Reynolds, C., Da Silva, J. T., Garzillo, J. M. F., Frankowska, A., Kluczkovski, A., Rose, D., Takacs, B., de Quadros, V. P., Holmes, B. A., Schmidt Rivera, X., and Bridle, S.
Introduction\ud FoodEx2 consists of a vocabulary of foods with assigned codes, allowing the classification and description of foods reported in different types of data (food consumption, composition, or production method) [1]. At least 56 food consumption databases have been coded with FoodEx2 [2]. The linkage of greenhouse gas emissions (GHGE) databases to FoodEx2 codes would allow rapid matching of GHGE data to any food database previously harmonised with FoodEx2. We have linked 4558 FoodEx2 codes to GHGE data [3]. In this work, we aim to assess the reliability of the linkage, by comparing it to similar databases.\ud \ud Methods or approach\ud The main database (“City”) was developed by matching 43 food categories from Poore and Nemecek (2018) [4] to the 4558 FoodEx2 codes, adjusting for edibility. The items were matched by hand, using the closest raw product; if it was a product with multiple ingredients, we took the largest ingredient by weight. The reliability of the matching was assessed by comparing “City” to three GHGE databases: 1) SHARP (Mertens et al 2019) [5] has linked GHGE to FoodEx1 (an initial version of FoodEx); 2) Rose et al (2019) [6] and Heller et al (2018) [7] have GHGE data linked to National Health and Nutrition Examination Survey (NHANES), which is coded with FoodEx2; 3) Garzillo et al (2019) [8] have GHGE data linked to the Brazilian Food Consumption Survey. We have compared “City” with the three databases by calculating Spearman correlation coefficients. The food items with GHGE in both “City” and SHARP were ranked and split into quintiles. We checked quintile rankings of agreement between “City” and SHARP by calculating weighted kappa statistics. For food items not ranked in the same quintile, we checked whether values from SHARP were within the p5 and p95 from “City”.\ud \ud Findings and interpretations\ud The “City” dataset was strongly correlated to all three comparator databases. However, the number of directly comparable food items between all datasets were low. The number of food items compared, the Spearman correlation coefficients, and p-values are as follows: 1) “City” versus SHARP: n=945, r = 0.699, p < 0.001; 2) “City” versus Rose/Heller: n=608, r = 0.572, p < 0.001; 2) “City” versus Garzillo: n=329, r = 0.610, p < 0.001. Of the 945 food items with GHGE in “City” and SHARP, 50% (n = 476) were ranked in the same quintile. The kappa statistics was 0.536 (p < 0.001). Of the 469 food items not ranked into the same quintiles, 44% (n=206) were within p5 and p95, while 31% (n=144) were lower than the p5, and 25% (n= 119) were higher than p95. The food items with the biggest differences between “City” and SHARP are into the following food categories: wheat and rye; fish and seafood; pig meat; fruits; nuts and pulses. These food items will be further investigated in the next update of the data, aiming to increase reliability to estimate GHGE from food consumption.\ud \ud Conclusions\ud The FoodEx2 database linked to GHGE by City presented a strong correlation with other GHGE databases and therefore could be considered as a tool to estimate the environmental impacts from food. However, further work is still needed to refine the data, in particular checking values categorised in opposite quintiles and that do not fall within p5 and p95. This database allows for a quick link between GHGE and multiple dietary databases harmonised with FoodEx2.\ud \ud References\ud EFSA (European Food Safety Authority), Nikolic, M and Ioannidou, S, 2021. FoodEx2 maintenance 2020. EFSA supporting publication 2021: 18( 3):EN-6507. 19 pp. doi: 10.2903/sp.efsa.2020.EN-6507\ud Karageorgou, Dimitra and Lara-Castor, Laura and de Quadros, Victoria Padula and de Sousa, Rita Ferreira and Holmes, Bridget Anna and Ioannidou, Sofia and Mozaffarian, Dariush and Micha, Renata, Harmonizing Dietary Datasets for Global Surveillance: Methods and Findings From the Global Dietary Database. Under review.\ud Livestock, Environment And People (LEAP) Conference, Oxford, November 2019. C.J. Reynolds , X. Schmidt Rivera, , A. Frankowska, A Kluczkovski, J. T. da Silva S. L. Bridle,R. Levy, F. Rauber, V. P. Quadros, A. Balcerzak, R. F. Sousa, M. Ferrari, C. Leclercq, B. Koroušić Seljak, Tome Eftimov "A pilot method linking greenhouse gas emission databases to the FoodEx2 classification"\ud Poore, J., & Nemecek, T. (2018). Reducing food’s environmental impacts through producers and consumers. Science, 360(6392), 987–992.\ud Mertens, E., Kaptijn, G., Kuijsten, A., van Zanten, H., Geleijnse, J.M. and van't Veer, P., 2019. SHARP-Indicators Database towards a public database for environmental sustainability. Data in brief, 27, p.104617.https://doi.org/10.1016/j.dib.2019.104617\ud Rose D, Heller MC, Willits-Smith AM, Meyer RJ. "Carbon footprint of self-selected US diets: nutritional, demographic, and behavioral correlates," American Journal of Clinical Nutrition 2019;109:526-534. DOI:10.1093/ajcn/nqy327.\ud Heller MC, Willits-Smith A, Meyer R, Keoleian G, Rose D. "Greenhouse gas emissions and energy use associated with production of US self-selected diets," Environmental Research Letters 2018;13 044004. DOI: 10.1088/1748-9326/aab0ac.\ud Garzillo JMF, Machado PP, Costa Louzada ML, Levy RB, Monteiro CA. Footprints of foods and culinary preparations consumed in Brazil. https://doi.org/10.11606/9788588848405 (accessed Dec 7, 2020)