high-consuming and lowincome individuals found large reductions in taxed beverage intake. 55 Although we did not conduct a quality assessment of these studies, when restricting the sample of included studies to those that analyzed both a pre-and post-tax implementation period as well as untaxed jurisdictional controls, 51,64-68,70-72 results were consistent with those described above.

| DISCUSSION
We conducted an umbrella review to summarize the highest level evidence of the effects of FPs on food and non-alcoholic beverages in improving diet and preventing overweight, obesity, and diet-related NCDs. We included four systematic reviewsone including a metaanalysisin our final analysis. [38][39][40][41] We found evidence to support the effectiveness of FPs in changing consumption of taxed/subsidized goods and some evidence of taxes leading to substitution towards untaxed products or those with a lower tax rate. However, there was a lack of RWE supporting the effectiveness of FPs in improving population health. There was little evidence from the included reviews as to the distributional impacts of FPs. One review, in their meta-analysis, found no significant difference in SSB consumption following an SSB tax between adults and children while, in their narrative synthesis, found some evidence for differences in consumption across SEP; 40 two studies in Mexico reporting significantly greater reductions among lower income households 74,75 and one study in Chile reporting greater reductions among higher income households. 76 In our robustness check, we examined those reviews that were closest to being included (only two critical-domain flaws in their AMSTAR 2 rating). These reviews supported the effectiveness of FPs in changing consumer behavior, [45][46][47][48] improving population health, 46,48 and reducing disparities in health 18,46 or at least not increasing these disparities. 49 They also found that taxes are likely to be regressive, though not substantially. 46 To increase the timeliness of our findings, we also conducted a literature review of primary studies, published between January 2020 and November 2021. We found further evidence that taxes on unhealthy nutrients or products reduce their consumption, along with further evidence to support the contention that there is substitution towards untaxed sugary products or neighboring untaxed jurisdictions that often negated the impact of the tax. However, there were also examples of positive substitution with individuals switching to healthier beverages (e.g., water) or reductions in net sugar consumption in spite of negative substitution. 57,66 Reductions in consumption of unhealthy products were largest among lower socioeconomic groups or those with high baseline intake. 54,57,61,65 This suggests that the benefits of the tax may be more likely to occur for those most at risk of the harms of unhealthy diets, highlighting the importance of tax design in improving public health and in reducing inequalities. This is supported by other reviewsalbeit excluded from our final samplethat showed that taxes reduce health inequalities and, while they may be regressive, are often more cost-effective for lower than for higher socioeconomic groups when considering avoided healthcare expenditure or improved health. 8,18,77 Much of this evidence came from modeling studies (which often fail to account for substitution effects) highlighting the relative paucity of detailed real world analyses on the impact of FPs. Although modeling studies have a greater risk of bias, they can also model higher tax rates than those actually implemented by governments given that these are usually met with opposition from affected groups, such as manufacturers. 41 Teng et al. in analyzing RWE found that higher taxes were associated with larger impacts on consumption. 40 It remains important to treat the results of modeling studies with caution given their dependence on assumptions including the political will to adopt FPs at levels that might be more effective.
As reviews continue to be published, it is important to discuss those published after our search window in relation to our results.  82 Although sufficient price increases have been critical to the success of tobacco and alcohol-related FPs in improving health, consideration of supply-side factors (e.g., two-tiered excise taxes to encourage product reformulation) or earmarking tax revenues for cointerventions (e.g., public awareness campaigns) were also seen as important. 83 It is plausible therefore that policies that change the price of a good can improve health; however, as the evidence base evolves, so too must the design of FPs to reflect the findings of research in this area.
FPs have the potential to correct market failure such as externalities, where the consumption of harmful goods imposes a societal cost such as on the health system, or internalities, 1 where consumers take action that are not in their own best interest. 84 In the case of SSB  54,57,61 and are likely more cost-effective for these groups. 85 Farhi and Gabaix showed that where externalities and internalities are heterogeneous across SES groups, subsidies alongside taxes may be more effective in correcting market failure, 86 for example, a 10% reduction in the price of fruits and vegetables. 4 FPs are still unlikely to be a silver bullet, and we found evidence that to improve public health, they may be more effective when implemented alongside other interventions. 41 This could include public information programs around food content and health as well as labeling or interventions to promote lifestyle change, for example, physical activity, given the links between other lifestyle behaviors and NCDs 87 as well as the link between unhealthy diet and sedentary behavior. 88 When considering an intervention that targets volume or value of a good, an ad valorem tax, like a sales tax, increases the price as a percentage of the goods price, but these are likely to push consumers towards cheaper but not necessarily healthier options. 42,83 Ad quantum taxes on the quantity of the health-harming nutrient, for example, sugar, are preferable. Excise taxes on suppliers allow for this but require that manufacturers pass-through enough of the tax to the consumer to change behavior. The more own-price elastic demand there is for the product, which is a function of how many substitutes are available, the less likely a producer is to pass through the cost to the consumer, although as in Andreyeva et al., a substantial proportion (82%) of a SSB tax is often passed through. 78 Although this makes for a more complex intervention, they are more effective. 85 Denmark provides an important example where the nutrientsaturated fatwas taxed. However, this tax was repealed, 39 and acceptability of FPs to both policymakers and the public merits further research.
These principles generally assume a paternalistic government intending to correct market failure in relation to health; however, different tax designs may serve different aims; for example, an ad valorem tax may be the least burdensome administratively or an ad quantum tax on products rather than nutrients may generate more revenue. 89 In Mexico, an added-sugar and calorie-dense tax appears so far to have failed to correct such externalities because of substitution effects 62 but was maintained as the government revenue and the earmarking of revenues for improved drinking water improved its acceptability. 4,90 In relation to the issue of leakage, only one meta-analysistaxes on SSBswas included in our final sample, and the estimate of crossprice elasticity was insignificant but likely underpowered 40  While these challenges may be significant this should serve to encourage new high-quality longitudinal studies rather than deter their conduct.

| LIMITATIONS
The limitations in our umbrella review point to directions for future research. Firstly, as we intended to avoid FPs implemented in controlled settings, such as experiments in supermarkets, and to examine the distributional impacts of FPs across axes of differentiation, 33 we restricted our review to policies that were implemented across an entire population within a jurisdiction. Although there are no US government programs to support increased F&V consumption nationally for the whole population, 93 a number of reviews focused on the US Supplemental Nutrition Assistance Program (SNAP) that subsidizes healthy eating for low-income households that were excluded from our review (unless reviewed alongside eligible FPs).
Additional research is required to examine policies such as these that target subgroups of the population to encourage healthy eating, because these provide a figurative (and potentially literal) carrot alongside the stick of taxation to encourage consumption of important micro-and macronutrients. 92 Missing also from this review were reviews of efforts by governments to control promotional offers by private enterprises such as bans on buy-one-get-one-free or minimum pricing laws such as those related to alcohol in Ireland, which may be effective in the case of other health-harming goods. 94 Synergies may also exist in terms of policies related to climate change or environmental sustainability. 95,96 For example, many foods that are nutrient dense, such as fruits, vegetables, and nuts, also have a lower environmental impact than less nutrient-dense and more highly processed foods. 97 As noted above, our umbrella review was restricted to evidence within the search windows of the included reviews, which can create a lag in terms of research timeliness. We thus included some discussion of systematic reviews and meta-analyses conducted after our search date, examined reviews that were narrowly ineligible following our quality assessment, and conducted a brief literature review of more recent studies examining the effectiveness of taxes and subsidies to reduce NCDs. This also highlights the trade-off that must be made when conducting a review, between the highest quality review evidence and all potentially relevant evidence. For example, the two Cochrane reviews (that follow a specific and rigorous approach to systematic reviews of research in health care and policy to satisfy the criteria for publication in the Cochrane Database of Systematic Reviews 98 ) included in our review were of the highest quality but had the fewest primary studies. It is reassuring that similar results were observed between our sample of higher quality systematic reviews (no or only one critical AMSTAR 2 domain flaws); lower quality reviews (two critical domain flaws); and primary papers (literature review). However, this highlights key gaps in the literature regarding high-quality RWE on the effect of FPs in relation to final (health) outcomes; their distributional effects; the comparative effects of their design (e.g., ad quantum or ad valorem as touched on by Teng et al.) 40 ; population-wide subsidization programs; the implementation of FPs alongside other public health measures; and reformulation. We also found that when taxes were repealed or cut, consumption of unhealthy products returned to baseline levels and potentially even higher, which highlights the importance of further research to understand the political and public acceptability of such interventions.

| CONCLUSION
We find that taxes and, to some extent, subsidies are effective in changing consumption of taxed/subsidized items, but substitution is likely to occur. FPs present an important policy tool for improving public health but their design is critical. That substitution to unhealthy goods can negate the effect of the FP not only undermines its goals but may also undermine public confidence. This could reduce support for such policies or lead to their repeal, further undermining the success of FPs globally. The lack of evidence regarding the impact of FPs on health outcomes identified in this review is, therefore, a concern and more high-quality RWE longitudinal studies are needed as the health effects from FPs may take longer to manifest. The implementation of a FP to improve diet and health can act as a signal to the public as to the importance (both negative and positive) of affected products/nutrients in relation to diet. The revenues from taxes in particular can also improve public acceptability although further research is needed to better understand whether FPs can be implemented and optimized within the political arena so as to achieve their aims.