A limited ability to accurately assess water intake at a population-level has likely slowed progress in elucidating the impact of water intake on health. Some evidence suggests that low water consumption and underhydration are associated with adverse health outcomes including chronic kidney disease and diabetes (1–3). Similarly, the Institute of Medicine suggests dehydration may be related to numerous detrimental health outcomes including cardiovascular strain, urinary tract infection, and diabetes dysregulation (4). Conversely, increased water intake is associated with positive health outcomes including reduced risk for kidney stones (4) and urinary tract infections (5) as well as augmented glucose regulation (6) and adolescent cognitive performance (7). Thus, there appears to be an inverse relationship between water intake and health risk. However, evidence supporting these associations is not sufficient to establish total water intake (TWI) recommendations beyond an Adequate Intake, the least specific recommendation included in the Institute of Medicine's dietary reference intakes (4). Due to the wide range of TWI volumes that allowed individuals to maintain adequate serum osmolality, the Adequate Intake is the median value of the TWI volumes observed in the Third National Health and Nutrition Examination Survey (4). These high variations have largely been attributed to differences in culture, climate, and/or physical activity (4). However, measurement error in TWI assessment and lack of a standard assessment tool could exacerbate these variations.

The TWI Adequate Intake recommendations encompass water consumed from all foods (~20% TWI) and liquids (~80% TWI) (4). The current recommendation is based upon the National Health and Nutrition Examination Survey 24-h diet recalls, which were conducted before utilization of the United States Department of Agriculture's Automated Multiple-Pass Method, which is a validated method for energy and nutrient intake, but not TWI (8). Additionally, trained interviewers prompt participants to choose an occasion for every food item on the record, most of which are meals (9). However, beverage intake occurs more continuously throughout the day. Individuals have 0 – 19 drinking occasions per day (of water only), with an average time of 3 h between drinking occasions (range 1–17 h) (10). This has been observed when TWI was compared between a fluid-specific tool and the United Kingdom's National Diet and Nutrition Survey, which utilizes food diaries (11). The fluid-specific tool revealed that 70% of beverage consumption occurred outside of meals. In Indonesian populations, TWI estimated from a 7-day fluid diary was significantly greater than that from the 24-h dietary recall, by 382 mL (12). Additionally, the 24-h dietary recall captured 2.2 fewer drinking occasions (6.7 vs. 8.9 occasions). Consequently, current dietary assessments which have not been validated for water intake are not necessarily suitable for drinking behavior.

To date, investigators have not been able to identify a method to estimate TWI that is comparable to an objective reference value, such as that obtained from water turnover by dilution of deuterium oxide (D₂O) (13) corrected for metabolic water, which is costly and impractical for population-level use. Recently, our group and others have begun to advance the field through development and validation of fluid-specific assessment tools (14–19). Compared to 24-h recalls, which are subject to bias from day-to-day variation in consumption, frequency questionnaires are more likely to capture usual intake (20). However, only relative validation, via dual reporting, has been assessed for prior beverage frequency questionnaires with comparison of water intake estimates against self-reported 24-h records (14–17). As the 24-h diet record and new questionnaires under assessment for validation are both self-reporting instruments, sources of error will overlap between the instruments and be correlated. Additionally, validation through dual recording will not distinguish inaccuracies if they are reported on both assessments. Dual recording could also deceptively improve accuracy of the new questionnaire, and therefore falsely show validation, as recording intake in diet records in days leading up to a frequency recall will likely improve recall accuracy.

We recently utilized D₂O to validate Liq.In⁷, a 7-day fluid diary, to record all beverage intake over seven-days. While it has been shown to be an accurate recording instrument for TWI volume, the seven-days of recording impose substantial subject burden (18). Additionally, Liq.In⁷ only captures water from liquids, and not TWI. However, there is limited evidence from the US and Europe supporting the current belief that TWI is ~80% water from liquids. In fact, those with high and low TWI have been observed to consume a similar amount of water from food (~0.6 L·d⁻¹), resulting in substantially different contributions to TWI. For instance, water from food comprised ~23% of TWI in those with high TWI, while ~47% in those with low TWI (21). Consequently, liquid-only assessments may be preferable in studies where precise recording of fluid intake is important but may elucidate misleading results in terms of TWI. To address this gap, we developed a total water intake frequency questionnaire (TWI-FQ) that prompts individuals to recall water intake from food and beverages over a 7-day period. The purpose of this investigation was to assess the reliability and validity of the TWI-FQ to estimate TWI as compared to the value obtained with dilution of D₂O, corrected for metabolic water.

Body mass was consistent within weeks (% of change in body mass: week 1, 0.05 ± 0.99%; week 2, 0.35 ± 1.19%; week 3, 0.11 ± 1.22%) with low coefficients of variance between the three measurements during all weeks (week 1, 0.54 ± 0.36%; week 2, 0.62 ± 0.41%; week 3, 0.60 ± 0.50%). Mean daily water turnover and the components that contribute to water turnover computed using data from D₂O dilution and the TWI-FQ are presented in Table 2. Daily caloric intake during week 2 was 2,028 ± 523 kcal (range: 911–3,430 kcal).

The jack-knife analysis identified eight mild outliers and three severe outliers across eight participants (63% male; age, 37 ± 13 y; BMI, 26.5 ± 5.9 kg·m⁻²) (Supplementary Table 2). All three severe outliers were found in males in week 1, while four mild cases were identified in each week. TWI was overestimated by the TWI-FQ in five of the eleven cases, two of which were identified as severe outliers. No outliers were excluded from validity or reliability analyses.

The purpose of this study was to assess the validity and reliability of a TWI-FQ to estimate TWI as compared to the objective reference value, D₂O. The principle finding of this study is that the TWI-FQ consistently underestimated TWI. While this tool would not be suitable for individual assessment, the overall magnitude of bias may be acceptable for assessment at the sample-level. In this protocol, we utilized the isotopic tracer, D₂O, as the rate of disappearance of D₂O following enrichment is directly associated with water turnover and is not subject to homeostatic or inter-individual variations in metabolism (34). Accordingly, D₂O is an unbiased measure of water turnover that is not subject to measurement error commonly seen in self-report data. Furthermore, we utilized Bland-Altman statistical analyses that accounted for variation between methods, between individuals, and between occasions (33). Most prior studies (14–17) have utilized correlation and t-test analyses, which are not independently sufficient to assess agreement between two methods for validity assessment (31, 32).

Due to the robustness of the D₂O dilution method, we observed total body water as a percentage of body mass estimates within the ranges reported by the Institute of Medicine (43–73%, males 19–50 y; 41–60%, females 19–50 y) (4). Additionally, daily metabolic water production has been estimated to be ~250–350 mL·d⁻¹ for sedentary individuals (4). Although metabolic water (264 ± 107 mL·d⁻¹) was determined from diet records during the wash-out period in the current study, it still aligns with the aforementioned estimates. Confirmation of body weight stability for all participants across each week indicates that the assumption that TEE was equivalent to total energy intake was met for metabolic water calculations. Accordingly, we are confident the D₂O TWI estimates reflect actual TWI.

The TWI-FQ significantly underestimated TWI by −350 ± 1,431 mL·d⁻¹ compared to D₂O. While the mean difference is clinically adequate, there is considerable variation in bias as evidenced by the standard deviation of 1,431 mL and limits of agreement allowing for underestimation of −3,155 mL·d⁻¹ and overestimation up to 2,455 mL·d⁻¹. The magnitude of these differences is substantial considering the Adequate Intake for water is 2.7 L·d⁻¹ for women and 3.7 L·d⁻¹ for men (4). Based on visual examination of Figure 3, these large differences appear to be driven, in part, by individuals who consume high amounts of TWI (≥4 L·d⁻¹). In some cases, high amounts of TWI were accurately reported in the TWI-FQ (Supplementary Table 2). Large differences may in part be related to a learning curve as all three severe outliers were identified in the first week. Furthermore, most participants with outliers appeared to improve by week 3 (i.e., reduced from severe to mild outlier or no longer an outlier). Ultimately, outliers were a mixture of overestimation and underestimation with no clear association with subject characteristics (i.e., sex, age, BMI). Despite large variances, the TWI-FQ was still determined to be reliable due to moderate correlation between weeks (r = 0.725) and moderate test-retest reliability (ICC = 0.706). Systematic bias in TWI-FQ between weeks was not statistically significant as evaluated via Bland-Altman plot, in which the mean difference in TWI-FQ TWI estimates was 198 ± 1,180 mL·d⁻¹.

The mean difference (36 ± 593 mL·d⁻¹) between repeated D₂O TWI estimates was minimal and non-significant. However, the acceptable limits of agreement (-1,149, 1,221 mL·d⁻¹) are still large clinically and indicate a considerable degree of within-subject variance in week-to-week TWI. Additionally, mean D₂O TWI was distributed across a wide range of volumes, between 1,000 and 9,000 mL·d⁻¹, with the majority of mean D₂O TWI falling between 1,000 and 4,500 mL·d⁻¹. This indicates there is also a considerable degree of between-subject variance in D₂O TWI, which was also captured by the TWI-FQ, as can be seen in Figures 3A,B. This magnitude of variance in TWI is not surprising as daily water needs can vary greatly between and within individuals depending on age, sex, diet, physical activity behaviors, climate, and culture (4). We purposefully recruited participants who were well-distributed across sex and age. Therefore, although the limits of agreement for the TWI-FQ validity assessment were large, these data indicate that the variance observed was compounded by within- and between-subjects' differences in water consumption habits.

Previous liquid questionnaires have been developed to assess fluid intake but not TWI (14–17, 19). While this TWI-FQ was designed specifically to assess water intake volume at population levels, previous questionnaires were developed primarily to assess energy intake from liquids (16, 17), grams or fluid ounces of individual and total liquids consumed (16, 17, 19), water intake and voiding habits for treatment of urinary tract symptoms (14), and water balance (15). Additionally, validation protocols for these questionnaires utilized imperfect reference instruments, such as 24-h diet records, which are subject to intake-related bias and correlated error (35). We used methods similar to the previous study to assess validity and reliability of the Liq.in⁷, which is a 7-day fluid record that required participants to record liquids and foods with high water content as they were consumed (18). Compared to D₂O, the Liq.in⁷ underestimated water from liquids by −131 ± 845 mL·d⁻¹. However, this assessment was based only on one week of data, the Bland Altman statistical analysis utilized did not account for within or between subject variation, and water from food was not included in this analysis. TWI was also assessed between the Liq.in⁷ and a 24-h dietary recall in Indonesian adolescents and adults using a Bland Altman analysis (12). An overestimation of 382 mL·d⁻¹ was observed compared to the 24-h dietary recall with limits of agreement 1,600 and −2,300 mL·d⁻¹. Although the limits of agreement were narrower than those in the current study, the difference was determined to be significant as 11% of values fell outside of these limits. The mean difference also increased with greater TWI, with underestimation of 139 mL·d⁻¹ for the lowest quartile of TWI and overestimation of 1,265 mL·d⁻¹ for the highest quartile of TWI. Thus, it appears individuals are less able to recall fluid intake accurately with greater consumption.

Our approach does not come without limitations. Metabolic water production was determined through self-reported data in 24-h diet records. Self-report dietary assessments are subject to error (e.g., difficulty interpreting handwriting, day-to-day variation in consumption, or misreporting of consumption) and can be burdensome to participants. However, metabolic water is a small component of water turnover (250–350 mL·d⁻¹) (4) and over- or underestimation would not substantially impact the outcomes of this investigation. Furthermore, a prominent study in this field that determined water turnover in 458 adults (40–79 y) estimated metabolic water from the average macronutrient content of the diet based on a one-time 24-h recall in the general population in the US (26). In contrast, participants in the present investigation completed multiple 24-h diet recalls for metabolic water estimates.

The accuracy of the TWI-FQ may vary day-to-day, with TWI estimates that are more representative of days closer to the day of questionnaire completion. However, we were not able to evaluate this as participants are asked to recall consumption for the entire week rather than for each day of the week. Similarly, the D₂O method utilizes three urine samples to determine an average daily TWI for the 7-day period and does not allow for estimation for each specific day. Furthermore, we were not able to evaluate potential differences in validity or reliability of the TWI-FQ by age or sex as this study was not powered for these comparisons. Finally, we were not able to validate whether the TWI-FQ is sensitive to change in TWI. Therefore, this tool may not be suitable for use in intervention studies designed to change TWI, particularly if detection of small changes is desired.

In conclusion, the TWI-FQ may be a useful tool to assess population-level TWI behaviors. Due to the large variances observed, the TWI-FQ should not be utilized to assess individual-level TWI behaviors in which greater accuracy may be needed. Utilization of the TWI-FQ to assess population-level TWI may allow investigators to better determine relationships between liquid intake, hydration, and health. Moreover, the TWI-FQ could be utilized in conjunction with multiple 24-h diet recalls/records to better reflect water from food and subsequently TWI. Several studies have successfully improved accuracy of self-report dietary data through combining 24-h diet recall/records with food frequency questionnaires (36). The findings of this study can only be generalized to individuals 19–65 y. Further investigation is needed to assess application of the TWI-FQ in different geographical regions, climates, cultures, activity levels, and age groups.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

The studies involving human participants were reviewed and approved by University of Arkansas, IRB. The patients/participants provided their written informed consent to participate in this study.

The authors' responsibilities were as follows–SK, FP, IG, and EP designed research (project conception, development of overall research plan, and study oversight). EJ, LJ, CC-J, and JA conducted research (hands-on conduct of the experiments and data collection). AC, EJ, FP, SK, and AM analyzed data or performed statistical analysis. AC and SK wrote paper. SK had primary responsibility for final content. All authors have read and approved the final manuscript.

EJ had received grants from Danone Research and FP had served as member of the scientific advisory board of Danone Research. IG and EP were Danone Research employees. SK had served as scientific consultant for Quest Diagnostics and Standard Process and had served as a member of the scientific advisory board of Danone Research, and had active grants with Danone Research and Standard Process. The remaining 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.