To assess food industry commitments and practices, the ‘Business Impact Assessment on Obesity and population-level nutrition’ (BIA-Obesity) was applied, as developed by the International Network for Food and Obesity/Non-communicable Diseases Research, Monitoring and Action Support (INFORMAS) and previously described in detail by Sacks et al. [10, 18]. The tool assesses the transparency, comprehensiveness and specificity of commitments as well as practices across six domains, namely: ‘Corporate nutrition strategy’, ‘Product formulation’, ‘Nutrition labelling’, ‘Product and brand promotion’, ‘Product accessibility’ and ‘Relationships with other organisations’ .
All indicators within these domains relate to commitments that go beyond legislative requirements. As a result, indicators and scoring criteria need to be adapted to the local context prior to implementation of the tool. Indicators related to the on-pack disclosure of the ingredients list and nutritional declaration were removed as this is regulated by the European Union . As it is not common in France for supermarkets to have in-store restaurants, indicators relating to menu-labelling were removed for this food industry. Furthermore, non-alcoholic beverages containing added sugars or sweeteners in France are subject to a tax . Consequently, commitments to increase prices of sugary beverages compared to healthier drinks were not taken into account. Since the provision of unlimited refills was banned in France in 2017  the indicator relating to commitments of quick-service restaurants to not provide free refills was removed. Lastly, the indicator regarding the publication of political donations was removed as in France legal persons (including, and in particular, companies) are not authorized to pay any donation or any benefit in kind to political parties . The remaining indicators were adapted to suit the French regulatory environment and take into account relevant industry pledges and voluntary government-led initiatives (i.e. Nutri-Score).
This study was approved by the Human Ethics Committee of the University of Ghent (number: 2019/0780).
Selection of food companies
Food companies with a combined market share of over 34% among packaged food manufacturers (35%), non-alcoholic beverage manufacturers (52%), supermarkets (48%) and quick-service restaurants (50%) were selected using French Euromonitor 2018 market share data (Table 1) . For packaged food manufacturers, an additional selection was conducted based on companies’ market share within specific food categories to ensure that the most prominent companies per food category were covered by the selection (‘Breakfast cereals’, ‘Baked goods’ ‘Confectionery’, ‘Ice-cream and frozen desserts’, ‘Processed Fruit and Vegetables’, ‘Processed Meat and Seafood’, ‘Sweet biscuits and cereal bars’, ‘Drinking milk products’, ‘Yoghurts’, ‘Savoury snacks’ and ‘Ready meals’). Three additional companies were included based on this extra selection (Kellogg’s, Barilla and Bonduelle).
Data collection and analyses
Publicly available commitments and policies were collected between June 2019 and December 2020. Relevant information was collected from company websites, company reports, brand websites and relevant industry pledges and initiatives. Per selected company, screenshots were taken of relevant webpages and relevant documents were downloaded.
Subsequently, the information was entered in an Excel spreadsheet per BIA-Obesity indicator. A report summarizing the collected information as well as the preliminary scoring was compiled per company. Company representatives were contacted via various channels, including meetings with industry associations (ANIA and L’Alliance 7), phone call inquiries, contact information on company/brand websites and LinkedIn. Companies willing to verify and complete the collected data were sent the summary reports after signing a written informed consent. For all additional information they provided some kind of evidence was required. Upon request companies could sign non-disclosure agreements prior to sharing sensitive internal documents. For companies that refused participation or failed to share feedback in time, the assessment was based solely on publicly available information. Supermarkets were assessed as both retailers and food manufacturers (the latter for own-brand products).
The nutrition-related commitments were scored in Excel. Supplementary file 1 provides examples of how scores were assigned for BIA-Obesity indicators. All company commitments were scored by IVD and two companies per food industry (a total of eight companies) were blindly re-scored by YZ. Discrepancies were discussed until an agreement was obtained. The final BIA-Obesity scores per domain were weighted as recommended by INFORMAS (Supplementary file 2) .
Median scores (range and interquartile range IQR), overall and per BIA-Obesity domain, were calculated for each food industry and across food industries. For companies that verified and completed the publicly available information, median scores before and after their participation were calculated. A one-tailed Wilcoxon signed-rank test was conducted to compare scores before and after participation. The Wilcoxon signed-rank test was applied as the test assessed changes in a dependent outcome variable before and after companies had the opportunity to provide additional information. It was opted for a one-tailed test as companies could only improve their scoring by sharing extra information in addition to the publicly available evidence. A two-tailed Wilcoxon rank-sum test was used to compare scores of two independent groups, namely companies that engaged with the process and those that did not engage. Both tests are non-parametric tests.
For some of the BIA-Obesity policy domains, a set of key performance indicators was selected to assess company practices on population nutrition. The selected indicators, as well as the sources where the data were derived from and the years, are presented below in Table 2. For the domains on ‘Corporate nutrition strategy’ and ‘Relationships with other organisations’, no performance indicators (such as an assessment of companies’ corporate political activities) were included due to a lack of time and resources available to collect data within these domains. For the domains ‘Nutrition labelling’ and ‘Product accessibility’ no performance data were available at the time of assessment. For the other BIA-Obesity domains, specific indicators were included, dependent on data availability and feasibility of the assessment. An overview of the different performance indicators can be found in Table 2.
For packaged food and non-alcoholic beverage manufacturers and supermarkets (own-brand products), the healthiness of the complete product portfolios was analysed using Open Food Facts data for France in 2018. As Open Food Facts cannot guarantee the accuracy and completeness of the data, the nutritional data of all products that could be found on Mintel GNPD (Global New Products Database), on brand websites or supermarket websites were verified using the aforementioned sources. Duplication of products was avoided by ensuring that each barcode appeared only once.
For quick-service restaurants, the nutritional information per 100 g was obtained from the national brand websites in 2019, where possible (Burger King, Domino’s Pizza, McDonald’s and Paul). For KFC no nutritional information was available per 100 g and no portion sizes were specified on the national website, so an online table with nutritional information from 2018 was used. On the website of Brioche Dorée and Quick no nutritional information was available per 100 g and portion sizes were not defined. As a result, the product portfolios of Brioche Dorée and Quick could not be analysed.
The healthiness of the entire portfolios of all selected food companies was analysed using the Nutri-Score, which is the official front-of-pack labelling system in place in France since March 2017 . The proportion of products with Nutri-Score A, B, C, D and E was determined, as well as the median Nutri-Score across the company’s portfolio or menu. When calculating the Nutri-Score for non-alcoholic beverages, it was assumed that no juices had a fruit and vegetable content above 40% as the data sources and product ingredient lists did not allow for a distinction to be made between the fruit and vegetable content of different juices. To check the viability of this assumption, a Pearson correlation coefficient was calculated between the Nutri-Score available through Open Food Facts and the calculated Nutri-Score for non-alcoholic beverages. A strong correlation was observed between both Nutri-scores (R = 0.84, p < 0.0001). In addition, a correlation between the Open Food Facts Nutri-Score and the calculated Nutri-Score was also conducted for the entire dataset. A very strong correlation was observed between the calculated Nutri-Score and the Nutri-Score displayed within Open Food Facts (R = 0.98, p < 0.0001).
The company’s portfolios were also analysed in relation to the proportion of ultra-processed foods (according to the NOVA classification ). The NOVA-classification distinguishes products based on their level of processing (unprocessed or minimally processed foods, processed culinary ingredients, processed foods and ultra-processed foods) . The proportion of products within portfolios that are ultra-processed (NOVA) as well as the median Nutri-Score and the proportion of products with Nutri-Score ‘A and B’ and ‘D and E’, were examined by company. The results were reported as a proportion of products with Nutri-Score ‘A and B’ and ‘D and E’ as this was considered to reflect the healthiness of companies’ overall product portfolios. The proportion of products with Nutri-Score ‘A and B’ was deemed to represent healthier alternatives within the product portfolio while the proportion with Nutri-Score ‘D and E’ was considered to signify less healthy products.
Product and brand promotion
To assess the proportion of products within company portfolio’s (not-)permitted to be marketed to children the WHO-model was applied. The WHO-model determines per product category whether products should be (not-)permitted to be marketed to children. An overview of the 17 product categories included in the WHO-model can be found in Supplementary file 3. While a threshold for nutrients of concern determines if a product could be permitted to be marketed to children for most product categories, some categories are entirely permitted (such as ‘Fresh and frozen meat, poultry, fish and similar’ and ‘Fresh and frozen fruit, vegetables and legumes’) or not-permitted to be marketed to children (such as ‘Chocolate and sugar confectionery, energy bars, and sweet toppings and desserts’; ‘Cakes, sweet biscuits and pastries, other sweet bakery wares, and dry mixes for making such’; ‘Juices’; ‘Energy drinks’ and ‘Edible ices’) . From a public health perspective it would be expected that companies with a higher proportion of products not-permitted to be marketed to children would have stronger commitments in place to reduce such practices.
To specifically evaluate the products promoted by supermarkets, food promotions in the flyers of the six biggest supermarkets in France were collected online from the weekly/two-weekly circulars over a six-month period (October 2019 – March 2020). All promotions were entered into a database and manually classified according to the NOVA-classification and the 17 food categories of the WHO-model (Supplementary file 3). Per product the following information was recorded: product- and brand name, type of promotional character, the level of discount, type of incentive offer, if the product was a fresh fruit or vegetable, whether the product was a fresh meat or fish product and the Nutri-Score . The proportion of promotions for ultra-processed foods, foods with promotional characters, incentive offers or discounts and the proportion of promotions for fresh fruits and vegetables were calculated. Data were analysed separately per supermarket.
The relationship between commitments and practices
Correlations (𝝆-values) between commitments and practices were calculated applying the Spearman’s rank correlation coefficient, a non-parametric test that measures the direction and strength of a monotonic association between two variables. 𝝆-values range from − 1, indicating a perfect negative correlation between two variables to + 1, indicating a perfect positive correlation between variables. Correlations were calculated between commitments made within the domain ‘Product formulation’ and the proportion of products within the portfolio with Nutri-Score A and B and D and E. Correlations between the domain ‘Product formulation’ and the proportion of ultra-processed products were also calculated. Lastly, correlations between commitments within the domain ‘Product and brand promotion’ and the proportion of products not-permitted to be marketed to children according to the WHO-model were assessed.
𝝆 -values between 0.5 and 1 as well as between − 0.5 and − 1 were considered to represent a moderate to strong correlation. P-values < 0.05 were considered statistically significant. All analyses were performed using Microsoft Excel and SAS 9.4 (Cary, USA, 2018).