Food records and 24-h dietary recall methods provide detailed information regarding food and nutrient intake; they help determine dietary intake in population settings (1–3). They also examine sources of bias and random variation in collecting nutrient intake data (1). Several studies assessed the within- and between-person variation through the food record and 24-h dietary recall and demonstrated the need to collect dietary information for at least 2 days (1–7) to capture the within-person variation (1, 6). A follow-up study showed within-person variation in energy intake throughout the year, including a weekly residual variance of 2–26% of energy intake and a systematic variation of energy intake according to the days of the week, coincident with a slight variation in between-person variation (4). Other studies revealed different sources of variation per population settings (5–8), including the variation among days, weeks, or seasons (9) and well-marked differences per season in Brazil (5) that depended on sex and age (7).

Random variation is commonly observed in short-term methods. This variation affects statistical analysis and hypothesis testing because it increases the data dispersion, although the nutrient or food intake average does not change (9, 10). One day of dietary intake assessment does not represent usual food and nutrient intake (10, 11); therefore, collecting more than one repeated measure has become common practice in diet surveys to capture day-to-day variation, also known as random variation of food intake, which is usually large (11). The greater the number of days of dietary intake assessment, the closer the data dispersion approaches the variation of usual dietary intake (10–17). After several 24-h dietary recalls, the variation becomes more stable (10–17). Still, the number of repetitions required depends on the population in the study, gender, culture, diversity of food intake, and the level of accuracy expected.

Palaniappan et al. showed that the energy variance ratio was much lower among men (0.49) than women (1.76) (16). Oh and Hong reported that older adults required more dietary records in Korea than in western countries, resulting in culture-dependent eating habits. The authors concluded that the number of repetitions required for nutrients depended on the type of nutrients, gender, and the level of accuracy expected. For instance, to assess energy intake with 5% accuracy, it would be necessary to record 97 days of dietary assessments among men and 128 among women; however, for riboflavin, 240 days would be necessary for men and 455 for women. The number of days required to assess the usual nutrient intake was smaller if lower accuracy is accepted. For example, in the same study, energy intake among men could be captured with 1 day of assessment if 50% accuracy was deemed acceptable (17).

This method of evidence collection has guided study design, sample size calculation, and strategies for data collection to reduce bias and random variation across populations (4–18). It has emphasized collecting dietary data in non-consecutive days to capture the within-person variation for food or nutrient intake (10). Nevertheless, data collection using these methods is practically infeasible in large or prospective studies because of the number of replications required to ensure their accuracy, resulting in expensive data collection and complexity of the analysis to manage random variation and bias (4–9). Alternatively, 48-h dietary recall enables collecting 2-day dietary data in a single individual assessment, increasing the response rate and expanding the within-person variation. The paucity of studies using the 48-h dietary recall allied with specific characteristics of this methodology implies that its ability to capture within-person variation is not known.

Nevertheless, a few studies have already adopted 48-h dietary recall. A recent study comparing the derivation of food patterns using various methods for data collection of diet showed that 48-h dietary recall appeared to be superior to single 24-h dietary recall because it allowed the generation of food patterns with a reduced number of interviews (18). Compared to a Food Frequency Questionnaire, the 48-h dietary recall was found to estimate the protein intake accurately (19). Rehm et al. found that 48-h dietary recall was superior to a single food record (20).

Regardless of whether a 48-h dietary recall is a valid method for capturing the natural variation of food and nutrient intake in population settings, the method could gather information of 2 days (24-h dietary recall) in one interview, optimizing time and increasing response rate. Therefore, in the present study, we evaluated the sources of variation, estimating the sample size and number of days required to detect the usual intake of energy, macronutrients, cholesterol, total fiber, and micronutrients by gender, age, seasons, and periods of the week in data collected using the 48-h dietary recall method.

This was an exploratory analysis using data collected to validate a semi-quantitative Food Frequency Questionnaire in southern Brazil. Dietary intake was assessed using 48-h dietary recall to better capture the within-person variability in a single evaluation. This analysis hypothesized that a 48-h dietary recall would capture within-person variation better than a single 24-h dietary recall, but not as much as two non-consecutive 24-h dietary recalls. Because secondary data were used in this study, nutrient intake calculated with the 48-h dietary recall used in this data collection was compared with 2 days of 24-h food recalls according to the available literature. In this manner, our analysis aimed to support decision-making for future studies based on the strengths and limitations of 48-h dietary recall.

There were 237 study participants, including 129 were women, 126 children 11–19 years old, and 46 were elderly (60–90 years old). Among the entire group of participants, the within-person coefficient of variation was lower than between-person variation for most nutrients, except for calcium (CVw2 = 40.8 and CVb2 = 38.4%), magnesium (CVw2 = 27.4 and CVb2 = 18.7%), and monounsaturated fat (CVw2 = 20.0 and CVb2 = 17.3%). In women, there was a higher within-person variation than between-person variation for magnesium intake (CVw2 = 28.3 and CVb2 = 91.8%) and in men, for vitamin C (CVw2 = 83.7 and CVb2 = 36.1%) (Table 1). The magnesium intake for the entire group of participants had the highest variation ratio (λ = 2.14), followed by magnesium among women (λ = 2.11) and calcium in the entire population group (λ = 1.13). Men had a lower variation ratio in general, ranging from 0.005 for carbohydrates to 0.76 for calcium (Table 2). The correlation between the usual or true, and the estimated (as calculated based on this data) energy and nutrient intake was high for most nutrients; however, it was a little lower for magnesium (r = 0.69), calcium (r = 0.80), and iron (r = 0.84) intake in the entire group of participants, for magnesium (r = 0.70) and iron (r = 0.86) intake in women, and for calcium intake of men (r = 0.85) (Table 2).

Like for age (Table 3), within-person variance varied substantially for micronutrient intake. In general, the within-person variance was lower for participants 60–90 years old, except for vitamin C (CV_w = 71.6 in the elderly and CV_w = 19.3 in adults). In adolescents, the correlation between and observed intake was perfect for total fat and cholesterol. There was a perfect correlation between usual and observed protein intake, total fiber, calcium, potassium, saturated fat, and the elderly, for carbohydrate and total fiber in adults. Table 4 shows that the sample size to assess macronutrient intake in the whole population was smaller than for other nutrients. To estimate the energy and the total fat intake using a 48-h dietary recall to assess 2 days of dietary intake, 58 and 56 participants, respectively, would be necessary. More significant numbers of participants would be required to assess protein and carbohydrate intake (181 and 110 participants, respectively). The sample size differed substantially for each nutrient and participant group. To assess the intake of vitamin C using a 48-h dietary recall to assess 2 days of dietary intake, 1,641 men would be necessary but only nine women. This is a result of the much lower vitamin C within-person variation for women than for men that resulted in a perfect correlation between usual and observed intake for women (r = 1.00) and a lower correlation for men (r = 0.80) (Table 2). A similar phenomenon was sought for cholesterol intake. For cholesterol, it would be necessary to consider 636 adults and 653 older adults (Table 4), but only 28 adolescents, because, using a 48-h dietary recall to assess 2 days of cholesterol intake, the correlation between usual and observed intake was lower for adults (r = 0.88) and the elderly (r = 0.88) but higher for adolescents (r = 1.00) (Table 2). This finding suggests that the elderly population varied more in terms of intake of food sources of cholesterol from day to day than did adolescents.

Using one 48-h dietary recall to assess 2 days of dietary intake, capturing 10% of the within-person variation (the percentage of variation around individual averages) in the entire group of participants, 7, 8, and 18 days would be necessary to assess the usual intake of energy, total fat, and carbohydrate intake, respectively; however, a much larger number of days would be necessary to assess protein intake (151 days) (Table 4). The within-person variation had the most significant impact on the number of days necessary to assess usual dietary intake. The larger the within-person variance and the variance ratio (Tables 2, 3), the greater the number of days needed to assess usual nutrient intake. For example, in the entire group of participants, the number of days to assess the usual intake of magnesium (1,119 days), vitamin C (1,110 days), and iron (1,089 days) (Table 4) accepting an error of 10% around the individual average were similar. In these examples, the numbers of days necessary to assess usual dietary intake was more dependent on the within-person variance ratio (λ) than the between-person variance (magnesium CV_w = 27.4 and λ = 2.14; vitamin C CV_w = 52.8 and λ = 0.40; iron CV_w = 40.6 and λ = 0.69) (Table 2).

Adolescents needed 24, 54, 422, and 59 fewer days than adults to assess the intake of energy, total fat, cholesterol, and polyunsaturated fat, respectively. They required 42, 260, 443, and 1,684 fewer days than the elderly to assess the same nutrients (Table 4). The numbers of days necessary to assess carbohydrates and total fiber were higher among adolescents than adults (37 and 118 days, respectively) and the elderly (978 and 999 days, respectively). Overall, the adult population needed fewer days than adolescents and the elderly to assess intake of protein (44 and 549 days less, respectively than adolescents and elderly), calcium (714 and 248 days less, respectively), iron (1,124 and 81 days less, respectively), magnesium (1,554 and 642 days less, respectively), potassium (278 and 31 days less, respectively), vitamin C (479 and 1,876 days less, respectively), and saturated fat (567 and 83 days less, respectively).

We found that 48-h dietary recall provided ineffective measurement of within-person variation. This finding means that, for some nutrients, 48-h dietary recall informs as much as a single 24-h dietary recall, depending on age and gender, although the between-person variation was captured acceptably. Because the within-person variation was potentially underestimated, the sample size and number of days necessary to assess the usual intake of energy and nutrients are more likely to represent one instead of 2 days of dietary intake.

We found lower within-person variation than between-person variation for intake of energy, macronutrients, and micronutrients than other studies that used two or more non-consecutive (8, 30) and consecutive (31) 24-h dietary recalls for data collection. Oh and Hong, in an elderly population from Inchon, Korea, observed that within-person variation was twice as high in men and 1.5 as high in women than between-person variation (17). Compared to a study based on data collected using two 24-h dietary recalls in adolescents, the within-person variation of energy was 280-fold, and the variance ratio was 20 times higher than what we observed in the present study (30).

Several conditions may influence the variation ratio. We used an unadjusted statistical model that could have influenced the results. For example, in the analysis of data collected using 24-h dietary recall, the energy variation ratio of men was 1.07 in the analysis adjusted for the fixed effect of gender, age, education, smoking, size of family, and season; the variation was 0.49 in the unadjusted analysis. In the same study, the variation ratio was lower in the unadjusted models for women (16). Macronutrient intake influences energy balance, remaining constant throughout the days (1); therefore, the intake of each nutrient is expected to vary more than energy, in general (16, 30). Studies based on the 24-h dietary recall have shown that intake of protein, total fat, and carbohydrate is naturally higher than the energy intake in general (16, 32); however, there may be lower for women's intake of protein (17) and carbohydrates (30, 32). In the present study, cholesterol, fiber, and micronutrients had a higher variation ratio than energy and macronutrients, as observed in previous studies, although those studies were based on the 24-h dietary recall (8, 13, 16).

Consequently, we acknowledge that the 48-h dietary recall could capture lower energy and nutrient variability than two 24-h dietary recalls. In addition, 48-h dietary recall and the food record, both based on recall, performed better in men with significant differences for nutrient intakes when comparing the former to the food record. This finding suggests that males had less variation in their intake over the 48-h dietary recall than subsequent (5) days recorded in the food record (1). In this and other studies, the sample size and the number of days required to assess the energy and nutrient intake, accepting an error of 10% around the individual average, or based a correlation of 0.95 between the and the estimated energy and nutrients intake, differed according to gender and age group and ranged widely among nutrients (1, 8, 12, 17, 30).

Using the 48-h dietary recall method to collect the data was as much accurate as one 24-h dietary recall to assess the intake of vitamin C for women in contrast to seven nutrients (protein, total fat, cholesterol, total fiber, calcium, iron, magnesium, and potassium) for adult men and women. Two days of nutrient intake showed a perfect correlation between usual and estimated nutrient intake for a few nutrients, and this result varied with the population group. These perfect correlations suggest that one 48-h dietary recall and two consecutive 24-h dietary recalls generate the same results. Studies using non-consecutive 24-h dietary recall or food records showed lower correlations between the usual and the estimated nutrients (8, 17, 30).

Energy and macronutrient variability tend to be lower than those of fiber, cholesterol, and micronutrients; the overall variability is strongly influenced by gender and age group and can be captured using either the 24-h dietary recall or the 48-h dietary recall. The 48-h dietary recall might be helpful to assess energy and nutrients because it captures between-person variation. For example, it can be used to assess energy, protein, total fat, and polyunsaturated fat in the overall sample; however, among the elderly, it might introduce higher variation than in other population groups for the assessment of energy, protein, total fat, vitamin C, and polyunsaturated fat. Nevertheless, within-person variation is considerably lower than between-person variation when the data are collected using the 48-h dietary recall, and this finding contradicts the assumption that one 48-h dietary recall could capture the variability of two 24-h dietary recalls accurately. Our study did not intend to recommend how many days are necessary to assess dietary intake using the 48-h dietary recall to other populations; instead, its purpose was to highlight the limitations and strengths of this methodology, showing potential influences of the method and population characteristics on usual nutrient intake estimations, to support decision-making and research to assess the validity of this methodology.

The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author/s.

The studies involving human participants were reviewed and approved by Committee of Ethics in Research of Hospital de Clinicas de Porto Alegre, and approved under the register number GPPG-00176.

SR planned and performed the statistical analysis and wrote the manuscript. SF conceived and designed the data collection, took part in the analysis, and revised the manuscript. All authors contributed substantially to the work and approved the final version of the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.