Harmonising dietary datasets for global surveillance: methods and findings from the Global Dietary Database.

Objective: The Global Dietary Database (GDD) expanded its previous methods to harmonise and publicly disseminate individual-level dietary data from nutrition surveys worldwide.

Design: Analysis of cross-sectional data.

Setting: Global.

Participants: General population.

Methods: Comprehensive methods to streamline the harmonisation of primary, individual-level 24-h recall and food record data worldwide were developed. To standardise the varying food descriptions, FoodEx2 was used, a highly detailed food classification and description system developed and adapted for international use by European Food Safety Authority (EFSA). Standardised processes were developed to: identify eligible surveys; contact data owners; screen surveys for inclusion; harmonise data structure, variable definition and unit and food characterisation; perform data checks and publicly disseminate the harmonised datasets. The GDD joined forces with FAO and EFSA, given the shared goal of harmonising individual-level dietary data worldwide.

Results: Of 1500 dietary surveys identified, 600 met the eligibility criteria, and 156 were prioritised and contacted; fifty-five surveys were included for harmonisation and, ultimately, fifty two were harmonised. The included surveys were primarily nationally representative (59 %); included high- (39 %), upper-middle (21 %), lower-middle (27 %) and low- (13 %) income countries; usually collected multiple recalls/ records (64 %) and largely captured both sexes, all ages and both rural and urban areas. Surveys from low- and lower-middle . high- and upper-middle income countries reported fewer nutrients (median 17 . 30) and rarely included nutrients relevant to diet-related chronic diseases, such as -3 fatty acids and Na.

Conclusions: Diverse 24-h recalls/records can be harmonised to provide highly granular, standardised data, supporting nutrition programming, research and capacity development worldwide.