Systematic review of dietary salt reduction policies: Evidence for an effectiveness hierarchy?

Background Non-communicable disease (NCD) prevention strategies now prioritise four major risk factors: food, tobacco, alcohol and physical activity. Dietary salt intake remains much higher than recommended, increasing blood pressure, cardiovascular disease and stomach cancer. Substantial reductions in salt intake are therefore urgently needed. However, the debate continues about the most effective approaches. To inform future prevention programmes, we systematically reviewed the evidence on the effectiveness of possible salt reduction interventions. We further compared “downstream, agentic” approaches targeting individuals with “upstream, structural” policy-based population strategies. Methods We searched six electronic databases (CDSR, CRD, MEDLINE, SCI, SCOPUS and the Campbell Library) using a pre-piloted search strategy focussing on the effectiveness of population interventions to reduce salt intake. Retrieved papers were independently screened, appraised and graded for quality by two researchers. To facilitate comparisons between the interventions, the extracted data were categorised using nine stages along the agentic/structural continuum, from “downstream”: dietary counselling (for individuals, worksites or communities), through media campaigns, nutrition labelling, voluntary and mandatory reformulation, to the most “upstream” regulatory and fiscal interventions, and comprehensive strategies involving multiple components. Results After screening 2,526 candidate papers, 70 were included in this systematic review (49 empirical studies and 21 modelling studies). Some papers described several interventions. Quality was variable. Multi-component strategies involving both upstream and downstream interventions, generally achieved the biggest reductions in salt consumption across an entire population, most notably 4g/day in Finland and Japan, 3g/day in Turkey and 1.3g/day recently in the UK. Mandatory reformulation alone could achieve a reduction of approximately 1.45g/day (three separate studies), followed by voluntary reformulation (-0.8g/day), school interventions (-0.7g/day), short term dietary advice (-0.6g/day) and nutrition labelling (-0.4g/day), but each with a wide range. Tax and community based counselling could, each typically reduce salt intake by 0.3g/day, whilst even smaller population benefits were derived from health education media campaigns (-0.1g/day). Worksite interventions achieved an increase in intake (+0.5g/day), however, with a very wide range. Long term dietary advice could achieve a -2g/day reduction under optimal research trial conditions; however, smaller reductions might be anticipated in unselected individuals. Conclusions Comprehensive strategies involving multiple components (reformulation, food labelling and media campaigns) and “upstream” population-wide policies such as mandatory reformulation generally appear to achieve larger reductions in population-wide salt consumption than “downstream”, individually focussed interventions. This ‘effectiveness hierarchy’ might deserve greater emphasis in future NCD prevention strategies.


Introduction
Non-communicable diseases (NCDs) kill over 35 million people annually. Common cancers, cardiovascular diseases, diabetes, respiratory diseases and dementia together now account for over two thirds of the entire global burden of disability and death. [1,2] These NCDs are mainly attributable to just four major risk factors. Furthermore, the contribution from poor diet exceeds the combined contribution from alcohol, tobacco and physical inactivity. [3] This poor diet mainly reflects a predominantly unhealthy global food environment, dominated by processed foods high in sugar, saturated fat, trans-fat and, crucially, salt. [3] In the UK and other high income countries, over 70% of dietary salt is consumed in processed foods such as bread, breakfast cereals, processed meats, snack foods, soups and sauces. [4][5][6] This food environment contributes to excessive salt intake among adults, on average 10g/day or more, [7] far in excess of what the body actually needs. [8] High salt intake is a major risk factor for increasing blood pressure, [9][10][11] cardiovascular disease, [12][13][14] stroke, [15,16] and stomach cancer. [17][18][19] Moreover, a reduction in salt intake would substantially reduce this risk. [10] WHO recommends a maximum adult salt intake of 5g/day. [20] Different strategies and policy options have been proposed to achieve this goal. Individual level interventions often involve behavioural approaches, for example dietary counselling, leaflets or medical advice. These are sometimes termed "downstream" or "agentic" interventions, and are dependent on the individual responding. [21,22] Conversely, "upstream" structural interventions take place at the population level and typically involve policies such as regulatory approaches, taxes or subsidies. Finally, intermediate interventions target subgroups in worksites, schools or communities. [23] National salt reduction strategies were identified in 75 countries in 2015, a substantial increase from 32 in 2010. [24] However, the debate regarding the most effective and acceptable salt reduction strategy continues.
Notable policy approaches have been seen in Finland, [25] Japan, [26] and more recently, the United Kingdom. [27] In the UK, a combination of awareness campaigns, agreed target settings, voluntary reformulation from industry and population monitoring of salt consumption have led to a 1.4g per day reduction in population salt intake between 2001 and 2011 (the campaign started in 2003).
[27] However, health inequalities in salt consumption have persisted. [28,29] Furthermore, the introduction of the UK Responsibility Deal in 2010 shifted emphasis to 'downstream' interventions, coupled with ineffective voluntary agreements and, controversially, the direct involvement of the industry in policy decisions. [30,31] Geoffrey Rose famously advocated population wide approaches rather than targeting highrisk individuals.
[32] Furthermore, there seems to be some evidence for a public health 'effectiveness hierarchy' whereby "upstream" structural interventions consistently achieve larger improvements in population health, are more equitable and often reduce health inequalities [33,34] compared to "downstream" agentic interventions targeting individuals, for instance in tobacco control and alcohol policies. [35,36] Emerging evidence suggests that a comparable effectiveness hierarchy might also exist for salt reduction strategies, whereby upstream interventions apparently achieve bigger reductions in salt intake. [37,38]. To test this hypothesis and hence inform future preventive health strategies, we have systematically reviewed the evidence for studies focusing on the effectiveness of salt interventions to reduce salt intake.

Study design
We conducted a systematic review of interventions intended to decrease population dietary salt intake. To ensure proper conduct, we adhered to the PRISMA checklist (Preferred Reporting Items for Systematic Reviews and Meta-Analyses)(S1 Table). [39] We used a narrative synthesis and formally investigated evidence to support or refute an effectiveness hierarchy. The research protocol can be found in S1 File.

Search strategy
We first identified exemplar studies to define and refine search terms needed for targeted searches. The search strategy consisted of a combination of four sets of key words: 1) salt, sodium; 2) health promotion, nutrition education, campaigns, dietary counselling, regulation, legislation, tax, self-regulation, reformulation, social marketing, promotion, provision, labelling, marketing control, primary care advice, food industry; 3) public policy, health policy, nutrition policy, policies, interventions, strategies, initiatives, programmes, policy option, actions; and 4) effectiveness, effect, intake, consumption, reduction, cost-benefit analysis, and cardiovascular diseases.
A pilot search was conducted to determine appropriate databases, identify relevant studies and highlight potential issues to be addressed. This process identified six databases which were then used for the targeted searches: Ovid MEDLINE, Science Citation Index, SCOPUS, Cochrane Database of Systematic Reviews, The Campbell Collaboration Library of Systematic Reviews and the CRD Wider Public Health database. We searched for all studies published in the last four decades (from 1975 onwards). The final searches were conducted on 30 October 2015. All papers identified by the searches were imported into the Zotero data management programme to identify duplicates and help screen titles, abstracts and full texts as appropriate. The reference lists of included studies were scanned for potential additional papers and topic experts (FPC and SC) were also consulted for additional data sources. [40,41]

Study selection and inclusion criteria
Studies were included if they investigated the effectiveness of specific interventions on population dietary salt intake and contained quantitative outcomes. Only studies in English were included. We included a wide range of study designs including meta-analyses, trials, observational studies and natural experiments. Empirical studies and modelling studies were analysed separately, in view of their profound differences. The retrieved studies were assessed using the PICOS approach (Participants, Interventions, Comparators, Outcomes and Study design), summarised in Table 1. The primary outcome was salt intake (g/day). Studies reporting urinary sodium excretion (mmol/day) or sodium mg/day were converted to g/day. Where necessary, we simultaneously considered studies reporting solely on salt intake data in a specific population with the corresponding studies describing the interventions during that same time period.
One reviewer (LH) conducted the searches; extracted potential papers and removed duplicates. Two reviewers (LH and AEG) then independently screened titles and abstracts for eligibility using the inclusion and exclusion criteria. Full text was retrieved for all papers deemed potentially eligible and these were also screened independently by the two reviewers. Any discrepancies were resolved by consensus or by involving the senior author (SC).

Data extraction and management
Pre-designed and pre-piloted tables were used to extract data from all included studies. To ensure that all relevant information was captured, extracted data included: first author; year of publication; funder(s); study aim(s); sample size; study design; methods; participants; policies analysed; geographical scope; length of follow-up; outcomes, effect and response;

Include Exclude
Studies for all age groups from all populations, from high-, middle-and low-income countries

Interventions
Systematic Reviews and primary studies evaluating the effects of actions to promote salt reduction by government policy or adopted in specific real or experimental settings Studies evaluating the effect of a general or specific diet

Comparators
Systematic and non-systematic reviews where actions to promote salt reduction were evaluated or compared No comparisons of different actions to promote salt reduction presented

Outcomes
Primary outcome of interest was dietary salt intake (g/day). Studies including urinary sodium excretion as an outcome were converted to g/day. Secondary outcomes included changes in clinical/physiological indicators related to NCDs and behaviours associated with a healthy diet Process evaluations reporting on implementation of interventions/policies without any quantitative outcome data; feasibility or acceptability without an assessment or primary outcomes (intake); studies on individuals as opposed to populations; data on cost only and BMI authors' assessment of limitations and our own assessment of potential risk of bias. The sources referenced for the effect sizes used in each modelling study were also specified in the tables (recognising that some modelling studies are based on empirical studies, potentially some included in this review). This data extraction was done independently by two reviewers (LH and AEG).

Quality assessment of included studies
Two reviewers (LH and AEG) independently assessed the methodological quality of each study (poor, fair or good). We used the National Heart, Lung and Blood Institute (NHLBI) tools specific for each research design (i.e. RCTs, cross-sectional studies, before and after studies, and systematic reviews).
[42] Several questions were asked for each study design (varying from 8 to 14) and depending on the points scored, the studies were labelled as good, fair or poor. However, we also took into consideration as to which questions points were allocated. For example, if an RCT scored 10 out of 14 points, but did not conduct an intention to treat analysis, it would be rated as fair rather than good.

Data synthesis and effectiveness hierarchy continuum
The evidence was summarised as a narrative synthesis according to intervention type, ranging from downstream to upstream interventions, to facilitate comparisons between the interventions. Summary tables of the studies included in this review can be found in Tables 2-10 for  empirical studies and Table 11 for modelling studies. A more detailed data extraction of these studies can be found in S2 Table. We defined UPSTREAM interventions as those targeting the entire population (not a subset, however large) and creating structural changes (effectively removing individual choice from the equation). This accorded with the Nuffield's ladder taxonomy, [44] and with McLaren's structural/agentic continuum. [21] Conversely, we defined DOWNSTREAM interventions as those where the principal mechanism of action is "agentic", being dependent on an individual altering their behaviour.
Interventions were then categorised according to their position in the McLaren et al. (2010) continuum from "upstream" to "downstream" (Fig 1). [21] Multi-component interventions were considered separately.

Patient involvement
Individual patients were not involved in this research; this is a secondary analysis of published data.

Results
The literature search identified 3336 potentially relevant papers. An additional 26 papers were identified through other sources, including reference lists and key informants. After removing 836 duplicates, 2526 publications were left to be screened by title and abstract, after which 134 full-text papers were assessed for eligibility. A total of 70 papers were finally included (49 empirical studies and 21 modelling studies, Fig 2).    reported an overall reduction in salt consumption of 2.8g/day at 12 months and 2g/day up to 60 months.
[45] The two meta-analyses overlapped in respect of only three studies. One additional RCT found a statistically significant net reduction of 0.6g/day between the groups,[48] whilst a second RCT found no effect between the control and intervention group. [50] All three modelling studies predicted that dietary advice is less effective in reducing the disease burden of high salt intake, only gaining 180-2,600 quality-adjusted life years (QALYs) compared to other interventions (7,900-195,000 QALYs). [54][55][56] Dietary counselling-school based and worksite interventions (Table 3) Three school-based interventions (one of good quality;[40] one of fair quality; [57] one of poor quality [58]) and three worksite-based studies (all of fair quality) were included. [59][60][61] No modelling studies were identified for this section.
Schools. A nutrition programme in schools aimed at distinguishing between healthy and less healthy choices reported a non-significant reduction. [58] In the second school based RCT, the practical intervention group achieved a significant net reduction of 0.7g/day compared with the control group. [57] In a cluster RCT in China, education and training significantly reduced salt intake by a mean of −1.9 g/day in 279 school children (and −2.9 g/day in adult family members). [40] Worksites. A randomised trial of a chronic disease prevention programme achieved a net reduction of 1.2g/day between the intervention and control group (P = 0.01). [59] A factorybased intervention study in China assessed health education aimed at altering diet, together with a high-risk strategy of hypertension control. Salt intake was reduced by 3.9g/day from a mean of 16g/day (P<0.05). [60] Dietary counselling-community level (Table 4) Four empirical studies and one review, all of fair quality, [62][63][64][65][66] investigated community based dietary counselling. One study reported a statistically significant difference of -0.4g/day in salt intake between the intervention and control groups. [62] Two intervention trials of nutrition education reported significant reductions of 0.7g/day and 2.2g/day reductions Systematic review of dietary salt reduction policies respectively in salt intake after 12 months. [63][64] One RCT reported a favourable trend; however, this was non-significant and could have been caused by contamination between the groups. [63] Mass media campaigns (Table 5) One empirical study of fair quality [67] and five modelling studies; four of good quality [56,[68][69][70] and one of fair quality [71] were included.

Review Ghana and China
Aim: to evaluate the effectiveness of the community-based interventions for CVD prevention programmes in LMIC Outcomes: BP and salt intake (g/day and n, %)

Huang et al. (2011)
• BP: reduction prevalence HT 12.9%* pre vs post • Salt: reduction in salt intake 30%* (n, %) The UK FSA salt reduction programme involved media campaigns to discourage table salt use, plus sustained pressure on industry to reformulate. Although salt consumption declined by 0.9g/day using spot urinary sodium readings from 2003-2007, the media contribution was unclear but likely modest. [67] The modelling studies likewise suggested media campaigns were generally considered less effective than food labelling or reformulation. [56,[69][70][71] The Change4Life campaign in the UK was predicted to reduce salt intake by 0.16g/day, less than labelling or reformulation. [68] Gillespie et al. (2015) similarly estimated that social marketing might modestly reduce salt consumption by 0.03g/day to 0.13g/day. [69] Nutrition labelling (Table 6) Two empirical studies, both of poor quality, investigated the effect of nutrition labelling on salt intake [72][73]. Reduced salt intake was not observed in participants who reported frequent vs. non-frequent label use (7.7g/day vs. 7.6g/day). [73] Ten modelling studies also examined labelling, four of good quality [56,[68][69][70] and two of fair quality. [71,[74][75][76][77] These suggested that labelling might modestly reduce UK salt intake by  [68,69]; much less than the 0.9g/day estimated by Roodenburg et al. (2013). [77] Another study suggested that salt intake might be lowered by 1.2g/day if the population were to choose products labelled as low-salt, or increased by 1.6g/day if they choose products labelled as high salt content. [74] Reformulation (Table 7) Very few studies which focused on reformulation included quantified results of salt intake. In one empirical Taiwanese study of fair quality, [78] salt was enriched with potassium in the intervention group and their outcomes were an apparent reduction in cardiovascular deaths by 41%, compared to the control group rather than salt intake. Furthermore, people in the intervention group lived 0.3-0.9 years longer. [78] Fourteen modelling studies evaluated reformulation, eleven of good quality[41, 54-56, 68-70, 79-82] and three of fair quality [71,83,84]. Mandatory reformulation could consistently achieve bigger salt reductions than voluntary reformulation; 1.6g/day compared with 1.2g/day; [68] and 1.4g/day versus 0.5g/day. [69] Mandatory reformulation might also achieve higher reductions in disability-adjusted life years (DALYs) and QALYs compared to voluntary reformulation. [54,56,79]      In the Netherlands, reformulation of processed foods was predicted to reduce median salt intake by 2.3g/day, [84] compared with a 0.9g/day from a two-year salt reformulation initiative in Argentina. [82] Fiscal interventions (Table 8) Two systematic reviews of fair quality [85,86] included three modelling studies eligible for this review. Furthermore, three additional tax modelling studies were included, all of good quality. [56,81,87] Two studies included in Niebylski et al's. systematic review (2015) modelled a 1% tax on salty snacks or on cheese and butter; neither reduced salt consumption. [86] Another modelling study suggested that a very high (40%) tax might achieve a 6% reduction in salt consumption (0.6g/day). [81] One modelling study predicted that a 20% tax on major dietary sodium sources might prevent or postpone 2000 deaths annually, [87] whilst Nghiem et al. (2015) predicted that a sodium tax might gain more QALYs than other interventions. [56] Multi-component interventions (Table 9 and Table 10) Fifteen papers were included under multi-component interventions. Most studies came from Japan, Finland and the UK. Two were of good quality; [88,89] ten of fair quality; [24,43,[89][90][91][92][93][94][95][96] and four of poor quality. [97][98][99][100] Four studies were included which presented dietary salt intake and linked to papers describing the interventions; (one of good quality; [25]; two of fair quality; [101,102] and one of poor quality. [103] Japan. The Japanese government initiated a sustained campaign in the 1960s. [26] Over the following decade, mean salt intake fell from 13.5g/day to 12.1g/day overall (and from 18g/ day to 14g/day in Northern Japan). Miura et al. (2000) reported that salt intake subsequently decreased from 14.5g/day in 1972 to 10.6g/day in 2010, a fall of almost 4g/day [103]. Stroke mortality was predicted to fall by 80%. [90,93] Finland. Starting in 1978, Finland pursued a comprehensive salt reduction strategy using mass media campaigns, mandatory labelling and voluntary reformulation by the food industry. Population salt consumption was monitored regularly by using 24h urinary assessment and dietary survey data. [72] By 2007, salt intake had reduced by approximately 4g/day, from 13 to 8.3g/day in men, and from 11 to 7g/day in women. [24,25] Stroke and coronary heart disease (CHD) mortality fell by over 75% during that period. [90] United Kingdom. The UK salt reduction strategy included voluntary reformulation, a consumer awareness campaign, food labelling, target settings and population monitoring. [95] By 2011, population salt intake, measured by 24h urinary sodium excretion, had decreased by 1.4g/day (9.5g/day to 8.1g/day) [88]. He et al. (2014b) estimated that this might reduce stroke and coronary heart disease mortality by some 36%. [88]     Salt labelling If the entire population were to choose low-salt breads, cheeses, processed meat and fish, fat spreads, and breakfast cereals, then salt intake could be lowered by 1.5 g in men and by 0.9 g in women. If everybody was to select high-salt products, then salt intake would go up by 1.9 g in men and by 1.2 g in women. Thus, the potential difference between the low and the high alternatives would be 3.4 g in men and 2.9 g in women. If all prepared foods had a reduced salt content, the mean salt intake would go further down by 2.3 g in men and by 1.7 g in women. Excluding under reporters, the mean salt intake was 11.1 g in men and would go down to 9.5 g if all men chose lightly salted products and further down to 6.8 g if also all prepared foods would have a lower salt content. In women, the respective numbers are 7.8, 6.7 and 4.9 g Fair Temme et al. Modelling study Brazil Aim: to evaluate the impact of introducing products that are in agreement with the Choices criteria in the usual diet. Outcomes: salt intake Food Labelling Salt would still be considered an important reduction, 36% in relation to the typical menus (TM), but it would be 7.6 g/day, which is above the recommended by the program (6.5 g/day).

Salt (g/day)
• Typical menus (TM): 11.9 Modelling study The Netherlands Aim: to evaluate these nutritional criteria by investigating the potential effect on nutrient intakes Outcomes: salt intake Labelling A reduction of -23% for sodium was seen for sodium compared to the 'actual scenario'.

Salt (g/day)
• Reformulation (restaurants and manufacturers) Restaurants and manufacturers reaching agreed upon sodium targets. Expansion of the initiative to ensure all restaurants and manufacturers reach agreed-upon sodium targets would be expected to avert from 0.9 to 3.0 MIs (a 1.6%-5.4% reduction) and 0.5 to 2.8 strokes (a 1.1%-6.2% reduction) per 10,000 Americans per year over the next decade, after incorporating consumption patterns and variations in the effect of sodium reduction on blood pressure among different demographic groups. The expanded NSRI covering both packaged and restaurant food items would be expected to reduce mean daily sodium intake by 447 mg per person per day on average, or 13.0%. If the NSRI included only restaurant food items, the program would lower MI and stroke mortality by an estimated 2.7% (95% CI, 1.4-4.0) and 2.1% (95% CI, 0.4-3.9), respectively. Hence, most of the benefit from the program would likely be due to sodium changes among packaged foods Good Murray et al. Other countries have implemented several strategies including labelling, media campaigns and voluntary reformulation and effect sizes ranged from -0.4g/day in France [24,93] to -4.8g/ day in China [24,102].
Modelling studies of combined interventions. Six modelling studies investigated the effect of multi-component interventions, three were of good quality; [70,104,105] whilst three others were of fair quality. [70,106,107] Several modelling studies consistently suggested that multi-component salt reduction strategies (e.g. labelling, health promotion and reformulation) would be more effective than any single intervention. [70,71] For instance, Gase et al. (2011) suggested that using labelling, promotion, subsidies and provision of low sodium options could lead to a 0.7-1.8g/day reduction. [106]

Main results
This systematic review of salt reduction interventions suggests that comprehensive strategies could generally achieve the biggest reductions in salt consumption across an entire population, approximately 4g/day in Finland and Japan, 3g/day in Turkey and 1.3g/day recently in the UK. Mandatory reformulation alone could achieve a reduction of approximately 1.4g/day, followed by voluntary reformulation (median 0.7g/day) school interventions (0.7g/day) and worksite interventions (+0.5g/day). Smaller population benefits were generally achieved by short-term dietary advice (0.6g/day), community-based counselling (0.3g/day), nutrition labelling (0.4g/day), and health education media campaigns (-0.1g/day). Although dietary advice to

Comparison with other research
Geoffrey Rose famously argued that a greater net benefit came from the population-wide approach, (achieving a small effect in a large number of people) when compared with targeting high risk individuals (a large effect but only achieved in a small number of people).
[32]  Multi-component interventions. Multi-component salt reduction strategies involving a series of structural initiatives together with campaigns to increase population awareness have been successful in Japan and Finland where they substantially reduced dietary salt consumption, and associated high stroke and cardiovascular disease mortality rates. In Finland, some credit should also go to other dietary changes e.g. fat quality. [108] Between 2003 and 2010, a multi-component approach in the UK including voluntary reformulation and political pressure on industry to agree category-specific targets achieved some success (1.3g/day reduction in population salt consumption over 8 years to 8.1g/day in 2011). Interestingly, pre-existing health inequalities in salt consumption persisted.
[29] However, from 2010, the Responsibility Deal simply advocated a voluntary scheme. This was ineffective, and MacGregor therefore subsequently recommended mandatory reformulation.
[31] Other useful reductions were demonstrated in other countries mostly using dietary surveys and some from grey literature. However, the -4.8g/day reduction reported in China appears extra-ordinarily large and perhaps merits some caution [24]. Multi-component interventions clearly have more potential than single interventions, and synergies might be anticipated. [13,93] Similarly powerful benefits have also been observed with comprehensive strategies for tobacco control and alcohol reduction. [35,36] Reformulation. In high income countries, the majority of dietary salt intake comes in processed food (75%) and reformulation can be very effective in reducing salt consumption. [109] Though mandatory reformulation is more powerful, most countries currently use voluntary reformulation. [54,56,68,69,110] Success may then be very dependent on the degree of political pressure applied to the food industry and on regular, independent monitoring, as recently achieved in the UK. [111,112] Food labelling. Nutrition labelling can be potentially effective, as demonstrated in Finland [72] and Brazil. [74] Nutrition labelling allows consumers to make informed choices whilst also putting pressure on the food industry to reformulate. [89] However, interpretation of labels depends on health literacy and different labelling systems may confuse consumers, [113] and reinforce inequalities. [29]. Consumers generally want simple (traffic light) labels which are easier to understand. [76,77,113,114] Dietary interventions in diverse settings: communities, worksites, schools and homes. Dietary interventions can be delivered at different levels, such as communities, worksites, schools or to individuals. However, effectiveness varies widely.[45,47,50] Furthermore, the benefits of dietary counselling decrease over time and are thus generally not sustainable; much smaller reductions might therefore be anticipated in unselected individuals in the general population.
[44] Furthermore, for many individuals, issues such as competing priorities and financial constraints might reduce compliance and adherence, [8,13,21,22] and thus reduce net population benefits.
Mass media campaigns. Few empirical studies have examined salt media campaigns. However, benefits appear to be generally modest. [56,67,68,69,115] or negligible. [111] Many individuals may not perceive any personal relevance and hence fail to engage in any behaviour change. [22,116,117] Taxation. Price increases can powerfully reduce consumption of tobacco or alcohol. [35,36] However, salt is cheap, and a substantial tax of at least 40% might be needed to reduce consumption by just 6%. [81,118] Public health benefits and cost-effectiveness Most economic analyses have consistently predicted substantial reductions in cardiovascular mortality, and consequent gains in life-years, QALYs, DALYs and healthcare savings. This is consistent with the growing evidence that population-wide prevention policies can often be powerful, rapid, equitable and cost-saving.[38, [119][120][121][122] Several modelling studies also investigated the cost-effectiveness of the salt interventions described above. Mandatory and voluntary reformulation appeared far more cost-effective than labelling or [54,55,68] dietary advice targeting individuals. [122]

Strengths and limitations
This systematic review has multiple strengths. Firstly, two independent reviewers screened all papers and assessed quality using appropriate validated tools. Secondly, the inclusion of modelling studies (presented separately) adds value by allowing the evaluation of certain interventions where empirical studies failed (e.g. labelling). In addition, we recorded the effect size used in each modelling paper together with the source reference. Furthermore, most of the better quality modelling studies confirmed the superiority of upstream approaches. Finally, the studies reviewed included a wide variety of interventions, thus providing a useful spread of estimates.
Our review also has limitations. We were unable to conduct a formal meta-analysis due to the profound heterogeneity of the diverse studies, many of which included multiple interventions. Furthermore, studies were only included if the full text was available in English (15 non-English papers were excluded). We also had to exclude two potentially relevant studies which lacked the full text. [123,124] Publication bias remains possible, potentially over-estimating the true effect of some interventions. The primary outcome of this study was dietary intake (consumption); we excluded studies considering other dietary behaviours such as awareness, knowledge, preferences or purchasing behaviour. Also, the positive benefits of policy changes may sometimes appear larger if favourable underlying secular trends have not been formally considered. Furthermore, we did not contact authors for missing data. However, all the key information was presented in all but two papers. [123,124] Finally, generalization of the results should be cautioned as countries may vary in baseline salt intake.

Future research
This review highlights the greater power of combined (multi-component) strategies, mandatory reformulation and traffic light labelling. Most were cost-effective and many were cost-saving. However, the feasibility of implementing policy changes also deserves further study. Many factors can facilitate or obstruct successful policy development, notably including political feasibility and stakeholder influence. [114,131,132] Stoeckle and Zola's "upstream"/"downstream" concept was disseminated by John McKinlay, [133] critiqued by Krieger, [134] and then refined as a structural/agentic continuum by McLaren et al 2010. [21] To test our effectiveness hierarchy hypothesis, one ideally needs to quantify the "average" effect of each category of salt reduction intervention. Yet, the limited number and heterogeneity of these studies precludes a formal meta-analysis. However, the consistency with the effectiveness hierarchies demonstrated by tobacco and alcohol control interventions is encouraging. The effectiveness hierarchy hypothesis now clearly needs to be tested in other fields.

Conclusions
There are clear implications for public health. The biggest population-wide reductions in salt consumption were consistently achieved by comprehensive multi-component strategies involving "upstream" population-wide policies (regulation, mandatory reformulation, and food labelling)."Downstream" individually-based interventions appear relatively weak (e.g. dietary counselling to individuals and school children, and media campaigns in isolation).
This 'effectiveness hierarchy' might deserve greater emphasis on the agendas of the WHO and other global health organizations reviewing action plans for NCD prevention.
Supporting information S1