Usual dietary intake estimation in the German National Cohort (NAKO)

Accurate measurement of dietary intake remains challenging in large-scale nutritional studies. This study aimed to develop and evaluate both a practical dietary assessment strategy and a computationally efficient statistical method for estimating usual dietary intake in the German National Cohort (NAKO Gesundheitsstudie).

We developed a blended approach using data from NAKO. During baseline (2014-2019) and first follow-up examinations (2019-2024), up to four 24-h food lists (24 h-FLs) and one food frequency questionnaire (FFQ) were collected. We combined these dietary intake data sources using an adapted Multiple Source Method (MSM) and supplemented them with estimated consumption amounts based on data from the German National Nutrition Survey II (NVS II, 2005-2007) to generate measurement-error-corrected estimates of dietary intake. The adapted MSM was empirically evaluated against conventional logistic linear mixed-effects model (LLMM), which can be computationally complex for large datasets due to lengthy processing times. Additionally, a simulation study evaluated how varying the number of 24 h-FLs and FFQ assessments affected the consumption probability estimates.

The adapted MSM showed high statistical agreement with LLMM (correlation ≥0.97). The usual intake of 90 EPIC-SOFT food groups, 124 nutrients, and energy intake was estimated for 152,304 participants (75% of the cohort) who had at least one 24 h-FL and an FFQ available. Furthermore, the simulation showed that including repeated 24 h-FLs alongside an FFQ improved the accuracy of individual consumption probability estimates, particularly when only one or two 24 h-FLs were available.

The adapted MSM offers a computationally efficient, practical alternative to LLMM, generated dietary intake estimates for over 150,000 NAKO participants to support future research. By integrating repeated 24 h-FLs, an FFQ, and external consumption data, this blended approach balances logistical feasibility with statistical precision, providing a scalable, cost-effective framework for large-scale nutritional studies.