A Methodological Framework for Aggregating Branded Food Composition Data in mHealth Nutrition Databases: A Case Presentation

TL;DR

This paper presents a framework for aggregating branded food data to update nutrition databases used in mHealth apps.

Contribution

The paper introduces a three-step framework for aggregating branded food composition data into generic food names.

Findings

347 new generic food names were proposed from branded products.

Aggregated energy, protein, and carbohydrate values showed high homogeneity.

Baked goods and milk products had higher heterogeneity in nutritional values.

Abstract

Background/Objectives: Up-to-date, relevant and detailed food composition databases (FCDs) are a central component of mHealth apps. Thus, the expansion and/or update of such FCDs though the aggregation of branded food data (BFCDs) could prove as a cost-efficient methodology. However, a framework for data aggregation from BFCDs has yet to be documented. Methods: Products (n = 3988) available in the HelTH BFCD were grouped following a three-step process. Firstly, foods were grouped based on their name, and then the aggregated nutritional composition was tested for heterogeneity using a coefficient of variation cut-off of 20% followed by a search of the ingredient list and other product characteristics to identify descriptors that reduced heterogeneity. Results: Following a three-step process, n = 347 new generic food names were proposed, each derived from at least three branded products, of which n = 235 were populated with aggregated nutritional content values. We found that 95.3%, 88.6%, 86% and 82.6% of aggregated energy, protein, carbohydrate and sodium values, respectively, had a coefficient of variation <40%. Aggregated saturated fatty acid and total sugar values were less likely to fall in the homogeneity level (76.3% and 65.3%, respectively). The heterogeneity was concentrated in specific subcategories like baked goods, milk products and milk imitation products, primarily. Conclusions: BFCDs can be used as a resource to expand existing databases with relatively homogeneous and up-to-date nutritional composition data. The application of this framework on larger datasets could improve the generic food name yield and homogeneity and support mHealth apps and other uses.

Figures