Estimating days needed for dietary assessment in pregnancy: a modeling study

Dietary assessment is essential for understanding associations between diet and health. The number of days of dietary data collection must account for high variation in daily intakes, while balancing accuracy with participant burden. Although several methods exist for determining the optimal amount, few have been applied to pregnancy.

This study aimed to algorithmically determine the number of days needed to accurately estimate key dietary characteristics: energy, macronutrients, macronutrient density, diet quality, and intake timing during pregnancy.

We analyzed dietary data from 147 pregnant individuals in the Temporal Research in Eating, Nutrition, and Diet during Pregnancy study. Each participant provided ≤28 d of image-based dietary records to determine nutrients. Using mixed-effects models and assuming the only source of error is day-to-day variation in diet, we calculated the number of days required for the correlation between estimated and true intake to be ≥0.90 (N_R) and for estimated intake to be within 20% of true intake (N_C) for: energy, absolute macronutrients, macronutrient density, healthy eating index 2020 score (HEI 2020), and intake timing. Using bootstrapping, we extended N_C into a probabilistic framework linking methodological choices to practical consequences.

Within-person coefficient of variation (CV_W) exceeded between-person coefficient of variation (CV_B) for all characteristics, with fat (g) having the largest difference (CV_W = 40.6; CV_B = 26.2). Macronutrients required the most days to estimate usual absolute intake (fat: 17 d), whereas macronutrient density (% calories from fat: 5 d), HEI 2020 (5 d), and intake timing (2-6 d) required fewer. Individual-level analysis showed that cohort-based estimates underestimated the number of days needed for accurate assessment, as only 56% of participants met the cohort-level requirement for HEI.

We provide a new approach to estimating required dietary days to inform future study design. Existing studies may be underpowered, and cohort estimates may overstate individual-level accuracy.