Between September 2016 and April 2017, three-arm parallel group randomized controlled trials (https://clinicaltrials.gov/ct2/show/NCT02769455) were conducted sequentially using an experimental online supermarket and including individuals engaged in grocery shopping. The design of the first trial among students and the recruitment have been described elsewhere [21]. Briefly, the first trial included students from multiple French universities, aged between 18 and 25 years old. The second trial included active adults from the French NutriNet-Santé cohort [22], aged between 30 and 50 years old, and having a monthly income below 1200€ per consumption unit (i.e. corresponding roughly to the second decile of income). The third trial focused on individuals over 50 years old within the NutriNet-Santé cohort also, suffering from at least one nutrition-related cardiometabolic diseases (obesity, type 2 diabetes, dyslipidaemia, arterial hypertension, cardiovascular disease). The three trials were conducted following the same protocol as the trial on students and differed only on the targeted population [21]. Participants were invited to fulfill an inclusion questionnaire to collect data on the eligibility criteria mentioned above, as well as information on various sociodemographic characteristics and nutrition-related behaviors. Eligible participants were then randomly allocated to one of the three arms using a random block method, and invited to simulate a shopping situation as if they were in their usual supermarket. For the three trials, the final sample size was calculated considering an effect size of 0.2 (for the main outcome, the FSAm-NPS score of the shopping cart), a power of 90% and a p-value of 0.02 considering the three-arm design, resulting in 1956 individuals, i.e. 652 participants per arm (Supplemental Figure 1 in Additional file 1). To reach this final sample size while considering the non-respondent rate, the number of participants validating their shopping cart was monitored.
In the experimental arm of the three trials, the Nutri-Score was affixed on the front of the package of all pre-packed foods in the online supermarket – no label was applied on unpacked foods (e.g., fresh fruits and vegetables, butcher meat). Briefly, the Nutri-Score is a summary graded scale indicating the overall nutritional quality of a food product, based on the United Kingdom Food Standards Agency Nutrient Profiling System, adapted to the French context by the High Council for Public Health (FSAm-NPS) [23]. Considering the nutritional content of food in unfavorable (energy, Saturated Fatty Acids (SFA), sugars, sodium) and favorable elements (proteins, fibers, fruits, vegetables, legumes and nuts), the FSAm-NPS score ranged from − 15 points for products with higher nutritional quality to + 40 points for foods with lower nutritional quality. The Nutri-Score is then calculated using the FSAm-NPS score and expressed through a graded scale between “A” in dark green for healthier products (e.g. fresh fruits, vegetables, whole-grain bread) and “E” in dark orange for unhealthier products (e.g. processed meat, butter, chocolate or cookies). In the three trials, two control arms were included: (i) no label, and (ii) the RIs affixed on the front of pre-packed foods. The RIs label is a nutrient-specific FoPL providing the content in energy, fats, SFA, sugars and sodium in gram per serving, as well as their percentage contribution to the guideline-based daily intakes [24]. The two schemes that were tested are displayed in Supplemental Figure 2 (Additional file 2).
The experimental online supermarket was created to resemble existing online supermarket and allowed participants to simulate a purchasing situation, though without any payment. The supermarket included 751 different foods and beverages, with raw and unpacked fresh products (N = 56 foods) and pre-packed foods (N = 695 foods). Selection of products was made in order to reproduce the food offer available in online grocery stores. Information on the nutritional composition, ingredients list, and the price of the product was provided for all foods on the experimental supermarket, with in addition the FoPL on pre-packed foods in the Nutri-Score and RIs arms. The products from the experimental online supermarket were classified in 36 food categories, including four categories containing foods with various degrees of processing. Therefore, for the categories of “fruits”, “vegetables”, “meat” and “fish”, products were categorized according to their degree of processing, using the NOVA classification which is based on the extent and purpose of industrial food processing [25]. The NOVA classification categorizes foods into four groups: the group 1 includes products with no or little processing, the group 2 gathers processed culinary ingredients (e.g., sugars, oils, butter), the group 3 includes processed products (i.e. foods containing usually two or three ingredients, and which have been transformed through various methods of preservation or cooking), and the group 4 contains ultra-processed foods, for which specific industrial processes were applied (i.e. hydrogenation, hydrolises, extruding, moulding, etc) or some substances added (i.e., flavoring agents or food additives such as colors, emulsifiers, humectants, non-sugar sweeteners). In the present study, the four food categories for which products were classified according to the degree of processing did not contain any foods from NOVA group 2. Distribution showed an overall balanced distribution (33–33-34% split) across NOVA groups 1, 3 and 4 for fruit and vegetables (NOVA 1 fresh fruits and vegetables, NOVA 3 fruit purées, canned vegetables, NOVA 4 prepared fruit and vegetables with additives); a 17–29-54% split for fish (NOVA 1 fresh fish, NOVA 3 canned fish, NOVA 4 prepared fish or fish patties) and a 40–60% split between NOVA 1 and 4 for meat (NOVA 1 fresh meat cuts, NOVA 4 processed meat).
Main socio-demographic characteristics of participants collected at inclusion were described according to each of the studies performed (i.e. students, working adults with low incomes, subjects with chronic diseases). The mean proportion of unpacked and pre-packed products that were purchased in each arm was calculated. For pre-packed foods, products were distinguished according to their Nutri-Score class. Then, in each trial arm, the distribution of foods across the different food categories, considering additionally the degree of processing (NOVA group) for the four categories mentioned above, was calculated and expressed with mean proportions and standard deviations. The mean proportions were then compared overall between arms using one-way ANOVA. Then, the two-by-two differences between arms were calculated and pairwise comparisons were performed using Tukey tests to consider multiple comparisons. Analyses were conducted on the overall sample with an adjustment for the population, using the SAS software (version 9.4; SAS Institute, Inc). Tests were two-sided and a p-value below 0.05 was considered statistically significant.