What is the impact of food reformulation on individuals' behaviour, nutrient intakes and health status? A systematic review of empirical evidence

Food reformulation aimed at improving the nutritional properties of food products has long been viewed as a promising public health strategy to tackle poor nutrition and obesity. This paper presents a review of the empirical evidence (i.e., modelling studies were excluded) on the impact of food reformulation on food choices, nutrient intakes and health status, based on a systematic search of Medline, Embase, Global Health and sources of grey literature. Fifty‐nine studies (in 35 papers) were included in the review. Most studies examined food choices (n = 27) and dietary intakes (n = 26). The nutrients most frequently studied were sodium (n = 32) and trans fatty acids (TFA, n = 13). Reformulated products were generally accepted and purchased by consumers, which led to improved nutrient intakes in 73% of studies. We also conducted two meta‐analyses showing, respectively, a −0.57 g/day (95%CI, −0.89 to −0.25) reduction in salt intake and an effect size for TFA intake reduction of −1.2 (95% CI, −1.79 to −0.61). Only six studies examined effects on health outcomes, with studies on TFA reformulation showing overall improvement in cardiovascular risk factors. For other nutrients, it remains unclear whether observed improvements in food choices or nutrient intakes may have led to an improvement in health outcomes.

NCDs. Exposure to a heathy food environment has been shown to be a stronger driver of healthy eating than health promotion and education efforts. 2,3 Food environment interventions include changes to the availability, price, information or composition of food products.
For example, food environments can be changed through reformulation of packaged and processed foods. Reformulation, in the context of the prevention of NCDs, is defined as a change in the nutrient profile of a food with the goal of making it healthier for consumers. If successful, food reformulation strategies will improve dietary intakes by changing the composition of foods without changing consumers' eating habits.
Food reformulation strategies became popular in the mid 2000's when governments and manufacturers focused on the removal of industrially produced trans fatty acids (TFA) and the decrease in salt content of manufactured products. Given that an average of 46% of daily energy intakes in Europe * is from processed foods, 4 reformulation has the potential to improve the dietary intakes of the population, if applied consistently.

| The role of food reformulation in public health
Food reformulation can take many forms and can be driven by different motivations and incentives. A public health model of food reformulation typically involves the enactment of policies deploying regulation or incentives leading manufacturers to improving the nutritional properties of food products by changing their composition. Reformulated food products would replace, rather than add to, preexisting versions of the same products and would be marketed in ways that would preserve consumers' acceptance and liking of the products after reformulation, for example, through gradual and 'silent' reformulation that would not be perceived by consumers as altering the food product's sensory characteristics.
Reformulation would, in this way, change dietary intakes without significantly altering food choices, potentially leading to health improvements.
In practice, however, food reformulation initiatives often deviate from the model described above, even when they retain a public health goal. A common deviation is that reformulated products add to preexisting versions (which should be more appropriately labelled as diversification of food products, rather than reformulation) compounding food choices and their impacts on dietary intakes. This strategy can be favoured by manufacturers as adding products create product diversity that can be a source of increasing sales. When this happens, marketing strategies may actively encourage consumers to switch to the reformulated versions of their products, which may trigger a wider range of substitutions in consumers' food choices. In addition, food reformulation may not be prompted by an explicit government policy but rather be driven by voluntary agreements, industry pledges, corporate social responsibility motivations, which often involve less systematic approaches and a reduced scope for food reformulation.
This review is not limited to one or another approach to food reformulation. However, the approach taken may contribute to determining the effectiveness of food reformulation in improving dietary intakes and health outcomes; therefore, this dimension was assessed qualitatively in the review.

| Existing systematic reviews
Several systematic reviews have studied the effect of reformulation (sometimes amongst other policies) on the consumption of specific nutrients, mainly sodium, [5][6][7] TFA, [8][9][10] and more recently sugar. 11 Reformulation was effective to reduce populations' sodium intakes, 5,7 with a greater reduction observed in countries where reformulation is mandatory compared to countries where it is voluntary. 7 Similarly, the reduction of TFA intakes was greater in countries where TFA reduction is mandatory, than in countries where it is voluntary, or in countries implementing solely labelling policies. [8][9][10] Overall, these reviews show that reformulation is most effective in improving dietary intakes when part of multiple-component interventions designed to change the food environment. 5,7,9 The systematic review on the effect of sugar reformulation reported that reformulation could lead to reductions in sugar intakes and in body weight. 11 However, these findings derive from studies in controlled environments; the effect of reformulation implemented as a population-wide intervention may be different. Previous reviews on food reformulation do not investigate consumers' reactions to reformulated foods, such as consumers' long-term acceptance of reformulated products. Consumer acceptance is a key driver of the effectiveness of reformulation in changing dietary intakes, in the absence of which unwarranted substitutions may take place. Also, no review included at the same time outcomes related to consumer choices, dietary intakes and health outcomes, the three steps through which reformulation may have an impact on public health.

| Aim of this review
This review aims to assess the empirical evidence of the impact of food reformulation on food choices, nutrient intakes and health status, with no restriction on the type of reformulation strategy employed, nor the nutrient targeted. When possible, we also report the effect of reformulation strategies on children.

| METHODS
The systematic review methodology was adapted from the Cochrane methodology for systematic reviews. The protocol is published in the PROSPERO database (CRD42019127624), 12  theoretical reformulation. The term 'reformulation' refers to a change in food composition being part of a specific food environment. One form of reformulation consists of replacing the original product with a similar looking (including its name) and tasting product that has an improved nutritional profile. Other methods used by manufacturers include adding a new product derived from an existing one, or reducing the portion size of products. To study similarities or dissimilarities across all examples of reformulation, there was no restriction on the type of reformulation strategy employed (i.e., studies were included regardless of which nutrient change the strategy was focused on changing).
Control condition could be the same group before the intervention started (single group pretest and post-test), or a comparable group not facing the intervention. Outcomes of interest included those that investigated the individuals' response to reformulated foods, such as purchase patterns and food choices, nutrient intakes or resulting health outcomes. Studies describing solely the composition of the food environment but no outcome relating to individuals' choices were excluded.

| Search strategy
We employed keyword searches through EMBASE, MEDLINE and Global Health for peer-reviewed studies published until December 2018 (the search strategy for Medline is available in Appendix A).
The search strategy was refined by conducting sensitivity analysis in EMBASE with a test set of 15 key papers. Adjustments to the search strategy concluded once 85% of the key papers were identified. We also included references from four relevant systematic reviews. 5,[8][9][10] Grey literature was searched using the NOURISHING database of reformulation initiatives and following advice from experts.

| Screening, data extraction and data synthesis
Studies retrieved from the searches were sequentially screened by title, abstract and full text by one author (MG). A second author (ABS) independently reviewed a sample of 10% of studies based on titles and abstracts. The full text of all studies included by the first author were screened by the second author.
We extracted data about intervention, study design and outcome from the included papers; data on subgroups of the population (defined by age, sex and socio-economic status) were extracted when available.
For studies similar enough to be pooled, we performed a metaanalysis using inverse-variance weighting with random effects. Mean differences were used when outcomes were reported in the same unit. Standardized mean differences (Hedge's g) were used otherwise as a measure of the effect size. Between-study variance was estimated using restricted maximum likelihood as recommended for continuous outcomes. 14 Heterogeneity was assessed with the I 2 statistics, and meta-regression was performed to explain heterogeneity. Sensitivity analyses were conducted by restricting the sample to studies with the least risk of bias (see section below). Publication bias was assessed with funnel plots. Otherwise, a narrative review was used to qualitatively analyse extracted data.

| Risk of bias assessment
Risk of bias for each study was assessed using an adapted version of the Newcastle-Ottawa Scale (NOS) for assessing the quality of nonrandomized studies, given that most studies were designed as natural experiments. Studies were rated on four domains: selection (representativeness of the sample and ascertainment of exposure), comparability of the two groups, outcome (ascertainment of outcome and T A B L E 1 PICO (population, intervention, comparator, outcome) table for the selection of studies duration of follow-up) and adjustment (control for confounders and evaluation of reformulation independent of other mechanisms). An overall score was then calculated (out of seven), reflecting decreasing levels of risk of bias.

| RESULTS
The search strategy retrieved 11,315 studies from the three data- in the context of reformulation. 5,18,24,25,29,[31][32][33][34][35][36][37][38][39][40][41][42] Only six studies reported on outcomes linked to reduction of NCD burden such as diet-related risk factors or diseases. [43][44][45][46][47][48] The majority of studies were conducted in high-income countries; only three of the 59 studies were conducted in upper-middle-income countries (Costa-Rica and Turkey) (Appendix C). Three of the 32 initiatives were mandatory (i.e., bans on the use of TFA), 19 resulted from a rise in consumer awareness (through public health campaigns or nutrition labelling), six were designed as public-private partnerships (with targets to incentivize manufacturers to reformulate their products), two were manufacturer-led and one was a real-condition experiment led by a research group (Appendix C).
The adapted NOS risk of bias tool was effective in differentiating the risk of bias between studies (Appendix D). The domain representing the higher risk of bias was the adjustment domain, with only seven papers (16%) having a point in the two criteria of the domain (adequate control for confounders: 34%, and evaluation of reformulation prevented from other input 27%). Other domains from the risk of bias scale had better compliance, with between 64 and 80% of papers with a low risk of bias.

| Effect on consumer behaviour and food choices
Overall, reformulated food products were accepted and consumed by the population, that is, when food products were reformulated; those products were purchased, resulting in an improved composition of the population purchases. This means that reformulation did not trigger specific behavioural response. A favourable change in the composition of a food category or of household's basket of purchase was observed in 22 of 27 studies after products were reformulated ( Table 2). In those studies, the nutrient content of individual products was weighted by their sales (or purchased quantities) to have a measure representative of households' intakes.
Studies reporting on the acceptability of reformulated products analysed sodium (n = 10), TFA (n = 3), energy reductions (n = 3), fibre (n = 2) or whole grain (n = 2) increases, or an improvement in several nutrients at the same time (n = 5). The extent to which F I G U R E 1 PRISMA flow chart showing the screening of records foods were reformulated changed depending upon the nutrient targeted by reformulation initiatives (i.e., the relative change was different across nutrient targeted). For example, the percentage reduction of TFA in reformulated products (between 80% to almost 100% 16,43 ) was substantially larger than that of reductions in total energy (reductions of around 0.4%-3% 27,30 ) or sugar (reductions of around 2%-3% 17,21 ). The extent to which foods were reformulated also depends on the food category. For example, categories that had the most important salt reduction were cereal products, soups, sauces or cold cuts. 17,24,25,30 There was evidence that the larger the improvement of the composition of the food, the larger the effect was on consumers' purchases were observed. For example, intense reduction in food TFA led to large reduction in TFA purchase (restriction of restaurants TFA to 0.5 g/serving led to a 85% reduction in TFA purchases from restaurants in New York City 16 ); whereas modest reductions in food energy density led to smaller reduction in energy density of total purchases (Clapp et al 23 observed an overall reduction of 1.1% in total purchase, following reductions in foods of around 0.1% to 4.4%).

| Effect on nutrient intakes
Twenty-six studies reported the effect on nutrient intakes; 70% of these studies concluded that reformulation led to improved intakes of the relevant nutrient ( Table 2). Studies that used biomarkers (from blood or urine) or reported measures (dietary surveys or total purchases taken as a proxy) of nutrient intakes showed similar positive results, with 77% and 69%, respectively (Appendix E). Most of these studies were focused on sodium (n = 20); and of these studies, 13 reported reduced sodium intakes, ranging from a decrease in sodium intake between 4% and 15% per year (Table 3, and meta-analysis below for estimates of mean differences). Five studies evaluated the effect of TFA reduction and again reported positive effects, ranging from a total decrease between 38% and 85%, measured over a range of seven to 19 years (Table 3)  Note. Positive results were defined as a significant change in average nutrient density of purchased products or a change in nutrient intake going in the direction of an improvement for public health (i.e., reduction for sodium, TFA, energy or sugars, increase in fibres or whole grains and improvement of the nutrient profile of foods), or a reduction in disease risk or mortality. aa Acceptability of reformulated products was either measured using sales/purchases of the reformulated product before and after reformulation and the evolution of market-share weighted averages of a nutrient content before and after reformulation. A positive effect on intake was defined as a significant change in average nutrient density of purchased products or a change in nutrient intake going in the direction of an improvement for public health (i.e., reduction for sodium and TFA and increase in whole grains). b Significant at (at least) 5%. Statistic tests were not performed in the study.

| Effect on health status
Six studies investigated the empirical impact of reformulation initiatives on health status (i.e., morbidity and mortality), with five of these studies showing an improvement in health status (Table 3). Of these six studies, five focused on the effect of TFA, whereas only one focused on the effect of sodium reduction. The morbidity or mortality from cardiovascular diseases was reduced in four of the five studies that evaluated the effect of TFA bans in packaged foods or restaurant foods, where mortality was reduced by between 4.3%-6.2% (Table 4).
The sodium reduction study concluded that the effect of the UK intervention had a positive impact blood pressure.
F I G U R E 2 Pooled estimate of the effect of initiatives including salt reformulation on populations' salt intake (in g/day), by measurement method used to measure a change in dietary sodium

| Isolated effect of reformulation
It was difficult to isolate the effect of reformulation, as many stud-

| Effect on subgroups of the population
Only three studies evaluated the effect of interventions specifically in children or adolescents. 31,32,38 Studies showed similar effects across age groups. The majority of studies did not report results by gender.

| DISCUSSION
Overall, reformulated products were accepted by consumers, although evidence differs by nutrient. Several studies (10) showed that salt-reduced products were accepted, whereas a limited number of studies found that products reduced in sugar or increased in fibres were less likely to be accepted by consumers. Acceptance in this context means that the reformulated food product led to an improved nutrient composition of total food purchased and hence in intakes. Compensation (overconsumption or a change in dietary patterns such as a switch towards non-reformulated products) did not offset the benefits of reformulation (70% of studies showed a positive overall effect of reformulation). Nonetheless, compensation did occur (e.g., for salt) in some studies in which reformulation did not lead to decreased intakes or led to a smaller decrease than predicted from changes in food composition. 18,35 Two studies showed that the effect of reformulation was partly offset by consumers switching to less-healthy options. 20,39 Studies using experimental settings found that while abrupt reformulation was noticed by consumers and led to compensation, 49,50 silent reformulation did not led to compensatory behaviours hence reduced intakes of sodium or saturated fatty acids. 51,52 Due to the variability in initiatives' nutrient focus and scope

| Health outcomes
Five of six studies examining health outcomes show that reformulation can provide health benefits by improving the nutritional properties of foods. 53 However, a link between reformulation and health outcomes was only observed for TFA reformulation, and it cannot be generalized to other nutrients. The extent of the reformulation was also a factor driving impact on health outcomes.
The combination of a complete removal of TFA from the food supply in some countries, in addition to the strong correlation between TFA intake and health, 54

| Intervention designs
This review suggests that to have the expected favourable impact on consumers' intakes and health, reformulation interventions should have large scope across and within food categories that are the major sources of the targeted nutrient. This is because the effect of reformulation can be offset if consumers switch to similar products that have a worse nutrient profile. For all nutrients, it appeared that a reformulation across the maximum of food products, and covering most of food categories (and at a measurable size) was needed to see significant changes. 18,21,35 The absence of effect was often explained by a small proportion of products reformulated, or a too slow pace. 18,35,57 Strategies are generally designed to promote the reformulation of the foods that represent the leading sources of the targeted nutrient. Foods reformulated to reduce TFA were mostly margarines, fried products, or biscuits although most countries had a limit (either a ban or a 2% of fat limit) applying to all processed foods sold in retail and restaurants.
Foods most often targeted in reformulation programmes for the reduction of sodium intakes were cereal products, processed meats and soups. Foods such as biscuits or prepared meals were included in strategies focusing on multiple nutrients, as these foods can be at the same time sources of salt, sugar and TFA. Reformulating a comprehensive range of food products limits the possibility for consumers to switch to alternative sources of the targeted nutrient (e.g., nonreformulated foods). show that high-TFA products still remained in the market, meaning that some individuals were still exposed to high levels of TFA. 31 In

| Reformulation in the context of wider policies
Reformulation initiatives were often employed in conjunction with other initiatives, most commonly, initiatives to inform and educate consumers. Nation-wide reformulation initiatives for the reduction of sodium and TFA were deployed with public health campaigns that informed consumers about the harmful effects of excessive intake of those nutrients. 5,7 Further, front of pack labelling, claims or logos incentivized manufacturers to reformulate their products so that they attract consumers. 28,30,38,63 Systematic reviews on sodium and TFA showed that multicomponent strategies including a reformulation scheme were the most promising to improve diets. 5,7 In particular, interventions including regulations to change the price or composition of foods

| Reformulation alone is unlikely to reduce energy intake
Although reformulation is a promising strategy to improve the nutrient profile of foods, it is unlikely to lead to major reduction in the energy density of the food environment (i.e., reformulation is not contributing to obesity prevention). Most often, reformulation strategies included in this review led to products with the same energy density.
For example, if a strategy aimed to reduce sodium, TFA or sugar, the energy density of the product would be unchanged following the reduction of the respective nutrient. Reformulating foods to decrease a certain macronutrient is often done by substituting it with a macronutrient of the same energy density (e.g., TFA was replaced with other types of fat and sugars by carbohydrates, with the total energy density held constant 9,55 ). One way to decrease energy density of a product is to substitute its sugars with fibres, as fibres provide less energy than sugar for the same weight. 55 Given that reformulation is a strategy designed to gradually improve the food environment, reformulation may be an inadequate strategy to reduce total energy intake of individuals via decreasing the energy density of foods. This is mainly because it is unlikely that major reductions in energy density of products can be achieved in without changing the sensory characteristics of that product, especially texture and taste. 55 An exception could be reformulation of sugar-sweetened drinks, where the replacement of sugar with no-calorie sweeteners decreases the energy content of the drink. This reformulation is easier to implement than reducing sugar in solid foods, as highlighted by the latest report from Public Health England about their sugar reduction programme: the sugar was reduced by 3% in foods, but by 29% in beverages. 64 To reduce energy intake of populations, other strategies are needed, such as reducing portion sizes of products, and promoting healthier alternatives to energy-dense foods. 65

| Strengths and limitations of the review
This review is the first to evaluate the impact of reformulation inter-

| CONCLUSION
The evidence base examined in this review shows that food reformulation has the potential to improve people's diet and health.
Changes in the nutrient composition of food products translate into changes in the balance of nutrients from food purchased by consumers.
Overall, the evidence shows that a reduction in sodium or TFA contents  The effect of reformulation was estimated to be estimated on its own when the effect of the reformulation initiative was isolated from potential confounders (such as educational campaigns aiming at changing behaviours). b Positive results were defined as a significant change in the outcome measured, going in the direction of an improvement in public health (i.e. reduction of intakes for sodium, TFA, energy or sugars, increase for fibres or whole grains), or a reduction in disease risk or mortality.
c Positive (+) results were defined as a significant change in average nutrient density of purchased products or a change in nutrient intake going in the direction of an improvement for public health (i.e., reduction for sodium, TFA, energy or sugars, increase in fibres or whole grains and improvement of the nutrient profile of foods), or a reduction in disease risk or mortality. Negative or the absence of impact are denoted with The number of the reference is specified when the short citation is ambiguous.  F I G U R E G 1 Sensitivity analysis of the pooled estimate of the effect of initiatives including salt reformulation on populations' salt intake (in g/day). Only studies with a NOS Score >2 were selected [Correction added on November 28, 2020 after first online publication: Figure G1 has been updated.]