Different types of ultra-processed foods have opposite associations with type 2 diabetes. Grouping these foods into such a coarse category may undermine interventions and can hide individual associations, meta-analysis of 635,332 participants finds