Data types and sources

Data on the salt content of foods were acquired from 2 sources, depending on the date. Data for the years 2015 to 2018 were sourced from a commercial third-party supplier, Edge by Ascential (previously known as Brand View). Edge by Ascential collects product information, including nutrient composition data from the websites of 4 leading UK retailers: Asda, Morrisons, Sainsbury’s, and Tesco, with 2015 being the earliest data available. All data were collected on the same data (13 December) for each year. For 2018 to 2020, similar data were sourced from foodDB, a research database developed by the University of Oxford [19]. These data were collected in November and December for the same 4 retailers for 2018, 2019, and 2020. For 2018, data from both sources were used in the analysis.

The data from each source were combined to form a composition database with the following information: year, product name, brand name, company (manufacturer) name, ingredients, and salt content in g per 100 g. Products that did not have a salt per 100 g value were excluded, as were products that did not contain values for at least 2 other nutrients. No missing data or salt values were imputed. As barcode data were not available, duplicate products found in multiple retailers were removed by filtering on year, category, brand name, product size, kcal per 100 g, and salt per 100 g. An extra analysis was done to test the validation between the 2 sources of composition data (Edge by Ascential and foodDB) for 2018. The sales-weighted mean salt content by category was calculated for each data source, and a Wilcoxon signed-rank test was conducted in R to test for differences between the 2 independent samples, with the unit of analysis being the brand.

There were 374,849 individual products included in the original composition database for all years. A total of 80,960 products were excluded due to missing salt per 100 g values; 205,003 were excluded because they did not belong in one of the included categories (as outlined in Table 1); and 105 were excluded because they were missing values for 3 or more other nutrients. A total of 88,781 products were included in the final composition database for all years.

The existing categories and retailer shelf names in the composition database were used to assign products to the new sales categories (see Table 1) where possible, otherwise, key word searches were conducted on product names to ensure that all products were assigned a sales category. This process was done in Microsoft Excel and R Studio (version 2022.02.3). Canned fish was not included in the Edge by Ascential database and so was excluded. Products with a relatively low salt content, including sweet biscuits, cakes, pastries and plain pasta, rice, and noodles, were excluded. There was no information provided on whether the salt content of dried sauces, gravy, and stock cubes was given per 100 g as sold or as served; therefore, these products were excluded to remove inconsistencies.

Pairing composition data with sales data

Weighting composition data by sales volume allows the total volume of salt sold to be estimated, which can be used to estimate average salt purchases (as a proxy for consumption) within the population. For this study, brand-level sales data were sourced from Euromonitor International via the Bodleian Library, University of Oxford. Euromonitor is a private market research company that provides retail sales data in value and volume terms and is representative of the whole packaged food market. More details on Euromonitor sales data used in this study are provided elsewhere [20].

Euromonitor only measures sales by brand and so individual product sales data were not available. A brand was defined as a set of products that have the same core name and are manufactured by the same company. For example, the company Kellogg’s manufactures multiple brands, including Special K (one brand) and Coco Pops (another brand) breakfast cereals. Within each brand, there may be multiple individual products (e.g., Special K Original and Special K Red Berries). The brand-level sales data were matched with product-level salt content data based on product name, brand name, company name, category, and year. Where brands were matched with more than 1 individual product, mean salt content was calculated. This process is outlined in the flow chart in Fig 1 below. Each year, any new brands that were added to the market place were included, and any brands that were removed were excluded (i.e., this study included all brands, not just those present throughout the whole time series).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Data processing and analysis flow chart.

https://doi.org/10.1371/journal.pmed.1004114.g001

Corresponding product information from the composition database could not be found for 58 brands in the sales database, either because these brands were not sold in the 4 retailers included in the database (n = 18) or manufactured in the given time period (n = 40). These brands were therefore removed from the sales database (Fig 1). Euromonitor classifies small and local brands under the umbrella of “others,” representing 8% of sales overall and ranging from 3% (breakfast cereals) to 24% (savoury snacks) for category-level sales. The remaining products in the composition database that had not been matched to a brand were assigned to “others” within each category, with the mean salt content being weighted by the sales value for “others.”

Data analysis

The unit of analysis was the brand, and the sales-weighted mean salt content (g/100 g) was the primary outcome. Total volume of salt sold was also calculated in tonnes. The category, brand, and company variables in the datasets meant that analyses were stratified by company and category. Change in sales-weighted mean salt content was calculated between 2015 and 2020. In order to put the results in context of dietary recommendations, results for total volume of salt sold are presented in grams per person per day instead of tonnes. These values were calculated by dividing tonnes by population size using annual population estimates from the Office for National Statistics [21] and 365 to give daily rather than annual figures.

The change in total volume of salt sold was split into the change in the mean salt content (an indicator of reformulation) and change in total volume sales using a decomposition formula described elsewhere [20,22]. In summary, the annual percentage change in the total volume of salt sold is equal to the sum of the annual percentage change in the mean salt content and the annual percentage change in the volume of foods sold. This was calculated at the total, category, and company level. As the data were not normally distributed (as assessed visually), differences in the sales-weighted mean salt content of each category in 2020 compared to 2015 were tested using a Kruskal–Wallis test, where brands were used as the unit of analysis and p-value cutoff set at 0.05.

Sensitivity analysis

Euromonitor provides sales data for the leading brands, but each category is assigned an “others” volume that represents the small and local brands not covered in its brand list. We assumed that all products that were not matched with a brand were included under “others” for the relevant category. In order to test what impact this had on the results, we conducted a sensitivity test by recalculating the sales-weighted mean salt content overall and by category with “others” excluded.

Comparison with other studies

To our knowledge, this is the first peer-reviewed study that has reported on company-level changes in the mean salt content of foods in the UK, although other studies have reported on category-level changes in the salt content of foods over time. A study by Ni Mhurchu and colleagues in 2011 examined the sodium content of 44,000 foods purchased by 21,000 UK households between 2008 and 2009 [23]. This study also reported that processed meat and bread were the largest contributors to sodium purchases, although they did not look at changes over time. A 2013 study by Eyles and colleagues found that the sodium content of 47,000 processed foods sold in 18,000 households fell by 7% between 2006 and 2011 [12]. This suggests that the rate of change we have seen in our study is in line with historic changes. Another smaller study looked at the salt content of bread in 2001 (n = 40) and 2011 (n = 203) and found that the salt content of these products fell by around 20% over time, with supermarket own products having a lower salt content than branded products [24]. This previous large reduction in the sodium content of bread may explain why our study found little further change in more recent years. However, the salt content of bread in 2020 was 0.93 g/100 g in 2020, above the category target of 0.90 g/100 g for 2017 [14] and with further progress needed to reach the target of 0.85 g/100 g for 2024 [11]. A study using household panel data between 2005 and 2011 found that meat and cheese, dairy, and fats were the main 2 categories contributing to salt intake at 23% and 19%, respectively, and were in line with the findings presented here, at 23% for meat and 17% for cheese and butter and spreads combined [25]. This study, and that by Gressier ad colleagues [13], supports this study’s findings that that reformulation and industry behaviour were the main drivers of reductions in population salt intakes, as opposed to changes in consumer behaviour.

To our knowledge, there are no recent peer-reviewed studies that overlap with the time series presented here, although PHE has published a number of progress reports on the salt content of foods compared to the salt reduction targets. However, it is hard to make any direct comparisons between these reports and this study’s findings, as the PHE salt reduction target categories are so specific it is not possible to align categories. For example, in this study, we presented results for the “cheese” category, although data are available for hard cheese, processed cheese, and soft cheese. The PHE targets for cheese include “soft white cheese,” “cottage cheese,” “mozzarella,” “blue cheese produced in the UK only,” “cheese spreads,” and “other processed cheese such as sliced and string cheese with emulsifiers” [11]. The sales data we used in this study were not granular enough to calculate the sales-weighted mean salt content of these specific PHE categories.

Results of a meta-analysis suggest that a 4.4-g/day reduction in salt intake leads to a 4.4-mm Hg fall in systolic blood pressure [6]. The change in total volume of salt sold observed in this study was −0.16 g/day; while any decline in salt intake is welcome, it seems unlikely that this change would have led to any substantial reductions in disease prevalence and mortality.

Strengths and limitations of this study

By combining salt content data with sales data, we have been able to analyse the salt content of foods that have been sold, not just available products. This provides insight into how categories and companies have reduced the salt content of their products, and how sales—a marker of consumer behaviour—has changed over time.

The total volume of salt reported here initially appears low. For example, in 2018, we report 2.22 g salt/person/day compared to 8.4 g/person/day estimated in the urinary sodium excretion surveys. However, this discrepancy is due to a number of incremental factors. For example, we include only packaged foods sold through retail and exclude food sold for consumption outside the home in restaurants, cafes, and takeaways (30%; [26]), salt added at the table (20%; [27]), and sodium naturally occurring in food and drinks (10%; [27]). Furthermore, we were unable to analyse data from some categories (e.g., drinks, sweet bakery). With these adjustments, our estimates of salt per day are much more in line with the urinary sodium survey. It is also important to note that the sales data do not provide any split by demographics, and, therefore, any change in population composition (e.g., increase in proportion of older people) are not captured, making the daily per capita estimates presented here relatively crude. They include both adults and children of all ages and are therefore not directly comparable to the results of the urinary sodium figures, which are for adults aged 19 to 64 only.

The advantages and limitations of using Euromonitor sales data to analyse the nutrient content of foods over time have been outlined in detail in a previous study [20]. In summary, Euromonitor sales data avoid reliance on individual recall and underreporting in scan data [28,29], and it also allows for analysis of individual brands and companies. However, its limitations mean that table salt and salt sold from fast food are both excluded, making the total volume of salt sold likely to be a significant underestimation. Euromonitor’s reporting of brand sales, and not individual product sales, is a major limitation and means that any heterogeneity that occurred between products that belong to the same brand would have been missed, given that the method used assumes all products under the same brand are sold equally. A previous study has examined this limitation in detail, concluding that while the overall findings are unlikely to be affected, there may be some significant misrepresentations at the brand and potentially company level [20]. Euromonitor is also limited in its granularity, meaning that is groups small and local brands for each category under the umbrella of “others,” although the results of the sensitivity analysis show that this is unlikely to have a significant effect on the results overall but may have had a small effect on the ready meals, pizza, and soup and processed beans, potatoes, and vegetable categories. The lack of granularity in the sales data also means that these results are not to directly comparable with PHE’s salt reduction targets, which have very narrow categories; therefore, this study is not a specific evaluation of the salt reduction targets. However, the authors do have future research in this area planned, including modelling the targets’ potential health impact.

There are several limitations to the nutrition composition database used in this study. The first is that we assumed that the products in the composition database were entirely representative of the brands that they matched with in the sales database. However, this may not be the case for several reasons: (1) any products that were out of stock on the day that the data were collected will not be included in the composition data; (2) it is possible that some brands have product variants, products that are only available in convenience stores, markets, or restaurants and cafes, which are not captured here; and (3) products sold exclusively in-store will not be captured by these data as these were only collected online. However, a previous validation study using the same foodDB database found that 85% of products sold in supermarket stores could be matched to those sold by the same retailer online, with nutrition information being identical [30].

Dried sauces and gravies were excluded from the dataset as no data were given to indicate whether the salt content values were as sold or as given. After investigation, most products had their own preparation instructions and attempts to devise a generic formula that could be used to convert all values over a certain threshold to as sold (as the sales data are given as sold) caused jumps in the category results across years; therefore, it was decided to exclude dried products. While the high salt content of dried products makes them a category of interest, results from the NDNS data show that the total sauces category represents 6% of salt intake overall, and it is not expected that their exclusion will have a large impact on the overall results.

Policy implications

This analysis suggests that salt intake from processed foods purchased in retail settings fell by 6.7% between 2015 and 2020, although this decrease was not statistically significant. We highlight categories where progress is slow and the heterogeneity between companies within categories suggesting there is considerable scope for further reductions. In some categories, the salt content (e.g., bread and ready meals) has not changed in recent years. This is likely due to a combination of reasons: Many in the food industry cite the technical challenges to salt reduction, especially in products where it also acts as a preservative; there can be consumer resistance to low-salt varieties, although this can be overcome by gradual salt reduction; and there has been a shift in the efforts of both industry and policy makers towards sugar and calorie reduction instead. For savoury snacks and ready meals, reductions in salt content were counterbalanced by increases in sales. This highlights the need for policies that encompass consumer behaviour change and purchasing (such as reducing consumption of high-salt products and shifting towards low-salt varieties) as well as industry-led salt reduction targets and the launch of new low-salt varieties. The salt reduction programme in the UK was initially credited with such success because it included a public awareness campaign and strong political support [31], both of which have waned in recent years.

The specificity of the salt reduction targets means it is very hard to assess progress towards the targets using available data sources. Mandatory reporting of data to common standards would allow policymakers, academics, investors, and others to better track companies’ progress towards public health goals, as has been called for in the National Food Strategy [32]. It would shine a spotlight on parts of the food system, which were leading or lagging behind policy targets, allow for a better understanding of how effective the current targets are at improving public health outcomes, and allow alternative measures to be considered, such as mandated maximum salt targets (as in Argentina) and taxes on high-salt foods (as in Portugal) [33].