|Home | About | Journals | Submit | Contact Us | Français|
Data currently available on drinking water intakes do not support dietary exposure estimates for contaminants that have acute effects lasting less than 24 h. Realistic exposure estimates for these types of contaminants in drinking water require detailed information on amounts and time of consumption for each drinking occasion during a day. A nationwide water consumption survey was conducted to address how often, when, and how much water is consumed at specific times during the day. The survey was conducted in two waves, to represent two seasons, and the survey instrument consisted of 7-day water consumption diaries. Data on total daily amounts consumed, number of drinking occasions per day, amounts consumed per drinking occasion, and intervals between drinking occasions show larger between-subjects variation than within-subject variation. Statistically significant associations were also observed between drinking water consumption patterns and participants’ ages and sex and geographical regions in which these participants live. The number of drinking occasions on a given day varied from 0 to 19, with the majority of respondents reporting 6 or less drinking occasions per day. The average interval between drinking occasions varied from 1 to 17 h, with 57% of the person-days reporting average intervals at least 3 h apart. The mean amount consumed per drinking occasion showed little association with the number of drinking occasions and fluctuated between 8 and 10 oz. To our knowledge, this survey is the only source of information on within-day patterns (i.e., when and how much) of drinking water consumption for a nationally representative sample of the US population. The detailed water consumption data from this survey can be used to support less than 24-h dietary exposure estimates for contaminants in drinking water.
Dietary exposure to contaminants in food and drinking water is defined as the product of the amount consumed times the residue present in the food and drinking water. Dietary exposure estimates to account for acute toxicity effects are based on the general assumption that they occur over a 24-h time period. Therefore, the total dietary exposure for a 24-h period is typically compared to the appropriate daily dose for the contaminant in question to estimate risk. In contrast, some contaminants are known to have acute effects that occur and are reversible in a much shorter time period (e.g., carbamate insecticides). Exposure assessments for these contaminants must reflect the reversible nature of the acute toxicity to provide realistic risk estimates.
The USDA Continuing Survey of Food Intake of Individuals (CSFII (USDA, 2000)) is the consumption database used by current pesticide dietary exposure models (e.g., DEEM™ (Exponent, 2008) and CARES® (ILSI, 2004)). This database was originally designed and conducted for use in the food and nutrition area. For foods, the survey recorded the time of day and the amount of food that was consumed. The time of day data allow one to use the data for shorter exposure time periods, even down to minutes. However, for drinking water (non-food based water), only the total amount of water consumed in the 24-h period was recorded in CSFII. Data for drinking water that reports time of day and amount consumed per drinking occasion did not, to our knowledge, exist at the time of the survey.
A nationwide water consumption survey was therefore conducted to address how often, when, and how much water is consumed during the day. The survey was conducted in two waves, to represent two seasons, and the survey instrument consisted of 7-day water consumption diaries. This database of detailed hourly water consumption can be used to supplement the CSFII water consumption database, or other data, to more accurately determine exposure to contaminants that have acute toxic effects shorter than 24 h. Some trends and associations observed in the data on amounts consumed, numbers of drinking occasions, and intervals between drinking occasions are explored. Participants’ ages, sex, and geographical regions as related to trends are examined. Finally, a comparison with CSFII water intakes is undertaken.
The Drinking Water Consumption Survey (DWCS) was funded by Bayer CropScience1 and conducted by the National Product Database (NPD) group (NPD, 2004), with guidance from Exponent Inc. Exponent also coordinated the survey, managed the data, and conducted the initial data analyses. The objective of the survey was to obtain a distribution of water consumption patterns for a 24-h time period, for a nationally representative sample of the US population. The NPD group was chosen to conduct this survey because of its experience in tracking the consumption habits of the US population since 1980 through its National Eating Trends (NET®) service2 (NET®, 2004).
The DWCS was conducted in two waves, the first in summer 2000, during the month of August, and the second in late winter/early spring 2001, mostly during the month of March. Two nationally representative samples (one for each wave) were extracted by NPD from a core sample of households from NPD’s Home Testing Institute (HTI) consumer panel. The sampling frame consisted of the list of households participating in the HTI consumer panel. At the time of the survey, the core sample of the HTI panel consisted of 250,000 pre-recruited households available for custom research. The HTI core sample is set up to provide a return sample that is geographically and demographically representative of the US population with respect to a number of variables, including household size, age, income, and census region. All members in the selected households were asked to participate in the survey. The sample for wave 1 included 3000 households randomly selected from the core sample of 250,000 households, whereas, in an effort to increase the number of children in the survey, the sample for wave 2 included 650 households, randomly selected from households with children less than 6 years of age, in addition to 3000 households randomly selected from the core sample3. The households were selected using stratified random sampling, with strata defined by household size, age, income, and region. All household members in the selected households were asked to fill out a 1-week diary in which they recorded all their drinking water occasions. The female head of the household (or male, if no female was present) coordinated the recording of all of the drinking water consumption for themselves and their household members and assisted younger members of the household. To increase participation in the survey, participants were offered a small gift incentive and a cash prize drawing4. There was no attempt to include the first wave households in the second wave.
Each household member received a four-page diary, where each page had recording space for 2 days. A separate cover page included the greeting and study instructions. To enhance reporting accuracy, a filled out sample diary was included in the cover letter and clip-art of various glasses and cups, and the corresponding sizes/ounces were displayed on the first page of the diaries (Figure 1).
The following information was collected in the diaries:
In addition, demographic data and information on the household composition and socioeconomic status were also available for each household and were linked to each participant’s diary data.
The raw data for each individual were provided to Exponent in ASCII format. The analytical results presented in this study were conducted using SPSS® V.7.5 for Windows, Microsoft® Excel 2000, STATATM 10.0, and SAS® for Windows Version 9.1.
Seven days of water intake data were collected for the majority of the participants in the DWCS. Participants in the DWCS recorded their water consumption amount in 2-ounce bins (e.g., 3–4 oz, 5–6 oz, etc.) with the exception of the lowest bin, which ranged from , and the largest bin, which was open-ended (25 + ounces). The upper limit of the recorded bin was used in this analysis (e.g. 3–4 oz = 4 oz). Fifty ounces was used for the maximum intake bin. These data treatments are likely to result in an overestimation of the actual amount of water consumed, but it should not affect the within-day intake trends or differences observed between seasons, regions, and age and sex groups. The total daily water intakes for each survey person-day were obtained by adding the amounts reportedly consumed during the day. An estimate of the average time between drinking occasions was derived for each person-day by calculating the time between consecutive drinking occasions and by averaging these times for each person-day. The last drinking occasion specified on the diary forms provided to the survey participants, “Midnight to 6:00 am,” was not time-specific and was excluded from this calculation. An estimate of the average amount consumed per drinking occasion was estimated for each person-day by averaging the amounts consumed per drinking occasion for that person-day.
Individuals within the same household were treated as independent observations. Although individuals living in the same household are likely to show some association in their food consumption patterns because they could be sharing one or more meals per day and hence potentially eating the same foods at these meals, their water consumption patterns are likely to be less correlated, as drinking water is expected to be available every day.
Demographic characteristics of participants in the DWCS were summarized and compared with the demographic characteristics of the US population. The demographic variables considered in this study were age and sex, region, and race. Combined age and sex groups were created for this analysis as follows: (1) <2 years male/female; (2) 3–5 years male/female; (3) 6–12 years male/female; (4) 13–19 years male; (5) 13–19 years female; (6) 20–49 years male; (7) 20–49 years female; (8) >50 years male and (9) >50 years female. No formal statistical tests were conducted in these comparisons. Other variables available from the survey, such as education, income, occupation of the household head, and source of drinking water, were not examined in this paper.
Total daily and within-day water intake patterns were also analyzed for DWCS participants. Specifically, the total amount of water consumed per day, the number of drinking occasions per day, the average amount consumed per drinking occasion, and the average interval between drinking occasions were summarized for each survey wave. A random diary day was selected for each participant and analysis of variance was used to evaluate the difference in water consumption by season (wave), region, and age and sex groups for the following variables: the total amount consumed within a day, the average amount consumed per event within a day, and the average interval between consumption events, whereas Poisson regression was used to assess differences in the number of drinking occasions per day.
To assess inter- and intraindividual variability for all participants and all reported consumption days, a repeated measures analysis of variance was conducted on total daily amount consumed, average amount consumed per drinking occasion, and average interval between drinking occasions, whereas a mixed effect Poisson regression was conducted on the number of drinking occasions per day. The analysis was repeated adding an indicator variable representing weekend days.
Questionnaires and diaries were mailed to 3000 households in the first wave and 3650 households in the second wave. One thousand nine hundred ninety-two participants in 994 households (33% response rate) completed the first wave of the survey, and 2950 participants in 1320 households (36% response rate) completed the second wave of the survey.
Some diaries were filled out incorrectly, with duplicate or missing person IDs within a household, or with multiple entries per time interval. The treatment of these diaries depended on whether it was possible to track down the demographic information on the subjects with duplicate or missing IDs and on the number of entries per time interval. If it was not possible to correctly identify the age of the participants from the demographic data fields, the diaries were discarded. Also, if it was not possible to identify the sex of participants’ ages 13 years or more, their diaries were discarded. Further, if participants had diaries with more than 3 entries per time interval, they were discarded, whereas participants who had diaries with 2 or 3 entries per time interval had their diaries “corrected” by deleting the lowest amount(s) entered in these intervals. Specifically, in wave 1, there were 544 (0.72% of all diaries) diaries with multiple entries per time interval, but 153 of these were for 8 individuals from the same household where we deemed these entries as unreliable5. The data from this entire household were dropped. Of the remaining 391 diaries with multiple entries per time interval, one diary had 9 entries, one had 8 entries, and one had 7 entries. These 3 diaries were dropped. The remaining (16 with 3 entries and 372 with 2 entries) diaries were recoded as described above. In wave 2, there were 328 (or 0.48% of all diaries) diaries with multiple entries per time interval. Seven of these had 4 entries and were excluded. The remaining diaries, 23 with 3 entries and 298 with 2 entries were recoded as described above. The resulting data file contained data from 4198 individuals (1740 from wave 1 and 2458 from wave 2) from 2154 households (921 from wave 1 and 1233 from wave 2). Eighty-two percent of these participants completed 7-day diaries, and less than 10% had fewer than 5-day diaries, resulting in an average of 6.5 diary days per study participant. The total number of DWCS person-days in the data file analyzed in this paper was 27,282 (11,347 for wave 1 and 15,935 for wave 2), that is, 93% of the total number of diaries that would have been expected if the survey participants had returned all 7 diaries. It is not clear if the missing diaries were due to participants not filling out the diary information on days on which they did not consume water, or due to non-response.
The age and sex distribution of the survey participants is summarized in Table 1. Approximately 16% of all survey participants were children less than 13 years old, 6.7% were teenagers, and 34.5% were adults 50 years of age or older. The proportion of older females (>50 years) among survey participants was higher than that in the US population (based on the 2000 Census), but the proportion of participants in the other age and sex groups is comparable to the age and sex group distribution of the US population. The regional distribution of the survey participants is similar to that of the US population, whereas the racial breakdown of survey participants indicates that a larger proportion of the survey respondents were white as compared to the corresponding proportion in the US population.
Figures 2--55 compare the mean total daily water intakes, number of drinking occasions, average interval between drinking occasions, and average amount consumed per drinking occasion derived from the total number of person-days available for analysis to the sub-sample formed by selecting a random survey day for each person. The two sets of distributions were comparable.
The distribution of total daily water intakes on the randomly selected day ranged from 0 to 396 oz and exhibited a skewed pattern, with more than 95% of the observations below 100 oz. Higher intakes were reported in the summer season, that is, the first survey wave, as compared to the winter season, that is, the second survey wave for all age and sex groups, except for females aged 13–18 where the reverse was observed (Table 2). The ANOVA was conducted on the log-transformed data and showed that region was not associated with total water intake, whereas both season and age and sex groups were significant predictors of total amount consumed (P-values <0.0001).
The number of drinking occasions on the randomly selected diary day varied from 0 to19, with the majority (>82%) of respondents reporting 6 or less drinking occasions (Figure 6). The drinking occasions were spread throughout the day, with a small peak observed in the 2400 hours to 1300 hours interval and a decline in the number of drinking occasions after 2200 hours (Figure 7). There were slightly more drinking occasions being reported in the summer wave than in the winter wave (Table 3). Season and age and sex group were significant predictors of the number of drinking occasions per day (Poisson regression, P-values = 0.0004 (season) and <0.0001 (age and sex groups)).
The mean of the distributions of average amount consumed per drinking occasion on the randomly selected day was higher in the summer season than in the winter season (Table 4). The ANOVA conducted on the log-transformed data showed that season and age and sex groups were significant predictors of the average amount consumed per drinking occasion (P-values <0.0001), whereas region was not a significant predictor.
Average times between drinking occasions varied from 1 to 17 h, with 57% of the randomly selected diary days reporting average intervals at least 3 h apart, and were slightly longer for the winter season than for the summer season (Table 5). The ANOVA conducted on the log-transformed data showed that both season and age and sex groups are significant predictors of time between drinking occasions (P-values = 0.0006 (season) and 0.0066 (age and sex groups), respectively), whereas region was not a significant predictor.
Higher total daily intakes were associated with higher numbers of drinking occasions (Figure 8). The mean of the distributions of average amounts consumed per drinking occasion fluctuated around 8–10 oz for person-days with 12 or less drinking occasions per day and showed a small decrease for person-days with more than 13 drinking occasions per day (Figure 9). An increase was observed for person-days with 18 and 19 drinking occasions per day, however, this could be due to the presence of a few large values among the relatively small number of respondents reporting 18 or 19 drinking occasions per day.
Inter- and intraindividual variability was assessed using repeated measures ANOVA and mixed effects Poisson regression within strata defined by wave and age and sex groups. The ANOVA models could not be run on the entire database owing to limitations of the statistical software used in the analyses and hence were run within these strata, because the analyses conducted on the randomly selected survey days had showed significant wave, and age and sex differences. Specifically, for each drinking water consumption variable (total daily amount consumed, number of drinking occasions, average interval between drinking occasions, and average amount consumed per drinking occasion), 18 analyses (2 waves × 9 age and sex groups within wave) were conducted.
For total daily amounts consumed, significant day differences (i.e., within subject differences) were observed for females 13–19 (P-value = 0.0003), females 20–49 (P-value = 0.0038), females 50 + (P-value = 0.0076), and males 20–49 (P-value = 0.0034) in wave 1, and for all age and sex groups except children <2, children 3–5, and males 13–19 in wave 2 (P-values ranging from <0.0001 to 0.0256). For the average amount consumed per drinking occasion significant day differences (i.e., within subject differences) were observed for females 13–19 (P-value = 0.0461) and females 50 + (P-value = 0.0436) in wave 1, and for children 6–12 (P-value <0.0001) in wave 2. For the average interval between drinking occasions, significant day differences (i.e., within subject differences) were observed for females 20–49 (P-value = 0.0003), females 50 + (P-value <0.0001), and males 20–49 (P-value = 0.0115) in wave 1, and for children 6–12 (P-value = 0.0076), females 20–49 (P-value <0.0001), females 50 + (P-value <0.0001), and males 20–49 (P-value = 0.0290) in wave 2. Finally, the mixed effects Poisson regression models showed significant survey day differences for females aged 20–49 (P-value = 0.012), females 50 + (P-value = 0.040), and males 20-49 (P-value = 0.011) in wave 1, and for all age and sex groups except children <2, children 3–5, and males 50 + in wave 2 (P-values ranging from <0.001 to 0.019).
The analyses including an indicator for weekend days generally showed the same significant survey day effects. In addition, there was a significant weekend effect for several age and sex groups with lower total intake amounts and fewer numbers of drinking occasions being reported on Sundays.
Witschi (1990) notes that the cooperation of subjects and validity of the records in dietary record surveys decline in relation to the recording process. The response rate of the DWCS survey (33% for wave 1 and 36% for wave 2) is slightly higher than the 25% response rate that was anticipated by NPD and is similar to what one expects from mail surveys involving a relatively heavy participant burden such as that associated with filling out the 7-day diaries. For instance, the response rates in 7-day diary surveys included in the Multinational Time Use Study (Institute for Social and Economic Research, 2004) varied between 20% and 79%. It is possible that the survey could have resulted in a higher response rate if it had focused on the 2000 households covered by NPD’s NET survey, rather than the sample of households that were randomly selected from the HTI consumer panel specifically for the survey, as members of the NET panel are used to responding to lengthy questionnaires. However, NPD’s standard approach to custom surveys is to address the survey to a random subsample of their HTI consumer panel.
Diaries with 4 or more entries per time interval were discarded and those with 2 or 3 entries per time interval were recoded by excluding the lower amounts. The total number of diaries with multiple entries per time interval was less than 1% (0.72% of wave 1 diaries and 0.48% of wave 2 diaries). It was not possible to determine (1) if the multiple entries per time interval really meant that the participants drank the various amounts during the time interval and hence should have instead recorded the total of these amounts, or (2) if the multiple amounts belonged to different time intervals and were entered incorrectly in the same interval, or (3) if the multiple amounts were errors that should have been deleted. Thus, our treatment of the data would have resulted in underestimating the total daily amounts (if (1) or (2) are correct), and would underestimate the interval between drinking occasions (if 2 is correct). However, given the very small fraction of diaries involved, it is unlikely that our data treatment would have had a significant impact on our results.
No drinking water consumption was reported for 90 (0.3%) of the person-days6 in the DWCS. This percentage is lower than that reported in USDA’s 1994–96 & 1998 CSFII (17.5%). However, as pointed out earlier, about 17% of the DWCS survey respondents did not have complete 7-day diaries, and it is not clear if the missing diaries are due to participants simply not filling out the diary information on the days they did not consume water.
The mean daily intake estimates in the DWCS are higher than the estimated intakes derived from USDA’s 1994–1996 and 1998 CSFII for both seasons among consumers of water (Table 6). This comparison was limited to consumers only owing to the difference in the percent of participants in each survey that reported no drinking water consumption. This difference may be due, in part, to the fact that the upper limit of the recorded bin was used in the analysis of the DWCS data, or to the difference in the two survey instruments. Specifically, whereas the CSFII survey collected that information through the question: “How many fluid ounces of plain drinking water did you drink yesterday?,” the DWCS provided participants with a time grid to report their water consumption, thus potentially helping them to remember all their water consumption occasions, in contrast to the CSFII general 24-h total consumption recall question.
Regional and age and sex intake patterns were comparable for the two surveys. Specifically, higher total daily intakes were observed in both surveys for the West region as compared to the total daily intakes in the Northeast region (Table 6). Both surveys showed similar patterns with intakes increasing with age until age group 20–49 years. The CSFII data show a flattening or a small decline for the older age and sex group (Figure 10).
As the CSFII only collected total daily water intakes, it is not possible to compare the within-day patterns observed in the DWCS to CSFII within-day patterns.
Current model capabilities for dietary risk assessments (for food and drinking water) use a 24-h time period for conducting exposure assessments for contaminants with acute toxicity effects. To support models and methods for assessing contaminants with acute effects that last less than 24 h, data are needed on amounts consumed and time of day for each eating occasion. Food consumption information collected in the CSFII, which is the basis for current aggregate and cumulative pesticide risk assessment models, includes not only the amounts consumed at each eating occasion but also the time of consumption, whereas plain drinking water consumption information in the CSFII is available only on a total 24-h basis. It may be possible, using the information collected by the DWCS, to “allocate” the total daily water consumption amount reported in the CSFII into various drinking occasions. Specifically, if each subject in the CSFII survey was randomly matched to subjects in the DWCS on the basis of survey season, region, age, sex, and total amount of drinking water consumed per day, then the total amount reported by that CSFII participant could be allocated to the same number of drinking occasions as those reported by the matching DWCS participant. Similarly, the proportion of the total daily water consumption allocated to each of these drinking occasions can be assumed to be similar to that reported by the matching DWCS participant. This approach would then allow a less than 24-h assessment of both food and drinking water (aggregate assessment) for a pesticide.
In an effort to ease the reporting burden on survey participants, the DWCS used a “grid” format in the questionnaire, with pre-determined time intervals and pre-determined amounts. The use of hourly time increments and 2-oz amount increments to create the grid results in an uncertainty regarding the exact amount consumed and the exact consumption time, and some assumptions may be necessary when using the data from the DWCS to supplement the CSFII survey data. For instance, should the upper limit of the consumption bin be used to represent the amount consumed, or should the midpoint, the lower limit of the bin, or a random value between the lower and upper limit be used? Similar questions apply with respect to the 1-h time bins used in the grids.
To our knowledge, this survey is the only source of information on within-day patterns (i.e., when and how much) of drinking water consumption for a nationally representative sample of the US population at this time. NHANES surveys after 2004 are reported to be collecting time and amount of drinking water consumed along with the food consumption (Moshfegh and Goldman, 2007). These data are essential to the pesticide industry and government branches, such as EPA OPP and OPPTS, who want to be able to conduct realistic acute aggregate assessments (an FQPA requirement) for rapidly reversible pesticides (i.e. with less than 24-h receptor reversibility). By examining drinking occasions per day, amount consumed per drinking occasion, and intervals between drinking occasions, as detailed in this paper, the stated goals for risk assessment can be met when incorporated into available aggregate and cumulative models. This is one specific use for the data generated by this drinking water survey. However, there are many additional variables to be exploited in this data set. Other information in the survey includes source of drinking water (community water system versus well water), patterns of consumption for weekdays versus weekends, whether water is typically consumed with or without food, where the water was consumed, and whether the water consumed on a given drinking occasion is bottled or tap water. Many other industries or risk assessment areas such as the bottled water industry or drinking water contaminant assessors may benefit from exploring these data.
These analyses were supported, in part, by Bayer CropScience.
1For access to the data from the survey, please contact DWCS Baver CropScience, Product Safety Management, P.O. Box 12014, RTP NC 27709-2014.
2The NET maintains a diary panel of 2000 households to track the consumption habits of the US population. For a 2-week period, panel households complete a daily diary, recording consumption information for all members in the household.
3We have no information from NPD relating to the potential overlap between the sample of households drawn from the HTI panel for our survey and the 2000 households included in the NET panel.
4All study participants who returned a completed diary were eligible for a drawing to win one of twenty-five $100 cash prizes
5For seven subjects, all the amount entries in most or all time intervals were filled, and for one subject the entries were recorded in a “diagonal” pattern, with amounts increasing by one category at each time interval.
6Contributed by 40 participants in the DWCS.