We present the results of the FoodBench-QA 2026 shared task at the CL4Health workshop, collocated with LREC 2026. FoodBench-QA challenges systems to answer food and nutrition questions using evidence from food composition databases and food-related ontologies. The shared task comprises three main tasks: nutrient estimation from recipe ingredients, evaluated using EU Regulation 1169/2011 tolerance thresholds; FSA traffic-light classification for fat, salt, saturates, and sugars; and food named entity recognition and linking to three ontologies, namely Hansard Taxonomy, FoodOn, and SNOMED CT. We received submissions from five participating teams across all tasks. For nutrient estimation, the best system achieved accuracy rates of 93.57% for protein, 86.50% for sugars, 84.65% for fat, and 86.26% for saturates. For FSA traffic-light prediction, the best macro F1 scores ranged from 0.65 to 0.90 across different nutrient-color combinations. For named entity linking, the best systems achieved macro F1 scores between 60.71% and 80.89% for natural text and 87.75% and 95.75% for artificial NEL datasets, depending on the ontology.

The accelerating growth of big data in the biomedical domain, with an endless amount of electronic health records and more than 30 million citations and abstracts in PubMed, introduces the need for automatic structuring of textual biomedical data. In this paper, we develop a method for detecting relations between food and disease entities from raw text. Due to the lack of annotated data on food with respect to health, we explore the feasibility of transfer learning by training BERT-based models on existing datasets annotated for the presence of cause and treat relations among different types of biomedical entities, and using them to recognize the same relations between food and disease entities in a dataset created for the purposes of this study. The best models achieve macro averaged F1 scores of 0.847 and 0.900 for the cause and treat relations, respectively.