The National Diet and Nutrition Survey: adults aged 19 to 64 years, (NDNS) is a nationally representative survey of the diet and health of adults living in private households in Great Britain in 2000/2001 21 . It was commissioned jointly by the Department of Health and the Food Standards Agency and ethical approval was obtained via the normal channels. Fieldwork was conducted over a 12-month period in 2000/2001 to cover any seasonality in eating behavior and in the nutrient content of foods. Overall, 61% of the eligible sample (n = 3704) completed the dietary interview (responding sample, n = 2,251) and 77% of those who completed the dietary interview completed a full seven-day weighed dietary record (diary sample, n = 1724, representing 12068 person-days of data) 21 . There was no evidence of serious non-response bias, although inevitably surveys of this type favour willing participants whose diets may be less extreme, or less variable day to day.

Respondents were asked to keep a weighed record of all food and drink consumed, both in and out of the home. Each respondent was issued with a set of digital food scales and instructed how to weigh and record items in two diaries, “Home” and “Eating out”. A description of each item was recorded, including the brand name of the product and if appropriate, the method of preparation. Also recorded was the time (to the nearest 5 minutes) and location. Everything consumed by the respondent had to be recorded, including drinks of water and medications and supplements. Respondents were asked to weigh everything they could as separate items in order that the nutrient composition of each could be calculated. Recipes for all home-made dishes were collected. Where it was not possible to weigh the food, respondents were asked to record as much information as possible, particularly the portion size and an estimate of any leftover and, for meals out, price and place of purchase. In certain circumstances duplicate portions were purchased for weighing. Interviewers visited workplace canteens to collect information on portion sizes, cooking methods and ingredients.

Each interviewer called back 24 hours after placing the diaries to check that the items were being recorded correctly, to give encouragement and re-motivate where appropriate. Further visits were included as necessary and any apparent omission of meals or snacks was probed. Interviewers were trained in coding the diaries and could therefore identify the level of detail needed for different items. As fieldwork progressed, new codes were added for homemade dishes and items appearing in the diaries. Interviewers were asked to assess the quality of the dietary record and how far they thought it reflected the respondent’s normal diet. Further detailed checks were carried out for completion and consistency at the Office for National Statistics and then information from the dietary records was linked to the Nutrient Databank. This holds information on 56 nutrients, including water content, for each of the 6000+ food codes.

Respondents also completed a diary detailing the time spent on specified activities, and the intensity of activity, over the same 7 days as the dietary record. A physical activity score was calculated by multiplying the duration of each activity by the metabolic equivalent value (MET) for activities of that intensity. The average MET score represents an estimate of hourly energy expenditure per kg of body weight. Body weight was measured by trained personnel at the nurse visit after completion of diaries. Further details are given in Appendix D of the published report 22 .

Computerised raw data files and documentation from this survey, comprising weighed food records, questionnaires and biochemical and medical data, were obtained under license from the UK Data Archive (http://www.data-archive.ac.uk) and analysed using IBM® SPSS statistics v19 (SPSS IBM Inc Chicago, Illinois, USA).

Data were first analysed at the level of the individual food diaries, comprising weighed records for 12068 days (7 days for each survey respondent). Individual food items are recorded both with their original food codes and a higher level classification into 115 food groups for the published reports. The present analysis focused on total water intake (TWI) of all food and drink, as determined from food composition tables in the database. Metabolic water (water derived from oxidation of substrates) was not included as the focus was on comparison with dietary water requirements. Water in coffee or tea was coded as coffee/tea, while water for diluting concentrated soft drinks was coded as soft drink. Bottled water was coded separately from tap water. Milk includes all liquid milk, including that in tea and coffee, and milk consumed with cereal. Discretionary table sugar added to beverages was counted as food, rather than drink, as it was not possible to differentiate sugar use in drinks from sugar sprinkled on cereal or fruit consumed at the same time. For this project beverages were combined into 8 categories for further analysis: hot beverages (tea/coffee); milk; 100% fruit juice; caloric soft drinks (including sodas, juice drinks, sports and energy drinks); diet soft drinks; alcoholic drinks (including beer, cider, wine, spirits, alcopops); bottled water; and tap water. To investigate trends over the day, consumption occasions were aggregated into hourly intervals and into 5 periods, approximately corresponding to breakfast (<1000 hrs), lunch (>1000 to 1400 hrs), afternoon (>1400 to 1800 hrs), evening (1800 to 2200 hrs) and night (>2200 to 0200). Mean daily consumption, calculated over 7 days, was used to compare between individuals. TWI was compared with EFSA Dietary Reference Values (DRV) for Adequate Intake of Water (AI) for adult men and women of 2.5 L and 2.0 L, respectively 4 . The EFSA value for adequate intake is more conservative than other recommendations from WHO or IOM 5 . Nordic and German speaking countries take the approach that water intake is considered inadequate when it is less than 1 g per kilocalorie of energy requirement 4 . Therefore a combined classification (TWI-2), based on dual criteria, (i) the AI, and (ii) the ratio of water intake in grams to energy intake in kcal 4 was used to provide a more conservative estimate of the proportion of adults consuming low amounts of water (although with no implication of inadequate hydration status).

Univariate analyses were carried out for men and women separately because of known differences in water consumption and recommended intakes. Crude differences in TWI and beverage consumption between groups were assessed using T-tests with Bonferroni correction for multiple comparisons. Non-parametric tests were used for variables that were markedly non-normal. Chi-square tests were used for categorical variables. All tests were 2-tailed and P < 0.05 taken as indicating statistical significance.

Multivariate analyses included adjustment for sex, age, bodyweight, smoking, dieting, physical activity level and mis-reporting (a recognized problem in dietary surveys with the potential to affect results) 23 . We attempted to adjust for implausible energy reporting using individuals’ ratio of energy intake (EI) to basal metabolic rate (BMR) 24 and applied Goldberg cut-offs corresponding to confidence limits of plausible intake based on a 7 day diet record at a physical activity level (PAL) of 1.55 25 . Adults whose reported EI was <1.05 times their BMR were classified as under-reporters, while those with reported EI >2.28 times BMR were classified as over-reporters. Rather than exclude misreporters entirely, which can introduce bias, we included the classification variable in the analysis.

Analysis of covariance was used initially to examine how energy intake varied as the proportion of “liquid calories” increased (compared with food calories); covariates in the model were: sex, age, bodyweight, smoking, dieting, mis-reporting and physical activity. Multiple linear regression was used to estimate the effect on energy intake of varying the type of beverages consumed, whilst controlling for the effect of confounders. The effect of substituting non-caloric beverages by caloric beverages was estimated by including caloric beverages (as percentage of total beverage weight) as the main independent variable, with total beverage weight (g) held constant. This necessarily implies an equal and opposite change in other beverages. A further model included energy from food, thus disallowing compensation (reduction of food calories). Finally, we used the within-person daily consumption data to explore the effect of changes in consumption day-to-day, with each person acting as their own control. The independent variables in this analysis were all difference variables (deviation from the subject’s 7 day mean, deviation in total beverage consumption from subject mean) with deviation in energy intake as the outcome.

Survey respondents ranged from 19 to 64 years (mean 42 years), with slightly more women than men completing dietary records (Table 1). Around one third were current smokers and 25% of women (12% of men) were on a diet to lose weight. Obesity measures were representative of the UK adult population in 2000, with 60% overweight or obese.

Total water intake from all sources averaged 2.5 L and 2.0 L and men and women, respectively (Figure 1, Table 2). This concords with the estimated AI set by EFSA 4 . Beverages accounted for 75% of total water and 66% of the total weight of food and drink consumed. Mean beverage consumption was 1779 g/d (2012 g/d among men, 1593 g/d among women). Hot drinks (principally tea and coffee) were consumed by 97% of men and women, similarly for milk (95-96%) (Figure 2). Among men, alcoholic drinks were more popular than tap water (79% vs. 60% consumed, respectively), while 66% of women consumed alcohol and 73% drank tap water. More than half the sample drank caloric soft drinks at least once during the week, while less than half drank fruit juice, diet soft drinks or bottled water.

Tea and coffee accounted for over 40% of the total daily weight of beverages while milk provided an additional 12% (232 g/d for men, 202 g/d for women) (Table 2). Men consumed much higher quantities of alcoholic drinks (mean 501 vs. 141 g/d for women; P < 0.0001) and slightly more caloric soft drinks (mean 129 vs. 95 g/d, P = 0.001). By contrast, women consumed more tap water (243 vs. 187 g/d for men; P < 0.0001) and diet soft drinks (93 vs. 82 g/d, P = 0.002). Bottled water and fruit juice were consumed in small amounts by both sexes (< 60 g/d). Younger men and women consumed less hot beverages and milk, but significantly more soft drinks (both caloric and non-caloric) and alcohol than older adults (Table 3). Mean consumption of caloric soft drinks over the total sample was 110 g/d or 2 cans per week, but younger women and men (<35y) consumed 140 g/d and 216 g/d, respectively. Consumption of alcoholic drinks in these groups averaged 179 g/d and 579 g/d, respectively. Beverages contributed approximately 16% of total energy overall (men 17%, women 14%) and alcohol was a major contributor (9% of EI for men; 5% for women) (Table 2).

Results are based on two-sided tests assuming equal variances with significance level 0.05.

Tests are adjusted for all pairwise comparisons within a row using the Bonferroni correction.

TWI was very highly correlated with the weight of beverages and water from beverages (r 0.97) and more weakly correlated with food intake (r 0.29) (Table 4). Hot beverages, tap water and alcohol were most highly correlated with TWI (r > 0.4). Alcoholic drinks (r 0.17), caloric soft drinks (r 0.15) and milk (r 0.14) had the highest correlation with EI, while for tap water and bottled water the coefficients were essentially zero.

* Water from food and beverages.

Out of a maximum of 8 different types of beverages in our classification, the mean “variety score”, averaged over 7 days, was 3.3 for both men and women. Adults consuming fewer than 3 types had lower TWI and were less likely to meet the AI compared with those drinking 3 types or more (mean TWI 1.9 L vs. 2.4 L ; 73% vs. 43% below AI). Variety score was positively correlated with TWI (r 0.34, P < 0.0001; Table 4) and with EI (r 0.16, P < 0.0001), suggesting that beverage variety is an indicator of higher consumption of food and drink generally.

Total water intake and beverage consumption (g) was significantly higher on Fridays and Saturdays than on other days of the week, especially among men. This appeared to be attributable to higher consumption of alcoholic drinks at weekends, especially Saturday (Figure 3). Women showed weaker trends, but consumption on weekend days was still twice that on other days of the week. Saturday consumption of alcoholic drinks averaged 262 g among women and 949 g among men. Smaller differences in the opposite direction were observed for hot beverages and water. Consumption of soft drinks and fruit juice did not vary greatly by day of the week.

Beverage consumption peaked at 0800 hrs and 2100 hrs for both men and women. When total consumption of each type of beverage was plotted as a separate line (Figure 4), a peak of hot beverages and milk in the morning and of alcohol in the evening was particularly noticeable, the latter being more marked in men. Other beverages were more evenly spread throughout the day, with slightly higher consumption around lunchtime (1300 hrs) and evening (1900 to 2100 hrs). Additional file 1: Appendix shows the 24 h time trends of beverage consumption for each day of the week (“24/7”).

Older adults consumed more beverages in the morning, young adults in the evening (Table 5). For example, men aged 50y + consumed on average 30% of their beverages before 1000 hrs, while young men (19-35y) consumed less than 22%; older men consumed 30% after 1800 hrs compared to 36% among younger men. Similar trends (all significant at p < 0.05) were observed among women. Evening consumption appeared to be driving the higher total beverage consumption on Fridays and Saturdays. Beverages consumed between 0200 hrs and 1000 hrs contributed 25% of total daily consumption and adults who drank nothing before 1000 hrs only partly compensated for the deficit, having a 16% lower beverage intake overall (data not shown).

Adults with the highest TWI tended to consume most of their beverages in the evening. Figure 5 illustrates the correlation between timing of consumption (i.e. the proportion of beverage weight consumed in different time periods) and total water intake; the coefficients change from negative in the morning (r-0.24 for men, r-0.16 for women) to positive (r 0.22, r 0.17) for evening consumption. This effect is partly explained by the dominance of alcohol in evening, as (particularly for men) the correlation coefficients were attenuated when alcohol consuming days were excluded.

In conclusion, alcohol may be a major influence explaining the higher total water intake, higher evening consumption and higher weekend consumption of beverages observed in this study. On days alcohol was consumed, total water intake was 42% higher among men and 24% higher among women (data not shown).

Men and women with water intakes below the EFSA AI (2.5 L or 2.0 L) who also had a low water to energy ratio (<1.0) were classified as having low water intake in both an absolute and a relative sense (low TWI-2), whilst those who were above the cut offs for both definitions were classified as high TWI-2. The remaining adults were excluded from the analysis. On this basis approximately 23% of all women and 33% of all men hade low TWI-2, while 45% and 41%, of men and women had high TWI-2. Men with high TWI-2 had slightly higher EI than men with low TWI-2 (2470 vs. 2287 kcal/d), but this was due to a higher contribution from alcohol (330 vs. 99 kcal/d) (P < 0.0001)(data not shown). For women, mean EI was not significantly different between the TWI-2 categories (1753 vs. 1716 kcal/d; P = 0.32). On a proportional basis, women with low TWI-2 drank significantly less water and hot beverages, while men consumed less water and alcohol compared with the high TWI-2 adults. Adults with low TWI-2 had diets of higher energy density and this applied to foods as well as beverages.

There was no difference in total energy intake when beverage calories increased at the expense of food calories (Table 6); (i.e. “a calorie is a calorie”). This is one interpretation of the liquid calories hypothesis. However, a more relevant question for public health is to what extent replacing caloric beverages by non-caloric beverages would help reduce energy intake and (possibly) reduce obesity.

(estimated marginal means, adjusted for covariates: age Sex, current smoker, weight, physical activity, mis-reporting , dieting status).

To address this, our first models (Table 7) included 7 covariates (sex, age, bodyweight, smoking, dieting, physical activity and misreporting) and also total beverage intake, hence mathematically equivalent to substituting caloric beverages (sum of milk, fruit juice, caloric soft drinks and alcoholic beverages) for an equal weight of non-caloric beverages (hot beverages excluding milk, diet drinks, tap water and bottled water). The predicted effect of replacing 100 g caloric beverages with non-caloric drink was a reduction of 15 kcal (allowing food intake to vary) (Table 7, Model 1). When food energy intake was constrained as constant (i.e. disallowing compensation) the net impact of 100 g caloric beverages was estimated at 34 kcal (Table 7, Model 2). These estimates are based on the consumption patterns of other adults in the survey with similar patterns.

* 4caloric beverages = sum of milk, fruit juice, caloric soft drinks and alcoholic drinks.

The second analysis (Table 8) used a within-person change model to address whether a change in an individual’s beverage consumption habit on any day was associated with a change in their total energy intake (compared with their 7 day mean). Modelling each beverage separately, with total beverage weight held constant, each of the 4 non-caloric beverages was negatively associated with energy, while the 4 caloric beverages were positively associated with energy (Table 8, Model 3). Combining caloric beverages together, the final models (Table 8, Models 4 & 5) gave an estimated effect of 29 kcal per 100 g of caloric drinks substituted, or 35 kcal if food energy (kcal) was held constant. By comparison, the mean caloric content of the caloric beverages consumed (calculated from the database) was 47 kcal/100 g.

*Model 3 is a composite of 8 regressions, one for each beverage, adjusted for change in total weight of beverages; coefficients express the change in energy intake associated with a change in one beverage type and a corresponding reduction in others, i.e. constant total beverage intake.