SAB Report on EPA's Per Capita Water Estimates for the United States


      United States        Science Advisory       EPA-SAB-EC-00-003
      Environmental       Board (1400A)          December 1999
      Protection Agency   	Washington DC	www.epa.gov/sab

vvEPA AN SAB REPORT ON

      EPA's Per Capita Water
      Ingestion Estimates for the

      United States
       A Review by the Executive
       Committees' Drinking Water
       Ingestion Subcommittee

-------
                                    ABSTRACT

       The Drinking Water Intake Subcommittee (DWIS) of the Science Advisory Board's
(SAB) Executive Committee reviewed a report on the Estimated Per Capita Water Consumption
in the United States. The document presents estimates of drinking water ingestion for the total
U.S. population and a number of subgroups of interest. Estimates are given for many age,
gender, and other descriptors. The Subcommittee was pleased with the report's use of a
substantial existing data base to improve upon the current EPA estimates for drinking water
ingestion.  The current Report is largely descriptive and contains little discussion of factors
embedded within the original survey and the Agency's analytical method for deriving estimates
that inform the reader of important factors that should guide use of the estimates. The
Subcommittee noted its desire to see a greater level of discussion on these elements so that
unintended misuse of the data can be minimized.

Keywords: Drinking water ingestion, exposure factors, drinking water consumption, drinking
water intake.

-------
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
OFFICE OF THE ADMINISTRATOR
SCIENCE ADVISORY BOARD

December 20, 1999

EPA-SAB-EC-00-003

The Honorable Carol Browner
Administrator
United States Environmental Protection Agency
401 M Street, SW
Washington, DC 20460

Subject: An SAB Report on EPA's Per Capita Water Ingestion in the
United States

Dear Ms. Browner:

This report presents the results of a review by the Drinking Water Intake Subcommittee,
a special subcommittee established by the EPA Science Advisory Board (SAB) Executive
Committee in response to a request from the Agency's Office of Water to review its report
entitled Estimated Per Capita Water Consumption in the United States. The review was carried
out during two meetings with representatives of the Agency during July 1999. The
Subcommittee concluded that the EPA report will be an important reference with extensive
utility both inside and outside the Environmental Protection Agency. In addition, the
Subcommittee believes that the U.S. Department of Agriculture's (USDA) Continuing Survey of
Food Intake by Individuals (CSFII) was the best available information source for the Agency to
use in developing its estimates of drinking water ingestion by the U. S. population. The CSFII
also provides a valid data set for estimating water ingestion for a limited number of subgroups
within the population. However, the Subcommittee has concerns about the descriptive nature of
the EPA report because it contains no explicit discussion of how these estimates might
reasonably be used by the Agency in its scientific assessment and policy considerations.

-------
       Even though the report will be invaluable in providing information about the distribution
of water consumption among the general population, it may be of limited value in providing
information about the drinking water consumption of certain subpopulations that may be of
interest to the Agency and to other users of the report.  However, this limitation is due to the
characteristics of the CSFII survey and not because of the Agency's analysis and interpretation of
the data. The CSFII survey was aimed at characterizing the food intake of the general
population and was not designed to gather information on specific subgroups or situations (e.g.,
very young children, Native Americans, individuals with diseases which impact their water
consumption or workers in hot environments). As a result, although specific groups of interest
are represented in the survey in proportion to their occurrence in the general population, the
information needed to identify them may not be present and, even when it is, the sample sizes in
the subgroups that can be identified are often too small to provide useful information on their
water intake (e.g., for young children in certain ethnic  or socioeconomic groups). Further,  even
though Native Americans are represented in the survey, the information gathered in CSFII  does
not allow one to differentiate which of the Native Americans who were included in the survey
follow traditional Native American culture and lifestyle and which of them practice
contemporary urban and suburban lifestyles.

       Several approaches are possible if the Agency finds that it needs information on the
distribution of water intake in subgroups, or for situations that are not adequately described by
CSFII. One is to commission special surveys designed to gather the needed information about
these groups.  A second approach would be to rely on current understanding of the physiological
need for water by individuals in different situations (e.g., developmental stages, physiological
states, or environments) to characterize the likely water consumption and then to couple this
information with survey information on the distribution of these developmental stages,
physiological states, and environments in the population.  Each approach has its strengths and
weaknesses.

        The draft report could be considerably strengthened, and the potential for
misinterpretation of its findings  could be reduced substantially, if the Agency provided
information on the statistical significance of differences in water consumption between major
subgroups of the population. Without such information, users of the report may be inclined to
emphasize the differences in water consumption among subgroups which may in fact be artifacts
of small sample sizes.

-------
      The SAB is prepared to provide additional review and assistance as EPA further develops
these estimates.  We look forward to the response to these comments from the Assistant
Administrator for the Office of Water.

                                      Sincerely,
                               Dr. Joan M. Daisey, Chair
                               Science Advisory Board
Dr. Henry Anderson, Cochairman                Dr. Richard Bull, Cochairman
Drinking Water Intake Subcommittee             Drinking Water Intake Subcommittee
Science Advisory Board                         Science Advisory Board

-------
                                      NOTICE
       This report has been written as part of the activities of the Science Advisory Board, a
public advisory group providing extramural scientific information and advice to the
Administrator and other officials of the Environmental Protection Agency.  The Board is
structured to provide balanced, expert assessment of scientific matters related to problems facing
the Agency.  This report has not been reviewed for approval by the Agency and, hence, the
contents of this report do not necessarily represent the views and policies of the Environmental
Protection Agency, nor of other agencies in the Executive Branch of the Federal government, nor
does mention of trade names or commercial products constitute a recommendation for use.
Distribution and Availability: This Science Advisory Board report is provided to the EPA
Administrator, senior Agency management, appropriate program staff, interested members of the
public, and is posted on the SAB website (www.epa.gov/sab').  Information on its availability is
also provided in the SAB's monthly newsletter (Happenings at the Science Advisory Board).
Additional copies and further information are available from the SAB staff.

-------

                        U.S. Environmental Protection Agency
                               Science Advisory Board
                        Drinking Water Intake Subcommittee
          Panel for Review of the EPA Report on Drinking Water Consumption
                               July 8 and 19-20,1999

Co-Chairs
Dr. Henry A. Anderson, Wisconsin Bureau of Public Health, Madison, WI

Dr. Richard Bull, Battelle Pacific Northwest Laboratories, Richland, WA

Panelists
Dr. Judy Bean, Children's Hospital Medical Center, Cincinnati, OH

Dr. Cynthia Bearer, Case Western Reserve University, Cleveland, OH

Dr. John Evans, Harvard School of Public Health, Boston, MA

Dr. Anna Fan-Cheuk, California Environmental Protection Agency, Oakland, CA

Dr. Richard Gilbert, Battelle Washington Office, Washington, DC

Dr. Barbara L. Harper, Yakama Indian Nation, Richland, WA

Dr. Michael Jayjock,  Rohm and Haas Co., Spring House, PA

Dr. Kai-Shen Liu, California Department of Health Services, Berkeley, CA

Dr. Edo Pellizzari, Research Triangle Institute, Research Triangle Park, NC

Dr. Barbara Petersen, President, Novigen Sciences, Inc., Washington, D.C.

Science Advisory Board Staff
Mr. Thomas O. Miller,  Designated Federal Official, U.S. Environmental Protection Agency,
       Washington, DC

Ms. Dorothy Clark, Management Assistant, U.S. Environmental Protection Agency,
       Washington, DC  20460

-------
TABLE OF CONTENTS
1. EXECUTIVE SUMMARY AND CONCLUSIONS 1
1.1. EPA's goals and objectives as stated for the Report, and the analyses it contains,
are too limited 1
1.2. The report should state that EPA did not have information that would allow
calculation of confidence intervals for subpopulations 3
1.3. The Agency should develop a strategy for the analysis, presentation and interpretation
of the Report's data that is consistent with the intended uses of the data 3

2. INTRODUCTION AND CHARGE 4

3. SUBCOMMITTEE COMMENTS 6
3.1. General Comments 6
3.2. EPA's goals and objectives as stated for this Report, and the analyses it contains, are
too limited 7
3.3. The EPA Report should state that EPA did not have information that would allow calculation
of confidence intervals for subpopulations 10
3.4. The Agency needs to develop a strategy for the analysis, presentation, and interpretation
of the data that is consistent with the intended uses of the data 10

Appendix A: Responses to Specific Agency Charge Questions A-l

Charge Question 1. Statistical methodology A-l
Charge Question 2. Confidence limits A-l
Charge Question 3. Short-term data and long-term estimates A-l
Charge Question 4. Data conventions A-2
Charge Question 5. Subpopulation distributions A-3
Charge Question 6. Subpopulations included A-4
Charge Question 7. Indirect water A-5
Charge Question 8. Food types not covered A-6
Charge Question 9. Intrinsic water A-6
Charge Question 10. Other issues A-7
Charge Question 11. Zero values A-8
References
R-l
nt

-------
1. EXECUTIVE SUMMARY AND CONCLUSIONS

The U. S. Environmental Protection Agency's (EPA) Science Advisory Board (SAB) was
asked to perform a peer review of the Agency report Estimated Per Capita Water Consumption
in the United States (hereafter referred to as the Report). The SAB Executive Committee
established the Drinking Water Intake Subcommittee (DWIS) to conduct this review. The DWIS
reviewed the Report during two meetings: one, a telephone conference meeting, on July 8, 1999
and the other, a face-to-face meeting, on July 19 to 20, 1999. Major Subcommittee comments
on the EPA Report are contained below in this SAB report. Specific responses to the 11 charge
questions are provided in Appendix A to this SAB report.

EPA is commended for seeking out databases that can be used for estimating ingestion of
drinking water on a national scale. The database selected as the analytic basis for the report, the
U. S. Department of Agriculture's Continuing Survey of Food Intake by Individuals (hereafter,
the CSFH or the Survey), is the best available and has critical attributes that allow advancement
of our understanding of ingestion of water by the general population of the United States. The
Agency's efforts to develop ingestion estimates from the CSFII survey data were significant.

The committee believes that the EPA Report will be an important reference resource with
extensive utility both within and outside the Agency. However, the following issues must be
considered if the Report is to achieve its full potential.

1.1 EPA's goals and objectives as stated for this Report, and the analyses it contains,
were too limited.

The Subcommittee is concerned that the Report is only descriptive and that it does not
explicitly discuss how the estimates might be reasonably used. The Agency has both scientific
and policy reasons for estimating water ingestion for the overall population, and for
subpopulations, that are not discussed in the report. Some of these respond to the statutory
mandate in the Safe Drinking Water Act. But there are other needs for information on drinking
water intake in risk assessment and regulation which involve establishing default values for
water ingestion, estimation of risks to highly exposed and/or sensitive subpopulations, and
characterization of the distribution of individual risks or the impacts of specific control
strategies. Important implications to these uses are not discussed in the current report.

EPA often uses default values for water ingestion levels when it develops allowable
concentrations for contaminants in drinking water. The Subcommittee is encouraged that this
EPA report provides information that will permit analysts to use specific data for water ingestion
in many future situations where allowable concentrations must be developed. For others, the
Report will provide assistance for developing information on the distribution of drinking water
ingestion by individuals that includes new information and the relationship of ingestion to
factors such as age, gender, and disease status.

-------
While the report does a good job of characterizing the distribution of drinking water
consumption in the entire US population, and in the major subdivisions of the US population
(i.e., by age, sex, race, and geographic region), it (iocs not provide the information that some
users may want on drinking water ingestion by smaller subpopulations. Further, certain groups
may have higher than normal water ingestion levels or they may be more sensitive to the effects
of contaminants in drinking water. Examples of these would include very young children and
workers in hot and/or dry climates.

This limitation exists because the CSFII data upon which the Agency relied for
generating its estimates were collected in an effort to characterize the patterns of food
consumption in the general population. They did not target certain subgroups that are now of
heightened interest to EPA. Therefore the samples in certain subgroups are so small that the
CSFII estimates of water consumption in these groups may be quite imprecise. Compounding
this problem is the Report's omission of statistical confidence intervals for most of the ingestion
estimates among subgroups of the population. If legislative mandates or regulatory analysis
require information on the water consumption of these subgroups, further studies will be needed.

Many of the results presented in the report may be sensitive to assumptions made during
data analysis. Examples of such data analysis conventions include the choice of regional
boundaries and the assignment of a principal source for ingested water. Currently the report
does not include a section analyzing the sensitivity of key results to these assumptions. The
Subcommittee urges the Agency to conduct a sensitivity analysis and to add a section to the
report describing the key findings from the sensitivity analysis.

Another key issue influencing the interpretation of the CSFII data is the choice of
averaging time. We know that in many other settings (e.g., air pollution exposure assessment)
heterogeneity tends to decrease as averaging time increases. The exact nature of the relationship
between averaging time and observed heterogeneity depends on the features of the data being
explored. For certain purposes (e.g., cancer risk assessment) the population distributions of long
term average exposures may be of interest. The current EPA report provides information about
drinking water intake averaged over TWO days. Therefore, to minimize the potential for misuse
of the data in the EPA Report, users might benefit if the Agency clearly stated the averaging
time on all tables and graphs in the report. Further, it may be necessary to more fully explore the
sensitivity of results to alternative choices of averaging time.

Therefore, the Subcommittee recommends that the EPA Report discuss the
characteristics of the EPA methods for estimating ingestion, and the USDA method for
conducting the CSFII, that have important implications for those who must use the ingestion
data.

-------
1.2 The EPA Report should state that EPA did not have information that would allow
calculation of confidence intervals for sub-populations.

A discussion point for the subcommittee centered on the question of whether it was
appropriate to provide data without meaningful confidence intervals. The design of the CSFII
survey requires use of an ultimate cluster methodology which is an aggregate of sampled persons
within each primary sampling unit. Smaller subpopulations within the sample (e.g., the less than
one year olds) did not meet these criteria. This prevents the calculation of confidence intervals
using the ultimate cluster methodology. It would be good to clarify this point for the readers of
the report.

13 The Agency should develop a strategy for the analysis, presentation and
interpretation of the Report's data that is consistent with the intended uses of the
data.

The Agency has taken a purely descriptive approach to the analysis and presentation of
data. This results in numerous tables containing drinking water ingestion estimates for many
conceivable combinations of attributes examined (e.g., Native American males by age group and
by geographic region, etc.). While this superficially exhaustive presentation of data may seem
attractive, the Subcommittee is concerned that this strategy for analysis, interpretation, and
presentation of the data is inadequate and potentially misleading. We urge the Agency to
develop a strategy for data analysis which, at a minimum, provides only those estimates of
drinking water intake for which estimates of uncertainty can also be developed, and preferably
which includes formal hypothesis tests of the significance of differences in the water
consumption of various groups. Further, the number of tables presented in the report should be
substantially reduced and limited to only those which support Agency needs and for which valid
estimates of precision can be provided. If the Agency feels that certain tables for which valid
estimates of precision can not be produced are necessary, this fact should be prominently
displayed on each such table.

-------
2. INTRODUCTION AND CHARGE

The Drinking Water Intake Subcommittee was asked to conduct a peer review of the
Agency Report that provides estimates of per capita water intake in the United States. The
Report contains estimates of the amount of direct and indirect water consumption. Direct water
consumption is defined as plain water consumed directly as a beverage. Indirect water is that
water added to foods and beverages during final home or restaurant preparation.

Empirical distributions of estimated water consumption were generated by water source
and by the respondent's demographic and physical characteristics. Water sources include: a) the
community water supply, b) bottled water, c) other sources including the respondent's own well,
rain cistern, spring, or public spring. Physical and demographics characteristics include: age,
gender, race, socioeconomic status, and geographic region. Estimates were also generated
separately for pregnant and lactating women.

The distributions of estimated water ingestion include point estimates of the mean and
the following percentiles: 1st, 5th, 10th, 25th, 50th, 75th, 90th, 95th, and 99th. Confidence
intervals for the mean and bootstrap intervals for the upper percentiles are provided for only the
larger subpopulations.

The charge to the Drinking Water Intake Subcommittee from the Office of Science and
Technology, US EPA Office of Water included the following questions:

a) The distributions of estimated water intake were generated using standard
statistical methodology for surveys with complex designs such as the 1994-96
CSFII. Is the statistical methodology used to generate the estimates appropriate?
Should we consider rounding?

b) We have limited the calculation of confidence intervals about the mean and boot
strap intervals for percentiles to the distributions for the larger sub-populations.
The complex sample design makes the calculation and interpretation of results for
smaller sub-populations virtually impossible to calculate and interpret. Is this an
appropriate decision?

c) The CSFII survey is based on short-term survey data. Upper percentile estimates
may differ for short-term and long-term data because short-term survey data tends
to be inherently more variable. Is it appropriate to report upper percentile
estimates such as the 99th percentile?

d) Are the data conventions used to identify direct and indirect water appropriate?

e) Do the data support estimates of sub-population distributions?

f) We have provided distributions of estimated water intake for numerous sub-

-------
populations. Should any additional sub-populations be added? Should any be
excluded? Specify sub-populations.

g) USDA has identified two types of indirect water in foods. They are:
i. The amount of water in food as consumed.
ii. The amount of water used to prepare food.
The water intake report provides estimates of the amount of indirect water in food
as consumed. If resources permit, we could expand our report as a future
addendum to include estimates of the amount of indirect water used to prepare
food. Would this be desirable?

h) Additional water intake estimates associated with types of food may be useful for
specific risk-exposure analyses, e.g., cold beverage intake. Such analyses are
feasible using the CSFII data. We could expand our report as a future addendum
if resources permit. Are any such targeted analyses of significant interest at this
time?

i) Intrinsic water is the water contained in foods and beverages at the time of market
purchase. Intrinsic water includes commercial water (added to food products by
food manufacturers) and biological water (found naturally in foods). Intrinsic
water is not included in our current analysis. If resources permit, we could
expand our report as a future addendum to includes estimates of intrinsic water.
Would this be desirable?

j) What are the scientific limitations to the use of the water consumption estimates
provided in this report (i.e., what other issues has the Subcommittee noted with
the estimates that are not covered elsewhere)?

k) The water intake estimates provided in this report are based on all respondents,
including those who did not report consuming water during the two survey days.
If resources permit, we could also generate estimates of water consumption which
exclude the zero consumers of water. We noticed that for some sub-populations,
especially the less than one-year-old infants, a substantial proportion consumed
zero or minimal amounts of tap water per day (presumably those who were breast
fed or drank undiluted formula or milk); these zero consumers of water can
contribute to lower estimates. Would this be desirable?

-------
3. SUBCOMMITTEE COMMENTS

3.1 General Comments

The Drinking Water Intake Subcommittee (DWIS) of the EPA Science Advisory Board
(SAB) has reviewed the Agency's report entitled Estimated Per Capita Water Consumption in
the United States during two meetings: one on July 8,1999 and the other on July 19 to 20,1999.
Specific responses to the Agency's charge questions are provided in Appendix A to this SAB
report.

EPA used the Continuing Survey of Food Intake by Individuals (CSFII) as its data source
for use in deriving its drinking water ingestion estimates. EPA is commended for seeking out
databases that can be used for estimating drinking water ingestion for the population at a national
scale. This was the best data source available and it had critical attributes that allow
advancement of our understanding of drinking water ingestion in the general U. S. population
and the Agency made good use of the data. Several strengths of the USDA 1994-96 CSFII
database are worth highlighting.

a) The database is large, recent, and it is a population based survey.

b) The database permits the categorization of various sources of ingested water.

c) The convention used to estimate the fraction of water in each food as consumed
was scientifically defensible.

d) The database permits a breakdown of the US population into some major groups
based on age, gender, special populations of females, regions, and broad
classifications of ingested water source.

The committee believes that the EPA Report will be an important source of information
on drinking water ingestion. The report will enjoy extensive use as a reference resource for
those within and outside of the Agency.

Even though this report will allow EPA to better understand contaminant exposures
associated with drinking water ingestion it does not, nor was it intended to, provide insight into
exposure to drinking water contaminants associated with dermal exposure (e.g., during bathing
or showering). Further, even though it provides estimates for some combinations of attributes,
many such attribute combinations are possible. Most of these are not included nor could all
possible combinations of potential interest be covered (e.g., infants who live in hot climates and
have health conditions which affect water intake). Therefore, to fully estimate contaminant
exposures associated with drinking water, EPA will need to go beyond projections that are based
solely on information contained in this specific Report.

Notwithstanding the strengths of the Survey and the EPA Report noted above, the

-------
Subcommittee does have a number of concerns with the Report. If the Report is to achieve its
full potential there are a number of issues that require further attention. These are discussed in
Sections 3.2 through 3.4 that follow.

3.2 EPA's goals and objectives as stated for this Report, and the analyses it contains,
are too limited.

The report was constructed only as a descriptive report without an explicit discussion of
how the estimates in the Report might reasonably be applied by users. The subcommittee has
several recommendations for revision to address this current shortcoming:

a) The Report needs a prominent and early explanation of the logic used in the
survey design and in the analyses used to develop the Agency's estimates. This
explanation should be understandable by the educated layperson. This is not a
criticism of the technical logic used in the analysis, rather, it simply recognizes
that most users of these estimates will not have the specialized knowledge of
statistics needed to understand fully the approach used.

b) The report must provide a much clearer indication of which estimates are reliable
and which ones are not as reliable. The extensive tables of statistics that appear
to break down the population to several subgroups provide potential users of the
data with a false sense of security about the precision of the estimates. This
practically guarantees that the results will be applied in ways not supportable by
the database.

The Agency has both scientific and policy reasons for estimating water ingestion in the
overall population and in subpopulations of interest. Some of these come from the statutory
mandates of the Safe Drinking Water Act (SDWA), but others come from the broader
environmental health community, such as: a) risk assessment; b) development of default values;
and c) sensitive subpopulations. The implications of the survey characteristics and the analyses
supporting EPA's estimates on these uses of the ingestion estimates are not sufficiently
discussed in the Report.

Risk assessments are scientifically-based efforts to estimate the impact that exposure to a
contaminant, or groups of contaminants in water, may have on human health. For waterbome
risk scenarios, it is important to construct as complete a picture of water ingestion as is possible.
Some of the distinctions in the present estimates limit that capability. For example, as the
Agency rightly points out, direct and indirect water represent only part of potential tap water
ingestion (and therefore exposure to waterborne contamination). Commercial water (that water
added by the manufacturer prior to marketing-not now included in the EPA estimates) is
frequently taken from tap sources, although these are frequently far removed from the point of
consumption. This does not mean that the estimates obtained from the present study cannot be
used in developing risk assessments, but part of the exposure assessment may have to obtain
broader categories of water source than are identified in the present analysis of the data. Such

-------
limitations in the tabular data need to be clearly stated in the report.

In its current configuration, the report provides estimates that are composites of both
those who reported drinking water during both survey days and those who reported drinking
none on those days. As EPA noted in its charge to the Subcommittee, this could result in
underestimates of drinking water ingestion. EPA traditionally uses a default value for water
ingestion when converting a "safe dose" (mg/kg/day) to enforceable concentration limits in
drinking water. The Report permits EPA to use improved data in developing such limits.
However, the Subcommittee believes that such analyses should focus on those portions of the
population that actually ingest drinking water. The estimates needed in this circumstance should
not be diluted by including large numbers of individuals that reported no water ingestion during
the survey (see Question 11 in Appendix A). In its current configuration, the report provides
only the diluted estimates. Ingestion estimates should be developed by EPA to reflect only those
who actually reported water ingestion as well as the current composite situation. When
sufficient data are available to estimate confidence intervals, these Survey data can be used to
develop default values.

Some subpopulations of interest are adequately represented in the report (e.g., pregnant
women) but others identified included too few representatives (e.g., children of Native
Americans). For this reason, the Subcommittee strongly recommends that the Report make
explicit the limitations of the estimates. The breakout of pregnant and lactating women provides
at least a starting point for defining the amount of water that is consumed by populations that
may have special sensitivities. There are also some data that can be used to estimate water
consumption by individuals of varying age. However, it is important to recognize and identify
the limitations of these data for smaller populations (e.g., children of Native Americans/
Alaskans). In addition, other populations could be identified that consume higher amounts of
water (e.g., diabetics and individuals with kidney disease) that, while not rare in the overall
population, are well below the statistical power of the Survey to detect. If there are not sufficient
data to support development of relatively robust measures of confidence, the use of the data to
describe water ingestion by these smaller subgroups would be misleading and do a disservice to
these groups. If these groups are to be a source of particular concern in the Agency's regulatory
agenda, surveys should be conducted that are adequate to support such estimates. Some other
data sources might be superior for such purposes (e.g., NHANES).

The report provided insight into the 1.0 liter/10 kilogram default value for ingestion of
drinking water by children that is currently used by EPA. The analysis presented in the EPA
report shows that water consumption per unit body weight is very high at birth and falls off
sharply with age. The Subcommittee is encouraged that the EPA Report now provides
information that will permit analysts to use specific data for water ingestion in many future
instances where allowable concentrations must be developed. For others, the Report will
provide a better basis for developing reasonable defaults.

In the Agency's derivation of maximum contaminant limit goals (MCLGs) the
mathematical operation essentially converts consumption to ml/kg/day, the Subcommittee
8

-------
believes that there is significant value to be gained from expressing estimates in these units as
well as volume ingested. When shown in such units, the real differences in water consumption
by age become much more apparent than when given as volume measures alone. Ml/kg/day
figures are best used until ingestion stabilizes and then the daily volume becomes equally
appropriate.

Clearly, the EPA Report is not intended to answer questions about other critical
subpopulations (e.g., workers that consume very large quantities of water because of the exertion
involved in their work or because of working in hot and/or dry climates). This points to an
opportunity for future work in this area. Some of this information may already be available in
the literature. If not such efforts could involve designing a relatively simple hypothesis and
model of the determinants of water ingestion. Some independent variables for such a model of
water ingestion could include: a) level of effort or metabolic rate; b) average ambient air
temperature; c) average ambient relative humidity; d) body weight; and e)age.

Describing and capturing data for these predictor variables, and subsequent water
ingestion for subpopulations that share common (and relatively narrow) ranges of these
variables, could lead to the identification of the subpopulations of greatest concern for
contaminant exposures through drinking water. It might also lead to the development and
validation of a comprehensive mode! for the prediction of water ingestion from such parameters.
The resulting simple hypothesis and model of the determinants of water ingestion could be used
generically because it would reflect water needs of individuals. In some individuals most, if not
all, of that water requirement might come from tap water. Those are the persons that the SDWA
is intended to protect. If more accurate estimates of actual drinking water ingestion are needed,
appropriate data could be collected by targeted surveys. The results could always be
benchmarked against the basic water needs of individuals under different physiological
conditions.

The value of some of the tabular distributions provided in the analysis is not clear. For
example, water ingestion was provided by region. The Subcommittee's agrees with the need for
regional estimates; however, the political regions identified in the Agency Report were probably
too large. The within region variability of ingestion is probably much larger than that between
regions.

It is important to emphasize that risk is a function of both exposure and sensitivity.
Sensitivity is determined by genetics, developmental stage (old as well as young), lifestyle, and
preexisting disease conditions that are not addressed in the Report. The Agency should simply
point out that these other determinants of sensitivity are not addressed in the report.
3.3 The EPA Report should state that EPA did not have information that would allow
calculation of confidence intervals for sub-populations.

-------
A discussion point for the subcommittee centered on the question of whether it was
appropriate to provide data without meaningful confidence intervals. The design of the CSFII
survey requires use of an ultimate cluster methodology which is an aggregate of sampled persons
within each primary sampling unit. Smaller subpopulations within the sample (e.g., the less than
one year olds) did not meet these criteria. This prevents the calculation of confidence intervals
using the ultimate cluster methodology. It would be good to clarify this point for the readers of
the report.

3.4 The Agency needs to develop a strategy for the analysis, presentation, and
interpretation of data that is consistent with the intended uses of the data.

The report should contain a description of the methodology used for analyzing the data.
This would better explain the approach employed for those who are not experts in the
sophisticated statistical techniques. In addition, the report should contain a strategy for future
analyses of the data including some hypothesis testing.

Data validation and quality assurance procedures used in the development of the report
should be prominently documented, with especial attention to conventions that were developed
to handle some of the data.

The presentation of numerous tables containing estimates developed in the Agency
analysis are clearly not appropriate for many of the applications the Agency will have for this
information. Tables should be substantially reduced. Instead of numerous tables with estimates
having unknown confidence levels the report should be limited to tables with estimates that
support agency needs and for which valid estimates of reliability can be provided. These tables
should be displayed in a useful way with significant figures appropriate to the level of precision
in the estimates. The text surrounding these fewer tables should make clear the limitations of the
estimates and whether they can be applied with confidence to evaluations of the subpopulations
with which they are identified.

For example, the Subcommittee had very little confidence that the data reported for
Native Americans reflected a Native American lifestyle (see Question 6 in Appendix A). There
is a difference between "race" and "lifestyle". The reasons for different intake rates primarily
reflects lifestyle (secondarily SES), and probably not race per se. If the Agency is convinced that
this data reflects such a lifestyle, it should explain the rationale supporting the conclusion. A
contrary conclusion should also be clearly explained.

Similarly, separate tables should be provided reflecting ingestion estimates for those
respondents reporting water ingestion during the two days captured in the CSFII. This should be
in addition to tables that reflect estimates based on a composite of respondents reporting
tapwater ingestion and those who did not report such ingestion. Both sets of analyses provide
important perspectives depending upon the use that the data will be applied to by the Agency.
There are also good reasons to display data in both in terms of ml/kg/day as well as liters
consumed. In all cases these data should include some measure of the precision of the estimate.
10

-------
       It is extremely important to segregate estimates for children by age for the reasons stated
earlier.  However, it is much less important to separate estimates for adults by age because the
differences observed are much smaller. In adults the future analytical focus should be on
identifying subpopulations that consume more water for other reasons, such as preexisting
disease (e.g., diabetes mellitus), occupational conditions, or effects due to climate.
                                            11

-------
Appendix A

Responses to Specific Agency Charge Questions

1. Statistical Methodology

Charge Question 1: The distributions of estimated water intake were generated
using standard statistical methodology for surveys with complex designs such as the 1994-96
CSFII. Is the statistical methodology used to generate the estimates appropriate?

The methodology described in the document is an appropriate technique to produce
estimates from a multi-stage, stratified, clustered sample. The Agency, however, did not clearly
state that the estimates were generated from a summary tape containing only final weights
assigned to individuals. This means that the Agency was limited in what it could do with the
data. References to the documents describing estimating equations for the US Department of
Agriculture's Continuing Survey of Food Intakes by Individuals are needed.

2. Confidence Limits

Charge Question 2: We have limited the calculation of confidence intervals
about the mean and boot strap intervals for percentiles to the distributions for the larger sub-
populations. The complex sample design makes the calculation and interpretation of results for
smaller sub-populations virtually impossible to calculate and interpret. Is this an appropriate
decision?

Yes. However, the rationale for this is buried in the narrative. The Subcommittee
recommends that the Agency state more clearly, and in a prominent place, its reasoning for not
calculating such intervals throughout the report. Also, the convention of placing "zeros" as
entries in the tables for place-holders where no estimates have been generated is confusing. The
Subcommittee recommends inserting "dashes" in place of such zeros. This convention is used
by others reporting results from such efforts.

3. Short-term Data and Long-term Estimates

Charge Question 3: The CSFII survey is based on short-term survey data.
Upper percentile estimates may differ for short-term and long-term data because short-term
survey data lends to be inherently more variable. Is it appropriate to report upper percentile
estimates such as the 99th percentile?

The decision whether to report upper percentile estimates depends in part on whether the
quality of these estimates is sufficient for their intended use. Quality may be judged by the
number of individuals interviewed, the fulfillment of underlying assumptions, and the computed
A-l

-------
statistical precision, bias and confidence in the percentile estimates. Uses of the drinking water
ingestion estimates may be very broad and could include risk assessment, rule-development for
microbial contaminants of drinking water and disinfection by-products, as well as other uses not
now anticipated. Some uses of drinking water ingestion estimates may require the short-term
survey data available from the present CSFII survey data (i.e., estimates of daily averages based
on only two non-consecutive days of data), while other uses may need long-term survey data
(i.e., estimates based on more than 2 days of data).  For example, short term data and a
knowledge of the variability of such data can be useful for risk assessments of acute health
effects such as diarrhea due to microbiological contamination, whereas long term data and a
knowledge of its variability are needed for risk assessments of long-term health effects such as
cancer. As the short-term data available from the current CSFII survey are not ideally suited for
all uses, it is particularly important that the report adequately describe the quality of the
estimates so that users can judge if the results of the current survey are of sufficient quality. As
indicated above, this quality can be described in various ways such as by providing variances
and confidence limits for estimated percentiles, by carefully stating and explaining all
assumptions used in obtaining those estimates, and by the number of individuals  interviewed in
the various subcategories.

       The number of individuals interviewed in subcategories is sometimes very small in the
CSFII data.  This point is illustrated by reference to Table A-3b in Section 1 le of the EPA
Report. In this table, there is only one individual in the <0.5 year age category and only three
individuals in the 0.5-0.9 age category. Clearly, upper percentiles should not be reported for
categories for which the number of persons interviewed is so small. The National Center for
Health Statistics has issued guidelines on minimum sample sizes required to obtain credible
estimates. These guidelines should be considered by EPA as a way to decide when drinking
water estimates should be flagged as being of lower than acceptable quality.

       Taking these considerations into account, this Subcommittee believes it is appropriate
that the lower and upper percentile estimates obtained from the CSFII survey be reported, but
that additional guidance on their quality and when they should and should not be used should be
provided.

       4. Data Conventions

              Charge Question 4: Are the data conventions used to identify direct and indirect
water appropriate?

       A series of conventions was established to allow the estimation of water intake as a result
of water  consumed as a component of foods. The procedure is described in detail and is
essentially the same as that used previously by Ershow and Cantor (1989) and by the Office of
Pesticides Program (Tolerance Assessment System, 1985). The procedures as described are
appropriate and will allow EPA to account for moisture gained and lost during cooking and
allow the estimation of the proportion of water from home supplies versus from commercial
water sources. A quick check of the results  of applying the conventions to the CSFII food codes

                                           A-2

-------
indicates that the procedures worked well. The results appear to be in the anticipated ranges.
The data should be rounded to reflect the appropriate level of precision. It would also be useful
to note in the text and on any files containing the factors that these represent a factor that is a
composite of factors, e.g., that different types of rice, rice cooked different lengths of time and
by different consumers will have different amounts of moisture and therefore different factors.

The Agency did not conduct a quality assurance check on the data. Given the multitude
of uses for this information, the Subcommittee recommends that a formal QA/QC audit be
conducted to ensure that the conventions were actually applied to each code as described in the
methodology.

Where indirect water and intrinsic water are lost during cooking, it is necessary to
determine how much is lost from each source. This is an arbitrary decision and the proposed
approach seems reasonable. Validation of the estimates should be undertaken to verify the
results.

5. Subpopulation Distributions

Charge Question 5: Do the data support estimates of subpopulatlon
distributions?

The CSFII data were used to generate point and interval estimates of daily average per
capita water ingestion in the manner presented in Section 8b of the EPA Report. Point estimates
presented include the mean, 1st, 5th, 10th, 25th, 50th 75th, 90th, 95th and 99th percentiles.
Subpopulations defined are gender, age, region, race, economic status, residential status and
certain specific female Subpopulations of pregnant and lactating women of childbearing age.
The results are presented in section 11 by water source and by nine sociodemographic
categories.

Examination of the tables on pages 11-3 through 11-326 easily reveals many
subcategories without sufficient observations to support the point estimates. For example, Table
A-3b on pages 11-15 and 11-16 shows point estimates of community water intake by race and
fine age category. Between the American Indians and Native Alaskans, there is only one
individual under 6 months and there are only three individuals in each of three other age
categories. Presenting point estimates this way will likely mislead readers. Potential users
should be cautioned about the uncertainty of point estimates having small sample sizes.

Whenever possible, point estimates should be presented with confidence intervals. But
due to small sample size of some Subpopulations, not all confidence intervals can be computed
from the data. It is not clear how many interval estimates cannot be derived from the data
available to the Agency. Only Tables 1, 2, and Figure 9-20 in Section 9 include 90% confidence
intervals. A survey of over 15,000 individuals should allow more confidence intervals to be
calculated and presented.
A-3

-------
Although parameter estimation, hypothesis testing, and modeling are difficult because of
the complex nature of this survey, the valuable information collected deserves further
exploration. A strategy should be developed to analyze, interpret, and present data on sub-
populations in a systematic and meaningful way. The first set of tables presented should be for
major subpopulations such as gender (male vs. female), age (infants, children, youth, adults),
race (white, black, Asian/Pacific Islander, American Indian/Native Alaskan), and region
(northeast, Midwest, south, west) without further subdivision. Both point and interval estimates
should be provided for each category of these major subpopulations. Hypothesis testing should
be carried out to see if the differences among categories are statistically significant.

In Section 11, ingestion estimates for nine sociodemographic subpopulations are
presented by water source. No rationale is given for why, among all the possible combinations
of major subpopulations that could have been selected, these nine combinations of
sociodemographic variables were chosen for presentation. Further, without understanding the
meaning and limitations of the data, over 200 pages of tables are of limited usefulness to readers.
If the relative importance of various sociodemographic variables can be evaluated by modeling
and hypothesis testing, cross-tabulation can be focused on a limited number of significant
variables.

6. Subpopulations Included

Charge Question 6: We have provided distributions of estimated water intake
for numerous subpopulations. Should any additional subpopulations be added? Should any be
excluded? Specify subpopulations.

The Report provided distributions of estimated water intake for a relatively large number
of subpopulations. As discussed earlier, the available data do not support reporting of some of
the values that are placed in the tables. This does not negate the need to lay out water ingestion
rates for subpopulations that might be at greater risk from drinking water contaminants. There
are clearly examples that are at least as important as those reported upon. These are pointed out
by the Subcommittee with the recognition that the CSFII database will not provide the needed
data for such analyses. Nevertheless, the Agency is encouraged to seek better estimates of the
distributions for two broad categories:

a) Sub-populations with different lifestyles, occupations, or activities.

I) Infants and toddlers are not a homogeneous group. There is a
population of infants in the 0-3 months of age group that receive
constituted powdered formula exclusively. These infants could be
consuming as much as 180-200 ml/kg/day from the same source of tap
water.
ii) Dietary survey misses lifestyles of specific cultural groups (e.g.,
Native American, recent immigrants) that are still practiced
iii) People who live in hot climate areas.

A-4

-------
iv) People who consume large amounts of water because of physical
activity (can consume as much as 300-500 ml/kg/day)

b) Health conditions that affect water intake:

I) Diabetes
ii) Conditions requiring rapid rehydration needs (GI upsets, food
poisoning)
iii) Disorders of water and sodium metabolism.

The subcommittee also noted that there are aspects of water ingestion that might be better
addressed by taking a physiological approach. If total water ingestion is first thought of in terms
of the needs that are defined by physiological state, developmental stage, levels of activity
(reflected in metabolic rates), and environmental settings a general model could be constructed.
This approach will always capture the upper limit, as one can assume that all of the water that is
not intrinsic to food could be derived from the tap. Then more accurate estimates of sources of
the actual water consumed could be constructed from survey information that is targeted to the
sub-populations of interest. This could be a more efficient way of addressing drinking water
ingestion by subpopulations of interest to EPA, in particular those noted in 'a' above.

7. Indirect Water

Charge Question 7: USDA has identified two types of indirect water in foods:
a) the amount of water in food as consumed; and b) the amount of water used to prepare food.
The water intake report provides estimates of the amount of indirect water in food as consumed.
If resources permit, we could expand our report as a future addendum to include estimates of the
amount of indirect water used to prepare food. Would this be desirable?

The current ingestion report provides estimates of the amount of indirect water in food as
consumed. The amount of water used to prepare food may be greater, owing to evaporative loss
during preparation. This loss can result in a concentration of non-volatile contaminants. Such
increases are chemical specific. To be able to calculate the amount of residue concentration,
both the amount of indirect water in food as consumed, and the amount of indirect water used to
prepare food must be known. This analysis should be limited to only those foods where the
amount of water added to prepare the food is known. The amount of water which is first boiled,
then added to food such as that used to prepare infant formula, is not known.

The critical question is whether preparation leads to large changes in the distribution or
ingestion of water contaminants in the population. Certainly in some cases the losses of water
volume could be large, but are they consistent within individual consumers. In addition, it is not
clear how common a practice unattended boiling or extensive boiling might be. There are many
other more important variables that remain unaddressed with respect to sensitive populations.
Consequently, pursuit of this issue should reflect programmatic priorities with respect to
sensitive subpopulations.
A-5

-------
An omission in consideration of indirect water that could be significant appears to be soft
drinks prepared from syrup in restaurants, fast-food establishments, and bars. Again, the pursuit
of this detail has to set within the priorities of the program. However, some initial evaluations
might be made by contacting the appropriate industry representatives to obtain information on
the ratio of syrup to canned/bottled soda sold.

8. Food Types Not Covered

Charge Question 8: Additional water intake estimates associated with types of
food may be useful for specific risk-exposure analyses, e.g., cold beverage intake. Such analyses
are feasible using the CSFII data. We could expand our report as a future addendum if
resources permit. Are any such targeted analyses of significant interest at this time?

This question was withdrawn by the Agency during the discussions at the July 19-20,
1999 Drinking Water Intake Subcommittee meeting.

9. Intrinsic Water

Charge Question 9: Intrinsic water is the water contained in foods and
beverages at the time of market purchase. Intrinsic water includes commercial water (added to
food products by food manufacturers) and biological water (found naturally in foods). Intrinsic
water is not included in our current analysis. If resources permit, we could expand our report as
a future addendum to include estimates of intrinsic water. Would this be desirable?

Yes, this would be desirable, but the Subcommittee would like to point out that the
Agency's use of the term "intrinsic water" is unusual. In most instances intrinsic water is that in
the raw food product, not water added by processors. In some cases (e.g., NASA) the term
includes both free water and metabolic water that is derived from a food. There would be some
value of using another term (e.g., commercial water) to describe this category. Care would have
to be taken that it is not confused with bottled water, however.

The Subcommittee felt that one advantage of including intrinsic water (as the term is
used in the Agency report) in the analysis would be to enable the derivation of a fluid
requirement distribution by recognizing this additional source of water. This combined
direct/indirect water ingestion distribution will be less variable than direct use only, as it is closer
to a biological/physiological measure than one of lifestyle. However, this is only one of a
number of other sources of water ingestion that would have to be known to construct the
physiological need for water for individuals under different conditions. Knowing intrinsic water
does capture another tap water source, even though it may be removed from the consumer's own
tap water. The relative component of commercial water could then be calculated by examining
only those products with a major contribution to one or more subpopulations. Examples are
soda in cans, iced tea in cans, bottled soda, beer, milk, prepared infant formulas.

10. Other Issues

A-6

-------
Charge Question 10: What are the scientific limitations to the use of the water
consumption estimates provided in this report (i.e., what other issues has the Subcommittee
noted with the estimates that are not covered elsewhere)?

This report needs to be viewed as a key reference for population based information on
water consumption. The following are examples of the many potential uses of the information.

c) It will be valuable to programs where consumption of water estimates are needed.

d) It will be useful to support Agency rule making.

e) It can be used to evaluate existing default water consumption rates and to provide
new defaults for subpopulations.

f) It can serve as a reference to compare to other data sources containing similar
information.

The document needs to keep these uses in mind and the text and tables should be
designed to be user friendly for these purposes. Thus many users will prefer the data
summarized in a ml/kg body weight format while others will need the ml/day summary. Both
formats should be provided. Keeping the uses in mind, it becomes especially important that the
limitations of the Survey database and the ingestion estimates based upon it be clearly spelled
out in the introduction and that the report contain only statistically valid estimates. It should be
noted that some sensitive subpopulations are not in the database or cannot be identified in the
database. These are identified under other charge question responses. It should be explicitly
stated when data are sufficient (and give the criteria used) and when they are not. Where it is not
obvious why estimates are not provided, it needs to be explained.

A use-restricting limitation is the survey design that precluded estimating water ingestion
in subpopulations that either by choice, or access, utilize only one source of water for ingestion.
The survey data identify and provide descriptive tables for three significant sources of ingested
water; community tap water, bottled water and other (private wells, cisterns, etc). While the
report provides detailed ingestion distributions for each water source within defined
demographic groups, "sole source" subpopulations of water ingestion limit the utility of the
report for local risk assessments. Such "sole source" ingestion distributions would be especially
valuable to assessing health risks from ingestion. The overall national mean water ingestion
finds community tap water contributing 75% of the water ingested. The tables provided show
that the 75% contribution is not evenly distributed over the population. A valuable statistic not
provided is the percent of individuals obtaining virtually all their water from community taps or
all from the other two sources and their estimated ingestion rates. Such individuals may be
consuming nearly 1/3 more tap water than the national estimate provides. If the size of this
population is substantial, using the national ingestion estimate to characterize contaminant
exposure to this group could significantly underestimate tap water contaminant risks. This
underestimate may partially be seen in Section 9, figure 2 which shows that over 47 million US

A-7

-------
residents are estimated to consume no tap water. Since these individuals require fluid to survive,
they probably represent those ingesting only "other" water from private wells or bottled water
only. The inclusion of these "unexposed" individuals in the ingestion estimates leads to
underestimates of ingestion among those with access to the water source. The potential for
underestimating ingestion is even more pronounced for infants where Section 9 figure 3 shows
nearly half of the infants drank no tap water. This probably reflects the high percentage of
infants being breast fed or using bottled water to mix formula. This seriously reduces the utility
of the information provided on this vulnerable population. Whenever possible it would be useful
to many users to have confidence intervals around the estimates.

11. Zero-Values

Charge Question 11: The water intake estimates provided in this report are
based on all respondents, including those who did not report consuming water during the two
survey days. If resources permit, we could also generate estimates of water consumption which
exclude the zero consumers of water. We noticed that for some sub-populations, especially the
less than one-year-old infants, a substantial proportion consumed zero or minimal amounts of
tap water per day (presumably those who were breastfed or drank undiluted formula or milk);
these zero consumers of water can contribute to lower estimates. Would this be desirable?

Yes, it is desirable, probably necessary, to eliminate the non-consumers of community
tap water from the survey statistics for purposes of developing a set of consumption estimates for
use in predicting exposure to drinking water contaminants. The DWIS suggests that those data,
for which there are adequate numbers of individuals, should be displayed both ways. In other
words inclusive of the population and a second display of only those individuals that are
consumers of tapwater.

Based on the projections in Section 9, Figure 3, approximately 50% of the children under
1 year of age do not ingest community tap water. The mean and upper confidence limits
generated from data from which these projections were made will greatly reduce the estimated
ingestion rates in some groups. A rough arithmetic estimate can be made of how important this
would be by recognizing that removing half of the population that does not consume water will
increase the mean consumption of water in the under 1 year of age group to approximately 90
ml/kg body weight. This is roughly six times that of an adult. Thus, the differential between
adults and children is at least twice that which is derived from currently utilized defaults. The
subpopulation of children representing the highest tap water intake will be those fed
reconstituted powdered formula. This will result in the greatest dose (per kg) of water
contaminants. There is less impact in the general population, where only about 8% of the total
population does not ingest community tap water. Nevertheless, the principle is the same.
A-8

-------
                                 REFERENCES
Damiano, J. and J. R. Mulhausen (1999)4 Strategy for Assessing and Managing Occupational
      Exposures. (Second Edition).  AIHA Exposure Assessment Strategies Committee Edited
      by, American Industrial Hygiene Association, Fairfax, VA.

Ershow, A. G. and K. P. Cantor. (1989) Total Water and Tapwater Intake in the United States:
      Population-Based Estimates of quantities and Sources. National Cancer Institute.
      Number 263-Md-810264.

National Center for Health Statistics Analytic Working Group. (1993) Guidelines developed by
      the NCHS. As cited in Life Sciences Research Office (LSRO) (1994). Third Report on
      Nutrition Monitoring in the United States. Interagency Board for Nutrition Monitoring
      and Related Research.

Saunders, S. and B. Petersen. (1986) Introduction to the Tolerance Assessment System. U.S.
      Environmental Protection Agency. May.

US EPA (1999) Estimated Per Capita Water Consumption in the United States. US EPA/Office
      of Water.  Prepared under EPA Contract No. 68-C4-0046. June 1999.
                                        R-l

-------
Chair
Science Advisory Board Executive Committee
U.S. Environmental Protection Agency
Ariel Rios Building
1200 Pennsylvania Avenue, NW
Washington, DC 20460 Re: EPA-SAB-EC-00-003

Dear Sir/Madam:

This is in response to the review of "Estimated Per Capita Water Ingestion in the United
States," a report prepared by the Office of Water (OW), Office of Science and Technology. The
Science Advisory Board (SAB) had many helpful and insightful suggestions. Technical staff in
the OW have addressed the comments and recommendations in a revised document that will be
completed by March 31,2000. Our summary response to the SAB review and recommendations
is enclosed.

The objective of the report is to present current, technically sound estimates of water
ingestion by the U. S. population and certain subpopulations. These estimates will be useful in a
wide variety of the Environmental Protection Agency program applications including risk
assessments and regulation development. The SAB subcommittee expressed concerns about our
lack of explicit guidance on the use of the estimates. We anticipate that guidance on the use of
the estimates will be addressed in program-specific documents to be developed in the future.

The majority of the SAB comments concerned presentation and interpretation of the
analysis results. The fundamental statistical methodology employed in our study was found to
be "an appropriate technique to produce estimates from a multi-stage stratified clustered
sample." As a consequence, no changes were made to the methodology, and the numerical
estimates did not change. In response to SAB comments, changes were made in the organization
of the report, key water ingestion estimates were identified and highlighted in the Report text,
confidence intervals were provided for key subpopulations estimates, and the number of
subpopulations included in the report was reduced. We have noted the SAB's recommendations
for additional studies but are not able to implement these at this time.

The SAB suggested that formal inferential tests of differences between subpopulation
water ingestion estimates be applied. While we have not done these, we do provide interval
estimates about water ingestion estimates for major subpopulations. These interval estimates can
be used to assess the extent of differences in subpopulation estimates.

-------
       We are grateful for the ongoing involvement of the Drinking Water Intake Sub-
committee of the SAB in our efforts to revise and update our estimates of water ingestion rates
for the U.S. population.  The discussion and recommendations of the subcommittee have been
very beneficial and led directly to a substantially improved report.

                                        Sincerely,
                                        J. Charles Fox
                                        Assistant Administrator
Enclosure
4303/HJacobs/cah/3/28/00/250-5412/doc name sabl.wpd
AX-0000373

Identical letters sent to:

Dr. Henry Anderson, Co-Chair
Dr. Richard Bull, Co-Chair

-------
Dr. Henry Anderson
Co-Chair, Drinking Water Intake Subcommittee
Science Advisory Board
U.S. Environmental Protection Agency
Ariel Rios Building
1200 Pennsylvania Avenue, NW
Washington, DC 20460

Dear Dr. Anderson:

-------
       We are grateful for the ongoing involvement of the Drinking Water Intake Sub-
committee of the SAB in our efforts to revise and update our estimates of water ingestion rates
for the U.S. population. The discussion and recommendations of the subcommittee have been
very beneficial and led directly to a substantially improved report.

                                        Sincerely,
                                        J. Charles Fox
                                        Assistant Administrator
Enclosure

-------
Dr. Richard Bull
Co-Chair, Drinking Water Intake Subcommittee
Science Advisory Board
U.S. Environmental Protection Agency
Ariel Rios Building
1200 Pennsylvania Avenue, NW
Washington, DC 20460

Dear Dr. Bull:

-------
       We are grateful for the ongoing involvement of the Drinking Water Intake Sub-
committee of the SAB in our efforts to revise and update our estimates of water ingestion rates
for the U.S. population. The discussion and recommendations of the subcommittee have been
very beneficial and led directly to a substantially improved report.

                                         Sincerely,
                                        J. Charles Fox
                                        Assistant Administrator
Enclosure

-------

Response to Science Advisory Board Recommendations

In July 1999 the Drinking Water Intake Subcommittee (DWIS) of the U.S.
Environmental Protection Agency (EPA) Science Advisory Board (SAB) reviewed the report
Estimated Per Capita Water Consumption in the United States (hereafter referred to as the
Report). The SAB provided its written comments and recommendations in December 1999.
The following 16 points address the SAB's major findings and recommendations and describe
our responses.

1. Report objective: The SAB expressed concern that the Report is purely descriptive and
does not provide "an explicit discussion of how the estimates in the Report might
reasonably be applied by users." It was EPA's intent to limit the Report's objective to
the provision of current descriptive statistics on water ingestion for the population of the
United States and selected subpopulations. The Report does identify some of the broad
applications for these estimates including their use in the development of risk assessment
and regulations which involve default values for water ingestion and in the estimates of
risks to highly exposed and/or sensitive populations. We believe that more explicit
guidance on the application of these estimates is out of the scope of the study. However,
we anticipate that guidance on the use of estimates, will be addressed by EPA Program
Offices in documents to be developed in the near future.

2. Overview of logic and survey design: The SAB stated that "The Report needs a
prominent and early explanation of the logic used in the survey design and in the
analyses used to develop the Agency's estimates," We have made some modifications to
the text in the Report in order to respond to this SAB comment. The details of the
statistical methodologies, which were in the main text of the Report, have been moved to
an appendix. The initial chapters of the Report now provide summary descriptions of
both the survey design and methodology. These explanations should be understandable to
the general scientific audience. The user is also directed to the references to obtain
additional background on the survey design and statistical methods.

3. Reliability of estimates: The SAB recommended that the report "must provide a much
clearer indication of which estimates are reliable and which ones are not as reliable." In
response to this comment, we have amended the Report to state that estimates based on
small sample sizes may be less reliable than estimates based on larger sample sizes. As
suggested by the SAB, we applied the minimal reporting requirements provided in the
"Third Report on Nutrition Monitoring in the United States" published in 1995 by the
Life Sciences Research Office of the Federation of American Societies for Experimental
Biology. In accordance with this document, mean ingestions estimated with a sample
size of less than 48 are marked with an asterisk to indicate that they may be statistically
unreliable. Similarly, percentiles estimated with sample sizes less than
12.8/(l-percentile) may be statistically unreliable and are also marked.

4. USDA data: The SAB commented that the Agency did not clearly state that the estimates

-------
6.
were generated from a summary tape containing only final weights assigned to
individuals. In response to this comment, the Report has been amended to include a more
detailed explanation on how the survey weights were calculated by USDA. USDA
provided sample weights for each survey respondent in the three survey years with two
days of consumption data. Sample weights, which project the data from a sampled
individual to the population, are based on the probability of an individual being sampled
at each stage of the sampling design. The sample weights associated with each
individual were adjusted by USDA for nonresponse to correct for nonresponse bias as
discussed in Appendix D of the final Report. However, certain variables, for example,
region, are at a summary level. USDA has named the States within a region. Estimates
by State, however, are not trackable because USDA data do not contain a variable
identifying States. For this reason, water ingestion estimates by State are not possible.
Furthermore, variance estimating strata are numbered sequentially. The sequential
numbering prevents aggregation of strata with similar consumption patterns and thus
reduces the ability to generate certain subpopulation variance estimates. The USDA
documentation referenced in the Report provide the details on the calculation of the
survey weights.

Commercial water: The SAB recommends that the Report make clear that the estimates
do not include commercial water (water added by the manufacturer prior to marketing).
In response to this comment we have further emphasized throughout the report that the
water ingestion estimates do not include commercial water or biological water (water
found naturally in foods). Also, all tables of estimates now bear a footnote which states
"All estimates exclude commercial and biological water."

Water ingestion by "consumers only": The SAB recommends that the Report include
water ingestion estimates based on those respondents reporting water consumption
during the two days captured by the CSFII. In response to this recommendation, we
revised the Report to provide water ingestion estimates based on both the entire
population and on "consumers only." The estimates for "all individuals" use water
ingestion data from all survey respondents in the population (or subpopulation) including
those who reported no consumption of the water from the source under consideration.
The "consumers only" estimates include only individuals who reported ingestion of the
water under consideration.

Survey limitations: The SAB strongly recommends that the Report make explicit the
limitations of the estimates. Specifically, the SAB points out that some sub-populations of
interest included are not represented in the report. In response to this recommendation,
the "Discussion" chapter of the Report was amended to provide additional detail on the
survey strengths and weaknesses. This chapter now specifically states that the survey
design does not support generating water consumption estimates for certain sub-
populations of interest. Examples of such sub-populations are Native Americans with
traditional lifestyles, people who live in hot climates, people who consume large amounts
of water because of physical activity, and people with medical conditions necessitating

-------
increased water intake. The reason that the survey does not support estimation of water
ingestion by certain sub-populations is that estimation for these sub-populations was not
provided for in the design of the study. Rather, the survey is designed to support
ingestion estimates by the U.S. population.

8. Units of milliliters/kilogram of body weight/day: The SAB recommends that the
Report provide water ingestion estimates in both units ofmilliliters/person/day and on
miliiliters/kilogram of body weight/day. In response to this comment, the Report now
provides all water ingestion estimates in both units ofmilliliters/person/day and
milliliters/kilogram of body weight/day.

9. New studies: The SAB suggests that additional studies to collect current or retrospective
information on subgroups of interest could augment the report. We have noted the
recommendations but have limited the estimates to those supported by the USDA's 1994-
96 CSFII as this survey was designed to collect consumption data from the U.S.
population.

10. Confidence intervals: The SAB stated that the "report could be considerably
strengthened, and the potential for misinterpretation of its findings could be reduced
substantially, if the Agency provided information on the statistical significance of
differences in water consumption between major subgroups of the population. " In
response to this recommendation, all key tables of water ingestion in the Results chapter
of the Report now provide 90 percent confidence interval estimates about the mean per
capita water ingestion and 90 percent bootstrap interval estimates of upper percentiles
from the empirical distributions of per capita water ingestion. However, the limited
sample sizes for certain sub-populations in conjunction with the survey design do not
always support estimation of variance which is a necessary component of interval
estimation. This is a characteristic of the survey data reporting by USDA. In response to
the SAB recommendation, the Report provides detailed discussions of this limitation in
both the Methods chapter and the Discussion chapter. Population and subpopulations
with sample sizes that are large enough to support variance estimation have interval
estimates reported in the main body of the Report. For the smaller subpopulations, as
determined by the number of respondents in the survey, point estimates are segregated in
the appendices.

11. Hypothesis testing: The SAB suggests providing information on statistical significance
of differences in water consumption between major subgroups of the population. We
have not applied formal inferential tests of differences between subpopulation water
ingestion estimates; the objective of the Report was limited to presenting current per
capita water consumption estimates. However, we provide interval estimates about water
ingestion estimates for major subpopulations. These interval estimates can be employed
by the user to assess the differences in subpopulations.

12. Data validation and quality assurance procedures: The SAB suggests that data

-------
procedures should be prominently documented. In response to this suggestion, we have
added a brief discussion in the "Methods" chapter of the Report which describes the data
conventions and validation procedures applied to create the data subsets from which the
estimates were created. This chapter also identifies the variables used to identify water
consumption and sources. It also relates the file interrelationships and assumptions
applied to water-containing foods. Data convention and validation procedures described
in the Report are augmented with listings of pertinent survey questions, methods for
calculating indirect water and listings of water containing food codes. These augmenting
materials appear in the appendices.

13. Number of tables of estimated water ingestion: The SAB stated that the number of
tables should be substantially reduced to reflect a limited number of subpopulations.
EPA has done this. We deleted the tables for race/ethnicity, region, economic status, and
residential status because of sparse data in some cells. We retained the estimates based
on age (broad and fine), pregnant women, lactating women and women of childbearing
age. We placed key tables of estimates which we considered of major interest to the user
in the "Results" chapter. These tables provide 90 percent confidence intervals around the
mean and 90 percent bootstrap intervals around the upper percentiles. Each table has a
corresponding graphical display. A more comprehensive set of tables is provided in the
Appendix E of the Report. We have flagged estimates in all tables that do not meet the
minimal reporting requirements as defined in the "Third Report on Nutrition Monitoring
in the United States". We have also added footnotes to the tables which address data
limitations.

14. Averaging time: The SAB recommends that the Agency make clear to the user that the
water ingestion estimates are based on two-day averages. The 1994-96 Continuing
Survey of Food Intake by Individuals (CSFII) collected two non-consecutive days of
food ingestion data. Quantities of ingested water reported were averaged by participant
to generate a two-day average. Throughout the Report we have stated that the estimates
of water ingestion are based on the average of the two days water ingestion reported by
individual survey respondents. Additionally, in response to the SAB recommendation,
we have added a footnote to all tables of estimates which states: "Estimates are based on
two-day averages."

15. Age categories: The SAB states that it "is extremely important to segregate children by
age..." and that "... it is much less important to separate estimates for adults by age
because the differences are much smaller." The Report provides water ingestion
estimates for broad age categories and fine age categories. In response to the SAB
comment, we amended the broad age categories to include a single adult age group. The
broad age categories now cover babies (less than one year old), children (one to 10 years
old), young adults (11 to 19 years old), adults (20 years and older). The fine age
categories include 11 age groupings. These groupings are less than six months (<0.5
years), between six months and one year (0.5 to 0.9 years), 1 to 3 years, 4 to 6 years, 7 to
10 years, 11 to 14 years, 15 to 19 years, 20 to 24 years, 25 to 54 years, 55 to 64 years,

-------
      and 65 and older.
16.    Sensitivity analysis: The SAB recommends that the EPA conduct a sensitivity analysis
      of the data assumptions made during the data analysis. This comment pertains to the
      assumptions made about the source of drinking water (plain water ingested directly as a
      beverage). These assumptions, which are described in the "Methods" chapter of the
      Report, were necessary because the CSFII survey does not completely designate the
      source of the plain drinking water. We agree with the SAB's recommendation and a
      sensitivity analysis under is consideration for future work. Time and resources do not
      permit the Agency to conduct the analysis at this time.

-------

-------