A four-stage process was undertaken to identify the ultra-processed foods from both the adult and the youth FFQs. First, all food items in the FFQs across different waves of data collection were complied. Food items that were nearly identical between FFQs but were presented with minor differences were captured as separate items (e.g., ‘Cold breakfast cereal (1 bowl)’ and ‘Cold breakfast cereal (1 serving)’). This was done to make sure that no food item was overlooked. FFQs from every 4 years of the NHS-I (1986–2010), the NHS-II (1991–2015), the HPFS (1986–2014), from 1996, 1998, 2001 for GUTS-I and from 2004, 2006, 2008, 2011 for GUTS-II were used.
Second, three researchers working independently assigned foods in the adult (N.K, S.R, E.M) and the youth (N.K, M.D, E.M) cohorts to one of the four NOVA groups based on their grade of processing – unprocessed/minimally processed foods (G1), processed culinary ingredients (G2), processed foods (G3) and ultra-processed foods (G4). Food assignment was guided by the definition, examples and supplementary material published by the proponents of the NOVA classification(1 (link)). Categorisation was an iterative process requiring the review of the original FFQs used at each wave of data collection to contextualise food items within the larger food lists. Food preparations made from multiple ingredients or different food items that were presented jointly in the FFQ were not disaggregated into their different components. Additionally, the nutrient profile of food items, their actual amounts consumed by the study participants or participant demographics were not considered at any point in the categorisation process. Instead, the original food item as it was listed in the FFQ was categorised in its entirety.
At the third stage, categorisation between researchers was triangulated. Food items for which there was consensus in the categorisation among all researchers were assigned to their NOVA group. A food item was flagged for further scrutiny and shortlisted in case a researcher was unable to assign it to a NOVA group or in cases of disagreement in categorisation by any two researchers.
At stage four, an expert panel comprising of three senior nutrition epidemiologists (F.F.Z; T.F; Q.S) with substantial experience working with the dietary intake in these cohorts, was convened to review and discuss the categorisation of the short-listed products. All discussions were additionally informed by the following resources:
Consultations with the research dietitians. The team of research dietitians, led by L.S, was responsible for overseeing the collection of dietary data and for ascertaining the nutrient composition of food items across all Harvard cohorts. They shared their insights obtained from gathering supplementary data, tracking new and reformulated products available in the supermarket, and conducting multiple pilot studies with cohort participants.
Cohort-specific documents. These resources provided more insight into the extent of processing of certain FFQ food items by highlighting information on the specific ingredients used in recipes and food preparations, the proportion by weight of individual ingredients to the final recipe or a more detailed description of food items (whether the food was canned or salted or boiled, the brand name of certain packaged foods, etc.).
Supermarket scans. The ingredient lists of the first five brands of specific products that were displayed on the Walmart website in 2019 and 2020 were scrutinised. They served as a proxy for establishing the level of processing for a small proportion of food items for which limited information was available from the resources listed above.
