Abstract
BACKGROUND/OBJECTIVES: Accurate dietary assessment is essential for understanding diet-health relationships, yet day-to-day variability in intake complicates the identification of individuals' usual consumption. This study aimed to determine the minimum number of days required to obtain reliable estimates of dietary intake across various nutrients and food groups using data from a large digital cohort. METHODS: We analyzed dietary data from 958 participants of the "Food & You" study in Switzerland, who tracked their meals for 2-4 weeks using the AI-assisted MyFoodRepo app. Over 315,000 meals were logged across 23,335 participant days. We assessed day-of-week intake patterns using linear mixed models and estimated minimum days for reliable measurement using two complementary methods: (1) the coefficient of variation (CV) method based on within- and between-subject variability, and (2) intraclass correlation coefficient (ICC) analysis across all possible day combinations. RESULTS: Our findings indicate that water, coffee, and total food quantity can be reliably estimated (r > 0.85) with just 1-2 days of data. Most macronutrients, including carbohydrates, protein, and fat, achieved good reliability (r = 0.8) within 2-3 days. Micronutrients and food groups like meat and vegetables generally required 3-4 days. Linear mixed models revealed significant day-of-week effects, with higher energy, carbohydrate, and alcohol intake on weekends-especially among younger participants and those with higher BMI. ICC analyses further showed that including both weekdays and weekends increased reliability, and specific day combinations outperformed others. CONCLUSIONS: Three to four days of dietary data collection, ideally non-consecutive and including at least one weekend day, are sufficient for reliable estimation of most nutrients. These results support and refine FAO recommendations, offering more nutrient-specific guidance for efficient and accurate dietary assessment in epidemiological research.