It has been documented that nutrient intake can vary strongly across different days of the week [37,38,39]. To analyze the association of different days of the week and nutrient intake, we used the linear mixed models approach described in the methods section. The results are plotted as a heatmap (Fig. 1E) where the p-values indicate the statistical significance of the coefficients, with thresholds set at <0.001 (red), <0.01 (orange), <0.05 (yellow), and ≥0.05 (gray). The numerical values within the cells represent the coefficients of the day of the week for each nutrient, indicating the change in nutrient intake compared to the intercept (with Monday as reference). Furthermore, the results are segmented into groups as follows: all participants (Fig. 1E), age groups (Supplementary Fig. 1), BMI groups (Supplementary Fig. 2), sex groups (Supplementary Fig. 3) and cold/warm seasons (Supplementary Fig. 4) in order to observe the variations of dietary daily habits with respect to age and BMI.

We observed significant variations across different days for several nutrients and food groups, as shown in Fig. 1E. Alcohol intake showed notable increases from Wednesday through Sunday (all p < 0.001). Similarly, carbohydrate intake increases significantly throughout the week, peaking on Friday, Saturday, and Sunday (increase of 15 g, 29 g and 22 g respectively; p < 0.001). Energy intake followed a similar pattern, with substantial increases on all days, particularly on Friday, Saturday, and Sunday (+202 kcal, +371 kcal, and +250 kcal respectively, p < 0.001). The intake of sodium and sugar also increased significantly on the weekend, with sugar increasing by roughly 10 g on Saturday and Sunday, respectively (p < 0.001). These observations reflect more indulgent dietary habits towards the end of the week. In contrast, non-alcoholic beverages, especially water and coffee, exhibited a significant decrease from Friday to Sunday (p < 0.001). Notably, dairy was the only food group for which there was no significant difference across all days of the week, even when considering subgroups (Supplementary Figs. 1–4).

For individuals aged 35 years or younger (Supplementary Fig. 1A), carbohydrate consumption pattern was similar to the overall study population, with significant increases on Thursday, Friday, and Sunday (+15 g, +30 g and +18 g respectively, p < 0.001). Interestingly, these significant increments were also seen for other weekdays for the older age group, with an increase of about 11 g from Tuesday to Thursday (p < 0.05), shown in Supplementary Fig. 1B. Eaten quantity for the younger age group changed significantly over the weekend with the exception of Sunday where it decreased (−90 g, p < 0.01). In the age group 50+, this increase was not significant for any other days. Fruit consumption significantly increased during the weekend (between 35 and 40 g) as well among the older age participants. Note that fruit consumption in this study comprises consumption of both fruit juices and whole fruits, which might explain why the younger age group displayed a much higher baseline for this food category. Converse to the previous nutrient trends, intake of non-alcoholic beverages such as coffee and water decreased significantly, particularly during the weekend (all p < 0.001) in the younger age group, while this difference was either non-significant (coffee, tea and non-alcoholic beverages) or not as significant (water, p < 0.05) in the older age group. Vegetable consumption on Saturday and Sunday also decreased (−33 g and −28 g, p < 0.01) significantly in the younger age group but non-significantly in the older age group.

For individuals with a healthy BMI (between 18 and 25), carbohydrate and energy intake increased significantly from Friday onwards to Sunday (all p < 0.001), as highlighted in Supplementary Fig. 2A, akin to the observation for the full cohort. Interestingly, the intake of meat, and thus also protein consumption, steadily increased significantly over the week. Similar to the full cohort, intake of non-alcoholic beverages, including coffee and water, showed a notable decrease on the weekend (p < 0.001), while vegetable intake decreased significantly on Saturday (−26g, p < 0.01). For individuals with a BMI greater than 25 (Supplementary Fig. 2B), carbohydrate and energy intake patterns were similar to those with healthy BMI. Conversely, meat consumption was higher among those with high BMI, but its consumption over the week was not significantly higher than the baseline with the exception of Saturday. The consumption of vegetables and grains-cereals decrease was more pronounced over the week among high BMI participants. For potassium intake, the increment was observably the highest on Saturday in both the groups (p < 0.001 for healthy BMI; p < 0.05 for high BMI), while for sodium, the intake was substantially higher on the weekend (Friday to Sunday with p < 0.001) for the healthy BMI group. Sodium intake among high BMI participants was also higher during the weekend, substantially higher on Saturday (p < 0.001), but slightly lower on Sunday (p < 0.01) and Friday (p < 0.05). Finally, when comparing consumption patterns by sex, males had higher increases in consumption of meat, daily eaten quantity, and kcal energy across several days of the week compared to females (Supplementary Fig. 3), consistent with findings from earlier studies [40, 41].

Further examination of seasonal patterns revealed that the observed day-of-week effects persisted across both cold (November-April) and warm (May-October) months (Supplementary Fig. 4). The weekend effect on energy intake remained consistent across seasons (cold: +395 kcal; warm: +344 kcal on Saturdays, p < 0.001), similar to the overall pattern observed in the full cohort. The decrease in non-alcoholic beverage consumption, particularly water, was more pronounced during cold season weekends (−185 g on Saturday, p < 0.001) compared to warm season (−127 g, p < 0.001). Sweet and salty snack consumption showed strong weekend increases in both seasons but was more pronounced during warm months (warm: +256 g vs cold: +219 g on Saturdays, p < 0.001). Vegetable consumption decreased consistently across weekends in both seasons (cold: −25 g; warm: −35 g on Saturdays, p < 0.001), aligning with the overall pattern observed in the full dataset.

These seasonal variations, while noteworthy, are comparatively small, and ultimately reinforce the importance of weekend measurements in dietary assessment, as the fundamental pattern of weekend-weekday differences remains consistent throughout the year. When using mixed models to analyze dietary data, certain limitations may arise that suggest the need for complementary methods like ICC and CV. Mixed models estimate population-averaged effects, which can overlook subtle day-to-day variations within individuals. Additionally, skewed or non-normal data distributions can impact the model’s accuracy, and assumptions about random effects might obscure real patterns. Due to these potential limitations, using ICC and/or CV based approaches may help in capturing within-subject variability and providing a clearer picture of daily dietary behaviors.

While mixed models revealed weekly consumption patterns, determining the optimal duration for dietary assessment requires analyzing both within-person and between-person variability in nutrient intake. To address this central question, we employed the Coefficient of Variation (CV) method, established by Black et al. [17], to quantify the relationship between within-person and between-person variation to directly calculate the minimum days needed for reliable dietary assessment. By examining the variance ratio between these components, we can determine how many days of data collection are required to achieve specific reliability thresholds, where the correlation (r) between observed and true mean nutrient intake serves as our reliability metric (see Methods).

Figure 2 shows the minimum number of days required to achieve high to very high reliability thresholds of 0.8 (blue), 0.85 (yellow), and 0.9 (green) for various nutrients and food groups. Water, coffee, non-alcoholic beverages, and eaten quantity by weight can be reliably estimated (at r = 0.85) with just 1 day of data collection. Furthermore, with the exception of the eaten quantity feature, we observed for the other three features that their between-subject coefficient of variations (CVb) were quite a bit larger compared to their within-subject coefficient of variations (CVw), thereby reducing their variance ratios. This indicates that individuals have very consistent patterns in their fluid intakes and overall food quantity consumption, but simultaneously, these intakes vary across individuals.

These thresholds represent different levels of desired correlation between observed and true intakes, with 0.9 being the most stringent criterion. Top 30 nutrients are shown; sorted by the variance ratio. Additionally, variance ratio, between-subject coefficient of variation (CVb), and within-subject coefficient of variation (CVw) are shown for the top 35 nutrients.

Similarly, major macronutrient intake like carbohydrates, sugar, fiber, fat, and protein as well as energy consumption (measured in kilocalories), also showed high reliability with only 2–3 days of data (at r = 0.8). Micronutrients, on the other hand, tended to require longer periods for reliable estimation. Saturated fatty acid, beta carotene, vitamins C as well as certain minerals like phosphorus, potassium, calcium and iron typically needed 3–4 days of data (at r = 0.8 threshold) to achieve decent reliability. Moreover, iron intake also displayed high CVw. However, some micronutrients like folate and magnesium required only 2 days. Food groups such as dairy and fruits needed only 2 days for high reliability (r = 0.8), and about 5 days for very high reliability (r = 0.9). Meat and vegetables required longer, about 3 days for high reliability, and correspondingly longer, 7 days, for very high reliability.

The ICC values for various nutrient features were computed over different numbers of days to determine the optimal duration of dietary assessment for reliable estimates of average nutrient consumption. Figure 3 shows the top 24 distinct food features that reached an ICC threshold of 0.75, the remaining features are shown in Supplementary Fig. 5. The x-axis in both figures represents the number of days, while each boxplot demarcates the ICC scores for combinations of days for each minimum number of days on the x-axis. The ICC thresholds of 0.9 and 0.75 are shown as a horizontal red line in each subplot. The color of the boxplots represents different nutrient groups: blue for micronutrients, red for food groups, and green for macronutrients. The ICC values for all nutrients generally improve as the number of days of dietary data increases. This indicates that more extended periods of dietary assessment tend to provide more reliable estimates, as expected. Largest daywise variability for all nutrients occurs when only 2 days of data are used.

The x-axis represents the number of days, while the y-axis represents the ICC scores. Each subplot corresponds to a different nutrient, with boxplots showing the distribution of ICC scores computed from all possible combinations of days for each number of days on the x-axis. The horizontal red lines in each subplot denote the ICC threshold of 0.75 (good reliability) and 0.9 (excellent reliability). Boxplot colors represent different nutrient groups: blue for micronutrients, red for food groups, and green for macronutrients. Note that ICC does not generally reach 1 due to inherent within-subject variability.

It took 3 days for the median ICC of carbohydrate intake to attain a value of 0.75, which is widely considered the minimum threshold for good reliability [36]. For protein and fat intake, it took longer, about 4–5 days, to attain this threshold. Similarly, the median ICC for alcohol intake required 5 days, while sugar and fiber intakes required just 2–3 days to reach 0.75, and in about 7 days nearly reached the ICC threshold of 0.9, which is considered excellent reliability. Four days was sufficient for the median ICC of energy kcal consumption to reach the 0.75 threshold, while this was just 2 days for eaten quantity. Micronutrients, like magnesium, pantothenic acid, folate, phosphorus and calcium, were able to cross the 0.75 threshold at 4–5 days, as observed by their median ICCs from their day-wise combinations for these days. Interestingly, water and non-alcoholic beverages intakes cross the 0.75 threshold in just 2 days. Many combinations of food groups, like vegetables-fruits and oil-nuts, as well as dairy, attained the 0.75 threshold at 4 days, while for meat and grains-potatoes-pulses, reaching the 0.75 threshold required at least 5 days.

In summary, using this method, most nutrients show a stabilization of ICC values above the 0.75 threshold after 4–5 days, suggesting that to achieve high reliability for most nutrient intake measurements a 4–5 days of dietary logging is desirable. Certain nutrients, typically non-alcoholic drinks such as water, achieve higher ICC values even with fewer days of measurement, indicating that these nutrients may require less extensive data collection to reach reliable estimates.

In order to extract the best (and worst) collection day combinations for each nutrient or food group, we used another approach wherein we compare the ICC of different day combinations to the mean intake of the entire dataset (consisting of a 7 day week). Hence, under the assumption of an entire week being sufficient in capturing the nutritional variation, this method allows to elucidate which days of the week (or their combinations) lead to highest (or lowest) ICC values when comparing to the full week.

Figure 4 highlights the best and worst day combinations at different number of days across various nutritional features (for the full set of nutrition features, see Supplementary Figs. 6, 7). Since the assumption of one week being sufficient to capture variation already loses some precision, we set the ICC threshold in the assessment below at 0.9, indicating excellent reliability. In order to achieve this threshold, carbohydrate intake required 3 days that were spread across the week (Monday, Wednesday and Saturday) for the best combination. The worst combination of 3 days also reached the 0.9 threshold, but the days were continuous weekdays from Monday to Wednesday. Similarly, for protein and fat intake, it required 3 days minimum for the best combination to cross the 0.9 threshold and included two weekdays and one weekend day (Saturday for both protein and fat), while their worst 3 day combination was below the 0.9 threshold. For fiber, the best combination of 2 days (Monday and Wednesday) attained the 0.9 threshold precisely, although with 3 days, even the worst combination was above the excellent reliability threshold. For alcohol, the alternating day combination at 3 days resulted in being the best combination which surpassed the threshold, while the worst combination at even 4 days - which were all continuous weekdays - was below the threshold. Eaten quantity and water intake required only one day of collection (Wednesday) to reach the reliability threshold. The same can be observed for non-alcoholic beverages group (Supplementary Fig. 6) Coffee intake nearly achieved this threshold with just one day of data (Tuesday). Interestingly, when adopting a more conservative approach by increasing the collection period to 2 days, the optimal combination for all four nutritional features (eaten quantity, water, non-alcoholic beverages and coffee) was consistently Monday and Friday. Similarly, extending to a 3 day collection period, the best combination was Monday, Tuesday, and Friday for all these features (except for non-alcoholic beverages).

For each nutrient and at each number of days, the day combinations which yielded the highest (in color) and lowest ICC scores (in gray) are shown. The plot is ranged between ICC values of 0.75–1.0, with the ICC reliability threshold at 0.9 shown as a red line - points lying below this range are not shown.

A 3-day combination of Monday, Tuesday, and Thursday precisely reached the 0.9 reliability threshold for meat intake. Notably, when extending the collection period to 4 days, even the least optimal combination - consisting of four consecutive days - surpassed this threshold. This suggests that any 4-day period within a week is sufficient to accurately capture an individual’s meat consumption patterns. For dairy intake, the least optimal combination at 3 days, consisting of two weekend days, had exactly reached the 0.9 threshold, while the best combination consisted of Monday, Tuesday and Sunday. Bread consumption needed 4 days for its best combination to surpass the 0.9 threshold. The collective vegetables and fruit group required just 3 days for the worst day combination to reach the threshold, while the best combination included only weekdays (Monday, Tuesday and Thursday).

Interestingly, the optimal 2-day combination for iron intake (Monday and Saturday) exceeded the 0.9 threshold. However, unlike other features, even some 6-day combinations that included both Monday and Saturday fell below this threshold. This unexpected result suggests that the observed patterns in iron intake may be subject to high variability or influenced by factors not fully captured in our data. For vitamin C, the best combination at 3 days (Monday, Tuesday and Saturday) reached the 0.9 threshold exactly, although a more conservative choice of 4 days yields ICC values above 0.9 even for its least optimal combination. Other vitamins like B1, B2, B6 and D required at least 4–5 days for the most optimal combination to cross the threshold, while vitamin B12 required just 3 days (Supplementary Fig. 7). For saturated fatty acids, the results were similar to vitamin C, with 4 days yielding ICC values above 0.9 even for the least optimal combination. Polyunsaturated fatty acids and monounsaturated fatty acids, however, required more days (Fig. 3 and Supplementary Fig. 7 respectively): only the best 5-day combination crossed the 0.9 threshold, while some 5-day combinations produced ICC values below 0.75, highlighting greater variability in intake.