EPA 570/9-85-001
NTIS Assession No. PBS4 21394:
TECHNIQUES FOR THE
ASSESSMENT OF THE CARCINOGENIC RISK
TO THE U. S. POPULATION
DUE TO EXPOSURE FROM SELECTED VOLATILE ORGANIC COMPOUNDS
FROM DRINKING WATER VIA
THE INGESTION, INHALATION AND DERMAL ROUTES*
C. Richard Cothern, Williain P. Coniglio and William L. Marcus
L. Lappenbusch
Chief, Health Effects Branch
Joseph A. Cotruvo
Director, Criteria and Standards Division
OFFICE OF DRINKING WATER (WH-550)
U. S. ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
U.S. Environments! Protection Agency
Region V, !JLv.. ,7
230 South D.;-:.- > "troet
-^o, !!"". ~ v "' -4
-------
Estimates of Population Risk Or Cancer Cases Expected From
Existing Concentrations in Public Drinking Water Supplies,
Estimates of Cancer Cases Averted And Number ot Systems
Affected By Drinking Water Regulation and Estimates Of The
in These Cases
Benzene
Carbon Tetrachloride
1,2-Dichloroethane
1,1-Dichloroethylene
Perchloroethylene
Trichloroethylene
Vinyl Chloride
JULY 25, 1984
* This report has been reviewed by the Office of Drinking
Water, u. o. Environmental Protection Agency, and approved
for publication. Approval does not signify that the contents
nece&saiily reflect the views and policies of the U. S.
Environmental Protection Agency.
U.S. Environmental Protection Agency
-------
EXECUTIVE SUMMARY
The risk of individuals experiencing an adverse health
effect after being exposed to a contaminant nay be calculated
by multiplying the level of personal exposure times a measure
of the hazard resulting from the toxic potency of the contam-
inant. The total number of cases of disease or death expected
among an entire society for one lifetime of its members is
called the population risk. For this report the present expo-
sure level tor drinking water contaminants among the U. S.
population is calculated and expressed as a population-concen-
tration figure. Population concentration is the sum of the
products of the number of people times their estimated exposure
level. Typical units of population concentration found in
drinking water contaminants are persons x micrograms/liter.
The carcinogenic potency for individual contaminants has been
expressed as the number of cases of cancer per person per
microgram per liter. This report considers only the carcino-
genic risk for specific contaminants and does not attempt to
deal with other manifestations of toxicity.
This report develops potency estimates and population risk
estimates for nine selected volatile organic contaminants
(VOC's) in drinking water. These nine contaminants were
selected for study because they offered a range of data of
potency and population exposure via drinking water. The com-
pounds include: benzene, carbon tetrachloride, 1,1-dichloro-
EX-1
-------
ethane, 1,1-dichloroethylene, perchloroethylene, trichloro-
ethylene, and vinyl chloride. The analysis provides a basis
for determining the direct effectiveness of national drinking
water standards to reduce illness and death from present levels
of these chemicals in water supplies. This portion of the
analysis could be called risk reduction analysis. It does not
consider the benefits the national regulation may have in
preventing further deterioration of water supplies and human
exposure.
The occurrence estimates were evolved from monitoring data
developed in several national drinking water surveys. The
surveys included are: the National Organic Reconnaissance
Surv ?y the National Organic Monitoring Survey, the Natior.nl
Screening Program for Organics in Drinking Water, the Community
Water Supply Survey, the Rural Water Survey and the Groundwater
Supply Survey. Taken together, these surveys provided data on
over 1,000 groundwater supplies and 1,000 surface water sup-
plies. This data base was drawn from 48,458 groundwater sup-
plies and 11,202 surface water supplies and is shown in Table
EX-1.
Drinking water contributes to the three routes ot expo-
sure, viz., ingestion, inhalation and dermal. An estimate of
the relative importance ot these routes for different age
groups for VOC's in drinking water is shown in Table EX-2.
Due to the chemical characteristics ot the VOC's, dermal
EX-2
-------
' VJ
as
W
H
5
*5
Z
r-4
irfj
Z
r-l
Oti
a
z
OS
t^ tx3
Z H
^C *"^
Z r-J
l-l ^
y co
*C 3c
z a:
° 0
U u
3- r"4
H £
O M
H 0
Z
o <;
1 CO
>< 0 Z
W CM O
01
^%
VJ
01
<*4
I-l
%^
B
CO
OC
o
o
1-1
B
c
o
1-1
4J
CO
VJ
4J
c
0)
o
c
_o
^J
0)
1 1
*J
CO
X ' ' OC
£i3 £u p-*
taj ^C
30 x-s02
<: co H
H 0 Z
Z M
< -o
co z
»3 o
o u
JJ
E-* £3
U3
Z H
r-J -xO
z a
o z
^ w
>J ^c
£3 ^
Ou
°«2
a
u
H
^J
3E
*-^
H
W
,c
J^J
o
1
o o
r» oo
1
00
,0 r--
^^
1
o o
ys^
1
00
1
oo
en -3-
/\
1
o o
CM m
/^
1
o o
r 1 CM
^v
o
C i
M-I in
c
1-1
Q
c
o
H
in
1
m
o
m
o
V
4J
CO
VJ ^
4J * 4-1 r-4
C 01 C CO CO
Ol 4J CO VJ 4J
O O 01 O O
c cz: 3 N
O 4J
cj o
o
O r-l
-r4 O> C 4-1 CO
4-1 0) CO CO 4J
CO CO O) 01 O
r^^r ca H
J ' a
< 1 0
H
O
H
PM
4J
c
CO
c
rJ.
E
CO
4-1
c
0
o
0
o
r-4
V
1 1
1 1
CM
n
vO
v£>
r-4
CM
vO
en
CM
r-l
in
in
t
^
vO
^
ON"
o
r^
oo
o
r-4
X
CM
r-l
r-
o
1 1
X
sO
1 1
01
c
01
N
c
Ol
CO
o
0
o
o
o
o
m
in
sO
f>
,^
ON
30
PO
r-4
o
^>
.
^
CM
ON
O
fc
00
l-l
00
o
1
X
&\
r-f
fx*.
o
r-4
X
in
0^
0)
T3
rJ
w
o
II
f
o
C CO
O VJ
rs 4J
VJ 0)
COH
U
0
o
0
o
o
0
o
f)
^^
l-l
o
o
ON
00
CM
r-4
f»^
ft
r-4
CM
O
CM
00
o
r-l
x
jj
p**4
l^»
O
X
oo
en
i
1-1
VJ
o
r 1
J=
O 01
r-l C
O co
1 JC
CM 4J
- 01
r-l
O
O
o
o
o
o
o
0
CM
in
in
ON
CM
k
CM
O
?rj
sO
ON
O
rsi
00
o
X
CM
CM
sO
0
4
X
t**
sO
1
O
u
o
i-l Oi
C C
O 01
1-1 1 1
Q 5S
1 JC
14 4J
- 0)
r-l
O 0
O
--4
/\
O
.n
o
o
o
o
CM
30
^
in
CM
^
m
.
m
CM
03
en
O
CM
r-l
CM
sO
CM
^
30
CM
k
m
CM
30
30
CM
ON
00
4
30
O
l-l
X
CM
CM
00
O
r-4
X
r-4
1
O 0)
VJ C
O 01
l-l I-J.
C >,
O JC
-4 4J
VJ 01
H
O i VJ
01 4J O
o
00
r-l
o
o
o
o
o
:M
r*.
^
r-l
cn
»
-4
f^.
ON
CM
r-l
CM
00
O
rH
X
CM
r-l
r*.
O
r-4
X
m
r-4
4-1 W 4J
CO 01
S 9 -o
O 0)
00 1 CO
-4 0
c oi a
-^ X X
C 4-1 01
1-1
VJ C 01
TJ-* V4
01
01-1 S
00-1
CO co r*
VJ »-4
01 4J CO
> CO
COX U
4J CO
W X
X CO 4J
4J oi
E co
TJ 3 01
C co B
cO co 3
co n
C m
O-l (B
^4 CO
4J 4-1 4-1
co O co
r-4 4J VJ
a 4-i $
O CO
aoi c
X co
0) 0)
X C E
4J CO
01 01
OJ^ J
**
Ol
WX TJ
^j Cw4 p
0 CO
^
^^ *
o-o vj
Vj CJ Ol
O- CO 4J
a-*
r-4
x x -^
4J 01 CO
E
IM CO CO
O co VJ
9 00
§0
C VJ
CO O O
01 4-i E Vj
X CO 01
4J 1 O AJ
3 .1-1
co ao 1-1
i-l O -^
O. O co
C 4J B
O 4J CO
»4 CO -0 VJ
4-1 X 01 00
CO 4J CO O
VJ 0 VJ
4JX 0.0
C O X -4
01 1-* 0) E
ox
C S oi m
O VJ
y o 0) o
4-1 3
C VJ
0 C C 0
4) -r4 0 0
g 4J -r4 --J E
i-l (6 AJ 4J 3
VJ rM S CO E
O 3 VJ VJ ~-i
X
o
,_(
^
c
f>
O C C fC
a. ai 01 E
o o
01 C C 3J
x o o -c
H o O 4J
*
-------
cr os
J> W
a, 3-
c. o
*-
n ft
B 2-
OJ 3
t*^QQ
A
0>
rt a
3" C
rt rt
0-B
O 09
ft »-
ID ft
*
3
O
rt
rt
ft
10
3"
Q
C
ft
09
H CM
O ft 3
B rt w o: n to y
H- o; £ Os B 3" 0>
O i-1 H- rt 09 O f-
n B 3* M C 09
O > B H- ft rr
Oo cr H- 3 09 n *
n te 3 oo cr o
§O 00 00 09 D
n Or--
o» cr n n o
^^ ft* *X3 J^i
7TO. rt
00 H- ft
>0 ° °
ao 3 fi
09 CO h-
s; ft o
09
ft
a
B
fr*
O
o b
Oc tc i rss i
n c ii i
B b
OS 0s*
-»v
B
d
n h- M
00 0
00 i ii i-
n N: uj i o
§c o c
o
-s.
B
!-
O
f^
O ^-'
00 ^ 1 1 C
no ii c
09
^
B
H-
n ^
o i i o
00 -»J ii O
i
09
B
o
n i *
O £- 1 N5 1
00
n *-"
1
^H ^3
3 1-
H. 3- C
3 (^> ^*
a^ a
O 0:
O rt M
O 00
09 3 ft
h" 09
no rt
L^. ^^
C
ft 3
3
^t.
1
a
1 00
c c
F"-*
gg
rt C
i-1
^v09
P' 1
Hi
£ ft
)0 0.
^^
h-1
I1 ^/t
O O
^v i^ Jfr
w n er
N> ft 09
i O
« r * f^
>o cr
^ ft
.j a
P"
o
CO
ft
ISS
^o ^^
0 0
*>
a ^
^e B
O^ ^^ ^*
O rt 0
n
7? "3 O
JO ft 00
09 09
h- B
ft n
0.
09
v<
to
cr> O
0 0
^ x^
O>
Q ^
OrT
EE
»
09
rc
VJl C
0 0
/^s
c o>
^J
?r ft
)o cr
^^
.p.
<
o
^
-3
r
TO
c ^^
r s
cr ^
H p^
5^ t^
z ^
H t-i
/~» P3
C S
OO
0
3> PJ
r*H f^
r:
PC C
tj^ ^2
c:
^f CA
R c
^*** ?c
r cc
i- p;
H 0
re
PC C
C£
i-i cr
z
C PC
pro
^
i-i
Z
C3
^
H
Cr
po
1-3
Jj>
Q^
r
n:
g
1
N;
(0
-------
exposure appears to be very small compared to the ingestion
and inhalation exposure potentials, which may be close in
magnitude. Respiration exposure is of increased importance
due to the enrichment of indoor air from water use.
The carcinogenic potency of these contaminants was esti-
mated from bioassay data developed from animal tests conducted
at high dose levels. The biological response at lower levels
of exposure was estimated using mathematical models. Figure
EX-1 shows a typical extrapolation. The initial starred points
are the animal data, and subsequent lines represent the pro-
jections from tour different models. The error bars on each
line represent the upper 95% confidence calculations from the
model.
Attempts were made to deal with the factors contributing
to uncertainty in these risk estimates, which can be very
large. The authors' estimates of contribution to the uncer-
tainty from generic toxicity factors are shown in Table EX-3.
Likewise, the contribution to uncertainty from the exposure
estimates is shown in Table EX-4. Using this method for
determining risk, it is evident that the largest contribution
to the uncertainty is from the choice of dose extrapolation
model. As can be seen in Figure EX-1, this uncertainty for
the region of interest can be five to six orders of magnitude
or more. This uncertainty is much larger than all other uncer-
tainties combined and therefore has been used to determine the
EX-5
-------
1
1
!
8
S I
§
8 I
5 I
2
- 1
55
1 £ c
2*-
a « o
11-
is 5
4J JJ -C
^ if
tr-c ^
!>.«
(C >-
C in
j= tn c
* g£
n-S
£^-^
g|5
o c
^ c
4J . .-
I §
O fe tr.
Sr-^ S£
(C 4J (C
w c
^8*E
0,0^
|l>
C 4J f,
^^ C
com
^ ill
8 c" R
V. C
(ft
se
of
H C AT
»-i J; tr
jj
C tt C
iS'H^
c C c
~* £
O-H F
fl.
(R
*i
I I
Ci
' NOI1S99NI
NOI1VTVHNI
1VWM3Q QNV NOI1VTVHNI 'NOI1S39NI VIA)
Q1SCMX3 NOSUa^/XSIH 3WIJ.3dn
-24-
^ ^ >
= tn JS
-------
TABLE EX-3
HEALTH EFFECTS UNCERTAINTY
Category
Contribution to Uncertainty*
1. Choice of endpoint
2. Personnel capabilities
3. Choice ot species, strain,
age and sex of animals
4. Test compound purity
(contamination, decay and
vehicle contribution)
5. Inappropriate statistical
t* ,t methodology
6. Distribution of animals
among doses and number
used
7. Selection ot dose levels
8. Lack preliminary tumor
change inrormation such
as hyperplasia
9. Experimental surroundings
10. Dietary considerations
Less than 10% of the time
chose the wrong endpoint--
some exception such as
chlorinated hydrocarbon,
where expect liver cancer (U)
All or nothing (E)
Ail or nothing (E)
For the VOC's is generally
trivial but could be two
orders of magnitude (0)
All or nothing (E)
One to two orders of
magnitude (E)
Up to two orders of magnitude
(E)
For VOC's is less than 10%
but could for other contami-
nants be as large as a factor
ot three (U)
Could be orders of magnitude
(0)
Factor of two (E)
* U-leads to an underestimate of the risk.
0-leads to an overestimate of the risk.
E-could lead to an overestimate or an underestimate
of the risk.
EX-7
-------
TABLE EX-3 (continued)
11.. GLP1 s (good laboratory
procedures)
12. Time-to-tumor
13. Microscopic tissue
examination
14. Diseases in test animals
15. Statistical noise
16.
17
Outcomes of cancer (it
include non-fatal cancers)
Conventional choice ot
p level (e.g., O.O5)
18. Synergism/Antagonism
19. No corresponding tissue
in humans
20. Most sensitive vs. average
21. Animal to man
22. Body weight vs. surface
23. Use ot upper 95% contidence
limit
24. Choice of dose-response
model
One to two orders of
magnitude (U)
Could miss the effect or
could be an underestimate by
a factor of two
Could be a factor ot two (E)
Ail or nothing (E)
Factor of 2 (E)
Could be two orders of
magnitude or more (0)
Ail or nothing (E)
Many orders of magnitude (E)
All or nothing (E)
Several orders of magnitude
(E)
Conservatively two orders of
magnitude (E)
One order of magnitude (E)
Up to an order of magnitude
(0)
5-6 orders ot magnitude when
considering risk levels in
the 10-4 to 10-6/lifetime
range (E)
* U-leads to an underestimate of the risk.
0-le"ads to an overestimate of the risk.
E-could lead to an overestimate or an underestimate
ot the risk.
EX-b
-------
TABLE EX-4
SOURCES OF UNCERTAINTY FOR OCCURRENCE, POPULATION CONCENTRATION
AND EXPOSURE ESTIMATES USED IN THE ASSESSMENT OF VOLATILE
ORGANIC CHEMICALS IN DRINKING WATER
Impact on estimate ofT
Factors
Occurrence
Population
Concentration Risk
Generation of monitoring data
Proportion of population
sampled
Representativeness of systems
selected
Geographic distribution,
system size ana source of
water
Sampling methods
Site ot sample collection
Time of sample collection
Method of sample collection
Container type
Stability during storage
Sample analysis
% recovery from sample
Compound identification
Accuracy of quantitative
determination
Assumptions during data analysis
Lower limits of quantitication
Oral exposure rates
Intake rate ot water
Pollutant level in
consumed water
(hot vs. cold)
X absorption for oral
intake
Respiratory exposure rates
Dermal exposure
5% (U)
10% (E)
20% (E)
20% (E)
10% (U)
10% (U)
100% (U)
. J% (U)
10% (E)
40% (E)
50%
Factor
of
2
Factor ot
3-4
Factor
of 2 (E)
10% (E)
50% (0)
10% (0)
Factor o±
3
Negli-
gible
U - leads to an underestimation of the risk.
0 « leads to an overestimation of the risk.
E - could lead to an overestimation or an underestimation of
the risk.
EX-9
-------
range of the population risk estimates.
Table EX-5 is a typical example of the calculation of
population risk and the risk reduction analysis. As seen in
Table EX-5, the population risk is in the range of less than
one to 400 per lifetime. (Fractional population risks result-
ing from calculations have been rounded off to the designation
of less than one). In most cases, the largest contribution to
population risk is from the population exposed below the detec-
tion limit. As this is an estimate, it may be an artifact ot
the analysis. A summary of the risk reduction analysis esti-
mates is shown in Table EX-6 for the nine selected VOC's.
The & iges of uncertainty in population risk estimates
and risk reduction analysis are quite large. If the actual
shape of the dose-response curve were known, the uncertainty
in these estimates could be reduced to about two to three
orders of magnitude. Data for human exposure, especially at
the lowest exposure category, could improve the uncertainty,
and data at lower dose levels could also reduce the uncer-
tainty. Neither ot these sources of uncertainty can be
influenced easily due to the extreme cost involved and because
of the unethical nature of human experimentation.
EX-10
-------
TABLE EX-5
BENZENE-POPULATION RISK ESTIMATES
FOR CURRENT LEVELS OF DRINKING WATER EXPOSURE
Mean Drinking
Water Concen-
tration
(Micrograms/
Liter)
0.25
2 .7!)
7.5
15
25
35
45
55
65
75
100
Number of
People Being
Served
2.1 x 108
4.5 x 106
1.2 x 105
6.2 x 104
1.6 x 104
6 x 103
3 x 10 3
2 x 103
1 x 103
1 x 103
1 x 10 3
Total Lifetime Individual
Risk For the Mean
Concentration*
Low
(Probit)
<10-10
<10-10
<10-10
2 x 10-9
1 x 10-8
3 x 10-8
4 x 10-8
5 x 10-8
7 x 10-8
1 x 10-7
5 x 10-7
High
(Weibull)
1.7 x 10-6
1.3 x 10-5
3.1 x 10-5
5.1 x 10-5
8.9 x 10-5
1.3 x 10-4
1.6 x 10-4
1.7 x 10-4
2.2 x 10-4
2.4 x 10-4
3.2 x 10-4
Lifetime
Population
Risk
<1 - 360
<1 - 60
<1 - 4
<1 - 3
<1 - 1
<1 - 1
<1 - 1
<1 - 1
<1 - 1
<1 - 1
<1 - 1
Totalt
400
Benzene-Risk Reduction Analysis
For Limiting Drinking Water Concentration
Maximum Allowable
Drinking Water
Concentration
(Micrograms/Liter)
75
35
7.5
2.75
Approximate Individual
Risk Rate for Maximum
Concentration
1 x 10-8
7 x 10-9
<1 x 10-10
<1 x 10-10
- 2 x 10-4
- 1 x 10-^
- 8 x 10-5
- 7 x 10-5
Cummulative
Cases
Averted0
<1 - 2
<1 - 6
<1 - 10
<1 - 70
t Rounded to one significant figure.
* The total individual risk was determined by assuming that
the risk due to inhalation exposure is equal to that due to
ingestion.
0 Number of cases averted for concentrations shown in the first
column.
EX-11
-------
8
CO
to z
l-l 3
0§
OS CO
H H
to <
06
to <
u
to
<
o
+-
T3
CO 0>
0> *->
CO J-i
CO 01
U ;>
i ^
C B 5
O Cd 4J
*4 * * **J
^ IM *^
£ 00 J
4-
o
CO 01
O* 4-*
CO VJ
CO O>
0 >
CM
1
^4
m
r^
0
i to
r-i
< <-l CO
V s^
_3
1 .
i
O CO Vj|
W E O>
O «0 4J
"< Jj '--1
£ 00 J
4_
O
CO 01
W ^J
CO i-i
CO 0)
o >
in
f*^
0
g
o
£^
I Z
w
t _J ^J
1 -*.
O CQ Ui
rJ E 01
CJ CO 4-*
1-1 Vj -i-l
£ OOrJ
c
o
** *-
4-> jrf
CO CO
r-j i-l
a~
o
Ou
H-l
u,
7^
in to
r- Z
r- t
ts
<
[H
<;
o o
o
1
r 1
V
0)
o
Id
o
1 i-l
x:
! O
CO
4J
c
CO
c
1-1
CO
4J
C
(.)
4J
0)
H
0)
c c
O) O
N X)
C VJ
pi o> «o
rjj i pQ ^J
o
o
^
1
m
r-4
0
o
^J
1
fO
iTi
«
f*x,
1
^^
0
o
o
CM
1
0
CO
in
f»
CM
o
o
o
*
o
tr>
1
o
>*
*
0)
c
CO
4J
0)
o
o
l-l
u
o
1
OJ
f-l
to
r-4
to
H4
J
1
OS
^^
e
H
z
r-l
r-l
(L4
C3
to
z
r-l
<
t-J
c^
V
c
0)
>.
x:
4J
01
o
^J
o
1
o
o
1
1 1
1 1
CM
1
^4
V
m
vO
PO
1
r-4
V
in
r**^
*o
1
f 1
\/
l/"\
^»*
CM
2
1
l-l
V
01
c
01
r-l
£.
4J
01
o
VJ
0
r-l
JC
u
V-i
01
PH
o
o
CM
1
,_(
V
o
0
^^
o
o
»
^^
1
^
V
m
fs*.
o
o
o
»
o
tN
1
l-l
v
in
r*-
CM
o
o
o
*
o
f"*
1
r-l
V
*
0)
c
01
1-1
£
4J
0)
0
t-l
o
£.
0
l-l
H
o
o
1
V
m
sO
o
o
to
-4
1
l-l
\/
in
f*^
0
o
CD
PO
1
r-4
V
in
f-».
1 *-s
c
1-1
0
o
o
^
1
J^
V
in
vO
o
o
55
»
-------
TABLE OF CONTENTS
PAGE
Executive Summary .... EX-1
V
Introduction ... 1
Individual Risk Rates b
Occurrence, Exposure and Population-Concentrations . . 29
Population Risk Calculations:
benzene, Carbon Tetrachloride, 1,2-Dichlorethane,
1,1-Dichloroethylene, Perchloroethylene,
Trichloroethylene, Vinyl Chloride 72
Uncertainty &3
a) Health ^tfects Uncertainty b6
b) Uncertainties in Occurrence Estimates and
Population Concentration Estimates 100
Discussion and Conclusion 117
Acknowledgements 127
References 128
Appendix A A-1
Appendix B b-1
Appendix C C-1
Appendix D D-l
-------
TABLES
-EX-1 Total Estimated Population (in thousands) Exposed to
the Contaminant in Drinking Water at the Indicated
Concentration Range
EX-2 Comparative Model ot Absorbed Dose From a Volatile
Pollutant In Drinking Water
EX-3 health Effects Uncertainty
EX-4 Sources of Uncertainty for Occurrence, Population
Concentration ana Exposure Estimates Used in the
Assessment of Volatile Organic Chemicals in Drinking
V.'ater
EX-5 Population Risk and Risk Reduction Analysis tor Benzene
EX-6 Cases Averted at a Total Lifetime Individual Risk Rate
1 Mathematical form and low dose behavior of 10
selected dose-response models
2 Comparative model of absorbed oose from a jl
volatile pollutant in drinking water
3 Selected drinking water impurities of raw water 34
4 Estimated amount of each VOC released to air, 3t)
land and water
5 Population concentrations 66
6 Population risk and risk reduction analysis tor 75
benzene
7 Population risk and risk reduction analysis tor 77
1,2-dichloroethane
b Population risk and risk reduction analysis for Ib
perchloroethylene
9 Population risk and risk reduction analysis for 7y
trichloroethylene
10 Population risk and risk reduction analysis for ttO
vinyl chloride (Maltoni)
ii
-------
TABLES (continued)
1,1 Population risk and risk reduction analysis tor 81
vinyl chloride (Feron)
12 Cases averted at a total lifetime Individual risk 82
rate
13 Health ettects uncertainty 87
14 Incidence rates for benzene ^2
15 Sources of uncertainty tor occurrence, population 101
concentration and exposure estimates used in the
assessment ot volatile organic chemicals in
drinking water
16 Sample storage time and conditions 1UB
111
-------
FIGURES
EX-1 Bioassay data and model extrapolations for
trichloroethylene
1 Flow chart showing the input ot information to 4
the risk estimation process for contaminants in
drinking water
2 Range of shapes for possible dose-response models 11
3 Converted aose-response data from experimental 13
exposure to trichloroethylene
4 Bioassay data and model extrapolations for 14
trichloroethylene
5 Bioassay data and model extrapolations tor benzene 15
6 Bicassay data and model extrapolations for 1b
1,2-dichioroethane
7 Bioassay data and model extrapolations tor 17
1,1-dichloroethylene
b bioassay data and model extrapolations tor 18
perchloroethylene (ingestion)
9 Bioassay data and model extrapolations for ly
perchloroethylene (inhalation)
10 Bioassay data and model extrapolations tor 2U
vinyl chloride (Feron)
11 Bioassay data and model extrapolations tor 21
vinyl chloride (Maltoni)
12 Converted dose-response data frou experimental 22
exposure to carbon tetrachloride
13 Bioassay data and model extrapolations for 25
carbon tetrachloride (rat)
14 Bioassay data and model extrapolations for 26
carbon tetrachloride (mouse)
15 Distribution ot a contaminant according to 30
household water use plan
iv
-------
(continued)
16 hypothetical dose-response curve demonstrating 11tt
the effect on the extrapolations of an
experimental point at 'A1
17 Dose-response curves for leukemia from atouic 122
bomb survivors
-------
INTRODUCTION
For over 400 years, we have sought to measure the effects
that contaminants, including chemicals, have on living tissue.
We have confirmed and reconfirmed one basic truth: all con-
taminants can cause cellular damage that may be expressed as a
loss of organ function or death if the contaminant is present
in living tissue at a concentration that is sufficiently high
or fcr a long enough period of time to interfere with normal
biochemical function. The amount of a contaminant entering
via all specific routes of absorption--the skin, the lung, the
intestinal tractmust be considered as exposure. The ability
of the contaminant to enter into biochemical reactions essen-
tial for life processes is a prime facto" in determining if
the living organism will exhibit stress, altered behavior,
loss of cell or tissue function that is diagnosable as disease
or other detrimental effects including death.
Water is the major constituent of living cells. Through-
out all recorded history, man has endeavored to assure a con-
stant and convenient supply of water not only to sustain his
life, but to enhance social concepts of personal hygiene and
provide recreation. A convenient and readily available supply
of water has been the driving force behind engineering advances
that constructed cisterns, catchments, aquaducts, indoor
plumbing, bathtubs, swimming pools, toilets, washing machines,
ice makers, showers, hot whirlpools, dishwashers and hutnidi-
1
-------
tiers. Hot water and cold water are brought directly to the
home for infants and aged, healthy and ailing alike to use
and enjoy.
How each hitchhiker in drinking water affects the individ-
uals in it depends, inter alia, upon many factors including:
(1) how much ot the chemical is absorbed per unit of body
weight; (2) how active the contaminant is within the cell and
(3) how susceptible the individual is to the chemical action
as determined by initial genetics, nutrition and sums of other
stresses including chemicals or prior diseases.
Some contaminants in drinking watrr are cancer-causing
agents. Examples of such contaminant include the Volatile
Organic Chemicals (VOC's) which are analyzed in succeeding
chapters. The particular VOC's being considered here were
selected because they appear to be the most important VOC's
presently known to be in the national water supply and the
dose-response curves for these and other carcinogens do not
have a threshold tor toxicity (Schneiderman, 1979). To develop
drinking water regulatory standards for these contaminants it
is necessary to estimate the risk posed by low concentrations
in drinking water. This estimation requires extrapolation of
dose-response curve from the experimental data at high doses
into the unknown region of low concentration. Coupling these
estimates with estimates of the number ot people exposed, the
number of cases of cancer that can be prevented by setting
2
-------
standards at difterent levels can be factored into the regula-
tory process.
Risk estimation for drinking water contaminants requires
the knowledge of the level present in the water and the amount
ot individual exposure resulting from a lifetime behavior pat-
tern. We must look upon the frequency at which various concen-
trations of the contaminant presently occur in water supplies
across the United States, the number of people being served by
those water supplies, the resultant exposure to these people
and a carcinogenic potency of this substance. Figure 1 shows
how this information is used in the risk estimation process.
An estimate of total expected excess cancer in a lifetime
for the U. S. from a given contaminant is called the popula-
tion risk estimate. This is an estimate and not a statement
of perfect knowledge. Thus an uncertainty or range of values
will be included to describe the estimate. If the frequency
distribution of the contributing factors controlling exposure
and toxicity were known, the propagation of errors can be
determined using standard and well-known methods. However, in
the present situation the frequency distributions are not known
and thus a range of values will be used to roughly describe the
situation.
The factors that are involved in the determination of
population risk from drinking water are:
3
-------
4J C
10 «H
C 0
O 4J
«W C
C O
H C>
r tr
^ « c
t) «H tH
H f. h
tb +J «D
£
-------
- The number of people exposed to contaminant i at
concentration c
- The amount absorbed from consumption of water
containing contaminant i at concentration c
- The amount absorbed from dermal contact of
contaminant i at concentration c
lie - The amount absorbed from inhalation of contaminant
i which evolved from water of concentration c
Ric ' The individual risk rate for contaminant i at
concentration c (this is expressed as the expected
number of excess cancers in a lifetime per person
per microgram/liter)
The total amount absorbed of contaminant i at concentration c
is:
Aic - Cic + Dic + Iic
The overall population risk for the ith contaminant will be:
PR = Summation over n contaminants of (P±c Ric Aic)
There are limits to our knowledge about any subject and
the current state of knowledge about the health effects of
contaminants in drinking water also has its limits. For some
aspects, much is known, for others little or nothing is known,
but in all areas there is uncertainty.
It is not possible to measure a quantity with 100% accu-
racy. There is uncertainty in any measurement. The tools of
scientitic measurement by their very nature involve both
systematic and random error. In addition, several other uncer-
tainties exist such as the effect of age, sex, species, organ
5
-------
involved and rate of exposure of the test animals or humans.
Almost always the dose-response data are known at high levels
of exposure with no information being available at the levels
of exposure for which a standard is being set.
If data only exist for animals, then uncertainty exists
when extrapolated to humans. If there is exposure to more than
one contaminant, the lack of information about possible syner-
gistic and antagonistic effects adds to the uncertainty. Some
of these factors tend to overestimate risk while others under-
estimate the risk.
In order to develop the present risk assessment, a number
of assumptions n ed to be made. These assumptions include:
-- The average human ingests 2 liters of drinking water
per day and inhales daily an average of 20 m3 of air
-- The health endpoint for these contaminants is the
same for animals and humans
-- 10~4 of that per liter in water is transferred to
a liter of air
-- 100% of that in water is released to indoor air
-- The only difference between man and the test animal
is one of scale
-- Dermal exposure is insignificant compared to oral
and inhalation exposure
-- The mathematical expressions used reflect the
biological dose-response curves at low doses
-- Everyone is exposed to some level of each contaminant
-- The benign and malignant animal tumors are indicative
of cancer
-------
-- Elevated human exposure during infancy does not alter
risk
-- National monitoring is representative of the existing
exposure profile
-- All of that ingested or inhaled goes to the bloodstream
-- There are no thresholds
The following sections describe the development of indi-
vidual risk rates, estimates of exposure and population concen-
trations anH calculation o± population risk rates, and a
detailed discussion of the contributions to the uncertainty.
-------
merit, it was converted to ingestion values by converting the
dose in ppm to microgram/raS. This conversion is accomplished
by multiplying the dose in ppm by 1.2 x 10^ times the ratio
(molecular weight of the chemical/molecular weight of air)
to give a dose in micrograms/m3.
Then, to convert the dose in microgram/m3 to ttg/kg/day,
multiply by:
20m3 mg 1_ - 2.86 x 10-4
day 10-3 micrograin 70 kg
The resulting human equivalent ingestion exposure bioassay
data is then fit with four analytical models; viz, logit,
multistage, probit, and . aijull. These four models are chosen
somewhat arbitrarily to be generally representative of models
currently popularly in use. The mathematical form of these
models is shown in Table 1. It is well-known that other models
could be used to fit the same data (tor more detail see Van
Ryzin, 1980, a and b; Rai and Van Ryzin, 1979 and 1981; Krewski
and Van Ryzin, 1981; and Krewski, 1982; and Munro and Krewski,
1981). There are no biologically based criteria for choosing
one model over another. The models are in the form of an ana-
lytical expression that can be used to mathematically fit the
high dose data. Figure 2 shows the shape of the curves for
supralinear, sublinear, and linear. The models discussed here
are generally sublinear in the low dose range. However, the
multistage model is often linear at very low doses. The fits
9
-------
INDIVIDUAL RISK RATES
The estimates of individual risk rates are based on bio-
assay data Derived from animal experiments. These data are
shown in Appendix A along with a description of how procedures
are used to convert to a continuous human equivalent exposure.
The bioassay data selected in this report to estimate carcino-
genic potency are from experiments reviewed and selected by the
National Academy of Sciences (1982) and the U. S. Environmental
Protection Agency Carcinogen Assessment Group (See, for ex-
ample, Anderson 1982 and Bayard 1983).
Some of the data was developed in study protocols calling
for four or five days/week ;xtosure or, in the case of inna"a-
tion experiments for a fraction of a day. Differences between
experimental conditions and daily lifetime human exposure were
compensated tor in mathematical conversion factors as was the
fraction of the animal lifetime over which the experimental
data were collected. For example, it test animals were exposed
five times/week and for fifty-two weeks of the one hundred-four
week animal lifetime the dose was multiplied by 5/7 x 52/104 to
extrapolate it to a continuous exposure. Extrapolation from
animals to humans was accomplished by multiplying animal doses
by the cube root of the ratio of the animal weight to human
weight.
If the carcinogenic data were from an inhalation experi-
8
-------
ec
f
i
is
D 2
co a
if
§
3
v
u. m
O w
I 8
H
fc <
t UJ
3 w
S z
2 £
o <«
E <
1
i
e
A
COL
1
"c
A
* I
f \
**T^*
pROBrr
V
A
ca.
7
ea.
"o
A
ea.
1
ca.
1
e
1
I
i
7
N
6.
«<
C
«<
ft
1 WEIBUU.
1
1
e
A
4
5
!
UH-3NO |
1
i
<£
e
A
^
"5
Al
CO*
1
I'
IU
i
£»
V
5
T
g
Jg
e
tf
%
0
c
0
«4
<0
I
c
B
f-l
0
10
E
0
8
4J
i
j!
-------
LU
«
i
§
JO
f4
«
o>
3SNOdS3d
jo Ainiavaoud
&
&
-------
of these four models to the nine VOC1s considered here are
plotted on Figures 3-16. The dose axis was changed to water
concentration by the following conversion.
The relationship between the mg/kg/day dose and microgram/
liter water concentration can be calculated for a 70-kg adult
human assuming 2 liters/day is ingested as follows:
1 me x day 70kg lp3 microgram
kg 3ay 2 liters 1 ing
« 35,000 micrograms/liter
Figure 3 shows the fit to the actual bioassay data for
tric'nloroethylene. Figure A shows a log-log plot of ^he entire
region of interest showing the two non-zero aua po.nvs. The
solid lines are the point estimates and the error bars show
the upper 95% confidence limits. The computer programs used
were GLOBAL from Crump (1982) for the multistage model and
Krewski (1982), tor the iogit, probit and Weibuli models.
Figures 5-11 show the model fits to the next seven VOC's.
The numerical values used to plot these groups are shown in
Appendix B. As can be seen, no one model is consistently the
highest or the lowest and the curves span a wide range of
risks. Figure 7 for 1,1-dichloroethylene shows only one model
because there was only one non-zero data point available from
the experimental data. The other models require at least two
non-zero data points.
12
-------
Figure 12 shows the actual bioassay data for carbon tetra-
chloride. The loglt, probit and Weibull models show an in-
creasing risk for decreasing dose. The same behavior is shown
on the more complete plots of Figures 13 and 14. This is
clearly non-representative of reality the assumptions used.
It should be further noted that the curves shown in
Figures 3-14 are not the only models that could be tit to the
df>ra. Thorp is no assurance that the actual curve is in the
range or those shown. It is possible (although highly un-
likely; Van Ryzin, 1984) that the actual curve lies outside
these suown. ine probit model generally decreases extremely
fast relative to dose and the Weibull model generally decreases
-evy slowlv relative to dose. This pretty much covers all
possibilities.
Figure 4 presents the animal test data and model projec-
tions for trichloroethylene. Note that although dose-response
curves projected by each model initiate from the same points,
they diverge significantly at lower dose levels. At a drinking
water concentration of 50 micrograms/liter the Weibull model
provides a risk estimate approximating 1 x 10~2 while the
probit model provides an estimated risk of 1 x 10-10. The
estimates made using these techniques differ by 8 orders of
magnitude, an uncertainty equivalent of not knowing whether
one has enough money to buy a cup of coffee or pay the national
debt. Figures 5-14 provide information on the other VOC's.
27
-------
Note that the uncertainty or projections from dirferent models
is from 3-5 orders of magnitude at a standardized drinking
water concentration of 50 micrograms/liter. None of the
models, including the multistage model, consistently provides
either the highest or lowest risk estimates for all of the
VOC's. Figures 12-14 show that the probit, Ueibull and logit
models project an increasing cancer risk tor decreasing expo-
sure to carbon tetrachloride. This suggests that mathematical
extrapolations can be very misleading. Since the response data
has not been adjusted for time-of-death (age adjusted) this
may be part of the reason. Also, administered dose may not
be organ dose (see Van Ryzin and Rai, 1981).
When it enters the body, carbon tetrachloride is converted
to an active species by hepatic P450 enzymes. The enzymes are
then destroyed by these active species. This phenomenon is
called suicide substrate and generally describes the situation
where a metabolite kills the enzyme which form it and thus
reduces further formation of that species (Dambrauskas, 1970).
This could possibly explain the behavior of the dose-response
curve shown in Figures 12-14 for carbon tetrachloride.
28
-------
TRICHLOROETHYLENE
MICE
.7
.1
Multistage
Probrt, Logh
fr WcIbuH
600 1000 1500 2000 2500
Dose (mg/kg/day)
t- FIGURE 3
Oanverted dose-response data from experimental exposure to trichloroethylene.
-------
o *
1- C
C) V-
8. E^-S
2 i Seu
Sci t c --
u 5 jr c
«j u re w F
iH iH fH O V-
0 >< r P- c
a.c 6 £ ?
4J 1-1 C< C C
CJ O W -1 K ^
' S ^ g C "
I 7 I ^
h o c -
^
i
ac
i -
* K
- ii
s c
C ^
if
»^ (0
li
4J B i t -
c c. -
-ucc
4J tr T
m .2 £ s v
O W C r-
o « r c i-
S ti ;
£«- K - i
C V}
1-4
I
Bc£
aiflf'
3 W ^ C.
£ '5 S *S c
a * 3 * £
« ? t S >
«-> ic O i-
» ^ «'.-,*» ~, -,*,
isiinou IVMUBQ QNV NOIIVIVHNI 'NoiiSiONi VIA)
aiscuxa
-H 4-
-------
I C
&1
5 g .
S
tI
O QJ
ffi C O <
* 3 .C L
<-. ro
j
tn i
a;
8lTS|-,
^3|^5^
§> m -V £ .*H g
H Q ^ 5 I
c tn c 5 f
3 i-H UJ K £
R:
\ninou Tviwaa ONV NOIIVIVHNI 'NOUSIONI VIA)
OifCMXa NOSUBrf/XSIW 3WI13JH
CO CO X5 £ 05 J-
-------
III
I*
1 I I I
NOI1VTVHNI 'NOIISaONI VIA)
I
-i .28^;
* f- ti "-
4J (B ZJ 4J C '-
H 43 c c. E
fB 4J -H
E^I
fc II
4J 4JS.CC
10 «0 58.F ,
.-I O
5
|{p|~
g ^ c
O (c j; -
^^ ^^ «CZ ^* ^**"
fc \3 tr c
c n
H I
R3
. v c '-
4-> C c ;
*- K 4- I.
H -H ~
iH tt
JD i" t
O C c, ^
G. K C
Cj ^ '
C -
r:
^ {C- P
a c -c c ^
i o c
2 "^ - u
*- « fr--s u- c
V. T -
(1T!-S C -
Jc.?cf
CD «n "c *c £ £ ^
-------
Ill
!
!
I
I.
B w
. I
e K
8
4J 1C
ns
ssf
* Q> to I
_ £ TO
W F
H 4-> :
4J flj ic ;
J5 ? "d i
fH 55 O S
4J
S
*C £ -H
II*
p ^
c n
H tH O
O^H »
CL-H o
:a§
(0
; ** C
5 5
.. B'c
&*=
(saxnoM ivwwaa aiw NOIIVTVHNI 'NOIISIONI VIA)
aascMxa Nosua^/xsiu iwiiajn
£ E
0 § -43 O
X JJ -Q * J
o 10 73 t; ^
-------
f
i
§
9 -S
§
8 5
i
e v
~ I
8
fi.i-
£ ~
_ u: jc "~ -
.?*?**
4J U
S.S
>- -
5 -3 £ r i
J«&*ll
3 w j; j-
W 43 C -
11 Eli
»«1§:
5 J: tr c ^
-1 F K
3 c c
«j .p :
^IF =
I
5^? ^ '
gf^?:
- g S §- :
^ c x ,-
C C K C r
D ^ ^
f * £
lllfj
TJ o x £ '
flj vi_' 5; .^ i-
i a f s i
C-r £ c "
c o = :
c > CL.T i
It (C \
j: w ^ -
K K (T i-
jj g3 -p £ :
III II \
I I
Uainou tvwaao ONV NOiiyivHNi 'NoiisaoNi VIA)
Q3SMX3
-------
H r
4J I
I I I I \ I
(sainoa ivivuaa QNV NOIIVIVHNI 'NOUsaoNi VIA)
1
o (
5.£
I
-^ n
k 4.
8s.
« r
52
frS
c
i
C tt
H
%l
(0
M
H 4J
Q »
-------
c »»w -H 5 c £
a?U £~ £
F 03 g t*
~£o
Itainow ivMwao aw NOIIVTVHNI 'NOUSIONI VIA)
NOSM34/XSIW
8J
^ fl) C K j_
*1 ^K c SL
^a Rs- *
JS^
S-c
fig
^-?
.-,
» *> jj 5 c. -
f-|g*|.c
C P C~ T IT
s * c e « e
-------
* *
!
s
!
1
1
g,
« fc-
1 J3
4J T;
rH C
I -H
.£.
%*
Sl
I*.
K-8
£>.
»W
rft tr
Jl.
i |
H* 1=1
S!
IQ X)
iH
0 OJ
lf
$ c
31 '^
*w O
i i
i
i i i >
E
i
i
e -H ~ c
R^i
ill
^C
K I
i i %
(sainoM TVHwao ONV NOIJ.VTVHNI 'NOUSIONI VIA)
in i
81
H C
-------
12i
CARBON TETRACHLORIDE
RATS, MALES & FEMALES
.10
.08
.06
.04
.021-
Probh, Logrt T
& WelbuII
FIGURE
50
100
Dose (mg/kg/day)
150
Converted dose-response data fron experioental
exposure to carbon tetrachloride.
-------
jS <*
K
g
I it 11
n 4-i 4
4J 0) ft)
C *H «-i C
Z -H C - 4-
8 p "s -H
S ^i 0.7
ffi r-l * F
r *d "
Ho . o
4J -f-i (U '^
w ju C
^^ » *n r
'il'i'itl'ill 1115
^ *j c £
feainOH TVWaaO QNV NOI1VTVHNI 'NOIASaONI VIA) ££ G C
aascMxa Nosusd/xsib awiisjn S r - -
n w 4J r
I
I
§
8
3 >- f
M J2 «
-------
I
C 03 C
5
!
i
8
8
ivwuaa GNV NOIIVIVHNI 'noiisaoNi VIA)
awuajn
n
o
*
P S..
-------
OCCURRENCE, EXPOSURE AND POPULATION-CONCENTRATION
The average adult in the United States uses 48-56 gallons
or water per day (Watson, 1967, Linaweaver, 19b9). This means
that approximately 255 gallons of water per day would flow
through a household inhabited by two (2) children and two (2)
adults (Bailey, 1969). Under these conditions, 1 milligram/
liter or one ppm concentration of a contaminant in drinking
water means that nearly 1 gram of each substance is passing
through the home. The remaining quantity of each contaminant
is present in water used to launder clothes, clean the dishes
and house and remove biological wastes from the home (Figure
15).
The quantity of any pollutant absorbed by each individual
each day is the result of many personal choices and several
factors over which we have very little direct control. Where
one works and lives, what one eats and drinks, all have a pro-
found influence on the exposure level to pollutants. Our per-
sonal preferences also make a difference in the magnitude of
our exposure to the pollutants entering the home through drink-
ing water. Table 2 provides an example of the potential dif-
ferences in exposure of family members from drinking water
pollutants. This table has been prepared for a theoretical
contaminant (whose characteristics are similar to those of
trichloroethylene) which is one hundred percent absorbed for
all oral intake, tifty percent for respiratory intake and is
29
-------
FIGURE 17
DISTRIBUTION OF A CONTAMINANT QMG/L) ACCORD: -,G To
HOUSEHOLD WATER USE PATTERN
HOUSEHOLD
BATHING LALTJCRY CLEANING
333 mg 133 Bg
A
/\ \
965
cg/eay
i
A
f
38 =g
CCOKIN-
74 ng
^
*^ 379 as
|\ Jp r^P v ITT *
^^y
8 eg
LINKING
LUIDS
-------
CM
oc
u
ss
9
S
tt
£ > i
og
tfc
z
[s3 * <
CD
22
jO
OH
ecj
r? '
OT 2§
oa <
feO
OOS
*J ^H
w ?r
o
o o
£ o
p"^
CaJ >**/
V4 C^
HZ
2 H
e£3
00
U Cu
r«i
,j
i i
H
rs
o
^
<
/Zn
01 Jrf
(
2°
CO
01
f-l s-\
co at
r.*
f-i r^
TJ
^
X 0)
to i I
"O cO
"-.. E'-v
co ai oc
e r», »/
CO
l-i J-> O
n1 'a ^°
jj -^
O <
*H
E
^^
co C
O 4» ^
o cv oc
4J _*^
"C 1
0) 0) CM
Q Ul (f)
iJ Cb ^-^
O
to
Xl
^^v
T3 00
0> ^
U-l
«S
f-i
*J ^J
E^
cfl
O ^^
(h C
r-4
O 0
O HI
O O
O ul
O O
o r-«
CM
0 0
l^ ^^
^
o o
00 -<
o
0>
C
O
1-1
w
c
C 0)
0
i-* U-l
4J O l-l
W i-l
o> c w
00 O
C »^ ^J
f-l 4J O
CO O
"O *"" ^ T3
H CO C
3 x: -H
^ c
£x4 H^
CO
E
to co
1 CM 1 O- OC
I I O
u
E
CO
E
CO
O II I** OO
Oil O
CM ^
O
^
E
-^^
to
E
CO
Vi
O il O OC
O il -< O
C*"^ V^
o
»H
E
o w
0 E
o en o"
O II !M P
r-l 1 O 1 OC
f^ 00
f-l U
u
t-l
E
In
o w
W
1 II O OO
t CN 1 CM O
O »-<
0
i-l
E
o
0)
CO
O 0) VX
t-» to CC
o c 0*0
c o o-^.
0) fH 00
4J T3 ^!
y-i o. CD ^.
o u u x> to
f-i O *" E
C cO co oO O cQ
o xi ap c co vj
i-l Wi CO C f^ X> OO
AJ o> -HE < o
ce 3 -^ x; e «-
i~ < O co *J 1-1 ( (J
CO XI E *" 5 co "-i
x; co t co c/5 J-1 E
c o> o
i-i O H
W
. J
rC
E
4_)
-^
3
c
co
OC
c
QJ
3
O
u:
tc
-------
absorbed in direct proportion to water rlux across the skin.
These simplifying assumptions allow a relative comparison
between the exposure to family members of different ages and
activity patterns. Calculations using standard respiration
and fluid intake show that drinking water contaminants can be
responsible for higher levels of exposure per unit of body
weight among children than adults and that respiratory exposure
for volatile chemicals during and immediately after showering
can be of the same magnitude as those created by direct inges-
tion. Prolonged swimming with its associated oral and dermal
absorption also has the potential for significantly increasing
exposure to drinking water contaminants.
The analysis which follows pr st its the di t ibution of
concentration of selected volatile organic chemicals in a
profile ot U. S. drinking water supplies and provides a basis
for estimating the resultant levels of human exposure.
There are approximately 60,000 public water supplies in
the United States serving roughly 214 million people. Three
major federally sponsored monitoring surveys have been con-
ducted nation-wide to obtain information on the concentration
of selected volatile organic chemicals appearing as pollutants
in the potable water being provided to the nation's cities and
towns. During these surveys, water supplies serving as tew as
25 people were sampled. The data from these surveys have been
assembled into a national profile that has been projected to
32
-------
all public water supplies. No information is available from
which to estimate conditions in private wells or systems not
covered by the definition of a community water supply (serving
more than 2b people on 15 connections).
The risk analysis being presented in this paper deals with
a selected list of volatile organic chemicals which represents
a subset of the VOC's for which monitoring data are availalbe
(lexulc j'i . In general the level of volatile chemicals found
in surface water is lower than that in groundwater because of
dilution and volatilization in the atmosphere during transport
while *>"= *^oanr*o of these processes may cause groundwater
levels to be very high. See Appendix C for extrapolation of
the ;t iii fence data (Letkiewicz, 1983). A description of the
monitoring data assembled into an analysis of chemical occur-
rence and the subsequent national projections for each chemical
follow.
In compiling the data, existing data were averaged if they
were for the same site. No raw water data were used. Only
data for finished and distribution water samples were used.
The data were tested for biases by intercomparing the survey
data. Analytical functions were chosen that best tit the
exist1'"2 Hat-a.
The following discussion tor specific contaminants relies
on information and technical details from Letkiewicz, et al.
33
-------
r
1
',
"
*
«
li
|
CO 1
U3 3
>-i 1
H j
0£ ?
0,
t-i f
[
u I,
f-1
< '
3 :
0 J
z k
l-l f
«n .2 (
i "^
W OS i
J 0 f
*
0) r^ |
i-i f* O CO i-l
tV ^3 V * tJ £ OI
«8 ro <0 -r< i-i *
3 3 E OOrH
0-
0>
CJ
«0 C I ^.
U-l CO O CO l-l
Jj f* t-i E 4*
to 01 fi u fi
r E oo-i
CO
g
»c
55
1 "^^
O co id
Xi_i B fit
i-i e vv
co O IB 4->
£ -H VJ --H
E 00-1
^ i --
CO ' f-l O CO i-i
i. ' « i \ i_i e fli
tw ^j *-* M t vy
0) W C 0 co 4J
4J CO CO -i-l l-i -^t
CO 3 E 00 r-l
3 CX
"O
C C i -*.
3 co O co l-i
o ._i p g a>
W AJ O w 4*^
O 0) -r< lJ f-l
£ E ooi
CO
o
fit
MM
X
z
CO
o
f-l
c
CO
30
^
0
>*
l-l
CM
o
CM
o
CO
[
CO
CM
CM
O
m
i i
i i
NO
CM
i 1
in
i i
CO
CM
01
C
01
N
C
CO
ro
CM
4-
1-4
m
o
^
CM
,_
m
CM
r^
^
CO
o
CM
CM
-1
O
co
^^
00
ON
01
C
Oi
f-l
y,
^ff
u
01
o
0
II
JC
u
H
o
1
CM
»
r-l
in
O
m
o
tf
o
0
CM
co
O
co
vO
ON
O
co
O
f*
CM
O^
CO
o>
0)
c
^^
4J
OI
o
i-l
o
JZ
u
f-l
o
1
1 1
«
-H
o
ON
1 1
CM
r-l
ON
O
CO
00
l-l
0
co
r-l
^_|
ON
O
NO
o
CM
rH
f-l
O
o
0)
c
CO
^3
4J
0)
O
l-l
o
1 1
rr
O
H
O
1
CM
H
O
CM
CO
ON
O
vO
O
m
«
vO
ON
CM
O
O
NO
«-<
f 1
o-
0
^
o
^
^^
ON
ON
0)
C
Oi
f-l
^^
J=
4J
01
o
o
f 1
JS
u
fH
Wi
H
0
CO
NO
f-l
^
ON
O
NO
H
1
CO
NO
r4
m
^
m
o
0
co
NO
VO
0)
o
f-l
1^4
O
f-l
£,
CO
4J
0)
H
C
o
JO
w
CO
u
CM
co
r*
O
tf
0
f-l
00
NO
CM
ON
NO
00
f-l
00
0
l-l
[^
f^
m
0)
c
01
t-4
w
f.
4-1
0*
o
o
r-l
.e
u
CO
.u
0*
H
NO
CM
0
CM
CO
o
rx
;M*
o
m
-4
NO
NO
ON
00
O>
m
ON
O
^^
00
o>
f-l
l-l
o
I 1
JC
o
1
^s^
c
f-l
>
co
ai
4-1
CO
>^
CO
f-l
0
p
01
r\
E
3
C
01
X-
4J
II
z
-------
(1963). A general overview of the amount ot each contaminant
released to air, land and water is shown in Table 4. The
tables in Appendix C list the estimated concentrations of
drinking water according to size of drinking water supply. To
generalize the analysis, the level 0.5 micrograms/liter was
chosen as the lower limit of quantitation. The level of 0.5
micrograms/liter was selected for these analyses because it
was a relatively common minimum quantifiable concentration
all surveys.
35
-------
TABLE 4
ESTIMATED AMOUNT OF EACH VOC RELEASED TO AIR, LAND AND WATER
(DATA FROM LETKIEWICZ, 1983)
Contaminant
Benzene
Carbon Tetrachloride
1,2-Dichloroethane
1,1-Dichloroethylene
Perchloroethylene
Trichloroethylene
Amount Released to:
Air Land Water
lii,00u-28b,000
43,000
11,900
2,000
153,000
103,000
770 2,600-13,500
1,200
100
0
104,000
7,500
25
250
2
400
190
36
-------
Benzene
Approximately 90% ot the benzene used in 1978 was a feed-
stock in the production of three chemicals (ethylbenzene,
cumene and cyciohexane). Other more minor uses included the
production of nitrobenzene, chlorobenzenes, alkylbenzenes,
maleic anhydride, and biphenyl, and use as a solvent and
pesticide ingredient.
Total releases ot benzene from production, indirect pro-
duction, and use processes in 1978 were estimated to be
111,000-288,GOO kkg to air, 770 kkg to land, and 2,600-13,500
kkg to water. Major benzene releases to air resulted from
incomplete combustion of gasoline. Major releases to water
occurred through oil spillage and solvent use. Benzene
entering water is expected to be volatilized and to undergo
degradation in the atmosphere.
Four Federal surveys were used to estimate the levels ot
benzene in the nation's public drinking water supplies: the
National Organics Monitoring Survey (NOMS), the National
Screening Program for Organics in Drinking Water (NSP), the
1978 Community Water Supply Survey (CWSS) and the Groundwater
Supply Survey (GWSS). Additional state data are also reported,
but were not used in developing the national estimates since
these data came from only a few states and were not well-
characterizea with respect to water type and system size
37
-------
sampled.
The data obtained from the tour Federal studies was com-
bined and sorted by source and size category in order to
develop estimates ot the number of systems nationally in each
source/size category containing benzene within various concen-
tration range. The methodology used is presented in Appendix
D. The national estimate of systems was used to calculate the
number of persons exposed to public drinking water containing
benzene levels in those ranges.
Using the combined Federal survey data and the delta dis-
tribution model fcr projecting national occurrence, it was
estimated that about 9b.7% of the groundwater systems (Table
C-l) of all sizes contain either no benzene at low levels and
it was not possible to estimate how many are free of benzene
contamination. Of the estimated 635 systems estimated to have
levels higher than 0.5 micrograms/liter, 55 (0.1% of total
groundwater systems) are estimated to have concentrations >10
micrograms/liter. and 2 systems «0.01%) to have levels >50
micrograms/liter. It is estimated from the Federal data that
no groundwater systems will have concentrations >80 micrograms/
liter. It should be noted, however, that the state data in-
dicate that some groundwater supplies may have benzene present
at substantially higher levels.
For surface water supplies (Table C-2), it is estimated
38
-------
that about 97% will have either no benzene present or levels
<0.5 mxcrograms/liter. It is estimated that 301 surface water
systems have levels between 0.5 and 5 raicrograms/liter (2.7%
of total surface water systems); none are estimated to have
benzene above 5 micrograms/liter.
It is important to note that some of the Federal data used
in computing the national estimates are from samples held for
a prolonged period of time prior to analysis, with possible
biodegradation of benzene. Therefore, these projections of
national occurrence may underestimate actual contaminant
levels.
Using the combined Federal survey data for surface water
and groundwater supplies, it was estimated that 209,590,000
persons (97.7% ot the population served by public drinking
water systems) are receiving water either free of benzene
combination, or having levels less than 0.5 micrograms/liter.
Of the 4,829,000 persons (2.3%) receiving water containing
benzene levels ^ 0.5 micrograms/liter, an estimated 155,000
«0.1%) are exposed to levels >5 micrograms/liter. About
3,200 (<0.1%) are estimated to be exposed to levels >30 micro-
grams/liter. Of the approximately 4.7 million people estimated
to be eA^uocJ uo levels ranging from 0.5 to 5 micrograms/liter,
81% obtain vatcr from surface water supplies. However, all
exposure to benzene in drinking water at levels about 5 micro-
grams/liter i6 projected to be from groundwater sources.
39
-------
Carbon Tetrachloride
Production ot carbon tetrachloride in 1981 amounted to
329,000 kkg. In addition to its direct manufacture, carbon
tetrachloride may be produced indirectly during the production
or breakdown of other chlorinated chemicals.
The major use of carbon tetrachloride is the production of
chlorofluorocarbons 11 and 12, which was estimated to account
for 87% of its use in 1978. Although chlorofluorocarbon pro-
duction has declined since 1978, their production still
accountec tor 91% ot carbon tetrachloride consumption in 1981.
It is also used in grain fumi',ant formulations and has numerous
other minor uses. Fumigant use was est m,.ted to account for
the majority of the environmental releases of carbon tetra-
chloride in 1978. Total releases of carbon tetrachioride troro
production and use processes in 1978 were estimated to be
43,000 kkg to air, 1,200 kkg to land, and 25 kkg to water.
Carbon tetrachloride is quite stable in the environment.
Quantities of carbon tetrachloride entering surface water are
expected to be volatilized and to undergo eventual degradation
in the stratosphere.
Six Federal surveys were used to estimate levels of carbon
tetrachloride in the nation's public drinking water supplies:
the National Organics Reconnaissance Survey (NORS), the
National Organic Monitoring Survey (NOMS), the National Screen-
40
-------
ing Program for Organics in Drinking Water (NSP), the 197«
Community Water Supply Survey (CWSS), the Rural Water Survey
(RWS), and the Groundwater Supply Survey (GWSS). Additional
state and miscellaneous data are also reported, but were not
used in developing the national estimates tor several reasons.
These data came from only a few states and were not well-
characterized with respect to water type and system size sam-
pled. Since states do not monitor water supplies routinely,
the reported data are believed to represent sites of recognized
contamination and could, therefore, bias the national estimates.
The data obtained from the six Federal studies was com-
bined and sorted by source and size category in order to
develop estimrtes of the number of systems nationally in each
source/size category containing carbon tetrachloride within
various concentration ranges. The methodology used is provided
in Appendix D. The national estimate of systems was used to
calculate the number ot persons exposed to public drinking
water containing carbon tetrachloride levels in those ranges.
There was an apparent relationship between system size
(based on population served) and the frequency of occurrence
of carbon tetrachloride at levels ^0.5 micrograms/liter in
the combined survey data for both groundwater and surface
water. Generally, a higher frequency of occurrence of carbon
tetrachloride was found in the large size categories than in
the small and medium size .categories.
41
-------
Using the combined survey data and the multinomial
approach for projecting national occurrence, it was estimated
that about 99% ot the groundwater systems (Table C-3) of all
sizes contain either no carbon tetrachloride or levels less
than 0.5 micrograms/liter. It is not possible, however, to
estimate how many of these systems actually contain carbon
tetrachloride at low levels and how many are tree of carbon
tetrachloride contamination. Of the 355 systems estimated to
have levels higher than 0.5 micrograms/liter, 110 (0.2% ot
total groundwater systems) are projected to have concentrations
>5 micrograms/liter; none are estimated to have levels >20
micrograms/liter. Although a greater percentage of the very
large systems are expected to be contaminated, in absolute
numbers more of the smaller groundwater systems are expected
to have contamination because of the very large number of
small groundwater systems in the United States.
For surface water supplies (Table C-4), it is estimated
that about 95% will have either no carbon tetrachloride present
or levels <0.5 micrograms/ liter. It is estimated that 575
surface water systems have levels X).5 micrograms/liter (5.1%
of total surface water systems); 28 supplies (0.2%) are
estimated to have levels >5 micrograms/liter. Only 9 «0.1%)
are expected to have levels above 20 micrograms/liter and none
are estimated to have carbon tetrachloride above 30 micrograms/
liter.
42
-------
Using combined data troro surtace water and groundwater sup-
plies, it was estimated that 187,609,000 persons (87.5% of the
population served by public drinking water systems) are receiv-
ing water with no carbon tetrachloride or levels at less than
0.5 nicrograms/liter. Of the 26,810,000 persons (12.5%)
receiving water containing carbon tetrachloride levels X).5
micrograms/iiter, an estimated 2,067,000 (1.0%) are exposed to
levels >5 micrograros/liter. About 655,000 (0.3%) are estimated
to be exposed to levels >20 micrograms/liter; no exposure above
30 microgranis/liter is expected. Of the approximately 27
million people estimated to be exposed to levels X).5 micro-
grams/ liter, 92% obtain water from surface water supplies.
All exposure to carbon tetrachloride in drinking water at
levels above 20 micrograms/liter is projected to be from u -
face water sources.
1,2-Dichloroethane
1,2-Dichloroethane (CH2C1CH2C1), also known as ethylene
dichloride, is a colorless, flammable liquid with a pleasant
odor. It has a high vapor pressure (63.8 mm Hg at 20° C) and
a low water solubility (0.87 g/100 g at 20° C).
Production of 1,2-dichloroethane in 1980 amounted to
5,050,000 kkg. In addition to its direct manufacture, 1,2-
dichloroethane may be produced indirectly during the production
or breakdown of other chlorinated chemicals or during the
A3
-------
chlorination of water and wastewater. The production of 1,2-
dichloroethane was estimated to account for approximately half
of the environmental releases of 1,2-dichloroethane in 1978.
The majority of 1,2-dichloroethane production facilities occur
in Louisiana and Texas.
The major use of 1,2-dichloroethane is as a feedstock in
vinyl chloride monomer production, which was estimated to
account for 85% ot its use in 197y. Other applications are
numerous and include use as a feedstock for other chlorinated
chemicals and gasoline lead scavenging. Total releases o±
1,2-dichloroethane from production and use processes in 1978
were estimated to be 11,900 kkg to air, 100 kkg to land, and
250 k g LO water. 1,2-Dichloroethane entering water is
expected to be volatilized and to undergo degradation in the
troposphere.
Five Federal surveys were used to estimate levels ot
1,2-dichloroethane in public drinking water supplies in the
United States: the National Organics Reconnaissance Survey
(NORS), the Program for Organics in Drinking Water (NSP), the
1978 Community Water Supply Survey (CWSS), and the Groundwater
Supply Survey (GWSS). Additional state data were reported for
comparison, but were not used in developing the projections
for several reasons. These data came from only a few states
and were not adequately characterized as to water type and
system size. Since states do not monitor water supplies
44
-------
routinely, the reported data are believed to represent sites
of suspected contamination.
The data obtained from the tive Federal studies were com-
bined and sorted by source and size category to allow for pro-
jections o± the number of systems in each source/size category
containing a specified level of 1,2-dichlorethane and the
number of persons exposed to public drinking water containing
1,2-u.Lt.uIui.vei.iieute at a specified level. The methodology used
is presented in Appendix D.
I" £o~o»-ai a relationship was seen between groundwater
system size and the likelihood of 1,2-dichloroethane contami-
nation. CI the groundwater systems serving 5 3,300 persons,
0.6% contained detectable 1,2-dichloroethane, while 1.3% of
the systems of intermediate size (3,301-10,000 persons) and
3.6% of the systems of large and very large size (>10,000
persons) contained detectable 1,2-dichloroethane.
Insufficient positive data were obtained for 1,2-dichloro-
ethane in both surface water (Table C-5) and groundwater (Table
C-6) systems to allow for statistical projections of the number
ot systems in each source/size category containing a specitied
level of 1,?-Hirhloroethane. However, since all 1,2-dichloro-
ethane measurements in groundwater and surface water were below
20 micrograms/liter, it was estimated that there is effective-
ly no i,2-dichloroethane contamination at levels above 20 raicro-
45
-------
grams/liter. In addition, most ot the systems in the United
States are projected to have levels of 1,2-dichloroethane
below 1.0 roicrograms/liter.
It was estimated that all 195,595,000 persons using public
drinking water supplies in the United States, including
3,129,520 bottle-fed intants, would be exposed to drinking
water levels at or below 10 micrograms/liter. Because of
limitations in the data, the number ol persons exposed to more
precise concentration intervals between 0-10 micrograms/liter
could not be estimated. However, the majority of persons
exposed to 1,2-dichloroethane in drinking water supplies are
projected to receive concentrations less than 1.0 micrograms/
'iter.
1,1-Dichloroethylene
1,1-Dichloroethylene (CH2=CCl2), also known as vinylidene
chloride, is a colorless liquid. It has a high vapor pressure
(495 mm Hg at 20° C) and a low water solubility (0.035 g/100 g
at 25° C).
Production ot 1,1-dichloroethylene in 1978 was estimated
to be 144,200 kkg. This figure included 1,1-dichloroethylene
captively produced for the production of methyl chlorotorm.
However, 1,1-dichloroethylene appears no longer to be used in
methylchloroform production. In addition to its direct manu-
46
-------
facture, 1,1-dichloroethylene may be produced indirectly
during the production of other chlorinated chemicals.
The major uses of 1,1-dichloroethylene in i97b were the
production of methyl chloroform and the production of copoly-
mers for use in resins, coating latex, and the manufacture of
modacrylics. A minor application was its use in the production
of chloroacetyl chloride, a component ot mace and tear gas.
Total releases of 1,1-dichloroethylene from production and
use processes in 1978 were estimated to be 2,000 kkg to air,
0 kkg to land, and 2 kkg to water. However, the production
and use of methyl chloroform was estimated to account for 1,300
of the 2,000 kkg of 1,1-dichloro :l /lene released to air in
197b, and methyl chloroform no longer appears to be produced
by this process. The production of copolymers is currently
estimated to account for almost all environmental releases ot
1,1-dichloroethylene. Quantities of 1,1-dichloroethylene
entering water are expected to be volatilized and to undergo
rapid degradation in the troposphere.
Two Federal surveys were used to estimate levels of 1,1-
dichloroethylene in the nation's public drinking water sup-
plies: the National Screening Program for Organics in Drinking
Water (NSP) and the Groundwater Supply Survey (GWSS). Addi-
tional state data are also reported, but were not used in
developing the national estimates. These data came from only
47
-------
a few states and were not well-characterized with respect to
water type and system size sampled.
Using the combined survey data and the multinomial
approach for estimating national occurrence, it was calculated
that about 9b% of the groundwater systems (Table C-7), of all
sizes contain either no 1,1-dichloroethylene or levels less
than 0.2 micrograms/liter. It is not possible, however, to
estimate how many of these systems contain 1,1-dichloroethylene
at low levels and how many are free ol 1,1-dichloroethylene
contamination. Of the estimated 858 systems expected to have
levels higher than 0.2 micrograms/liter, 81 (0.2% of total
groundwater systems) are projected to have concentrations >5
micrograms/liter; none it expi ;t ;d to have levels >10 micro-
grams/liter. The state data, however, indicate that there may
be some supplies with levels substantially higher than 10
micrograms/liter.
For surface water supplies (Table C-8), it is estimated
that about 99.7% will have either no 1,1-dichloroethylene
present or levels <0.2 micrograms/ liter. It is estimated
that 35 surface water systems have levels >0.2 micrograms/liter
(0.3% of total surface water systems).
It is important to note that some of the data used in
computing the national estimates are rrom samples held for a
prolonged period of time prior to analysis, with possible
48
-------
biodegradation ot 1,1-dichloroethylene. Therefore, these
projections of national occurrence may underestimate actual
contaminant levels.
Using combined data from surface water and groundwater sup-
plies, it was estimated that 209,630,000 persons (97.8% of the
population served by public drinking water systems) are receiv-
ing water with no 1,1-dichloroethylene or levels less than 0.2
micrograms/liter. Of the4,7B9,OOU persons (2.2%) receiving
water containing 1,1-dichloroethylene levels >0.2 micrograras/
liter, an estimated 52,OOU «0.1%) are exposed to levels >5
micrograms/liter. No individuals are estimated to be exposed
to levels >10 micrograms/liter. Of the approximately 4.7
million people estim ;e i to v>e exposed to levels ranging from
0.2 to 5 micrograms/liter, 52% obtain water from surface water
supplies. All exposure to 1,1-dichloroethylene in drinking
water at levels aboue 5 roicrograms/liter is expected to be from
groundwater sources.
Perchloroethylene
Perchloroethylene (CCl2*CCl2), also known as tetrachloro-
ethylene, is a clear, non-flammable liquid with an ethereal
odor. Commercial perchloroethylene has a high vapor pressure
(2U mm Hg at 26.3° C) and a low water solubility (0.11 g/100 g
at 25° C), two properties which make it a useful solvent.
49
-------
Production ot perchloroethylene in I9bl amounted to
313,000 kkg. In addition to its direct manufacture, perchloro-
ethylene may be produced indirectly during the production or
breakdown of other chemicals.
The major use ot perchloroethylene is in dry cleaning and
textile processing, which were estimated to account for 63% of
its use in 1978. Other more minor applications include metal
cleaning, fluorocarbon production, use as a specialty solvent,
and use as an antiheirointhic. Dry cleaning and metal cleaning
were estimated to account for almost all environmental releases
of perchloroethylene in 1978. Areas ot releases irom these
processes are expected to parallel areas of greatest population
density. Total releases of perchloroethyl .ie f ot production
and use processes in 1978 were estimated to be 153,000 kkg to
air, 104,000 kkg to land, and 400 kkg to water. Some per-
chloroethylene may enter drinking water systems through the
use of vinyl-lined A/C pipe. Quantities of perchloroethylene
entering water are expected to be volatilized and to undergo
degradation in the troposphere, while quantities in disposal
sites (e.g., landfills) appear to be capable of moving through
the soil to underground aquifiers.
Five Federal surveys were used to estimate levels of per-
chloroethylene in the nation's public drinking water supplies:
the National Organic Monitoring Survey (MOMS), the National
Screening Program for Organics in Drinking Water (NPS), the
50
-------
1978 Community Water Supply Survey (CWSS), the Rural Water
Survey (kwS), and the Groundwater Supply Survey (GWSS). Addi-
tional state data are also reported, but were not used in
developing the national estimates for several reasons. These
data cane trom only a rew states and were not well-characterized
with respecc to water type and system size sampled. Since
states do not monitor water supplies routinely, the reported
data are believed to represent sites of recognized contami-
nation and couia, therefore, bias the national estimates.
There is some relationship between system size (based on
pwpuic.'.;^ served) and the frequency ot occurrence ot per-
chloroethylene at levels >Q.5 micrograms/liter in the combined
survey it ror both groundwater and surface water. (The level
of 0.5 micrograms/liter was selected for these analyses because
it was a relatively common minimum quantifiable concentration
across all surveys.) The frequency of occurrence of systems
with levels >^Q.5 micrograms/liter in groundwater for various
system sizes was 3.6% (25-500 served), 2.2% (501-3,300 served),
8.5% (3,301-10,000 served), 5.6% (10,001-100,000 served), and
12.8% (>100,000 served). Statistical tests showed that the
difference between the two smallest size categories was not
significant, nor was the difference between the two largest
siic .at^.gcrics statistically significant. For surface water,
the freqve^y of occurrence of systems with levels >Q.5 tnicro-
grams/liter was 0% (25-50U served), 1.7% (501-3,300 served),
0% (3,301-10,000 served), 6.1% (10,001-100,000 served), and
51
-------
5.3% (>100,000 served). In the case of surtace water, the
differences among the three smaller size groups was not statis-
tically significant, nor was the ditterence between the two
larger size categories.
Using the combined survey data and the multinomial
approach for projecting national occurrence, it was estimated
that about 97% of the groundwater systems (Table C-b), of
all sizes contain either no perchloroethylene or levels less
than 0.5 micrograms/liter. It is not possible, however, to
estimate how many of these systems contain perchloroethylene at
low levels anu how many are free of perchloroethylene contami-
nation. Of the estimated 1,552 systems estimated to have
levels higher than 0.5 micrograms/liter, 322 (0.7% of total
groundwater systems) are projected to have concentrations >0.5
micrograms/liter; only 3 systems have levels >30 micrograms/
liter. It is estimated that no systems will have perchloro-
ethylene concentrations >70 micrograms/liter. Although a
greater percentage of the large systems are expected to be
contaminated, in absolute numbers more small groundwater
systems are expected to have contamination because of the very
large number of small groundwater systems in the United States.
For surface water supplies (Table C-10), it is estimated
that more than 98% will have either no perchloroethylene
present or levels <0.5 micrograms/liter. It is estimated that
180 surrace water systems have levels between 0.5 and 5 micro-
52
-------
grams/liter (1.6% of total surface water systems); none are
estimated to have perchloroethylene above 5 micrograms/liter.
The state and miscellaneous data generally support the
national projections. A few instances of groundwater contami-
nation at levels >70 micrograms/liter are reported for sup-
plies in New Jersey; however, the type of water sampled in
these cases is generally unknown. With respect to surface
water. two supplies in Connecticut were reported to have
levels >5 micrograms/liter (6 micrograms/liter; 31.2 micro-
grams/liter) .
It is important to note that some o± the data used in
computing the national estimates are from samples held for a
prolonged period ot time prior to analysis, with possible
biodegradation of perchloroethylene. Therefore, these projec-
tions of national occurrence may underestimate actual contami-
nant levels.
Using combined data from surface water and groundwater
supplies, it was estimated that 202,989,000 persons (94.7% of
the population served by public drinking water systems) are
receiving water with no perchloroethylene or levels at less
than 0.5 micrograms/liter. Of the 11,430,000 persons (5.3%)
receiving water containing perchloroethylene levels X).5 micro-
grams/liter, an estimated 874,000 (0.4%) are exposed to levels
>5 micrograms/liter. About 440,000 (0.2%) are estimated to be
53
-------
exposed to levels >10 micrograms/liter, and 105,000 (<0.1%)
are estimated to be exposed to perchloroethylene in drinking
water at levels in excess of 60 oicrograms/liter. Of the
approximately 10.5 million people estimated to be exposed to
levels ranging from 0.5 to 5 micrograms/liter, 67% obtain
water from surface water supplies. All exposure to perchloro-
ethylene in drinking water at levels above 5 micrograms/liter
is projected to be from groundwater sources.
Trichloroethylene
Trichloroethylene (Chul-CCl2) is a clear, non-flammable
liquid with an ethereal odor. Commercial trichloroethylene has
a high vapor pressure (58.7 mr, Hg at 20* C) and a low water
solubility (0.11 g/100 g at 25° C), two properties which make
it a useful solvent.
Production of trichloroethylene in 1981 amounted to
117,100 kkg. In addition to its direct manufacture, trichloro-
ethylene may be produced indirectly during the production or
breakdown of other chemicals.
The major use of trichloroethylene is metal cleaning,
which was estimated to account for 83% of its use in 1978.
Other more minor applications include use as a poiyvinyl
chloride chain terminator and uses in adhesives, textiles, and
paints. Metal cleaning was estimated to account ror almost
54
-------
all environmental releases or trichloroethylene (95%) in I97b.
The majority of metal cleaning operations occur in the north-
eastern and midwestern United States. Total releases of tri-
chloroethylene from production and use processes in 1978 were
estimated to be 103,000 kkg to air, 7,500 kkg to land, and 190
kkg to water. Quantities of trichloroethylene entering water
are expected to be volatilized and to undergo degradation in
the troposphere, while quantities in disposal sites (e.g.,
landfills, lagoons) appear to be capable of moving through the
soil to underground aquifiers.
Five Federal surveys were used to estimate levels of
trichloroethylene in the nation's public drinking water sup-
plies: the National Organics Monitoring Survey (NOMS), the
National Screening Program for Organics in Drinking Water
(NSP), the 197b Community Water Supply Survey (CWSS), the
Rural Water Survey (RWS), and the Groundwater Supply Survey
(GWSS). Additional state and miscellaneous data are also
reported, but were not used in developing the national esti-
mates for several reasons. These data came from only a few
states and were not well-characterized with respect to water
type and system size sampled. Since states do not monitor
water supplies routinely, the reported data are believed to
represent sites ot recognized contamination and could, there-
fore, bias the national estimates.
There was an apparent relationship between system size
55
-------
(based on population served) and the frequency of occurrence
of trichloroethylene at levels X).5 nicrograms/liter in the
Combined survey data tor both groundwater and surface water.
(The level of 0.5 raicrograms/liter was selected for these
analyses because it was a relatively common minimum quanti-
fiable concentration across all surveys.) The frequency of
occurrence of systems with levels >Q.5 micrograms/liter in
groundwater for various system sizes was 2.3% (25-500 served),
3.8% (501-3,300 served), 5.0% (3,301-10,000 served), 7.2%
(10,001-100,000 served), and 25% (>100,000 served). The
differences among the three smallest size categories were not
statistically significant, however. For surface water, the
frequency of occurrence of systems with levels X)«5 micrograms/
liter wes 0% (25-500 served), 1.7% (501-3,300 served), 5.1%
(3,301-10,000 served), 10.6% (10,001-100,000 served), and 16.8%
(>100,000 served). In the case of surface water, the differ-
ences among the three smaller size groups was not statistically
significant, nor was the difference between the two larger
size categories.
Using the combined survey data and the multinomial
approach for projecting national occurrence, it was estimated
that about 97% of the groundwater systems (Table C-12), of all
sizes contain either no trichloroethylene or levels less than
0.5 micrograms/liter. It is not possible, however, to estimate
how many of these systems contain trichloroethylene contami-
nation. Of the estimated 1,632 systems expected to have
56
-------
levels higher than 0.5 micrograms/iiter, 4^1 (0.9% ot total
groundwater systems) are projected to have concentrations >5.0
micrograms/liter, and 133 systems (0.3%) have levels >50 micro-
grams/liter. It is estimated that only 64 systems (0.1%)
nationally will have concentrations >100 micrograms/liter.
t
Although a greater percentage of the large systems are expected
to be contaminated, in absolute numbers more small groundwater
systems are expected to have contamination because of the very
large r.ur.bcr cf small groundwater systems in the United States.
For surface water supplies (Table C-13), it is estimated
that about yb/fc will have either no trichloroethylene present
or levels <0.5 micrograms/ liter. It is estimated that 49b
surface water systems have levels X).5 micrograics/liter (4.5%
of total surface water systems); only 9 «0.1%) are expected
to have levels above 5 micrograms/liter; none are estimated to
have trichloroethylene above 40 micrograms/liter.
The state and miscellaneous data tend to support the
national projections, showing several instances of groundwater
contamination at levels >100 micrograms/liter while these data
show no surface water contamination above 20 micrograms/liter.
It is important to note that some of the data used in
computing the national estimates are from samples held for a
prolonged period of time prior to analysis, with possible
biodegradation of trichloroethylene. Therefore, these projec-
57
-------
tions of national occurrence may underestimate actual contami-
nant levels.
Using combined data irotu surface water and groundwater
supplies, it was estimated that 189,288,000 persons (88.3% of
the population served by public drinking water systems) are
receiving water with no trichloroethylene or levels at less
than 0.5 micrograros/liter. Of the 25,131,000 persons (11.7%)
receiving water containing trichloroethylene levels X)«5 micro-
grams/liter, an estimated 1,844,000 (0.9%) are exposed to
levels >5 micrograms/liter. About 212,000 (0.1%) are expected
to be exposed to levels >50 micrograms/liter, and 42,000
(<0.1%) are estimated to be exposed to trichloroethylene in
drinking water at levels in ex'er s of 100 micrograms/liter. Of
the approximately 23 million people exposed to levels ranging
from 0.5 to 5 micrograms/liter, 76% obtain water from surface
water supplies. However, of the 1.8 million people exposed to
levels >5 micrograms/liter, 62% use groundwater sources. All
exposure to trichloroethylene in drinking water at levels
above 40 micrograms/liter is projected to be from groundwater
sources.
Vinyl Chloride
Vinyl chloride (CH2«CHC1) is a colorless, sweet-smelling
gas. It has a high vapor pressure (2,580 mm HG at 20° C) and
a low water solubility (0.11 g/100 g at 28° C).
-------
Production ot vinyl chloride in 1961 was 3,117,OUO kkg.
The major use of vinyl chloride is in the production of poly-
vinyl chloride, which is then manufactured into a wide range
of products. The production of polyvinyl chloride polymers
and copolymers is estimated to account for the majority of
environmental releases of vinyl chloride (greater than 54,000
kkg in 1975). Unreacted vinyl chloride monomer present in
polyvinyl chloride products accounts for -small releases to
water and land. Quantities of vinyl chloride entering water
are expected to be volatilized and to undergo degradation in
the troposphere.
Three Federal surveys were used to estimate levels of vinyl
chloride in the nation's public drinking water supplies: the
National Organics Monitoring Survey (NOMS), the National
Screening Program for Organics in Drinking Water (NSP), and
the Groundwater Supply Survey (GWSS). Additional state data
are also reported, but were not used in developing the national
estimates. These data came trom only a few states and were
not well-characterized with respect to water type and system
size sampled.
Using the combined survey data and the multinomial
approach for projecting national occurrence, it was estimated
that about 99.9% of the groundwater systems (Table C-14) of
all sizes contain either no vinyl chloride or levels less than
1.0 micrograms/liter. It is not possible, however, to estimate
59
-------
how many of these systems contain vinyl chloride at low levels
and how many are free of vinyl chloride contamination. Of the
31 systems estimated to have levels higher than 0.5 micrograms/
liter, 17 (0.9% of total groundwater systems) are projected to
have concentrations >5.0 micrograms/liter. Three of these are
expected to have levels as high as 60-70 tnicrograms/liter, but
none are estimated to have levels >70 micrograms/liter.
For surface water supplies (Table C-15), it is also esti-
mated that about 99.9% will have either no vinyl chloride pre-
sent or levels <1.0 micrograms/liter. It is estimated that 12
surface water systems have levels between 1.0 and 5.0 micro-
grams/liter (0.1% of total surface water systems); none ara
estimated to have vinyl chloride above 5 micrograms/lif r,
It is important to note that some of the data used in
computing the national estimates are rrom samples held tor a
prolonged period of time prior to analysis, with possible
biodegradation of vinyl chloride. Therefore, these projections
of national occurrence may underestimate actual contaminant
levels.
Using combined data from surface water and groundwater
supplies, it was estimated that 212,497,000 persons (99.1% of
the population served by public drinking water systems) are
receiving water with no vinyl chloride or levels at less than
1.0 micrograms/liter. Of the 1,922,000 persons (0.9%) receiv-
60
-------
ing water containing vinyl chloride levels !>1.0 micrograms/
liter, an estimated 591,000 (0.3%) are exposed to levels >5
m'icrograms/liter. About ilb.OOO (0.1%) are estimated to be
exposed to levels >50 roicrograms/liter. No individuals are
estimated to be exposed to vinyl chloride in drinking water at
levels in excess of 70 micrograms/liter. Of the approximately
1.3 million people estimated to be exposed to levels ranging
from 1.0 to 5 micrograms/ liter, 65% obtain water from surface
water supplies. All exposure to vinyl chloride in drinking
water at levels above 5 micrograms/liter is projected to be
from groundwater sources.
61
-------
Population-Concentration
One of the ultimate goals ot this analysis is to provide a
profile of conditions in the United States answering the basic
question: "How many people are being served drinking water
containing each level of the pollutant from zero to the maximum
quantity detected?" The national occurrence profile can then
be converted into population-concentration estimates which sum-
marize the national experience.
The population-concentration is calculated as the sura of
the population exposed (P) times concentration (c) at various
concentration levels, or
summation over
population-concentration - n concentrations of (c^ x P^)
For example, if 10,000 persons were exposed to a chemical at 5
micrograms/liter and 4,QUO additional persons were exposed to
the same chemical at 10 inicrograms/liter, the population-
concentration calculated would be:
population-concentration - (10,000 persons x 5 micrograros/
liter) + (4,000 persons x 10
micrograms/liter; - 90rOOO
micrograms/liter x persons.
These calculations would be straightforward if the number of
individuals exposed to specific concentrations of the chemical
were known. However, it is not possible to determine the
specitic concentration to which a population is exposed.
62
-------
Instead, projections have been performed which estimate the
number of individuals exposed at various intervals of concen-
tration (e.g., >5-10 micrograms/liter).
One way to express the uncertainty created by data grouped
into intervals is to create upper and lower bounds it clusters
occurred at the top and bottom of each interval. This becomes
critical for the category ot occurrence labeled "less than"
where we do not know if the findings represent occurrence
below uetecLiun or an artifact of the analytical methods. Mean
best case and mean worst case total estimates were calculated
using the lower and upper bounds or the tirst concentration
interval (i.e., the values below the detection limit), and the
mean value or the other concentration intervals. The lower
and upper bounds of the first concentration intei a. were used
rather than the mean value because of the higher uncertainty
in knowing where the values fall in this interval in comparison
to the other intervals. Example calculations tor vinyl
chloride follow:
Mean best case --
118 x 103 x 65 +
';?2 x 103 x 7.5 -H 1331 x ll)3 x 2.75 - 14.87U.25U
= 1.5 x 10? person x micrograms/liter
Mean worst case --
118 x 103 x 65 + 472 x 103 x 7.5 +
1331 x 103 x 2.75 + 212, 497 x 103 x 0.5 - 121,118,750
= 1.2 x 1U8 person x roicrograms/iiter
63
-------
Calculation or the mean best and mean worst population
concentration totals for each chemical are presented in Table
C-18.
Another way of dealing with uncertainty is to assume a
randon distribution of data within the intervals and compute
a standard error around the mid-point.
Using this method the population concentrations and stan-
dard errors can be calculated as rollows: Let k be the number
of systems sampled for a given contaminant. Suppose these k
systems arc broken down into ra distinct subgroupings as follows
Let i denote the ith sub-group, i*1 n. Let sub-group
i have: (1) ct a., the mid-point of the concentration (c^ =7.5
if in the 5 to 10 range); (2) m^ as the number ot persons
served by systems in the ith range group (this must be avail-
able); and (3) fi as the frequency ot systems in the ith
range group. Then, an estimate of the population concentration
is PC - population-concentration « summation over i from 1 to n
of (c^m^f^) and its standard error can be estimated by:
s.e. (PC) =
(summation over c^m^f. (1 -fi) - 1 summation c< CiiDjiDif < f^ )1 /2
i from 1 to n) X l 1 IT' V for i not l J i J 1 J
- to j
We have k = summation over f^. Also, the c^ for the first
interval (0 to .5) would again be .25. The advantage to this
64
-------
approach is that it also allows an estimate of the variability
(the standard error) of the estimated PC. The results of
using this relationship are shown in Tfble 5.
Factors in Human Exposure
Contaminants present in drinking water are transferred to
people through the fluids they ingest, the enriched indoor air
they inhale, and the contaminated water coming into contact
with their skin. The intake of air and water are governed by
metabolic rates reflecting age, sex, and physical condition
and modified by personal activity. The assumptions made in
this analysis about the magnitude of intake over a lifetime
for the U. S. population provide a ba.is for a population
risk analysis may be at wide variance for specific individuals,
Ingestion of Water
Risks projected for individual chemicals are based upon
the chronic ingestion of 0.03 liters of drinking water and its
contaminant, per kilogram of body weight and that 100% of the
ingested chemical as absorbed. The factor for ingestion was
developed using a standard of 2 liters intake per 70 kilograms
adult.
65
-------
TABLE 5
POPULATION CONCENTRATIONS
(in units o± persons x micrograms/liter)
TOTAL
BENZENE
ground (22 4-0.8) x 106
surface (34 + 1.7) x 10& (56 + 2) x
CARBON TETRACHLORIDE
ground (24 + 1.3) x 10)6
surface (86 + 7.1) x 106 (120 + 7) x 10&
1,1-DICHLOROETHYLENE
l,2-DICHLOROEr"H NE
ground (13 + 1.0) x 106
surface (49 + 1.2) x 106 (62 ± 2) x 106
ground (21 +0.8) x 106
surface (49 + 2.8) x 106 (70 + 3) x 10&
TETRACHLOROETHYLENE
ground (39 +3 ) x 106
surface (1.3 + 0.2) x 106 (40 + 3) x 10b
TRICHLOROETHYLENE
VINYL CHLORIDE
ground (70 +2.8) x
surface (74 +8.4) x 10& (144 + 9) x 10&
ground (4b + 2) x 106
surface (55 + 2) x 106 (103 + 3) x
66
-------
Individuals may experience many times this intake rate.
Formula-fed infants and young children have average intake
rates as high as 8 times greater than those of "average adults".
Adults in tropical areas may consume twice as much liquid as
the "average" as may the athletically inclined when engaged in
strenuous physical activity. The diseased may consume many
fold more water than is indicated. Social behavior ritualized
around the drinking of tea or coffee may lead to increased
water consumption.
A recent Canadian study shows that children under 5 years
of age, 10% of the children 6-17 years old and 2% of the adult
population consume more than 0.03 liters/kilograms/day of
drinking water (Canadian 1 81).
Inhalation
Each ot the chemicals mentioned in this analysis have been
shown to transfer from water into air if the water is heated
or aerated (Sorrell, 1982). Monitoring data indicate that
this process is continuous within the home leading to an im-
mediate enrichment ot respirable air at the point ot water use
and a diffusion throughout the home.
Showers taken within an enclosed bathroom using approxi-
mately 20-30 gallons of water containing volatile organic
compound result in the liberation of all or a portion ot the
67
-------
compound into the air. A person who showers inhales air con-
taining a highly enriched level erf the chemical while within
the bathroom.
Exposure is due to showering and other uses of water in
the hone. These sources mix with general home air resulting
with some higher level of the pollutant volatilized into the
indoor air.
Tne level of exposure for household members is determined
by their personal habits and the air exchange in the home.
Open windows and fans would make a large difference. Showering
may take thirty (30) minutes during which the person could be
inhaling air containing a large amour : ,t a volatilized water
pollutant. Calculations indicate that the resultant dose may
be equivalent (within a factor of two) to that obtained by
ingestion.
Very little data is available on the enrichment of general
indoor air with volatile organics. Data from New Jersey vali-
dates its occurrence at specific locations within the home
(Bishop, 1982). No work has been done on overall enrichment
of indoor air with the organics. However, research has been
conducted on radon enrichment of indoor air in Texas (Gessel
and Pritchard, I960). These data indicate that an enrichment
of indoor air (amount of radon per liter of air) is about 10-4
to the concentration found per liter of drinking water. This
6b
-------
model assumes an air change rate ot 1 per hour. Also see
United Nations (1977), Hess (1982) and Kahlos (1980) for more
discussion of this transfer process.
The total inhaled exposure is made up of two components,
one in the shower and the other a sum of all other inhalation
exposures indoors. Preliminary calculations indicate that
except tor shower exposure respiratory exposure would be equiv-
alent to 16-20% of the oral intake for individuals spending a
large amount ot time indoors. Time spent within the highly en-
riched environment of a shower stall and bathroom would contri-
bute LO *.a.L ui^te (83-125%) of the oral exposure.
For thii ^ nalysis it has been assumed that respiratory
exposure trom volatile organic chemicals present in drinking
water results in exposure equivalent to that from ingestion.
To compare the relative exposures trom inhalation and
ingestion, assume that for both pathways 100% of that which
enters the body reaches the bloodstream. Then if water with
a concentration of X micrograms/liter is ingested, the daily
intake to the bloodstream is:
X micrograms/liter x 2 liter/day - 2X micrograms/day
If water with a concentration of X micrograms/liter trans-
fers it ail to indoor air and it all goes to the bloodstream,
69
-------
then the intake to the bloodstream is:
X micrograms x 1Q-4 l(water) x 20 m3(air) x
liter1 airHay
106 cm3(air) l(air)
n->(air) 1000 cmJ
» 2X micrograms/day.
Thus the exposure under these conditions from ingestion
and inhalation* is equal.
Dermal Absorption
The skin provides an outer barrier preventing free passage
of foreign substances into the body. Water moves sio- ly
through the skin because ot its inability to pass through oil
and waxes covering the skin. Chemicals more easily dissolved
in oils pass through the skin more rapidly.
There are no data on the absorption of volatile organics
present in dilutee solutions found in bath water. It is evi-
dent that whole body emersion in 30 gallons of water places
the body in contact with a relatively large amount ot pol-
lutants present at 1 ppra concentration (Figure 15).
Lacking data upon which to make this judgment, it was
assumed that the contaminants passed through the skin at the
rate 01 flux tor water (0.2-0.5 mg/cm2/hr). Although dermal
70
-------
absorption during whole body emersion would be 2-3 times
greater per unit of body weight for infants rather than adults,
dermal absorption would contribute less than 1.5% of the oral
dose experienced by "average adults" and less than .05% of
the oral dose experienced by formula-fed infants (see Table
2). This analysis assumes that dermal absorption from bathing
is not a significant source of exposure and risk for these
compounds. Experimental data are needed to confirm this
assumption.
71
-------
POPULATION RISK CALCULATIONS
The population risks were estimated by using the procedure
described earlier in this paper. Risk reduction analysis is
also included. As can be seen in Tables 5 through 13 (Table 13
is a summary), the population risks are determined by a product
ot the number ot people attecteo and the individual risk.
Table 5 (benzene), for example, shows the estimated number of
people cxpc3ci to a given concentration of benzene in drinking
water. In each case the concentration is the average of the
range. The lowest concentration range (less than 0.5 micro-
grams per liter) represents the minimum reported concentration
for the national surveys and does not necessarily represent
the detection limit of the chemical in drinking water. While
we have used the rough average oi 0.25 micrograms/liter tor
this category, it is possible that all of the people in this
category could be exposed to 0 micrograms/iiter or all of them
could be exposed to the maximum of 0.5 micrograms/liter. For
each chemical the individual lifetime risks were determined
from the model fits described earlier.
In each case the curve giving the lowest individual risks
and the curve giving the highest individual risks were used to
determine the range ot population risks. See Uncertainty
section for discussion of rationale for this approach to
including the uncertainty in this estimate. The reasoning
for including only the uncertainty in choice of model is that
72
-------
it is by tar the largest uncertainty. When calculations
yielded less than one person, the number was listed that way
rather than the fractional number ot people. Also included in
these tables is the risk reduction analysis. The risk re-
duction analysis is a determination of the number of cases
averted if the standard is set corresponding to individual
risk rates of 10-4, i(j-5 or 10-6. jn these individual risk
rates the corresponding range of concentration and the cases
averted are listed in Tables 5 through 13. The concentration
range was determined from the graph shown earlier.
As seen in Table 5, the population risk estimates are
listed as ranges. The range is indicative o± the uncertainty
in this estimate. For benzene the lower estimate for each
concentration range is so small as to be less than one. The
sum of the upper population risk estimates is 3,000 when
rounded off to one significant figure. The majority ot the
population risk is contributed by the lowest concentration
category. This category represents calculations below the
detection limit and may be an artifact of the mathematical
analysis scheme used to estimate the population risk. The
range is largely a reflection of the uncertainty in choice of
the mathematical model selected to estimate individual risk.
The contribution to uncertainty from other factors is negli-
gible by comparison.
Drinking water concentrations corresponding to individual
73
-------
risk rates ot iO-6 and 10-5 can fail beiow the minimum detect-
able level ol 0.5 micrograms per liter. In the case of benzene
less than 1 to 70 cases are due to drinking water concentra-
tions above the "minimal reportable level". Therefore, if a
drinking water standard (MCL) was established at 7.5 micrograms
per liter, nationwide compliance would result in the aversion
of between less than 1 and 10 cases (Table 6).
Tables 7-11 provide similar information on the remaining
chemicals. Population risk estimates vary widely as do the
estimated number of cases averted. A comparison ot these
results is presented in Table 12.
Investigation into the actual mechanism ot the eftect of
vinyl chloride on health has revealed that it is actually a
metabolite that does the damage (Gehring and blau, 1977 and
Gehring, 197b). This mechanism has the consequence that the
responses predicted in Figures 10 ana 11 are two to three
orders of magnitude too high since the basis of that analysis
was that the vinyl chloride was the direct contributor to
health effects (Vay Ryzin, 1984). Thus, it appears that the
estimates for vinyl chloride are too high. The experimental
points for trichloroethylene and 1,2-dichloroethane appear
similar to those of vinyl chloride (see Figures 4 and 6).
Since all three of these estimates (see Table 12) were the
highest of the VOC's analyzed, it might be suspected that the
same mechanism is causing them to be high and this may be an
74
-------
artifact of the analysis.
76
-------
TABLE 6
BENZENE-POPULATION RISK ESTIMATES
FOR CURRENT LEVELS OF DRINKING WATER EXPOSURE
Mean
Drinking
Water Number of
Concentration
(Micrograms/
Liter)
0.25
2.75
7.5
15
25
35
45
55
65
75
100
People
Being
Served
2.1 x 108
4.5 x 106
1.2 x 105
6.2 x 104
1.6 x 104
6 x 103
3 x 103
2 x 103
1 x 103
1 x 103
1 x 103
Total Lifetime
Individual Risk for the
Mean Concentration*
Low
(Probit)
-------
TABLE 7
1,2-DICHLOROETHANE-POPULATION RISK ESTIMATES
FOR CURRENT LEVELS OF DRINKING WATER EXPOSURE
Mean
Drinking
Water
Concentration
(Micrograms/
Liter)
0.25
2.75
7.5
15
Number ot
People
Being
Served
2.0 x 108
1.3 x 107
0
1.4 x 105
Total Liretine
Individual Risk ror the
Mean Concentration*
Low
(Probit and
Multistage)
3 x iO-8
2 x 10-6
7 x 10-6
2 x 10-5
High
(Weibull)
3.8 x 10-4
1.3 x 10-3
1.8 x iO-3
2.6 x 10-3
Lifetime
Population
Risk
6 - 7b,000
2o - 17,000
0 - 0
3 - 360
Totalt 40 - 90,000
1,2-Dichloroethane-Risk Reduction Aralysis
For Limiting Drinking Water Concentration
Maximum Allowable
Drinking Water
Concentration
(Micrograms/Liter)
15
7.5
2.75
Approximate Individual
Risk Rate for Maximum
Concentration
6 x 10-7 - 8 x 10-3
1 x 10-6 - 1 x 10-2
1 x 10-5 - 2 x 10-2
Cumulative
Cases
Averted0
3 - 360
3 - 36U
29 - 17,000
t Rounded to one significant figure.
* The total individual risk was determined by assuming that the
risk due to inhalation is equal to that due to ingestion.
0 Number of cases averted for concentrations shown in the first
column.
77
-------
TABLE 8
PERCHLOROETHYLENE-POPULATION RISK ESTIMATES
FOK CURRENT LEVELS OF DRINKING WATER EXPOSURE
(USE INHALATION CURVE)
Mean
Drinking
Water
Concentration
(Micrograms/
Liter)
0.25
2.75
7.5
15
25
65
Number o±
People
Being
Served
2.0 x 108
1.0 x 107
4.3 x 105
2.5 x 105
8.2 x 104
1.1 x 1U5
Total Lifetime
Individual Risk tor the
Mean Concentration*
Low
(Probit)
<10-10
-------
TABLE 9
TRICHLOROETHYLENE-POPULATION RISK ESTIMATES
FOR CURRENT LEVELS OF DRINKING WATER EXPOSURE
Mean
Drinking
Water
Concentration
(Micrograms/
Liter)
0.25
2.75
7.5
15
35
45
55
75
100
Number o±
People
Being
Served
1.9 x 108
2.3 x 107
4.3 x 105
2.1 x 105
7.4 x 105
2.6 x 105
4.2 x 104
1.3 x 105
4.2 x 104
Total Lifetime
Individual Risk tor the
Mean Concentration*
Low
(Probit)
-------
TABLfc] 10
VINYL-CHLORIDE-POPULATION RISK ESTIMATES
FOR CURRENT LEVELS OF DRINKING WATER EXPOSURE
(MALTONI-OLD)
Mean
Drinking
Water
Concentration
(Micrograms/
Liter)
0.25
2.75
7.5
65
Number of
People
Being
Served
2.1 x 108
1.3 x 106
4.7 x 105
1.2 x 105
Total Lifetime
Individual Risk tor the
Mean Concentration*
Low
(Probit)
2 x 10-9
3 x 10-7
1 x iO-6
9 x 10-6
High
(Weibull)
2.1 x 10-4
1.1 x 10-3
2.2 x 10-3
3.5 x 10-3
Lifetime
Population
Risk
<1 - 50,000
<1 - 1,430
<1 - 1,030
<1 - 420
Totalt <1 - 50,000
Vinyl Chloride (Maltoni)-R^sk Reduction Analysis
For Limiting Drinking Water Concentration
Maximum Allowable
Drinking Water
Concentration
(Micrograms/Liter)
65
7.5
2.75
Approximate Individual
Risk Rate for Maximum
Concentration
6 x 10-6 - 6 x 10-3
3 x 10-7 - 2 x 10-3
1 x 10-7 - 1 x 10-3
Cumulative
Cases
Averted"
<1 - 420
<1 - 1,450
<1 - 2,880
t Rounded to one significant figure.
* The total individual risk was determined by assuming that the
risk due to inhalation is equal to that due to ingestion.
c Number of cases averted for concentrations shown in the first
column.
80
-------
TABLE li
VINYL CHLORIDE-POPULATION RISK ESTIMATES
FOR CURRENT LEVELS OF DRINKING WATER EXPOSURE
(FERON-NEW)
Mean
Drinking
Water
Concentration
(Micrograms/
Liter)
0.25
2.75
7.5
65
Number of
People
Being
Served
2.1 x 108
1.3 x 106
4.7 x 10 5
1.2 x 105
Total Lifetime
Individual Risk for the
Mean Concentration*
Low
(Probit)
-------
CM
1
CO
Z
o
a:
H
W
O
Z
8
til
Se
CO
to
a
H
to
<
(b
a
u
td
CO
w
CO
^>
o
CO 0)
0> 4J
CO U
CO 01
0 >
<
o to w
w e CD
O CO 4J
£ ooj
4-
o
CO «)
01 4J
to w
CO 01
0 >
1
O CO J-l
rJ e o>
U Cfl 4J
,_J «J -^J
^ ^4 *^
£ OCJ
4-
T3
CO 0)
U -u
CO VJ
CO CD
u >
1
O CO J-l
V* 0)
O CO 4J
n Id .,4
£ DOrJ
e
o
1-1
4-> +-
(0 .*
1 (0
3 *^4
a. ai
o
a.
4-1
c
CO
c
1-1
E
4J
G
o
0
CM
t
i-^
^
p*.
O
i-l
CO
1 M
CO
V t-9
^
m
^
g
o
r
H
i-4 Z
V U
t-l
1 1
u.
j*\ t,
r^ 3
CO
CM Z
t I
^£
0 H
° C§
|
^H
v
0)
o
Jj
o
r"
U
CO
4j
01
H
V
0) O
N £>
C W
0) CO
CO CJ
o
0
1
pr)
in
i i
o
o
1
m
f*»
o
o
0
»
o
rsi
i
O
tf\
^.
CM
O
O
o
o
OS
1
o
*
*
01
c
CO
JS
4J
0)
o
o
1 1
JS
0
f-J
a
CM
<-,
CO
M
CO
2
5
e
H
Z
i i
»-i
t*.
Cb
^i
CO
Z
^^
H
a
01
C
0)
^,
f.
4J
41
O
O
F-1
«r
U
a
i_i
-H
CM
1
i 1
V
*o
en
|
^j
^^
V
in
^
vO
i
^
V
m
^
CM
O
«->4
1
i-l
V
CD
C
01
1-4
^^
JC
01
0
w
o
1
^1
01
(X,
o
o
CM
1
-4
V
o
o
f-l
o
o
0
^
1
V
m
^
o
o
0
*
o
CM
1
-4
V
in
i^.
CM
O
o
o
o
r**
i
i-i
V
^{
01
c
01 *
-4 0>
>% "O
JC *
0) O
O -4
M JC
o u
r-4
JC ^
0 X
^ c
t-l >l-l
H >
o
o
1
,-)
V
in
^
o
o
0
r-4
1
V
m
f**
0
0
o
PO
1
<-4
V
in
^
CM
O
O
o
o
m
i
r-4
V
/-^
T3
pM^
O
1-1
c
o
4J
t
CO
^s
O
o
.
r-4
I
r-4
V
m
o
o
o
k
C^4
1
_j
^^
m
^
0
0
o
t
m
i
r-l
V
m
r**
CM
O
O
o
0
r-l
1
r-4
V
^^
I
01
Z
1
c
o
Wi
0)
(b
3)
9
4-1
4-1
C
CO
U
r4
4-1
i-l
C
00
1-1
01
c
o
o
4J
4-1
4-1
O,
4J
"O X
J) 0)
o J-1
§ O)
o v
0G CO
r- *
-------
UNCERTAINTY
.Since the uncertainty was so large in these estimates it
was felt to be important to closely examine its composition.
Thus it is in this chapter disected to determine what the
individual contributions are. In a scientific or technical
endeavor such as a risk estimate that produces numbers, the
most important part of the number is the uncertainty involved.
The actual number is meaningless without an indication or the
error or uncertainty. For example, to give an estimate as 100
cases/lifetime, it has a different meaning if it is 100 + 1 or
100 +80. In the first case, the tens and units places have
significance while in the second case they dj not.
There are many ways to determine the uncertainty in the
number, however, we will consider three here: viz, a statis-
tical, practical, and the socalled Delphi approach. Too often
these and other approaches are not used. It has been recom-
mended that the level of concern about uncertainty should be
increased (Morgan, et al., 1983).
The statistical approach requires knowing the frequency
distribution involved. To determine this, many measurements
must be made. When the data are sufficient and the frequency
distribution becomes defined the standard deviation can be
calculated and used as a. measure of the error or uncertainty.
This approach is the ideal one as it would allow tne deveiop-
83
-------
merit ot a risk profile or actual distribution of probable risks
as a function of the magnitude of the risk. The luxury of such
a plethora ct intoriaation is unlikely for risk estimation such
as those considered in this paper. It Is more likely that the
data is so sparse that the frequency distribution is unknown
and the statistical approach cannot be used.
The practical approach to determine the uncertainty is
well known ro those who have taken the laboratory in Physics I.
By a practical approach is meant the determination of the con-
tribution to the uncertainty from observation and experience
with the measurement tools. If we were measuring the length
of a table the uncertainty can be determined by noting the
smallest m-.a&ure on the measuring device.
For exposure some of the contribution to uncertainty that
can be observed are: known uncertainty for the analytical
method used, variation with respect to time, geography, well
depth, house size and ventilation rate. By simple observation
of the uncertainty the measurement tools and an approach to
combining caese errors in an overall uncertainty can be
estimated.
A third way of estimating the uncertainty is to ask the
expertssometimes called the Delphi approach. This can be
used to develop an estimate of the nature and magnitude of
uncertainty. As an example of this approach a recent study
84
-------
sought the judgments of experts concerning the environmental
effects of sulfur dioxide (Morgan, et al., 1983). In that
study the experts were asked to estimate the plume flight time
of sulfur dioxide from coal fired power plants and an estimate
of the health impacts. There was a small range of disagreement
(or uncertainty) among the experts concerning the flight time
but considerable uncertainty with regard to the health aspects.
Potential difficulties with the Delphi approach are due to
the different way people make judgments in the face ox uncer-
tainty (Whittraore, 1983). It has been found that anxiety about
uncertainty leads to overconfidence, that the seriousness of
the risk can lead to overestimates and the method of expressing
the rist c. n influence judf ie t (for example people react dir-
ferently if the risk is expressed as 1 in 1000 of dying or a
chance or 9y.9% of living).
This discussion will use a combination of the practical
and Delphi approaches to determine the uncertainty. Further,
the contributions to uncertainty that cannot be quantitied
will be listed and discussed.
85
-------
HEALTH EFFECTS UNCERTAINTY
This section is an attempt to list, define, and quantitate
the uncertainty that may be an integral part of toxicological
experimental design. Biology by derinition is not an exact
science. When an experiment is designed, a carefully reasoned
question is asked. If the answer is known, then one has an
80% chance of getting the same experimental results. Rarely,
and with good reason, are experiments repeated that take years
to perform. Therefore, the presentation below attempts to
catalogue the potential and actual uncertainty that is inherent
in the design, extrapolation and interpretation of such experi-
ments. The results ot this analysis are shown in Table 13.
We will attempt to focus on, but not limit the discussion
to, long-term carcinogenicity or chronic toxicity experiments.
This discussion will center on the VOC's because the authors
have extensive experience, access to data, and have performed
extensive analysis in these areas: 1) toxicology, 2) risk
estimation and 3) occurrence/exposure. Numbers, details,
estimates and information that are unreferenced originate from
the author's experience.
The probability of choosing the appropriate end-point is
about seventy percent using a chemical from a well character-
ized class such as the VOC's. The chance of being wrong is
larger in an unknown or poorly characterized chemical class.
86
-------
TABLE 13
HEALTH EFFECTS UNCERTAINTY
Category
1. Choice of endpoint
2
3
5
6
Personnel capabilities
of species, strain,
age and sex of animals
Test compound purity
(contamination, decay and
vehicle contribution)
Inappropriate statistical
test methodology
Distribution o: . liraals
among doses an number
used
7. Selection of dose levels
8. Lack preliminary tumor
change information such
as hyperplasia
9. Experimental surroundings
10. Dietary considerations
11. GLP's (good laboratory
procedures)
Contribution to Uncertainty*
Less than 10% of the time
chose the wrong endpoint--
some exception like chlor-
inated hydrocarbon where
expect liver cancer (U)
All or nothing (E)
All or nothing (E)
For the VOC's is generally
trivial but could be two
orders ot magnitude (0)
All or nothing (E)
One or two orders
magnitude (E)
Up to two orders of
magnitude (E)
For VOC's is less than 10%
but coula for other con-
taminants be as large as a
factor of three (U)
Could be orders of magnitude
(0)
Factor of two (E)
One to two orders 01
magnitude (U)
* U-ltads to an underestimate of the risk.
0-leads to an overestimate of the risk.
E-could lead to an overestimate or an underestimate of the risk.
87
-------
TABLE 13 HEALTH hFFECTS UNCERTAINTY (continued)
1'2. Time- to- tumor
13. Microscopic tissue
examination
14. Diseases in test animals
15. Statistical noise
16. Outcomes of cancer (it
include non-fatal cancers)
17. Conventional choice ot
p level (e.g., 0.05)
18. Synergism/Antagonism
19. No corresponding tissue
in humans
20. Most sensitive vs. average
21. Animal to man
22. Body weight vs. surface
23. Use ot upper 95% confidence
limit
24. Choice ot dose-response
model
Could miss the effect or
could be an underestimate
by a factor of two
Could be a factor of two (E)
All of nothing (E)
Factor of 2 (E)
Could be two orders ot
magnitude or more (0)
All or nothing (E)
Many orders of magnitude (E)
All or nothing (E)
Several orders of magnitude
(E)
Conservatively two orders of
magnitude (E)
One order of magnitude (E)
Up to an order ot magnitude
(0)
5-6 orders of magnitude when
considering risk levels in
the 10-4 to 10-6/lifetime
range (E)
* U-leads to an underestimate of the risk.
0-leads to an overestimate of the risk.
E-could lead to an overestimate or an underestimate of the ris
-------
This is illustrated by the unexpected occurence o± brain cancer
that was produced in animals exposed to acrylonitrile (Quast,
19&0 a and b).
The possibility o± choosing appropriate dose levels to
produce the desired effect is dependent upon several factors.
These include tht number of dose levels, the regimen of dosing
and the route of exposure. One can give the animals one or
more doses daily. The material can be given via ingestion (in
food or other vehicle) or inhalation. The inhalation exposure
could be for up to 24 hours a day for seven days a week. The
most common method used in inhalation, however, is 6 hours a
day iOr 5 days a week. This latter commonly used regimen
introduces an uncertainty becau e all of the extrapolations
are for lifetimes of continuous exposure. There are many
examples of choosing the wrong dose. The NCI bioassay program
adjusts their dose requirement if the animals begin to die or
exhibit morbidity at an unacceptable rate (NCI, 197bb).
The racilities in which acceptable experiments are per-
formed have many components. Most important are the quality
ot personnel and the adequacy of the physical plant. An ex-
ample of unacceptable data due to poor plant facilities is
from Gulf South Research Institute (Moore, 1983). In IBT
(U. S. vs. Kepplinger, 1983) the personnel though qualified
were dishonest. In the case ot GSK1 (Moore, 1983) the person-
nel were honest but incompetent.
-------
The experimental model (choict ot species, strain, age and
sex) Is an integral part of the design of the experiment. NCI
recommends the use of specific species and strains of rooents
(Weisberger, 1983). The NCI has determined that the use of the
Fisher 344 rat and the B6C3F1 mouse produce the best combina-
tion of sensitivity and low backround for use in carcinogenici-
ty studies. This is not true in every case. It is entirely
likely that for a particular chemical these animal models may
be inbeu&iuive. An example ot this kind of phenomena is
thalidomide. Only certain strains of rabbits and some monkeys
were capable of demonstrating phocomelia (Casarett, 1980).
One should select the statistical procedure before per-
forming the study. This is to eliminate the possibility of
introducing the personal bias and desire of the experimenter.
Without this objectivity, the experimenter may search tor the
statistics that confirm the significance of the already sub-
jectively determined result.
The experimental design nay play a major role in reducing
uncertainty. One can choose an unusual design such as a trun-
cated or pyramidal model in which the lower dose levels have
significantly more animals. This increases the sensitivity
approximately equal to the square root ot the increased number
(Land, 1980 and OTA, 1981). One must know ahead the shape of
the dose-response curve for this design to be successrul.
Therefore as the numbers of animals are increased (by squaring)
90
-------
one can reduce the dose level being investigated by one order
of magnitude (Land, 1980). This procedure is limited to a
total ot 20,000 animals because only a small number of facili-
ties can handle such large numbers of animals. The cost is
approximately 7 million dollars for an experiment using 20,OOU
animals. Using this large number of animals the lowest meas-
ured dose level is reduced by only one order of magnitude com-
pared to that achieved with the normal 1500 animal experiment.
Often similar experiments yield data that appear dif-
ferent. For example, the data for benzene shown in Table 14
(a-d) seem to be quite different. However as shown in Table
14e the concentration predicted for the 10~6/lifetime level is
within a factor of approximately two for the. > :our cases.
The maximum tolerated dose (MTD) is derived from an exper-
iment. It is the dose that produces a 10% depression in weight
gain over 90 days. The MTD is a surrogate that attempts to
predict the dose level at which no more than 20% of the animals
exposed will die over the duration ot a lifetime experiment.
The next standard level chosen is usually some fraction of the
MTD--say 1/2 or 1/4. This ignores the dose-response curve. It
may be wiser to choose the 5% and 1% projected response for 90
days as the second and third dose levels. If the MTD is not a
sufficiently high dose level, then the true carcinogenic re-
sponse will be missed. If it is too high, we will lose the
entire dose group, and not have enough dose level information
91
-------
TABLE 14a Incidence Rates o± Zymbal Gland Carcinomas in
Male Rats
Experimental Dose Human Equivalent Incidence Rates
(mg/kg/day)
0
50
100
200
TABLE 14b Incidence Rates
Female Rats
Experimental Dose Human
(nifc/kg/day)
0
25
50
100
TABLE 14c Incidence Rates
Gland Carcinomas
Experimental Dose Human
(mg/kg/day)
0
25
50
100
TABLE 14d Incidence Rates
Gland Carcinomas
Experimental Dose Human
(mg/kg/day)
0
25
50
100
0
6.1
12.2
24.4
of Zymbal Gland
Equivalent
0
2.5
5.1
10.1
2/48
6/50
10/50
17/50
Carcinomas in
Incidence Rates
0/50
5/50
5/50
14/49
of Malignant Lymphomas and Zymbal
in Female Mice
Equivalent
0
1.43
2.85
5.7
Malignant Lymphomas
Incidence Rates
4/49
9/4b
9/49
15/48
of Malignant Lymphomas and Zymbal
in Female Mice
Equivalent
0
1.43
2.85
5.7
Malignant Lymphomas
Incidence Rates
15 /4b
24/45
24/50
19/49
TABLE 14e
Table Number Concentration in Drinking Water that Produces
a 10-6/Lifetime Risk (tnicrograms/liter)
15a 1.7
15b 0.8
15c 1.0
15 d 2.3
92
-------
to use for extrapolation. The NCI, therefore, has made interim
changes in dose levels to compensate for the incorrect selec-
tion. The NCI then reports the dose level the animals weie
exposed to as a time weighted average. It is difficult to
determine the actual dose level that will reproduce the effect
using the time weighted average.
Prior knowledge of the historical background incidence of
the species and strain of animals under test is critical.
Initially one would like to design an experiment in which the
target organ, if known, normally has a low tumor incidence.
This is because a high backround incidence of tumors in the
target organ system would make a low-level carcinogenic re-
sponse inn js -ibie to detect (Watanabe, 19tt3) .
Many of the early NCI reported experiments were called
into question because animals who were exposed to different
volatile organic carcinogens were housed in the same room as
animals exposed to other volatile organic chemicals (hd,
1976).
The NCI has kept track of the historical control back-
ground cancer levels. They have segregated the population on
the basis of ditterent laboratories who routinely performed
their experiments. Over time and many experiments later, it
became obvious that certain of these laboratories had a
significantly higher incidence of background cancers (Haseman,
93
-------
19b4). One must assume, since the NCI supplied randomly se-
lected animals of identical species and strain, that differ-
ences in the way the animals were handled or housed were the
genesis of the laboratory difference.
These differences could be the result of:
1. Noise stress
2. Laminar flow vs. HEPA filtered air
3. Purity of drinking water
4. Cage cleanliness
Dietary considerations can be thought of in two categories
One and most often talked about is trace contamination by pesti-
cides or other organic chemicals. The NCI distributed a care
fully formulated and tested chow for mice and rats. This chow
was not generally available until 1979. Therefore, earlier
experiments are subject to questions when their results were
marginally significant. The second often ignored phenomena is
that diets deficient in one or another element can signifi-
cantly enhance disease processes (Revis, 1963).
There are certain housekeeping aspects to experimental
procedures that may have significant eftects. The first and
most ignored is the randomization of cage placement of animals
in their racks. There is some evidence that the location in
the racks had greater significance than the experimental
treatment.
94
-------
Good laboratory procedures (GLP's) have mandated that each
animal be given a unique identification. This is as a result
of the IBT experience in which animals died and were reborn
during a chronic experiment (US vs. Kepplinger).
Another source ox problems addressed by GLP's was the
handling of recently dead and moribund animals. Significant
amounts ot needed tissue were lost to gross and microscopic
examination due to inadequate procedures. GLP's now require
frequent enough examination to prevent cannibalism and auto-
lysis. Lack of these procedures produced major effects on the
experimental outcomes.
GLP's also specify that expe it iced and qualified person-
nel must supervise and perform autopsies. It is likely that
subtle etfects would be missed it only a technician alone
performed necropsies. (A small footnote to necropsy procedure
is included. The 'standard' is to preserve tissue in a zor-
malin solution. If one is looking for immune or autoimmune
ettects, they will be destroyed by formalin preservation.)
An attempt should be made to characterize time-to-tumor
phenomena. While this information does not change the outcome
of a cancer experiment, it provides a good measure for potency.
Unpalatable feed or drinking water can produce erroneous
results. This is due to reduced nutritional intake. One must
decide how often to examine and collect data from thu animals.
95
-------
Weight gain data should initially be collected very often and
then less often as the animals mature.
Microscopic characterization ot grossly observed lumps and
bumps is used to determine the carcinogenic character of the
lesion. Today, panels of 'experts' are assembled, duplicate,
double blind, coded slides sets distributed, and a consensus is
used as the tinai result. This does not take into account what
the 'true' riiihwer really is (NCI, 1977). It is not that these
talented experts differ in the characterization of the struc-
tural architecture that they are examining. It is that a clear
detinition ot wtiat a carcinogenic lesion in that tissue is
lacking. Benign tumors are added to any malignant tumors and
used in the calcu~.at ion of risk (Surgeon General, 9', J and
Plant 1Q7S'). Another example is to consider curable skin can-
cer due to arsenic exposure as life threatening (USEPA, I9tt3b).
Intercurrent animal diseases are difficult problems with
which to deal. Murine pneumonia can sweep through an animal
colony rapidly, negating two years of expensive and carefully
conducted experiments. Unfortunately, results are sometimes
reported in which the infection though noted, is ignored. The
results are often quoted without murine pneumonia cited. The
results of experiments in which the animals were or are sick
is not acceptable.
We will detine the concept or a statistical noise as the
96
-------
level at which one cannot determine the difference between
response and the natural occurrence In the general population.
This level will vary depending upon the endpoint chosen to be
measured. If the endpoint is an extremely rare occurrence,
e.g., angiosarcoma, mesothelioma, brain tumors, it is easily
shown to be significant. The second kind of endpoint that is
difficult to measure statistically is a normalized standard
measure--e.g., IQ. It is impossible to measure small changes
in limited populations even though they may be real. The
third example would be the yes or no parameter such as life or
death. These kinds of yes/no answers lead to easily charac-
terized statistical measures.
Another source ot uncertainty is the lack of well def? .e
decision criteria. For instance 1-5 Napthlenediamine is called
a carcinogen while Clonitracid and 2-(chloromethyl) pyradine
were not, even though all three had significant increases in
uterine turaors (Salsburg, 1983).
There is a major error in the evaluation of carcinogenic!-
ty data, that is all tumors that are cancer are considered
equal. This is shown in its most unreasonable form when one
considers the cancers produced by arsenic (cancer of the skin)
and those produced by benzene (leukemia). Skin cancer is
curable 99.99% of the time. Benzene induced leukemias are
incurably fatal. The risk model as applied predicts a much
higher incidence of skin cancer vs. leukemia. The CAG, there-
97
-------
fore, ranks arsenic as a much more dangerous carcinogen than
benzene (USEPA, 1983b).
The choice of 0.05 as a level of significance is an arbi-
trary one (Fisher and Yates, 1957). There is no reason that a
greater p would not also show significance. In those cases in
which a very rare cancer occurs as cited above, statistical
significance is not required since these are rare events.
Two tissues for which there is no human equivalent are the
Harderian gland and Zymbal's giana. However experience has
shown that these are predictive of cancer although in some
other tissue (haltoni, 1983). Animals and humans may not show
the same effect. For example benzene cau.jes cancer of the
hematopoietic system in humans and not in animals, benzidine
causes bladder cancer in humans and liver cancer in animals
and cadmium causes lung and prostate cancer in humans and
cancer of the testes in animals (Tomatis, 1979). Cigarettes
have been shown to cause lung cancer in humans and not in
animals and dioxin (2,3,7,8-tetrachlorodibenzo-p-dioxin) is one
of the most lethal chemicals to animals but the ettect observed
to date in humans is only chloroacne and some possible soft
tissue cancers. It would appear that to use animal bioassay
data to predict the effects on humans will be wrong for some
environmental contaminants.
There is a tendency to use the most sensitive animal model
98
-------
to determine the potential carcinogenicity oi chemicals. This
may overestimate the potential of actual carcinogenicity in
humans by several orders ot magnitude.
One of the major considerations in the extrapolation of
data froic animal to man are the differences in metabolic pat-
terns. The animal model may not produce a proximal carcinogen
ano one would miss characterizing a true carcinogen. An ex-
ample is arsenic (USEPA, 1983b).
Animals may produce more ot the active proximal carcinogen
and less of the material to deactivate or bind the carcinogen
than does man (USHEW, 197«) The distribution of the chemical
may be different in man vs. animals. An example would be the
binding ot heavy metals in plasma and in red blood cells
(USEPA, 1983a).
The choice ot taking the cube root or using the direct
ratio of the weights of humans to experimental animals can
lead to an error ot up to an order of magnitude.
The different models, as shown in Figures 4-14, predict
responses that are 5-6 orders of magnitude ditierent at the
10-4 to 10~6/lifetirae risk level.
99
-------
UNCERTAINTIES IN OCCURRENCE ESTIMATES AND
POPULATION CONCENTRATION ESTIMATES
The assessment ot the irequency of finding VOC's in drink-
ing water supplies across the United States and the concentra-
tion being served to various portions of the American popula-
tion is a major input to the risk assessment process. This is
called an occurrence estimate. The assessment process involves
the assembly of monitoring data from national studies in which
the criteria ot system selection did not alter the probability
of the system being representative of the pollution in other
systems within its class, thereby allowing the formation of a
distribution curve which can be used to project national occur-
rence estimates and popula i ns being served. The uncertain-
ties resulting from this process begin with conditions sur-
rounding sampling design which ultimately determines sensi-
tivity to low frequencies of occurrence and moves through all
of the technical procedures tor sampling, analysis, data veri-
fication and interpretation. Each of these procedures contri-
bute some uncertainty to the final estimates. The overall
list of uncertainties in estimating exposure and an estimate
of their impact on the risk analysis are shown in Tabj.e 15.
Site Selection of Systems to be Sampled
Two sources ot uncertainties must be dealt with during the
procedure for selecting sampling locations. These uncertainties
100
-------
TABLE 15
SOURCES OF UNCERTAINTY FOR OCCURRENCE, POPULATION
CONCENTRATION AND EXPOSURE ESTIMATES USED IN THE ASSESSMENT
. OF VOLATILE ORGANIC CHEMICALS IN DRINKING WATER
impact on Estimate oft:
Factors
Occurrence
Popula-
tion
Concen-
tration
Risk
Generation of Monitoring Data
Proportion of population sampled
5% (U)_
Representativeness ot systems selected
Geographic distrityition, system
size and source oi?1 v^ter 10% (E)
Sampling methods
Site of sample collection
Time of sample collection
Method of sample collection
Container type
Stability during storage
Sample analysis
% recovery xrom sample
Compound identification
Accuracy of quantitative
determination
Assumptions During Data Analysis
Lower ximits of quantification
Oral exposure rates
Intake rate of water
Pollutant level in consumed
water (hot vs. cold)
% Absorption for oral intake
Respiratory exposure rates
Dermal exposure
20% (E)
20% (E)
10% (U)
10% (U)
100% (n
10% (U)
10% (E)
40% (E)
50%
Factor
of 2
Factor
of 3-4
Factor
of 2(E)
10% (E)
50% (0)
10% (0)
Factor
of 3
Negligible
tU - leads to an underestimation of the risk.
0 - leads to an overestimation of the risk.
E = could lead to an overestimation or an underestimation of
the risk.
101
-------
are:
Was the selection ot systems to be sampled conducted
using criteria which provided an unbiased representa-
tion of the national universe of community water
supplies?
Was the study designed so as to detect levels ot
pollution which may occur infrequently?
The information used in this risk assessment to project
national occurrence rates was formed as an amalgam from
national survey data. In some studies the sampling locations
were selected randomly. In others sampling locations were
selected independently by individuals within dirrerent states
and regions throughout the country with the only guidance
being not to seller systems known to be contaminated. Prior
to forming an amalgam of these data sets, data we^e compared
with UOLCI xiuiu "random" sites, and it was found there was no
statistical difference between the two data sets. Thereafter,
the data sets were merged to increase the number ot samples
available from which to assess national occurrence.
It would appear that there is less uncertainty surrounding
the body o± the national frequency distribution formed from
these data than exists when one examines the high and low ends
of that curve. The exact number of systems experiencing an
extremely high level of contamination is unknown. Information
exists frora state agencies which demonstrates that such places
exist yet were undetected in the national monitoring survey.
102
-------
One possible source or this "false negative" could be caused
by the lower limit of sensitivity created by randomly selecting
l.OOU sampling locations from a population of 46,000 candidates,
The number of systems experiencing these high levels and
the number ot people they serve is probably quite small and is
expected to have a minor impact on population concentration
estimates. However it this distribution of systems having
elevated concentration was affected by some factor causing a
"clustering" ot systems within a category not sampled during
the survey the underestimate could be larger.
Uncertainty is also created because of our inability to
det> :t true concentration, espe i, Lly at the low end of the
frequency distribution. For this analysis it has been assumed
that places where the pollutant was not detected at a "minimum
quantifiable level" (<0.5 micrograms/liter) contained 0.25
micrograras/liter (the median concentration between 0 and 0.5
micrograms/ liter).
We know that certain solvents are ubiquitous in societal
use and appear at the micrograms/liter level in rain water.
For these substances there are probably tew it any, water
supplies which are absolutely free of contamination. However,
for other less ubiquitous chemicals, a number of water sup-
plies are probably absolutely clean. By far the largest number
of people tall into the "less than" category. The percent of
03
-------
risk associated with this "less than" category representing
the maximum overestimation, if all systems were truly zero, is
presented beiow tor each of the chemicals in this assessment:
Ration of total
Percent in the to fraction
"less than" above the "less
category than" category
Benzene 78% A.5
Carbon Tetrachloride 35% 1.5
1,2 Dichloroethane 60% 2.5
1,1 Dichloroethylene 90% 10
Tetrachloroethylene 51% 2.0
Tnchloroethylene 2y% 1.4
Vinyl Chloride 79% 4.7
Rough average 3.5
Sampling Location Within the System
The data available from national monitoring studies were
generated from a single sample taken at one point in time at
one location within the distribution system ot each water
supply.
How representative is data from a grab sample of the
drinking water quality being provided to users throughout the
system?
There are several components to this question. How do
104
-------
concentrations vary with time? Is the pollutant level created
by stable long-terra contamination of raw water? If so, then
the data could be used to produce fairly representative esti-
mates of exposure. However, if the pollutant level can effect
& portion of the system or nay change drastically over time,
the estimates will be highly uncertain.
There is no information to help answer these questions.
We know that some groundwater systems operate so that individ-
ual wells serve specific portions of a city. We also know
that some chemicals may be added to specific portions of a
system because of leaching from pipe linings (e.g., tetra-
chloroethylene). Under these conditions the exact sampling
point within a water system will make a large different n
the finding. For others, such as carbon tetrachloride, it is
felt that most of the problems arise rrom contamination oi raw
water or the use of contaminated chlorine disinfectant which
would arfect the water supply rather uniformly. Since this
effect of this source of uncertainty cannot be well defined,
it has been assumed that while results for specific systems
may be overestimated, unbiased selection throughout the systems
sampled has likewise produced an underestimate of conditions
in other systems which when combined, produce a reasonably
accurate national occurrence profile.
105
-------
Maintaining Sample Integrity Prior To Analysis
The problem of partial or complete disappearance of the
chemical pollutant contained in drinking water prior to analy-
sis due to physical or biological processes may produce a large
uncertainty in the analytical results. When partial disappear-
ance is suspected, the concentrations represented in data may
be taken as "minimum values" from an unknown but larger levei
present at the time of sampling. A problem arises in inter-
preting data reported as "zero or minimum detectable" found in
such a data set for the possibility exist that higher levels
were originally present which diminished in sample handling
produce false negative readings at th . time of analysis.
Under these conditions, it would be impossible to identity low
occurrences from false negatives. Preliminary data reported by
Click in 19/9 demonstrated a rapid loss ot benzene, ethybenzent.
and toluene from unutilized spiked tap water to background in
about 10 days. Chlorobenzene, m-xylene and p-dichlorobenzene
levels were found to diminish in 2-4 weeks of storage. The
loss of benzene was especially rapid with 50% loss being found
within 2 days. The situation with respect to the degradation
of unsaturated aliphatic hydrocarbons is not well understood.
What level of uncertainty has been created for chemicals
discussed in this report by loss during the storage of samples
prior to analysis? To address this question one must realize
that the only data used to project national occurrence were
106
-------
those developed troin water samples drawn after treatment in
the distribution system. Most of these samples, especially
those from systems using surface water, would have contained
chlorine disinfectant. Data on the level of VOC's in surface
water supplies was developed in studies predating the recog-
nition of possible degradation problems in which holding times
were long (see Tabie i6). Therefore, there exists a level of
uncertainty surrounding surface water contamination with each
of the VOC's, but especially benzene. Data trom which ground-
water predications have been made are also samples drawn from
the distribution system. However, since a significant number
of the smaller groundwater systems do not disinfect their
water, it is indeed fortunate rhat data were derived from a
more recent study (GWSS) whi .h was designed to ensure minimal
pollutant j-oss during sample storage.
The uncertainty for risk assessment on these VOC's is
greater tor benzene where occurrence predictions may under-
estimate national occurrence because of numerous .false nega-
tives in surface water supplies and lesser xor each of the
other VOC's.
Detection and Quantification
The process of chemical analysis is tilled with technical
uncertainties. Procedures have been established to control
the quality ot the resultant data. Continuation via alternate
107
-------
TABLE 16
SAMPLE STORAGE TIME AND CONDITIONS
Storage Time
Storage Conditions Before Analysis
NORSa Iced < 14 days
NOMS Ib Iced 7-14 days
NOMS II, III Room temperature " 21-42 days
NSPC 14-28 days
CWSSd Some not refrigerated 150-800 days
Some not retrigerated 300-700 days
GWSSf Acidiiied/bactericide <^ 28 days
added
*NORS - National Organics Reconnaissance Survey
t>NOMS - National Organics Monitoring Survey
CNSP - National Screening Program
dCWSS - Community Water Supply Survey
eRWSS - Rural Water Supply Survey
fGWSS - Ground Water Supply Survey
108
-------
analytical techniques (i.e., mass speculation) of a percent of
the qualitative findings has been practiced on the data used
on these surveys. Therefore, this source of uncertainty is
relatively low. In addition, quantitative determinations may
become a large surface of uncertainty. Once again, laboratory
procedures were instituted for these studies to assure that
the error was minimized (Taylor, 1982).
Uncertainties Exposure Assessment
The drinking water contribution to respiratory and oral
exposure has been estimated using assumptions about the rate
of water used during rluid intakes--bathing, the transfer rate
to air of VOC's transported in the diinP n^ water, the turnover
rate for air within the closed coiu ines of a bathroom and
house and the length of time with a corresponding volume of
air inhaled within each environment.
Definitive data are not available to define the variance
or average of any of these factors as they atfect the United
States population. The assumptions made are based on "reason-
able" extensions data confirmatory 01 a qualitative process
but insufficiently precise to validate quantitative conclusions.
10 y
-------
Oral Exposure
AssumptionIntake Rates
Procedures used in this risk assessment assumed that
average intake of drinking water over a lifetime approximates
2 liters/ 70 kg rate established tor "average adults" resulting
in 0.028 micrograms/liter/kg absorbed dose per day. This
consumption rate (0.03 liters/kg/day) has been selected as the
population mean for this report. Comparison with theoretical
average rates reveals that fluid intakes per unit body weight
generally decrease with age, beginning at 7-9 times as high in
intants and reaching a plateau bv early teens and adulthood.
A weighted average derived from a theoretical water consumption
curve indicates that the lifetime average may be 20% higher
than the current factor being used. Population studies on the
variability or water consumption have not been conducted in
the United States. Canadian data indicate that 13-16% of the
adults may consume more than 2 liters of water per day.
Assumption - Concentration of Pollutant in Fluids Consumed
Approximates the Concentration in the Tap Water Sampled.
This conservative assumption implies that individuals are
receiving that static concentration in drinking water over
their lifetime. It does not deal with population mobility nor
potential changes in concentration with time. The vast
110
-------
majority of the population does not remain in one city and is
not served by a single water supply over their lifetime. How-
ever,.the rate or movement from locations with higher pollutant
concentrations and vice versa is unknown. The same lack of
knowledge exists tor the stability ot drinking water concen-
trations over time, and therefore the uncertainty associated
with these tactors cannot be quantified.
If has been assumed that what is present in cold tap water
is present in consumed xiuids made with drinking water. V/e
know from the Canadian study that the preponderance of drinking
water consumed by children is in the form of cold drinks and
the amount of hot drinks increases with increasing ape. We
also know th t while the VOC's are transported in dr..nking
water they are rapidly transferred into air with aeration,
heating or upon standing. These factors woula diminish oral
exposure of adults significantly but tend to create point
source liberations into indoor air which might increase respir-
atory intake. Since the percent of absorption from respiratory
intake is assumed to be 50% of the oral absorption rate the
overall exposure of adults might be reduced by 30-50%.
Respiratory Exposure
This risk analysis recognized the transference of drinking
water borne pollutants into indoor atmosphere and the resultant
exposure 01 house-hold members to an enriched pollutant
.11
-------
atmosphere during time spent in the dwelling. It assumes that
respiratory uptake is equivalent to oral uptake with the major
contribution to exposure taking place during showering or bath-
ing and the time spent within the enclosed bathroom immediately
after these activities prior to dispersal of the pollutant
concentration. This conclusion is based on several assumptions
and their associated uncertainty:
-- Nearly 100% of the pollutant concentration of
VOC's in drinking water is transterred to air
upon heating, aeration or standing.
Laboratory and field testing data is available to support
this assumption. Engineering documents produced for the USEPA
by Environmental Sciences and Engineering of Gainesville,
Florida, document a high efficiency of aeration processes for
removing the VOC's iroro contaminated raw water. In addition,
laboratory tests show that hot water has lost virtually 100%
of its VOC contamination upon standing ror 3 hours or being
poured between containers 20 times. Cold water exhibits 20-80%
loss under the same conditions. Since bathing and showers
involve a large percent of heated water which is aerated during
impact on body or tub, the volatility is enhanced. An assump-
tion of 100% transference to air may be in excess by 5-10%.
Exposure During/After Showering and Bathing
Physical factors used to define the volume of air in a
112
-------
bathroom into which the pollutant would be dispersed were
chosen instinctively. The volume of air assumed to be present
in the bathroom was 9.4 m3 or 6" x 71 x b' --the 30 gallons
of water assumed to be used in a shower. The result of 100%
transfer of the water pollutant was 113 micrograms of pollutant
present at 1 microgram/liter in water tor 9.4 m3 of air or a
resulting concentration of 1.2 microgranis/mS. it was further
assumed that the individual spending 30 minutes in the bathroom
inhaled 0.42 m3 ot air containing 0.5 micrograms of pollutant
o± which 5U% was absorbed. The net absorbed dose was estimated
to be 0.0036 micrograms/kg/day.
The relationship between the volume of air contained in
the bathroom, including turnover rate w thin a 30 minute period
and "nationaj. averages" is unknown. Individuals in saunas,
steambaths or hot tubs would most probably exceed these time
estimates.
Volume of Air Inhaled Within the Confined Area of the
Showering Room
The estimated air inhaled while in the contined ar>.a of a
bathing room was .006 m3/kg of body weight. These figures
were developed as averages from an assumed respiratory rate ot
20 m3 per day per 70 kg individual. The uncertainty surround-
ing this factor is unknown.
-------
Concentration in Confined Air
It was assumed that the pollutant equally dispersed
throughout the air within the room remained in air throughout
the 30 minute time period.
There is some data rrom field studies in New Jersey to
indicate that pollutant concentration in a shower stall is
greatest at cne breathing zone. However, there are no aata
showing that the pollutant is removed from the air by binding
to fabric, wood or plastics in the rooio. Since the scenario
assumes 100% availability, it is likely to tend a worse case
estimate, however, the assumed rate ot absorption (5U%) was
an average from aaii al tests (30-80%) conducted at highv r iir
level«
Respiratory absorption at lower concentrations is probably
increased substantially creating an underestimate by 10-40%.
General Household Respiratory Exposure
An average of 250 gallons of water pass through an
American household each day At a pollutant concentration of
1 microgram/liter, 946 micrograms of a pollutant become avail-
able tor dispersal within the household. Knowing that tht
VOC's are easily transferred from the water column into air, we
also know that th*,- bulk or VOC's in drinking water get
114
-------
dispersed throughout the house. There are no measurements ot
the indoor air concentration resulting from this process. By
analogy, in estimates made for radon it was assumed that the
level in household air (in liters) would be equivalent to 10'^
of the level found in water (liters). At an initial concentra-
tion ot 1 microgram/liter in water concentration in general
household air would be .0001 microgram/liter (.001 microgram/
liter equals .1 microgram/m3). Individuals spending 12 hours
per day within the home environment inhaling 10 m3 of air
would be exposed to 1.0 micrograras of pollutant of which 50%
would be absorbed. The net absorbed dose is equivalent to .07
micrograms/ kg day.
There is no way to estimate the uncertainty associated
with this estimate except to state that a comparison of the
relative volatility ot radon and the VOC's would indicate that
the transfer of radon from water to air probably occurs more
rapidly and thererore more completely in the home than would
be expected for the VOC's.
Dermal Absorption
There is nothing known about the ratt of dermal absorption
for low levels of organics found in drinking water. It could
be assumed that the VOC's pass the dermal layer ±n proportion
to the flux of water (0.2-0.5 mg/cra2/hr). Several factors
might mitigate against this assumption:
115
-------
-- Much ol the VOC's in warm, hot or aerated
water would be transferred to air.
-- The VOC's remaining in solution would pass
through the skin at an enhanced rate due
to their lipid solubility and enhanced
blood flow to skin in warm water.
-- Soaps added to wash water would bind
with the VOC's remaining in solution.
Since reality is unknown, dermal absorption was assumed to be
proportional to water flux ranging from 2 x 10~6 to 9 x 1U~7
micrograms/kilograms per day, per part, per billion of the
pollutants present in drinking (micrograms/ liter).
Of all the uncertainties in the risk estimation process
founu here the largest is due to the choice of model. As
shown in Table 15, this can be as large as 5 ;o 6 orders of
magnitude. Since all ot the other errors, combined even when
in the maximum possible way, are not nearly this large, the
overall error can be considered to be that due to the choice
of model.
If a particular model can be chosen, the remaining uncer-
tainty in the risk estimates is 2-3 orders of magnitude. The
contribution to the overall uncertainty in either case from
the exposure estimate is less than one order of magnitude.
116
-------
DISCUSSION AND CONCLUSION
The number ot people likely to be arfected by a particular
contaminant in a lifetime due to existing concentrations is
called the population risk. To estimate the population risk we
need to know how many people are exposed and at what concentra-
tion level and what the risk is at those concentrations.
Both the exposure and toxicity information involve uncertain-
ties. Some ot these uncertainties can bt estimated and some
cannot.
Some, but not necessarily alx, of the uncertainties in-
volved are those due to: inherent measurement error, unrepre-
sentativeness ot estimating concentration in all supplies
from a data for only a few, lack of data for concentrations
below the limit of detection, estimating inhalation tox^c
effects from data derived by ingestion experiments and vice
versa (also errors involved in estimating effects due to aermal
exposure), using animal data to predict human effects and
extrapolating the expected risk level several orders of magni-
tude below the actual dose levels used in the experiments.
The largest uncertainty, at the present, is in the extra-
polation to low doses. Figure 16 shows such an extrapolation.
The starred points are the actual data points for this ricti-
tious contaminant (the figure is typical of those for most of
the volatile organic chemicals found in arinkiug water). It
117
-------
j 1
i
I
I
I
11
* *
«e o
*» «
4J O
c0-
4J t-l
> «
C 4^
l!
£6
H
« -P
*
^
s
-------
only thfc top two data points (along with a point at the origin
which cannot be shown in the figure) are used, they can be
fit to tour ot the currently popular models; viz, logit, multi-
stage, probit and V/eibull. If the point at 'A' in Figure 16
is excluded, the model tits are shown by the dashed lines lor
the Weibull and logit models. In that case Che multistage
and probit models predict curves close to the solid iii.es
shown. If another data point is added (denoted 'A'), the
multistage and prooit estimates change very little but the
Weibull and logit model predictions move down to the solid
lines.
At a drinking water concentration ot one microgram/liter
in Figure 16, the range of risk prediction is about seven
orders ot magnitude (or a tactor of iO.OOO.OuO) when point 'A1
is excluded, while adding the point at 'A' reduces this range
to a xactor or about tour orders of magnitude (or a tactor of
10,000). It is possible to develop another model which coula
predict a risk value outside the range predicted t>y these four
models. The probability of this cannot be estimated unless
the frequency distribution of the data is known. In order to
determine the frequency distribution, experimental points are
needed at the low dose and low risk levels of interest and
even at lower values. It is unlikely that this information
will ever become available.
Ii we have no information at alj. about a contaminant, we
119
-------
can predict with complete certainty that the population risk
lies somewhere between less than one person and all the people
(approximately 250 x J.O& people in the U. S.). The object
ot risk estimation is to reduce this range. Suppose we con-
sider Figure 16. Assume that about 5% of the U. S. was
exposed to this hypothetical contaminant at a relatively low
concentration of one microgram/liter. Then the range of
population risk would be approximately:
10~3 cases/person exposed x 5% of 250 x 10& persons
- 10,000 cases
io'"^ i_v.~«:t>/person exposed x 5% of 250 x 11)6 persons
- 0.001 cases
when rounded ott t, one signiticant tigur...
Does the above range have any meaning? Is it of any
value? What does O.OUl cases mean? It would make more sense
to express this as less than one case. Is this range of less
than one to 10,000 ditterent from another of say 100 to
10,000,000? Since the ranges overlap, the two contaminants
could not be said to be necessarily difrerent- larger or
smaller.
There is little experience in extrapolating dose-response
curves below the known data. In the area of health effects ot
ionizing radiation, etfects have been seen down to the region
120
-------
of one to ten rads whilt the regiun ot natural background and
that of regulatory interest is in the area of 0.1 rad (Land,
19bO). The dose-response curve tor alpha emitters appears to
be linear down to the lowest known effect level and thus to
get to the 0.1 rad xevei is an extrapolation of one or two
orders of magnitude. The available data and model fits for
health ettects ot ionizing radiation are found in tht dose-
response analysis of leukemia incidence from the atomic bomb
survivors (Land, 198U). As shown in Figure 17, the data exist
down to the 10"^/lifetime. Since this is the region of inter-
est for regulatory purposes, estimates of health tffects are
based on extrapolations of no more than an order of magnitude.
The megarnousc experiment involved about 30,000 mice
(exposed to 2acetylaminofluorine or 2-AAF) which is consider-
ably more mice than is normally used in a bioassay study
(Cairns, 1980). Although this large number was not enough to
add a point as tar down as that ot 'A1 in Figure 17 , it did
show that the dose-response curve looked linear for the late
appearing liver neoplasm. Using th^s large number of rodents
only extended knowledge one to two orders of magnitude into
the unknown region. In addition a new and unexpected effect
was revealed; viz, bladder neoplasm, which exhibited a minimal
eftect which could be interpreted as a threshold effect. Thus
there can be surprises when the effect of low doses is inves-
tigated.
121
-------
I
I*
§
I
tu
-I
I
B
0)
n
I
CO
0)
t-
03SMX3 IIOtt34XSlt!
ft
v-
-------
For tue volatile organic chemicals as potential drinking
water contaminants, the low dose region of interest is four
to six order* or magnitude fallow tne dose levels used in bio-
assay studies. There is no previous experience in extrapo-
lating the dose-response curve so far into the unknown. Such
extrapolations will involve considerable uncertainty as they
go far beyond any previous experience in this lield.
As can be seen from the extrapolation of the dose-response
data in Figures 3-14, no one model is consistently the high.st
or lowest. This is in conflict with the commonly held idea
that the multistage model predicts the highest risk or that
the linear model predicts the highest risk (Moghissi, 1983).
A strong point ot estimating risk as described here is
that it does produce a number. But the weakness of the proce-
dure is that it produces a large uncertainty. Thus, the number
could be anywhere in the range and could be outside the range.
If the range is i to i,00u, then 2 is the same number as 9y&
and we have the same confidence that either is correct. The
magnitude ot the uncertainty can be estimated by the range ot
model predictions shown in Figure 17, but the real value
could lie outside that range.
The actual number can be anywhere in the indicated range.
For the previous example of the range from 0.001 cases of
10,OUU,OOu cases, the estimate of the lower end is equally as
123
-------
possible as that of the upper end, and the probability of
0.002 cases is the same as the probability of 5,000 cases and
so forth.
The magnitude of the range of uncertainty in this example
can be compared to hitting a home run in a ball park. The
idea of a ball park is that the ball can be anywhere in the
park and the probability is the same for right field as left
field and so forth. For hitting a home run in a major league
park the uncertainty in its location is the order of several
hundred reet. That can be compared to the uncertainty in the
animal bioassay data. The corresponding uncertainty in the
extrapolation is larger than that of a sat .-liite position
after launch if we are ignorant of the launch time and it
could be anywhere in its orbit.
Although there are many interpretations that can be made
of the calculations in this report, only a few of the major
conclusions will be listed here. The report is an attempt to
clearly show how the calculations of risk were made. No policy
implications or discussion of the impact or connection to the
overall regulatory process is included here. This is not meant
to be a comment on the policy implications. The other contri-
butions to overall regulatory decision making are to be found
elsewhere.
The major observations that can be made from tht caicula-
124
-------
tions Described in this report
The largest uncertainty in the calculations is due to
the choice o± model and is 5 to 6 orders of magnitude.
No single model consistently is the highest or the
lowest on the dose-response plots.
If a single model were chosen, the overall uncertainty
in risk estimates would be 2 to 3 orders of magn.tude.
The exposure estimates contribute at most an order of
magnitude to tht uncertainty
It would appear that until a particular compound's
mechanisms ot cancer are better known, the uncertainty
in the toxicity will not likely be improved.
If exposures greater than 7.5 micrograms/liter were
eliminated, the number of lifetime cases averted would
be less than lO for benzene and perchloroethylene and
less than 400 for 1,2-dichloroethane, less than 2,000
for vinyl chloride and less than 4,000 for tnchloro-
ethylene. Except for 1,2-dichloroethylene (which is
3), the lower 1J ii s are all less than one.
Some research effort could help to reduce the uncertainty
in the risK estimates presented here. Clearly any information
that allows only one model would be the biggest contribution
to reducing the uncertainty Thus any research effort that
sheds light on the mechanisms of cancer would be the most
helpful. To determine what other research might be helprul,
it is instructive to refer to the table of uncertainties,
especially Table 15 for the uncertainties associated with
toxicity. The next largest uncertainties involve the distri-
bution or animals among doses, the aoses selected and the use
of averages rather than considering the most sensitive group.
In this latter ar^a some research regarding the importance of
125
-------
species, age and sex would De useful. Also the question of
the consequences of using homogenous, inbred strains could be
better defined by some research efforts.
126
-------
ACKNOWLEDGEMENTS
The authors wish to thank Mr. Charles V. Berry for help in
using the computer programs to analyze the data, Mr. Robert to
Bargh for help in developing a program to analyze uncertainty
in the occurrence data and Dr. Steven P. Bayard (of the Office
of Research and Development, USEPA) for many helpful discus-
sions. Also thanks to Dr. Fred Abramson (George Washington
University), Dr. John Van Ryzin (Columbia University) and
Dr. Jacqueline Michel (Research Planning Institute, Columbia,
SC) for their critical reading of the report and their many
helptul suggestions tor improving the presentation ot this
information.
127
-------
REFERENCES
Anderson, E. L. , 1982. "Quantitative Methods in Use in the
United States to Assess Cancer Risk", Paper presented at the
Work- Chemicals at Rome, Italy, July 12, 1982. For more
iniormation contact OHEA, USEPA, Washington, D.C 20460.
Bailey, J. R. , Beniot, R. J., Dodson, J. L. , Robb, J. M. and
Wallman, H. , 1969. A Study or Flow Reduction and Treatment ot
Waste Water from Households prepared by General Dynamics, for
the Federal Water Quality Administration, Department of the
Interior under Contract No. 14-12-428, Cincinnati, Ohio, Water
Pollution Control Research Series,
Bayard, Steven, 1983. Private Communication, Cancer Assessment
Group, USEPA, Washington, D.C.
Bishop, B. , 1982. In home evaluation of tne impact of benzene
in drinking water and indoor air, New Jersey Department of
Environmental Projection, Trenton, NJ , Personal Communication.
Boland, P. A. National Screening Program for Organics in
Drinking Water, prepared by SRI International for Otfict of
Drinking Water, ULEPA, Washington, D.C., Contract No. 68-01-
4666.
Brass, H. J., 1977. The National Organic Monitoring Survey:
A Sampling and Analysis for Purgeable Organic Compounds in:
Drinking Water Quality Enhancement Through Source Protection.
(Pojasek, R.B. , ed.) Ann Arbor, MI, Ann Arbor Science, pp.
393-416.
Cairns, T. , 1980. The EDQ1 Study: Introduct .on, Objectives
and Experimental Design, Journal of Environmental Pathology
and Toxicology, 3_:l-8, and succeeding articles.
Canadian Environmental Health Directorate, 1981. Tap Water
Consumption in Canada, Ministry of National Health and Welfare,
Publication number 82EHU80.
Casarett, L. J. and Doull, J., !9bU. Toxicology, Macmilian
Publishing Co. , NY.
Coniglio, Wm. , Miller, K. , Mackeever, D. The Occurrence of
Volatile Organics in Drinking Water a briefing paper for
USEPA. Deputy Assistant Administrator for Drinking Water,
March 1980.
Crump, K. , 1982. Science Research Systems, Rustin, Louisiana.
128
-------
Dambrauskas and Cornish, 1970. Toxicol. Appi. Pharacoi.,
17:83-97.
Doull, J., 1984. Dra, t report, Ad hoc Panel on Chemical
Cercinogenesis Testing and Evaluation, The University of Kansas,
Department of Pharmacology Toxicology and Therapeutics, Kansas
City, Kansas 66103.
Dow Chemical Company 198U. Bur.k, J. D., Nitschke, K. D.,
Bell, T. J., Wackerle, D. L. , Childs, R. C. , Beyer, J. E. ,
Dittenber, D. A. , Rampy , L. W. , McKenna, h J., 1980. Methyl-
ene Chloride: A Two-Year Inhalation Toxicity Oncogenicity
Study in Rats and Hampsters, Toxicology Research Laboratory
Health and Environmental Sciences, Dow Chemical Company,
Midland, Ml, December 31, 19»sj. Unpublished .
Feron, V. J., Hendrikson, C. F. M. , Speek, A. J., Til, H. P.
and Spit, B. J., 1981. Lifespan Oral Toxicity Study QJ. Vinyl
Chloride in Rats. Fd. Cosmet. Toxicol. 19:317-331 .
Fisher, R A. and Yates, F. , 1957. Statistical Tables, 5th
Edition, Tables VI, VI T and Vl2 , Oliver and Boyd, Edinburgh.
FRDS, 1983. Federal Reporting Data System, Facilities and
population served by primary water supply source (F"DS07)
/or «1 19, USEPA, Washington, D.C. 20460.
Gehring, P. J. and Blay , G. E. , 19/7. Mechanisms of Carcino-
genesis: DoseResponse, J. of Environmental Pathology and
Toxicology, JL:lt>3-179.
Gehring, P. J., Watanabe, P. G. and Park, C N. , 197b. Toxicol.
Appl. Pharmacol. , 44:581-591 .
Gesell, T F. and Pritchard, H. M , 1980. The Contribution or
Radon in Tap Water to Indoor Radon Concentration, Prepared for
the Third International Symposium on the Natural Radiation
Environment, Houston, Texas, April 23-28. Published in NTIS,
as Natural Radiation Environment III.
Baseman, J., Huff, J. A. and Borman, G. A., 1984. Historical
tumor incidence in the B6C3F1 mouse and F 344 rat, C*N Press,
See, G. A. Borman, National Cancer Institute, Box 27709
Research Triangle Park, NC 27709.
Hess, C. T., Weiffenbach, C. V. and Norton, S. A., 1982.
Variations of Airborne and Waterborne Rn-2z2 in Homes in Maine,
Environ. Inter. £:59.
JRB Associates, 1982. Evaluation of Drinking Water Data in EPA
Studies, Draft report, Prepared for the Office of Drinking
Water, USEPA, Washington, D.C. 20460, Contract #68-01-6185.
129
-------
Kahios H. and Asikainen, M., I960. Internal Radiation Doses
from Radioactivity of Drinking Water of Finland, Health Physics,
39:106.
Krewski, D., Koron, J. and Bickin, M , 1982. Optimal Experi-
mental Designs for Low Dose Extrapolating to Appear in Topics
in Applied Statistics. (T. W. Dwivedi, ed.), Marceli Dekker,
New York.
Krewski, D., I9o2a. Health Protection, Board of Health and
Welfare, Canada, Ottawa, Ontario, Canada.
Krewski, D. and Van Ryzin, J., 1981. Dose-Response Models for
Quantal Response Toxicity Data, In Statistics and Related
Topics, Csorgo, M. , Dowson, D. A., Rao, J. N. K. and Saleh, A
K. Md. E., eds. North Holand Publishing Company, New York.
Kuzmack, A. M., 1983. Characterization ot the Water Supply
Industry (FY82) Memorandum of May 16, 1983, Office of Water,
USfcPA, Washington, D.C. 20460.
Land, Charles, 1980. Estimating Cancer Risks From Low Doses
of Ionizing Radiation, Science, 209:li97.
Letkiewicz, F., Johnson, P., Macaluso, C.f Elder, R., Yu, W.,
and Be-son, C. , 1983. Series ot documents on occurrence of
1,1-dichloroethylene, 1,2-dichloroethane, dichloromethane,
vinyl chloride, carbon tetrachloride, trichioroethylene,
tetrachloroethylene and benzene, prepared by JRB Associates
for the Office ot Drinking Water, USEPA.
Lenaweaver, F. P., Jr. "Report on Phase One, Residential Water
House Research Project", The Johns Hopkins University Depart-
ment of Sanitary Engineering. October, 1963. Reported in "A
Study ot Flow Reduction and Treatment ot Wastewater from
Households", Bailey, J. R., General Dynamics Report for the
Federal Water Quality Administration, 19t>y.
Mabey, W. R., 1981. Aquatic Fate Process Data for Organic
Priority Pollutants. USfcPA, OWRS Washington, D.C. EPA-440/
4-81-014.
Morgan, M. G., Morris, S. C., Henrion, H., Amoral, D. A. L.
and Rish, W. R., 1983. Technical Uncertainty in Quantitative
Policy Anaiysis--a Sulfur Air Pollution Example, Manuscript
submitted to Science from the Department of Engineering and
Public Policy, Carnegie-Mellon University, Pittsburgh, PA
15213.
Munro, I. C. and Krewski, D. R. 1981. Risk Assessment and
Regulatory Decision Making, Fd. Cosmet. Toxicol. 19:549-560.
130
-------
Maltoni, C., Conti, B., Cotti, 0., Mandrioili A., Scearmato,
C., Valgimigli, L. , 1982. Benzene.- An experimental multi-
potential carcinogen, Institute ot Oncology BoxOgno and
Bentivogoio, Italy.
Maltoni, C., Lefimine, G., 1975. Carcinogenicity Bioassays
of Vinyl Chloride: Current Results, Annals of the New York
Academy ol Sciences 246:195-218.
Maltoni, C., Conti, B. and Cotti, G., 1983. Benzene: A
Multipotentiai Carcinogen, Results of Long-Term Bioassays
Performed at the Bologna Institute of Oncology, Am. J. of
Industrial Medicine, 4:589-5301.
Moore, J., 1983. Memorandum, Withdrawal of Gulf South Research
Performed NCI Studies, USJbPA, Office of Pesticides ana Toxic
Substances, TS-788, Washington, D.C. 20460.
National Academy ot Sciences, 1982. Drinking Water and Health,
Washington, D.C.
NCI (National Cancer Institute), 1976. Report on Carcinogenesis
Bioassay of Chloroform, National Toxicology Program, Box 12233,
Research Triangle Park, NC 27709.
NCI (National Cancer Institute), I976a. Report on the Carci' >
genie Assay of Chlorofoi , 3afcinogenesis Program, Division i
the Cancer Cause and Pre/ention, Technical Information Section,
Carcinogenesis Testing Program, National Toxicology Program,
Landow Building, Room A306, Bethesda, MD 20205. March 1, 1976.
NCI (National Cancer Institute), I976b. Carcinogenesis Bioassay
of Trichloroethylene, National Cancer Institute Carcinogenesis
Technical Report Series No. 2. NCI-CG-TR-2, Technical Informa-
tion Section, Carcinogenesis Testing Program, National Toxic-
ology Program, Landow Building, Room A3Uo, Bethesda MD 2U20i.
February, 1976.
NCI (National Cancer Institute), I977a. Bioassay 0*. 1,1,1-
Trichloroethane for Possible Carcinogenicity, National Cancer
Institute Carcinogenesis Technical Report Series No. 3.
NCICG-TR-3, January, 1977, Technical Information Section,
Carcinogenesis Testing Program, National Toxicology Program,
Landow Building, Room A306, Bethesda, MD 20205.
NCI (National Cancer Institute), I977b. Bioassay of Tetra-
chloroethylene for Possible Carcinogenicity, DREW Publication
No. 77-813, Technical Intormation Section, Carcinogenesis
Testing Program, National Toxicology Program, Landow Building,
Room A30t>, Bethesda, MD 2U2U5.
131
-------
NCI (National Cancer Institute), I977c. Bioassays or Nitrilo-
acetic Acid, Nitriloacetic Acid (NTA) and Nitrilotriacetic
Acid, Tnsodium Salt Monohydrate (NaoNTA H£0) ror Possible
Carcinogenicity, Technical Report Series, Number 6, U. S.
Department ot Health, Education and Welfare, Bethesda, MD
20014.
NCI (National Cancer Institute), i978a Bioassay of Ethyl^-ne-
dichloride for Possible Carcinogenicity, Technical Information
Section, Carcinogenesis Testing Program, National Toxicology
Program, Landow Building, Room A306, Bethesda, MD 20205.
NCI (National Cancer Institute), 197ob. Bioassay of Picioram,
Technical Report Series No. 23, U. S. Department of Health,
Education and Welfare, National Institutes of Health, Bethesda,
MD 20C1A. See also, reports numbered: 2, 3, 5, 42, 43, and
126.
NCI (National Cancer Institute), 1983. Study of Bioassay ot
1,1,1-Trichloroethane for Possible Carcinogenicity, Technical
Information Section, Carcinogenesis Testing Program, National
ToxieoS^v riMgrara, Landow Building, Room A306, Bethesda, MD
202U5.
Nutrition Foundation, 1963. Report of the Ad Hoc Working Group
on Oil/Gavage in Toxicology, 886 Seventeenth Street, NW,
Washington, D.C. 200Ufa.
Of-LJLuc w.i. IcJiuology Assessment (OTA), 19bl. Assessment of
Technologies for Determining Cancer Risks From the Environment,
U. S. Government Printing Office, Washington, D.C. 20402.
Plant, P. A., 1975. A Dangerous Definition, Chera., Engineering
News, Oct. 20, 1975.
Quast, J. F., Wade, C. E., Humiston, C. G., Carreon, R. M.,
Herman, E. A., Park, C. N. and Schwetz, B. A., 1980. A Two-
Year Toxicity and Oncology Study with Acrylonitrile Incor-
porated in the Drinking Water of Rats, Toxicology Research
Laboratory, Dow Chemical USA, Midland, MI 4b640. Jan. 22,
1980.
Quast, J. F. , Schuetz D. J., Balmer, M. F. , Gushow T. S.,
Park, C. N. and McKenna, M. J., 1980. A Two-Year Toxicity
and Oncology Study with Acrylonitrile Following Inhalation
Expofeuic vi Rdcs, Toxicology Research Laboratory, Dow Chemical
USA, Midland, MI 48640. Dec. 9, 1980.
Rai, K. and Van Ryzin, J., 1979. Risk Assessment of Toxic
Environmental Substances Using a Generalized hulti-Hit Dosu-
Respose Model, in Energy and Health, Breslow, N. E. and
WhitteiBore, A. S. eds. SIAM Press, Philadelphia, PA
132
-------
Rai, K. and Van Ryzin, J., lyttl. A Generalized Mulcihit Dose-
Response Model for Low-Dose Extrapolation, Biometrics, 37:341-
352. ~~
Rampy. L. W , Quast, J. F , Balmer, M. F. , Leong, B. K. J.,
Gehring, P. J., 1978. Results of a Long-Term Inhalation Toxi-
city Study on Rats 01 a Perchloruethylene (Tetrachloroethylent)
Formulation, Toxicology Research Laboratories, Dow Chemical
Company Midland, Ml.
Revis, N. Vi. , Douglas, b. H. , McCauley, P. T. , Witschj., H. P.
and Bull, R., 1983. The Relationship of Drinking Water Chlorine
to Corunary Atheroschlerosis, Pharmacologist, 25:Abstract 732.
Salsburg, D., 1977. Use of Statistics When Examining Lifetime
Studies in Rodents to Detect Carcinogenic!ty, J. of Tox. and
Env. Health, 3:617-628.
Salsburg, D., 1963. The Litetime Feeding Study in Mice and
Rats--An Examination of its Validity as a Bioassay for Human
Carcinogen, Fundamental ana Applied Toxicology, ^:63-b7.
Schneiderman, M. A., Decoufle, P. and Brown, C. C., 1979.
Thresholds tor Environmental Cancer.- Biologic and Statistical
Considerations. Annals of the New York Academy of Services,
329:92.
Sorreil, K. , 1982. Preliminary Experimental Retults on the
effect of aeration on the VOC content of Drinking Water, USEPA,
Municipal Research Laboratory, Cincinnati, Oh. Personal
Communication.
Sperling, F. and harcus W. L., 19bi. Thalidomide in Blood,
Pharmacologist, Jt:147.
Surgeon General, 197u. Report 01 the Ad Hoc Committee on the
Evaluation of Low Levels of Environmental Chemical Carcinogens,
Evaluation ot Environmental Carcinogens Re-
Symons, J. M., et al., National Organics Reconnaissance Survey
for Halogenated Organics. J. Amer. Water Works Assoc. 667
(ll):634-647. 1975.
Tomatis, L., 1979. The Predictive Value or Rodent Carcino-
genicity Test in the Evaluation of Human Risks, Annual Reviews
01 Pharmacology and Toxicology, 19:511-5JU.
United Nations Scientific Committee on the Effects of Atomic
Radiation, 197/. Report to the General Assembly. Sourc s and
Effects of Ionizing Radiation (United Nations, New York, NY).
USEPA: The National Public Water System Program. FY 62 Status
Report. Office of Drinking Water, USEPA, 1983.
133
-------
USEPA: National Interim Primary Drinking Water Regulations,
USEPA. 570/9-76-003, Office of Drinking Water, 1976.
USEPA: Community Water Supply Survey Computer Data File on
Inorganic and Organic Chemicals. Office of Drinking Water,
USliPA, Washington, D.C. , iy?o.
USbPA: Rural Water Survey Computer Data provided by Depart-
ment of Sociology, Cornell University for the Office of Drink-
ing Water, 1982.
USEPA: Community Water Supply Survey, Bureau of Water Hygiene,
Public Health Service, Department of Health, Education and
Welfare, Washington, D.C., I97u.
USEPA, 1983. Health Assessment Document for Carbon Tetra-
chiorioe Cancer Assessment Group, USEPA/ECAO, EPA 60U/8 82-UUi,
Revised Draft, March, 1983.
USEPA, 19o3a. Air Quality Criteria tor Lead, EPA-60u/8-»3-028A,
Washington, D.C. 20460.
UStPA, 19tt3b. Health Assessment Document lor Inorganic Arsenic,
EPA-600/ 8-83-021 A, Washington, D.C.
U. S Department of Health Education and Welf; -e i97b.
Special Occupational Hazard Review of Trichlorc«»tiiylene, Center
for Disease Control, NiOSh, Rockviiie, MD 20857.
U. S. vs. Kepplinger, et al., 1983. Northern District of
Illinois, #tt1CR235.
Van Ryzj.n, J., 19»Ua. Quantitative Risk Assessment, J.
Occupational Medicines, 22;321-326.
Van Ryzin, J. and Rai, K., !9bUb. The Use of Quartel Response
Data to Make Prediction, Witsch, H., ed., The Scientific Basis
of Toxicity Assessment, North-Holland Biomedial Press, New York
Van Ryzin, J., 1984. Columbia University, Biostatistics Depart-
ment, Private communication.
Waktanabe, P. G., 1983. Personal Communication, Concerning
Sprague-Dawley Rats, Spartan Substran, Dow Chemical Co., 1803
Building, Midland, Ml 48640.
Watson, K. S. Farrell, R. P. and Anderson, J. S., "Contribution
from Individual Homes to the Sewer System", JWPCF 30:12.
2039-2ui>4. December, 19o7.
Weisburger, J. H and Williams, G. M., 19tt3. Chemical Carcino-
gens, Chapter 2, Bioassay of Carcinogens: In Vitro and In
Vivo Tests, Washington, D.C.
134
-------
Whittmore, A. S , iy83. Facts anu Vaiues in Risk Analysis tor
Environmental Toxicants, Risk Analysis, 2h 23-33.
Westrick, J. J., Hello, J. Vi. and Thomas R. F. , The Ground
Water Supply Summary of Volatile Organic Contaminant Occurrence
Data, h?A Technical Support Division, Office of Drinking Water,
Cincinnati, OH. 1983.
135
-------
APPENDIX A
EXPLANATION OF HOW ANIMAL DATA WAS EXTRAPOLATED
TO A CONTINUOUS HUMAN EXPOSURE EQUIVALENT
-------
Benzene
Reference - Maitoni, et ai. , 1982.
The health endpoint considered for benzene was the zymbai
gland carcinoma in remale rats. The following is the bioassay
data for this contaminant:
Animal, mg/kg/day,
4 to 5 times/week Human Equivalent, Animals
for 52 weeks mg/kg/day Affected/Total
0 0 0/3U - 0
50 2.61 2/30 » 0.067
25U 13.1 8/32 - 0.25
To convert the given dose' to tog/kg/day continuous human
equivalent, multiply by 52/104, since the exposure was for one
year and the lifetime for rats is two years, and multiply by
the cube root of the ratio of the rat and human weights, or
(0.3/70)1/3 -0.162. To convert the four to five times
per week feeding, to continuous exposure, multiply by 4.5/7.
Thus, for the 50 rag/kg for the 4 to 5 times/week, the conver-
sion is:
50 mg/kg x (4.5/7) (52/104) (0.3/70)1/3
« 2.61 mg/kg/day human equivalent.
A-l
-------
Carbon Tetrachloride
Animal
TTIP /dsv
0
11
18
21
36
Animal
mb/kg/day
0
24
40
47
80
Human Equivalent
mg/kg/day
0
4.5
7.4
8.7
14.9
- wj , 1976a, 197bb, USEi-A, 1983.
The bioassay data from the NCI rat study is shown as
follows :
Animals
Affected/Total
0/37
2/45
4/4o
2/47
1/30
To convert from rag/day to mg/kg/day, divide by the average
body weight ot the ra 5, or 0.45 kg, and to get the human
equivalent, multiply by the cube root of the ratio of the rat
and human oody weights, or (0.45/70)1/3 » 0.186. The bioassay
data from the NCI mouse study is shown as follows:
Animal Animal Human Equivalent Animals
mg/day mg/kg/day mg/kg/day Affected/Total
000 t>/157
21 750 56 89/89
42 15UO 112 90/93
To convert from mg/day to mg/kg/day, divide by the average
body weight, or 0.02b kg, and to get the human equivalent,
multiply by the cube root of the ratio of the mouse and human
body weights, or (0.028/70)1 /3 - 0.074.
A-2
-------
1.2-Dichloroethane (Ethyiene Dichloride)
Reference - Draft CAG (USEPA) document quoting 1978 NCI bio-
assay report.
The following is tue bioassay data tor hemangiosarcomas in
the circulatory system of male rats:
Animal Human Equivalent Animals
mg/kg/day nig/kg/day Affected/Total
0 0 0/40
47 4.85 9/4b
95 9.80 7/27
To convert from animal to human equivalent dose, multiply
by 5/7 fcince the animals were dosed 5 days/week multiply by
78/104, since the experiment lasted 78 weeks and the average
lifetime is 104 weeks, and multiply by the cube root of the
ratio of the rat and human body weights, or (0.5/70)^/3 -
0.193.
A-3
-------
1.1-Dichloroethylene (Viuylidenc Chloride)
Reference - Maltoni, et al.. 1980.
The bioassay for the resulting kidney adenocarcinomas of
nice is shown below:
ppm/Daily
for
12 months
0
10
25
Liretime
Continuance
Equivalent, ppm
0
0.54
1.34
Human Equivalent Animals
mg/kg/day Affected/Total
0
0.023
O.OuO
0/i36
0/125
28/119
To convert from daily dose to lifetime continuous dose,
multiply by 'v/24 since exposure was 4 hours/day, multiply by
4.5/7 since ,;he exposure was 4 to 5 times per w ek and multiply
by 1/2 since the experiment was continued only tor 1/2 the
lifetime. Thus, 4/24 x 4.5/7 x 1/2 - 0.0536.
To convert from ppm to micrograms/m3, use the relationship
1 microgram/m3 - 10-3 ppm . Since the molec-
1.2 x IMW (VDC)/MW (Air)J
ular weight of (Mn) vinylidene chloride is 97 and that of air
is 2*5.8, then 1 ug/m3 - 2.5 x iO-A ppm. To convert trom
micrograms/m3 to mg/kg/day, multiply micrograms/m3 by (20m3/day
x mg/1,000 micrograms x i/70 kg) - 2.86 x 10-4; multiply by the
cube 70 kg root of the body weight ratios to convert to human
equivalent (0.03/70)1/3 - 0.075 and dividt by 2 on the
assumption that 1/2 of that air is absorbed. Thus:
A-4
-------
1,1-Dichloroethylene (Vinylidene Chloride) - continued
2.86 x 10-4 (Q.Q75) - 0.042.
2.5 x 10-<* x 2
A-5
-------
Perchloroethylene (Tetrachioroethylene) (Inhalation)
Reference - Rampy, et al., 1978.
The following are the bioassay results rrom this inhal-
ation study.
Animal Hunan Equivalent Animals
ppm mg/kg/day . Affected/Total
0 0 2/189
300 4.5 2/94
600 9.1 3/94
To change from ppm to mg/kg/day (human equivalent) multiply
by 1.2 x 103 (16i.«/28/8) - 6900, where 165.8 is the molecular
weight of perchloroethylene and 2b.8 is the mo .ecular weight
of air. Also, multipl) by the cube root of the ratio of the
body weights (0.35U/70)1/3 « 0.171 and 1/2 to adjust for 50% of
that inhaled being absorbed. Then multiply by (20m3/day x 1 rag/
1,000 micrograras x 1/70 kg) - 2.86 x 10-4. Multiply by 6/24
since the exposure was for 6 hours/day and by 5/7 since the
experiment was tor 5 days/week and by 52/104 since the experi-
ment lasted only 1/2 the lifetime. Thus to change from ppm to
mg/kg/day human equivalent, multiply by:
2.86 x 10-4 x 6900 x 0.171 x (6/24) x (5/7) x (52/104)
- 0.0151.
A-6
-------
Perchloroethylene (Tetrachioroethylene) (Ingestion)
Reference - NCI, 197', .
The bioassay data from resulting hepatocellular carcinomas
of female nice are shown below:
Animal Human Equivalent Animals
mg/kg/day mg/kg/day Affected/Total
0 0 0/20
386 18.02 18/48
772 3o.04 19/43
To convert from animal to human equivalent, multiply by
5/7 since the experiment was lor 5 days/week and by 78/90
since the experimer \ as 78 weeks long and the average J'fetime
was 90 weeks, and by the cub*, root of tht ratio or booy weights
(0.03/70)1/3 equals 0.0754 or multiply altogether by:
5/7 x 7tt/90 x 0.054 «= 4.67 x 10-2.
A-7
-------
Trichloroethylene
Reference - NCI, 1976b.
The roilowing are the bioassay data from hepatocellular
carcinomas in male mice:
Animal Human Equivalent Animals
mg/kg/day mg/kg/day Affected/Total
C 0 1/20
1530 56.3 26/50
2700 1^.2.6 31/4b
TO rnnvprt from animal to human equivalent, multiply by
5/7 since the experiment was done for 5 days/week, by 1.5/2.0
since the experiment lasted 1 1/2 years and the lifetime was 2
years and by the cube root of the ratio the body weights
(0.033/70)I/- or:
5/7 x 1.5/2.0 x (0.033/70)1/3 = 0.0417.
A-b
-------
Vinyl Chloride (Feron)
Reference - Feron, 1981.
The to1lowing are the bioassay data from female rats
exposed to vinyl chloride:
Animal Human Equivalent Animals
rog/kg/day me/kg/day Affected/Total
0 0 2/57
1.7 0.241 26/58
5.0 0.71 42/5?
14.1 2.00 56/57
To convert troro animal to human equivalent, multiply by
the cube root of the ratio of the body weights (0.20/70)1/3 =
0.142.
A-9
-------
Vinyl Chloride (Maltoni) (Inhalation)
Reference - Maltoni, 1975.
Animal Human Equivalent Animals
ppm mg/kg/day Affected/Total
0 0 W58
50 2.3 10/59
250 11 16/59
500 23 22/59
250U 1x0 32/59
To convert the animal doses to human equivalent chronic
doses in microgranis/ro3 , multiply by 1.2 x x03 162.5/28/8J
where 62.5 is the molecular weight 01 vinyl chloride and 26.8
is the molecular weigh . of air. Then multiply by 5/7 since
the exposure was 5 days/week and by 52/104 since the exposure
was for 52 weeks and the lifetime was 104 weeks. Then multiply
by the cube root of the ratio ot body weights or (0.35/70)1/3 =
0.171 .
Then to convert to mg/kg/day multiply by 2.86 x 10-4.
Thus to go from animal dose in ppm to human equivalent chronic
dose in mg/kg/day multiply by:
1.2 x 103 (62.5/28.8)(5/7)(52/104)(0.35/70)1/2 (2.06 x 10-*)
- 0.0455.
A-10
-------
APPENDIX B
Point Estimate and Upper 95% Confidence Limit Values For the
of Four Moaels (Logit, Multistage, Probit and Weibull)
to the Bioassay Data
-------
A
ZH
(d (ft
N W
Z C9
id Z
09 i-i
Z U
t-i &
5§
^2
O!><
u a
'
c
o
CO
o>
pi
ev
e
4)
4-1
jj
^^
y*
O
i-4
r-l
3
l-l
0)
rj
0) «J
IV 1-1 £
O-U-I -^
exc j
3 0
o
w
^J
*J ecj
c e
1-1 ft
o *^
CX, CO
H V
m o
ON C
l-l "O "H
oj-^ e
a>w-^
a c J
= o
4)
4-1
c e
O 4J
Ow (0
u
X V
n '^
y\ «
,
C CO
*-H "^i,
to at
c > J?
ec --J O ^
S 3 w ao
3 cro s
SC U Q N-'
r~
sO
1 o
o
O
1
1 1
m
CM '
o
1
t 1 '
1
1
\
1 !
1
1
t
(
\
1
t
l
<0
1
1
1
t
1 '
col ^
o
>,
(0
CO
CO
(0
0
<-(
CO
! ^0
1
Ol CM
r
m
^»
1
'
1
1
1
jlation
Extrap<
:M
o
M
\r\
r«.
en
0
X
a-
CM
*
1
O
X
in
en
en
i
0
.-i
X
en
en
<3-
l
O
X
o
m
.3-
l
. O
<-(
X
«n
^
i
0
X
CM
CM
CM
t
0
f^
1
O
X
r«.
en
00
i
o
X
vO
r-l
«a-
o
X
J-
en
»n
i
o
X
ON
en
-3-
i
o
I-t
X
vn
in
>n
i
o
X
in
\o
m
o
T-l
X
en
in
i
O
X
CM
CM
en
t
o
2
O
X
en
en
.CM
i
0
X
«o
0
in
O
X
r^
-------
i i |
l^*
1 fi CO
O CO 4J
{H CO (TJ
*SQ CC £^
i
1 '
1
f
1?
o
CO
(^
0-
1 <»
1
: o
M CD
.n o
JN C
1 CD 4J
4J
*H
JO
. u
(X,
M *^ *i^
t> ._, =
O.MJ i-<
ac fJ
=> o
o
O
0
1 !
*&
30
O
o
en
.^
o
o
1
1
i
j
en
en
0
o
1 '
!
1
i
H
1
_
1
1 O
i 1 l i-t
\ I
i
> i i
i
9)
4-1
4J CO
^c^e
O 4-^
Ou CO
btd
0)
hn 0
CT^ c
1 0) 4->
IfJ T3 fl
0> -i E
O.4-I !-(
i
! ! £
t
! 1
t
01 f 4->:r> o i i
3
4j
Ol 0)
0) ' nJ 4JI 1 >
U-l 4J CO
4J
CO
cd
r-l
O
Z
td
H : c E
.3
-^
CO
I-l !-!
O *J
!
i
1
& Wl
td
ci
H1 V
i,-i f 1
i) 4J
CO
fi
OS
1 W '"
a> 1-4 .
O.-W ^
i-i
^
3
«**
ai
3
ac J
=> 0
'^> i
0)
4J
4J CC
' ' c e
j 1 -4 -H
1 ! 0 4J
0
O*
3
|
*
HO
M
Z H
SCO
w
afi o
^tf 25
O M
CU (0
Cd
im o
>&i C
1
1
a>
M
cd
4.'
5
i-l
1 5>
£
a> w
cy **^ £
ac J
3 JO
iM W»
w
td i
H
3
HO ~
Z as
z id
M OS
y O
<3 C/J
1-4 ^^
Z Obi
0 X
o a
*-> t*J
C CO
1> T>
' ^*.
CO CtC
C > w*!
CO -f* 0) -«~
§3 co oo
CO
4J
CO
a
CO
CO
CO
(0
o
Q « i .^
a; w Q ^ CQ
i
i
1
1
i i
' 1
o
"7
1
p-.
00
^
i-4
C
o
^
4J
cd
-4
O
a
CO
f^
4J
X
1 (d
CM
1 1
CM
o
f-l
X
CM
00
1
0
iH
X
o
en
CM
i
0
f-l
X
CM
00
1
o
X
o
en
CM
i
o
X
CM
00
CM
1
O
f-l
X
f-l
f-l
en
i
o
X
vO
NO
o
slat ion
Extrap<
T
o
X
CM
f\l
1
0
X
CM
00
1
0
iH
X
o
en
CM
1
0
r-l
X
CM
00
1
O
X
o
en
CM
1
O
X
CM
00
CM
1
O
X
f-l
en
i
o
X
vO
NO
o
o
f-l
X
1
o
X
-1
1
o
f-l
X
en
00
i
o
*x
en
i-l
i
O
f-l
X
00
1
0
X
en
f^
en
i
0
X
-H
1
O
.-1
X
vO
vD
l
O
O
1
o
X
I-l
o
*
1
o
X
o
CM
o
f-l
o
X
i-l
CM
1
o
X
f-l
1
o
f-l
X
VO
CM
1
o
0
o
X
CM
o
f-l
1
o
X
0
en
o
^
1
o
X
CM
en
o
X
^
\o
1
o
f-l
X
vO
«O
en
i
o
0
1
o
X
en
o
1 1
o
X
CM
-a-
o
i
o
X
r^.
0
X
i-4
00
1
o
f-l
X
vO
vO
in
o
1
'
o; c1
t
i ' i
o c
X X
n >a-
m .D
i '
i
O C
^ _
1 ' 1
O' 0
f 1 [ 1
X X
vo! r-
( (
1
O1 O
*-4 n
\
1 1
C^ ' d^
X X
CN
-------
x-v
O>
CO
£
y
-.
z
Cd
0
1-4
OS
s
5
p*
Saz
HO
ZH
8w
Cd
Cti O
^C 55
O i-i
(d
H
HO
ZW
i i oi
y jg
< CO
Z a.
OX
u w
,
^
c
o
CO
J-l
0)
0-
1v
_e
4->
0)
"4-4
i-l
-^.
CO
01
CO
O
V
CO
1-4
Oi
4J
M
£>
It
O-
^j
i-l
O
^j
1 1
I-l
3
1-1
OI
o>
tt(
CO
u
^
4J
i-l
I
t
«
9
a:
^,
l CO CO
O co oj
^ w co
30 CO O
K 0>
in o
ON C
01 4J
W "O i-4
Oi 1-
CL-W-I
ac ,J
3 0
u
01
4J
W CO
C_E
O 4J
OH CO
w
K 0)
in o
ON C
a* 4->
Id tJ H-l
a*--, e
O>'4^ *M
ac j
=) 0
o
01
^J
4.) (0
C E
-v4 *H
O 4J
i CO
Id
fr< 0>
m o
ON C
a> f
kj 'U *^
01 -^ E
a-M 1-4
a c j
=> 0
0)
4->
4J CO
C E
1-1 H-4
O 4-1
Qj (0
t*3
fr4? 0*
m o
ON C
0) 4J
a»!S e
a 4-1 -r^
ac J
3 0
t>
01
JJ
4J CO
c e
O 4J
OH CO
td
4-> >i
C CO
cy ^3
r-l *^
co bC
^ ^
.^4 w **s»
3 w ofl
y o E
C0
4-)
2
>,
CO
CO
CO
CO
0
1-1
CO
3D
O
0
o
o
(1
'
1
vO
m
fS«
ON
0
CM
1
4
i
i
1
L
I
i
1
1
1
>lation
Extrapc
i
o
K
ON
ON
1
O
X
v£>
ON
1
O
X
ON
ON
0
r-4
X
vD
ON
1
O
X
ON
ON
1
O
X
\0
ON
CN
O
r-<
X
^4
NO
CM
1
O
X
ON
*
O
»»
l
O
X
ON
ON
l
O
X
sO
ON
l
O
X
ON
ON
l
O
cH
X
vO
ON
l
O
X
ON
ON
1
0
X
vO
ON
en
1
0
F-4
X
tr>
VO
*
i
o
,
o
X
ON
ON
O
X
vO
ON
1
o
X
ON
ON
1
O
i i
X
vO
ON
l
O
X
ON
ON
l
O
X
SO
O>
r^
i
o
i-i
X
en
vO
r>»
i
O
X
o
m
m
i
o
i
i
0
X1
?>
ON!
1
O
x
"°
o«
1
o
X
^
^
1
0
X
v£>
ON
i<
ON
ON
1
0
X
vO
ON
oo
1
0
1
X
en
vO
OO
o
X
o
m
vO
1
O
-------
1
' 1 « «
1 O CO 4J
1
1
"J co to
CO « Q
>4? 0)
I i.n o
i
0
l ;
a*
^m.
*
O
sO
CM
O
i i
i
ItTv C '.
0) 4^J
t
4J
'c'
O
CO
. ft
w
94
w i
Ol
W "O *"!
OJ j~l E
pv i_j i
«XC ,J
=3 0
01
4J
4J CO
O 4J
Cu CO
K 0)
m o
o> c
01 4J
VJ T3 -4
Ol-< E
^ ! lexcj
1
0) > 4J|Z> O '
e
4-1
0>
Ttl °
oc^~
o
-H
01
4J
4J efl
C E
l-J . -^ 'H
^ O *J
CO ! Oi CO
^5
N-'
^
CO
r4
06
U
25
^^
jg
^4
w
o
§
U
c^ a>
in o
t
i
1
1
1 1
J
1
1
t
: 1 i
i
1 '
1
1
1
'
i i
i i
ICTX c ! !
0) 4-1
1 J-l T3 "<
,_|
i-l
3
H-l
01
"If
Oi "* E
a-M -5
ac j
1 1
!
1
1 1
i
30 ii/ i
u
0)
4J
4J (0
C g
f4 -H
O 4J
CL, CO
tes
lo\ c
g-O
V 4-1
0)
01
CO
4-1
4J
C- 1 {-*
o to
i W
(NO
z
*
H
HO
z a:
^g
z u
^
JJ
V^ *O *^4
a» ^ g
O« ^4 *^4
ac j
3 0
u
Pu «
M
f
1
1
1
1
^
'
1
I
1
1 1
t
i
1
"
1
' CO
^^
4J ^
C (0
01 -Q
H^ afi ' -4 >».
£ 3 ' « M
< 1C
S' O
Oi
cj w
C > J£
co -^ D ^»
§3 w M
IT o e
S W Q C'
1 j
(0
Q
m
00
i t
^
cO
CO
co
efl
0
i-i
03
0 -3-
i
'
i
o
00
3\
1
1
i
elation
Extrap<
i
o
X
o\
CM
CM
1
O
<-4
X
sO
fv.
O
K
00
CM
«n
CM
i
o
X
o
r-
en
o
X
l-f
ao
^
o
i
X
-3-
m
<*
i
o
«H
X
\O
en
CM
1
o
CO
o
X
f-l
CM
«n
i
o
, 1
X
CM
r*
CM
O
X
Sv
i-l
cn
i
o
X
m
^-t
CM
1
O
X
CO
CN
\O
1
0
X
CM
CM
7
0
X
CM
sO
<
1
O
X
o\
«n
CM
i
o
X
o
1 t
«3-
l
O
X
0
SO
0
X
CM
CM
00
0
I-l
X
f^
en
7
o
X
y\
r-4
m
i
o
X
!JN
3%
m
0
X
r^
PO
sf
o
X
CM
CM
f«.
1
O
1
X
*»
m
r^.
0
X
sO
en
m
i
o
00
1
o
X
sO
CM
O
1
o
, 1
X
CO
co
<*
1
0
X
>o
in
ITI
10(
X
m
CM
co
i
0
X
co
1 1
«n
i
o
X
sO
sD
*
O
i i
X
<
in
00
i
o
r-l
X
sO
m
so
1
0
o
1
o
X
sC
-^
CM
O
«^
X
r^
i i
o
(
sD
l
O
X
>n
x>
o-
i
«=
X
<}
3-
m
j\
S
X
sD
m|
i
r^1
0,
1
-------
W
1*3
0
.JO
OH
O »-!
1 <
z
Hl-4
w
H
HO
11
i
Lme/Person)
>es/Litetl
«e
o
CO
2
i
i
t
H
JO
Wi
0,
4J
1H
1-1
3
0)
CO
4J
2
£
Human
1 W CO
O 03 W
H 03 CO
£Q ^3 C^
Upper 95%
Confidence
Limit
Point
Estimate
Upper 95%
Confidence
Limit
Point
Estimate
Upper 95%
Contldence
Limit
Point
Estimate
frt OJ
m o
ON C
0> 4-1
0. "2 -J
£§J
0
1 '
0)
Jj
E
O 4J
Ou co
4-J >i
C CO
CO 00
> J*
-i V^
3 « 00
,cr o e
to
Q
CO
CO
to
O
f-i
pa
o
o
0
m
CM
0
o
CM
O
O
sO
O
O
~
>lation 1
Extrapt
o
r-l
O
fH
O
1
o
f-l
X
«o
1
o
X
f-l
in
o
CM
1
o
f-l
X
«*
ro
1
O
X
CM
CM
1
O
CO
1
o
1
X
CO
1
o
X
CM
1
0
1
1
o
X
CO
1
o
X
CM
0
o
X
00
ro
0
X
CM
m
o
o
1 1
X
x>
co
1
o
X
CM
o
-------
)ETHYLENE
)N
PERCHLORC
INHALATIC
CONTAMINANT
EXt^SURE ROUTE:
i
t
^Person)
j/Lifetime/
0)
09
Q)
0
>«x
Jrf
n
r-i
0£
i
t
4J
T*
! A
)4
O4
4J
01
O
_}
i-l
f-4
^
.0
q-l
sV
3
01
01
CO
-P
1-1
w
i-i
I
Human
>,
1 03 05
O W J->
* W CO
aa to n
Upper 95%
Confidence
Limit
Point
Estimate
Upper 95%
Confidence
Limit
Point
Estimate
Upper 95%
Contidence
Limit
Point
Estimate
Upper 95%
Contidence
Limit
Point
Estimate
Equivalent
Dose
(mg/kg/day)
Bioassay Data
1-1
0
*
0
"
1
0
14
CM
0
*
o
in
d-
> i
en
o
*
O
r-4
ON
i
i
r
i
i
i
i
»
t
k
i
>
n
»
1
1
l>
1
alation
Extrap<
CM
i
^
X
r-l
CM
en
1
O
X
ON
r-(
CM
O
X
CM
CM
en
O
l-l
X
en
CM
CM
1
0
X
r-l
CM
en
i
o
r-i
X
en
CM
tf
O
H
X
^
vO
<
1-(
l-l
00
1
o
~^
X
r-4
CM
CO
o
X
en
sO
00
1
o
X
CM
in
i
°'t
1
X)
*-*
1
o
X
m
a-
vO
I T
o
X
fl
1 1
r*»
i
O
X
r\l
r-l
a\
t
0
i i
X
00
-H
l^s
o
X
m
i i
ON
1
o
X
o
CM
ON
I
O
X
en
vO
ON
1
o
X
CM
sO
1
0
ao
i
X
' in
-i
;
1
!
' oo
p
1 0
X
1 !M
-^
; o
i
o
X
r--
i i
1
; oo
1 i
0
i iS
-1
o
1
c
1 1
' X
1 ^2
CN1
O
t
0
X
c*1
sO
0
1
o
X
(SI
r*»
i
0
1
-------
)ETHYLENE
1
PERCH LORC
INGESTS
CONTAMINANT:
EXPOSURE ROUTE:
i
i
i
1
'Person)
I/Lifetime/
«
CO
10
0
~
.*
CO
1*
OS
,
i
i
i
1
u
1 *
»
O
>*
O
*o
en
1
1
I
1
i
1
i
i
? _.
jlation
Extrapc
i
0
r*
X
ON
9\
1
s
X
9«
CM
1
O
f-l
X
CO
9>
T
o
X
30
CM
1
O
f-l
X
9>
SO
O
X
r^
CM
CM
0
X
m
CM
CM
i
o
X
o
CM
O
o
1-1
1
o
X
CM
CM
0
fl
1
O
X
o
CM
O
f-l
O
X
i-4
CM
CO
o
X
«n
CM
en
t
2
X
o
CM
O
0
1-1
1
o
X
«»
f-l
o
1 1
1
o
X
m
1-1
o
-4
o
X
vO
f-l
sr
i
o
X
vO
CM
«*
1
0
X
o
CM
CM
1
O
0
i-l
1
0
X
.-1
f-l
o
f-l
1
o
X
1-1
f-l
o
1-1
1
o
X
CM
1-1
in
0
X
WD
CM
n
1
o
X
o
CM
CO
1
o
o
r-l
CM
1
0
X
en
r-.
1
0
1-1
X
^
ON
«M
1
o
X
U">
f^
o
l-l
CM
1
o
X
m
cr\
vO
0
X
v£>
CM
vO
1
o
X
o
CM
t
o
0
r-l
X
m
3\
CM
o
X
r^
«*
1
o
f-l
X
^
r>.
fM
1
o
X
CM
m
.-
1
o
fH
X
»N
00
CM
o
X
f-l
r^
7
o
X
tO
CM
r-
1
o
X
o
CM
in
1
0
,
1
1
1
1
i T :
o1
f-tl
X
1 ^>
"1
7
ci'
X
Jv
CM
r-
1
2
X
PO
vO
: 7
2
X
v£>
en
T
0
f 1
X
^
-
CM
1 1
2'
x,
^!
^i
^n
OO I
1 1
01
X
vO
CM
00 1
1 1
o
X
o
CM
i
vO 1
i!
I
i
-------
)ETHYLENE
i
TRICHLOR(
INGESTlOt
CONTAMINANT:
EXPOSURE ROUTE:
1
'c
o
CO
cu
fr.
j/LiJCetimey
0)
CO
CO
u
CO
r*
^J
00
3
rH
w
CO
4J
4J
fH
1
C
I
i i « CO
H CO CO
to m o
Upper 95*
Confidence
Limit
Point
Estimate
fr4? 1)
m o
5\ C*
^ ^J
Ui -O -H
O *H E
Q.-W -H
-T n *~
O
Paint
Estimate
m o
ON C
rJTJ -H
a) -H e
_a c j
0
Point
Estimate
Upper 95%
Confidence
Limit
Point
Estimate
Eqv' "ilent
Dose
(mg/kg/day)
Bioassay Data
l/"^
o
=>
o
CM
O
m
2
o
CM
rH
rH
1
1
1
1
1
I I
elation
Extrap<
i
0
K
r«»
t-4
CM
1
O
X
CM
CM
1
0
X
0
CM
CM
1
O
X
so
«o
1
o
r-4
X
cr>
(N
CM
1
o
X
CO
r-.
CM
1
O
r-1
X
en
en
i
o
cH
X
CO
c*
o
CN
1
O
X
a-
f-l
«*
o
X
CO
Ov
CM
1
O
X
f-l
m
m
i
0
X
0
%o
*~
o
f-l
X
CM
-H
CM
1
O
X
CM
CM
en
1
o
f-l
X
en
*
o
F-l
X
00
ON
1
O
sj-
i
O
X
H
«n
in
i
o
X
CM
,-1
^
0
X
f-l
<-4
en
i
o
*
f-l
CM
1
O
r-l
X
30
-*
en
i
o
X
m
vO
*
i
0
r-l
X
en
in
i
o
r-l
X
9\
9>
CM
1
O
\0
O
X
90
H
CO
O
X
.a-
-*
en
0
X
CM
CM
^~
i
o
X
30
f-l
CM
1
O
l-l
X
f»»
1
m
i
o
X
ON
<-l
in
0
i-4
X
en
\o
i
O
r-l
X
o
o\
7
o
o\
1
o
X
30
fM
*~
t
O
X
9\
*
1
O
1
X
X
<£>
m
0
X
ON
^~
00
1
o
l-l
X
en
ON
i
o
r-l
X
ON
ON
NO
1
O
O
1
O
X
0
m
r--
i
0
X
3-
rH
*
-H
ON
1
0
r 1
X
ro
i
0
0
J
X
ON
Ov
r^'
i
0
-------
1 : >^
1 ' i fl 03
1
'
i
m
! en
O W 4JI 1 O
1-1 w (B
33 (B O
X 0)
! |in o
ION C
1 A) -^ e
ICL4-( -^
^
JO
to
a,
ac J
=> o
y
01
4J
4J CO
c e
1-1 ~>->
0 *J
Oi W
w
x a>
m o
ON C
0) 4J
u -o «H
a> -^ 6
CO ' CL.U-I -«-i
U a C rJ
.
0
f-^
fs^
O
!
1
'
X i »* o
^^
a>
E
/B^
5
§
u
u-
w
o
OS
4 fM
si
OM
H
Jto
fcW
zu
i-tjs
>«-
..
oq
0
~4 J
4J
01
>M
a>
4J
4J 03
C
1-1 -i-i
f> : O 4->
J
-^.
a*
m
e«
u
_v
a
cH
OS
^^
t 1
«--1
£>
fl)
?
1
01
M
H-l
4->
1
I
O4 W
u
»« 01
m o
ON C
a* 4-1
JJT) **
0) -^ E
a. 4-1 -i-i
CLC rJ
3 O
30
v
0
;
1
! 1
i
i
i
t
i
p
t
1 !
i
i
i '
I i
1
' '
i
U i
41
4-1
jJ cd
C E
-4 -F-C
O -U
ft- CO
u
»* OJ
m o
ON C
0> 4J
|J T3 ft
V * B
a. -i -^
ac J
= 0
o
0)
u
*j CQ
c e
r-l -^
O *->
a. «
U3
i
i
1
I
i
*
1
1
W i '1
H
=>
HO
202
ZW
(D
4-1
X-N flj
4J >J Q
C nj1
4) T3 >,
i-tafi -! ~--
r^ (CM
! e > jc
HO
zo-
ox
ow
« -rJ V -^
§3 to oc
cro 6
W
CO
CO
«
o
*
i ^^
O
CM
O
s: w a s-x CQ
o
t
1
O
O
CM
1
1
*
P
1
)
>lation
Extrapc
i
o
X
ON
30
*»
1
O
iH
X
CM
SO
1
o
K
ON
30
*
1
0
iH
X
CM
00
1
O
X
ON
30
1
O
X
^-1
00
1
o
X
o
ON
1
o
X
m
00
o
i
i
0
X
en
01
i
0
»4
X
en
~4
1
O
X
en
CM
1
O
f-4
X
<*
l-l
o
X
ON
CM
1
O
X
ON
.-4
1
O
X
r-t
CM
O
X
r*.
f-4
i
0
en
i
O
X
NO
CM
-*
1
0
l-l
X
r»»
«n
CM
o
X
CM
L«-
fO
1
O
f-l
X
IT)
<*
CM.
1
O
X
en
in
CM
0
X
NO
CM
CM
1
O
X
en
CM
CM
1
0
X
ON
i-H
CM
1
O
r^
o
X
>3-
e*>
00
O
r-l
X
lf\
en
«»
i
o
X
30
a-
*tf
f
O
1 1
X
en
<
en
i
O
X
CM
00
en
i
o
X
l-l
m
en
t
0
X
en
CM
en
o
X
ON
i-l
en
1
o
CO
V*
0
X
en
**
«*
I
3
X
m
CM
tf\
1
0
X
r»»
i-i
sO
1
0
i-4
X
N£>
en
en
i
O
X
CM
r-t
&
O
X
30
en
-------
^v
ft
o
~
8
H
DC
95
VINYL CHI
INHALATIC
CONTAMINANT.
EXPOSURE ROUTE:
'Person)
i/Lifetime/
0)
CO
ce
u
. >~s
fX
TO
H
«
4J
H
J3
1 A
1 . W
1 0U
«
OC
0
fj
fH
rH
J3
i-l
01
01
at
ce
4J
1-1
4J
rH
P
jw
Human
i crj (0
O CO W
co ce o
Upper 95%
Confidence
Limit
Point
Estimate
Upper 95%
Contidence
Limit
Point
Estimate
Upper 95%
Confidence
Limit
Point
Estimate
Upper 95%
Coniidence
Limit
Point
Estimate
Equivalent
Dose
(mR/kg/day)
Bioassay Data
0
-4
O
0
r^
i-4
O
co
CM
r-»
CN
O
.H
iH
r-.
co
o
CO
CM
.*
in
O
o
d
1
!
i
i
i
*
f
)lation
Extrapt
CM
1
o
X
in
3A
CM
O
K
00
CO
1
0
K
O
ft
CM
1
O
t-4
X
00
-a-
T "
O
X
CM
^
CM
O
iH
X
r^.
in
co
i
o
X
CM
9\
CO
1
0
X
in
vO
o
CM
1
0
X
ON
rH
CO
1
O
X
r-
-4f
CM
1
O
X
m
CO
CM
i
s
X
CM
<-t
CM
1
O
X
r»-
<
CM
1
O
iH
X
00
I
*
r».
CO
i
O
X
\D
SO
CO
1
O
iH
X
30
fH
vO
1
O
X
CM
O\
SO
O
X
m
vO
co
i
O
vO
o
X
l-l
CM
f-
o
X
r-
i-l
3-
1
O
X
00
oo
-tf
1
o
H
X
30
fH
CO
1
O
X
CO
CM
st
o
fH
X
m
m
7
o
X
CM
ON
r-.
i
O
X
m
vO
T
o
""
00
1
o
X
>»
CM
ON
O
X
-*
fH
<3-
l
0
X
m
CM
m
i
o
fH
X
^
<
-*
1
0
X
^~
3D
sr
1
o
iH
X
r^
fH
00
1
o
X
CM
ON
00
1
0
X
m
vO
in
i
o
o
1
o
X
CO
rH
CM
1
0
X
o
vr>
m
i
o
X
r-.
s£>
m
i
0
*
1
O
X
<
CM
m
o
4
X
-a-
m
ON
1
o
X
CM
ON
ON
1
0
X
m
vO
«e
o
ro
l
0
X
m
r->
^
in
i
0
X
30
fH
vC
1
O
3
CM
m
i
0
1 x
CS)
oo
m
i
0
«
4
o
1
O'
X1
CM1
ON
O '
I
0!
X!
mi
vO1
!
1
ro
t '
0!
1
1
1
-------
APPENDIX C
OCCURENCE DATA FOR SELECTED VOC'S IN DRINKING WATER,
POPULATION CONCENTRATION AND STANDARD ERRORS
(FROM REFERENCE USEPA, JRB REPORTS)
-------
This Appendix is taken directly trora the Occurrence/
Exposure documents being developed for potential drinking water
regulation. This section describes what methodology is used to
develop estimates of national occurrence. The U. S. Environ-
mental Protection Agency (EPA) is currently considering the
proposal of national revised primary drinking water regulations
under the Sate Drinking Water Act, as well as non-federal regu-
latory approaches, to limit human exposure to high levels of
certain volatile synthetic organic chemicals (VOC's) that have
been detected in drinking water (Advanced Notice of Proposed
Rulemaking, 47 FR 9350, March A, 1982). The specific VOC's ot
immediate interest to EPA are:
0 Trichloroethylene
Tetrachloroethylene
Carbon Tetrachloride
0 1,1,1-Trichloroethane
0 1,2-Dichloroethane
0 Vinyl Chloride
0 Dichloromethane
0 Benzene -
0 Chlorobenzene
0 Dichlorobenzene(s)
0 1,1-Dichloroethylene
0 1,2-Dichloroethylene (cis and trans)
o
o
D-i.
-------
The objective ot the analysis of the occurrence ot the
VOC's listed above is to support EPA's consideration of the
need and alternatives tor controlling VOC's in public water
supplies in two principal areas:
0 As input to the health risk assessments tor the
VOC's, the analysis provides estimates of the
number of individuals in the United States
exposed to various levels of VOC's in drinking
water rrom public water supplies.
c As input to the assessment of the economic im-
pact or the regulatory and treatment alterna-
tives being considered, the analysis provides
estimates of the number 01 public water supplies
of various source (i.e., groundwater and surface
water) and size (based on population served)
categories having VOC's present, and the distri-
bution of VOC levels in those water supplies.
The methodology for preparing these estimates involvec tae
creation ot a data base drawing from the results of several
Federal surveys on the measured occurrence of the VOC's in
public water supplies as a function of water source and supply
size. Statistical models were then used to extrapolate from
the observed frequency of occurrence 01 the VOC's in the sup-
plies sampled in .the Federal surveys to the universe of public
water supplies having similar source and size characteristics.
A separate report has been prepared for each of the VOC's
listed. Section A is included in each report to provide detail
on the sources of data and the methodology used to obtain these
estimates. Some specitics presented in Section A may not be
applicable to the VOC that is the subject of this report.
Section B provides additional detail on the selection of an
D-2
-------
appropriate modei for estimating the national occurrence of the
VOC that is addressed in this report.
D-3
-------
A.I SOURCES OF DATA
A.1.1 Number of Public Drinking Water Supplies in the United
States and Size of Populations Served
It is currently estimated that there are approximately
60,000 public drinking water supplies in the United States
serving approximately 214 million people. Table A-l summarizes
the estimated number of surface water and groundwater systems
of various sizes and the associated populations served by then.
These data, which correspond to the "FY 82 Characterization ot
the Water Supply Industry" presented by Kuzmack (1983), as
updated by Schnare*, were derived from the Federa1. Reporting
Data Systems (FRDS) for FY 1982 (FRDS, 1983).
It should be noted that FRDS (1983) does not provide a
breakdown ot tht number of systems nor of the population served
for the 10,001-25,000 and 25,001-50,000 size categories, but
rather tor 10,001-50,000 as a single category. The estimated
number of systems in the 10,001-25,000 and 25,001-50,000 size
categories as presented in Kuzmack (1983), were estimated by
Dr. David Schnare of the Office of Program Development and
Evaluation, EPA Office of Drinking Water tram additional FRDS
data. (Data for these additional categories are needed for the
economic impact analysis.) The population served by systems in
* Personal communication between David Schnare, Office of
Drinking Water, U. S. Environmental Protection Agency, and
Frank Letkiewicz, JRB Associates, May 25, 19o3.
D-A
-------
these size categories were estimated by JRB as shown in TaDle
A-2.
D-5
-------
r-l
|
<
3
A
£
«Q
«
>
«
W U
CO U
OS
Z«M -_
»*J ^^
O 05-
rH CO OS
H 0
^^ K
i-j iJ u
S Ou H
o-o, <
0 = 0
PM CO
Z
COS O
Z M rH
g
|
I
C?
0)
5
-j
CO
c
o
CD .H
it? CB "S
P rH >
0> 3 >->
> O. 01
< O W
CM
e'en
O-o
eo co
f_l 3
9 0
ax;
O 4J
0- ^^
CO
U-l g
O CD
4J
CO
0 >,
z to
c
o
CD -r»
004J-0
CO CO CD
0) 3 P
> ao>
< 0 W
OH
C 'w'
OT5
-r< C
4J CO
rH 3
3 O
ax:
O 4-1
p- >-x
CO
IH £
O CD
4-1
CO
O r*i
Z CO
y-S
o
4)
«t
N 0)
i-l CO
CO
c
e o
4-> 4J
(0 CO
cols
o
a
^^
i
^ en so o
! vn < en o
CM P-s. ,£>
_H
rH -it O l*»
o oo m o
_. en en r*.
m «* r*. o
CM r-. r«. o
r-l vO 00 »^
« . Bk
t^ m ^ ^
r"^ r-4
so so en O
m oo so o
CM t^ r*-
i
\o o »^ in
gp C?N in c?v
\O O CM
r-l m
vn CM r-». m
CM i r*. «*
in ^ c*) &\
» » * *
,-1 CSJ i-H ^
0 0
0 O
o o o m
o o
r^ in t-1 CM
i i i i
CM O O O
-i m o
*
r-l
_,
r-l
g
CO
4
^ 1
P CO
V g
> to
o
o
CTN
CM
00
m
CM
^.
ON
00
0
o
CM
m
rH
in
^
0
O
en
i
o
m
CM
3 O
0 0
1 CM
-* r*.
O «*
>> 0
^ oo
«n oo
vO sD
^5 ^ ^
«
-^ r-t
O O
0 O
* *
^kf P^
^ CO
^5 ^^>
fO \O
^^ C-J
^ CO
r** ON
o
o o
=> o
o
« s
1 1
0 O
en O
en m
^
TJ
OJ
2
O
O
o
m
x>
CM
CM
cd
vn
en
00
o
o
o
^
t
m
m
r-l
CO
m
,».
Cj\
O
O
fi
«
m
CM
i
o
o
»
o
r-l
O
o
o
30
en
x>
o
s
a
o
CM
0
O
o
^*
en
X)
«n
en
cfl
O
o
^
o
o
o
t
o
«n
i
0
0
»
m
CM
O O O
O O O
o o o
r-i m o
sO 30 sD
CM
r- «tf so m
f oo 30 r*
en ^^ ^ en
rH r»*
, 5
u P H
CO (DO
iJ > H
rH
CO
C
CD -
c CD
o -u
ir^ *»"^
> U
c o
Ii3 co
CO
CO
tt
OS
» »
P N
CD cj
4J -F-I
CO 3
-1
CO CO
0 C
o
>,m
X> P
CD
o a
CD
4J CO
a)
o c
a-r)
s .
CO >,
gft f i
CO W
- CD
en 60
CO ^C
o\
rH C
..2
i*- ^
O U
CO U
-------
TABLE A-2
ANALYSIS FOR ESTIMATING THE POPULATION SERVED
BY SURFACE WATER AND GROUNDWATER SUPPLIES IN THE
10,001-25,000 and 25,0)1-50,000 POPULATION SIZE CATEGORIES
System Size
(population
served)
10,001-25,000
25,001-50,000
TOTAL
Source
Surface Water
w
y
29,54Ub
Groundwater
X
z
23,250°.
Total
27,870*
24,920*
5*,7yO
a Estimated total population for surface water and groundwater
provided by K' zmack, (1983).
b Population served by s n uce water and groundwater supplies
in the 10 001-50,000 Ci,t«£ory from FURS, (19o3).
w - 29.540 x 27.870 - 15,595.
52.79U
x - 23.250 x 27.870 - 12,27b.
52,790
y 29.540 x 24,920 - 13,945.
52,790
z - 23.250 x 24.920 - 10,977.
52,790
D-7
-------
A.1.2 Data on Measured Occurrence ot VOC's in Public
Water Supplies
There are three primary sources ot information avail-
able on the occurrence of VOC's in drinking water supplies that
were considered for use in preparing the national projections:
0 Federal surveys
0 State data
0 Miscellaneous data
Only the Federax survey data were ultimately used for
the national projections. While a substantial amount of state
ana miscellaneous published data were available for some of the
VOC's examined, these la .s u~ually did not provide adequate de-
tails on tht water source, size ot the population served, or
type of sample taken (raw, finished, or distribution), which
precluded incorporating them into the analysis. Furthermore,
there was usually no detailed information available on the
sampling and analysis methodology used to obtain the data,
which precluded their being subjected to a quality assurance
review (performed as a separate task) (JRB Associates, 19b2).
In addition, much of the state and miscellaneous data appeared
to have been obtained in response to spills, citizen com-
plaints, or other evidence of contamination and, therefore,
were not considered to be representative data tor preparing the
national projections. While the state and miscellaneous pub-
lished data were not used tor deriving the national projections,
D-8
-------
these data are presented and discussed in the individual VOC
occurrence reports.
The Federal survey data generally provided the informa-
tion on water source, population served, and sample type
studied that was necessary to perform the analyses in this
report. These surveys also provided sufficient information
on trie sampling and analysis methods to be subjected to the
qualicy assurance review. The following six Federal surveys
were used for the national projections:
0 National Organics Reconnaissance Survey (NORS)
The National Crganics Reconnaissance Purvey
(NORS) was conducted early in 1975 for t e purpose
of determining levels of four trihalomet.ianes
(chloroform, bromodichlororoethane, dibromochloro-
methane, and bromoform), carbon tetrachloride, and
1,2-dichloroethane in finished water supplies from
80 cities across the country (Symons et al., 1975).
A population base of 36 million individuals was
covered during the study. Analysis ox samples was
performed by the Water Supply Research Laboratory
of EPA in Cincinnati using purge and trap gas
chromatography with an electrolytic conductivity
detector.
0 National Organic Monitoring Survey (NOMS)
The National Organic Monitoring Survey (NOMS)
was conducted to determine the frequency o± occur-
rence of specific organic chemicals in finished
water supplies or 113 cities across the country
(Brass et al., 1977). Among the chemicals surveyed
were trTHaTomethanes, 1,2-dichloroethane, carbon
tetrachloride, trichloroethylene, benzene, vinyl
chloride, dichlorobenzenc, and tnchlorobenzene.
Data from three phases (referred to as NOMS I,
NOMS 11, and NOMS 111) or the study were collected
over an eleven month period (March 1976 to January
D-9
-------
1977) to retlect any long-term or seasonal varia-
tions. The analytical treatment of the samples was
similar to that tor the NORS samples. (Gas chroraa-
tography/mass spectrometry analyses were done for
benzene.)
National Screening Program for Organics in Drinking
Water (NSP)
SRI International conducted a study from June
1977 to March 1981, entitled "National Screening Pro-
gram for Organics in Drinking Water" (NSP), in which
raw and finished drinking water samples were collect-
ed from 16o water facilities located in 3i states
(Boland, 1981). The compounds sampled were 23 halo-
carbons, 6 aroinatics, and 22 pesticides, phenols, and
acids. The methods used for anaylsis included gas
chromatography with electron capture detection tor
purgeable halocarbons and the base/neutral extraction
fraction, and gas chromatography with flame ioniza-
tion detection for purgeable aromatics.
0 Community Water Supply Survey (CWSS)
The Community Water Supply Survey (CWSS) ex-
amined 106 surface water supplies, 330 groundwat^r
supplies, and 16 supplies with mixed water or pur-
chased sources o.n 1978. Trihal one thanes and other
volatile organic chemicals, including carbon tetra-
chloride, chlorobenzene, 1,2-dichloroethane, cis-
and trans-1,2-dichloroethylene, tetrachloroethylene,
1,1,1-trichloroethane, trxchloroethylene, benzene,
toluene, and xylenes were measured. One to five
samples were coleicted from each system, including
raw, finished, and/or distribution water. Gas
chromatography with an electrolytic conductivity
director was used for halocarbons and a flame
ionization detector for aromatic analyses.
Rural Water Survey (RWS)
The Rural Water Survey, conducted in 1978, was
carried out in response to Section 3 of the Safe
Drinking Water Act, which mandated tnat EPA "conduct
a survey of the quantity, quality, and availability
of rural drinking wat«r supplies'. Samples collected
from 855 households in rural areas from across the
D-10
-------
United States were analyzed for trihalomethanes and
for carbon tetrachloride, 1,2-dichloroethane, cis-
and trans-1,2-dichloroethylene, tetrachloroethylene,
1,1,1-trichloroethane, and trichloroethylene using
gas cnromatography with an electrolytic conductivity
detector. The majority of the 855 samples were from
households using private wells or small supplies
serving fewer than 25 people. Using information pro-
vided by Dr. Bruce Brower at Cornell University,
Department of Rural Sociology, it was determined that
the RWS file had data tor up to 207 groundwater and
up to A3 surface water supplies serving more than 25
people.
Groundwater Supply Survey (GWSS)
. The Groundwater Supply Survey (GWSS), conducted
between December 19bO and December i9ttl, involved the
national sampling of 945 public water supply systems
w^ir.^ groundwater sources tor 5 trihalomethanes and
29 other organic chemicals. Analyses were done using
purge and trap gas chromatograp.iy with an electrolytic
conductivity detector for halocarbons and a non-
destructive photoionization detector for aromatics.
There were 466 randomly selected supplies and 479
selected with state and EPA regional input based on
th: likelihood of finding some VOC contamination.
Table A-3 inaicates which of the VOC's listed earlier
were examined in each of the six Federal surveys used for the
national projections.
D-11
-------
TABLE A-3
VOC'S OF INTEREST EXAMINED IN THE SIX FEDERAL SURVEYS
USED FOR NAiIONAL PROJECTIONS
Trichloroethylene
Tetrachioroethylene
Carbon Tetrachloride
1,1, 1-Trichloroethane
1 ,2-Dichloroethane
Vinyl Chloride
Dichlorome thane
Benzene
Chiorobenzene
Dichlorobenzene(s)
Trichlorobenzene(s)
1 , 1-Dichloroethylene
1 ,2-Dichloroethylene
(cis and trans)
NORS NOMS
X
X
X X
X
X X
X
X
X
X
X
NSP
X
X
X
X
X
X
X
X
X
X
X
X
X
CUSS RWS
X X
X X
X X
X X
X ' X
X
X
X
GWSS
X
X
X
X
X
X
Xa
X
X
X
X
X
a Dichloromethane data for the GWSS were not used due to a
sample contamination problem.
D-12
-------
A.2 GENERAL METHODOLOGY
A.2.1 Development of Survey Files
To facilitate the handling of data for preparing the
national projections, suitable machine-readable files were
developed for each ot the six Federal surveys. JRB was pro-
vided access to existing computer files for the CWSS, RWS, and
GWSS through the EPA Office of Drinking Water's Technical Sup-
port Division (OWD/TSD) in Cincinnati, Ohio. JRB used the
published results of the NORS (Symons et al., 1975) and the
NSP (Boland, 1981), and printed results for the NOMS provided
by PDt ,/TSi) to create macuine readable flies of thosf surveys.
Three separate files were created for NOMS, one for each of
the three phases. (In effect, NOMS I, II, AND III were treated
as three separate surveys.) The final files for all chemicals
were in SAS format. All computer erforts for this project
utilized EPA's NCC-IBM (IBM 370) computer at Research Triangle
Park North Carolina.
It was necessary to prepare working versions ot each
survey file containing the following minimum information for
each of the sampled water supplies:
0 Location of the supply (state and city)
0 Population served by the supply
0 Water source (groundwater, surface water, mixed,
purchased, etc.)
1-13
-------
" A single concentration value for each VOC sampled
With the exception of the RWS and NSP, the existing
files and printed sources provided adequate information on the
location of the supply sampled. The RWS design involved the
collection ot drinking water samples from households in rural
areas of the United States. With the assistance of Dr. Bruce
Brower at Cornell University's Department ot Rural Sociology
(responsible for the preparation of a detailed analysis of RWS
results on inorganics, pesticides, and other parameters), it
was possible to determine which of the 855 households for which
VOC analyses were done obtained water from public water sup-
plies. However, because of confidentiality restrictions on the
RWS data, it was only possible to determine tne location or the
household and the public water supplies sampled at the state
and county leve^, but not at the city or town level.
For NSP, the locations were not reported in Boland
(19bl); however, the Ofrice or Drinking Water, Science and
Technology Branch (ODW/STB) was able to provide copies of data
sheets on the supplies sampled in NSP which provided inrortna-
tion on location.
The existing files ot the CWSS and GWSS each provided
data on the size of the population served by the supplies
sampled. For NORS, it was necessary to estimate the population
served using information presented in Symons e_t al. (1975) on
D-14
-------
tht location ot the supply and recent population aata ior those
areas from other sources. It should be noted that most of the
supplies sampled in tne NORb fall into the large and very large
size categories. Consequently, errors in the precise number
used for persons served by those systems would not alter their
classification or their impact on the national projections.
For NOMS data on the populations served by supplies
were provided by ODW/TSD. In the case of NSP, the population
served by each supply was not reported in Boland (19»1).
Again, those data were obtained from the data sheets provided
by ODW/STB (There were three NSP locations tor which tht
population was n^t specified.)
For RWb', data on the size or the population served by
supplies were not collected. However, data were obtained on
the number ot service connections for each supply. With the
assistance of Dr. Bruce Brower at Cornell University, it was
possible to estimate the population served by each supply xrom
the service connection data and data on the average number of
individuals per household observed in the survey (3.034).
The identification of water source as groundwater, sur-
face water, mixed, etc., was clearly designated in the CWbS,
RWS, and GWSS files. For NORS, the source was determined from
the descriptive information in Symons et al. (197^). For NOMS,
source designations were provided by ODW/TSD. For NSP, source
D-15
-------
information was given on the data sheets provided by ODW/STB.
Some public water supplies use a mixture of groundwater
ana surface water sources. Although these "mixed" supplies are
counted as groundwater or surface water among the 60,000 sup-
plies in FRDS based on the predominant source used, such -sup-
plies were excluded from the survey data for developing the
national projections because the predominant source was rarely
indicated in the survey file. Similarly, water supplies iden-
tified as purchasing water from another, usually unspecitied,
supply were excluded from the survey data for the national pro-
jections.
A 2.2 Computing Average Values for VOC's in Each Survey
In order to prepare the national projections, it was
necessary that a single value be obtained for each VUC in each
supply sampled. This requirement presented certain diffi-
culties zor scverai 01 the files where multiple sample results
were reported for the supplies. NORS and NSP provided data on
raw (i.e., untreated water sampled at the supply) and finished
(i.e., treated water sampled at the supply) samples; CWSS pro-
vided data on raw, finished, and distribution (i.e., water
sampled at a user's faucet) samples; NOMS and GWSS used fin-
ished water only; ana RWS usc-d distribution water only In
order that the national projections be derived from data on
drinking water samples most representative ox what people
D-16
-------
actually consume, all data on raw water were excluded troro this
analysis. Initially, consideration was given to excluding fin-
ished water sample data in the CUSS tor supplies also having
distribution samples. However, ODW/TSD staff indicated that
inconsistencies in coding data tor the CWSS resulted in some
errors in designating water as finished or distribution.
Therefore, all CWSS tinished and distribution samples were in-
cluded in the analysis as being equally representative of the
Lu wiiicli consumers are exposed.
The NORS, NOMS and GWSS provided a single analytical
result tor each VOC in linisned water. For NSP and RWS , there
was generally only a single va? IB reported for each VOC,
although a few systems had mult pi; samples. In CWSS, most
supplies had multiple finished and/or distribution samples for
each VOC. Where multiple samples occurred, a single "supply
value" was computed for each VOC using the following rules:
0 If positive values were reported for all
samples, the supply value was computed as
the arithmetic mean.
0 If both 'positive values and values below
trhp minimum quantifiable concentration
were reported, the supply value was com-
puted as the mean ot the positive values
and the minimum quantifiable concentra-
tion values.
0 If values Dtlow the minimum quantifiable
concentration were reported for all
samples, the supply value was computed
as the mean of the minimum quantifiable
limits reported for each sample. (These
means were recorded as "negative" values
to inuicate that the VOC was not observed.)
D-17
-------
Computing a single value tor supplies where the value
for one or more of the samples was reported to be below the
minimum quantitiabie concentration was problematic. These
"negative" values imply that the analyte in question may or may
not be present and, it present, is so at a concentration below
that measurable by the analytical method. In other words, the
actual value is greater taan or equal to zero and less than the
minimum quantifiable concentration.
Where a supply was reported to have samples with ooth
positive and negative values, two major alternatives were con-
sidered for treating the negative values. The first was to
ignore or eliminate the negative values from the computation
of the mean. This was considered unacceptable because it
implies that the negative data are less valid than the positive
data, which is not the case. This alternative would also
necessarily result in a higher average value for that system
than would bu the case if the actual value nor the negative
data were known.
The second alternative was to assign the negative data
a specific value for computing the supply average. Three
possibilities were considered 0, tue minimum quantifiable
concentration, and the midpoint between 0 and the minimum
quantifiable concentration. Assigning the negative samples a
value equal to the minimum quantifiable concentration was
selected since this gives the most conservative estimate ot the
D-18
-------
supply vaiue. That is, ii the analyte was in tact
the maximum possible concentration it could have in the sample
would be approximately (actual.y slightly less than) the min-
imum quantifiable concentration. For example, if a supply was
reported to have one sample with a VOC present at 0.3 micro-
grarcs/liter and another sample in which the VOC was not
observed at a minimum quantifiable concentration of 0.1 micro-
grams/liter, a supply value of 0.2 micrograms/liter was re-
corded in the working rile. Using the minimum quantifiable
concentration with other actual positive values to compute the
mean results is the most conservative estimate of the supply
value utilizing all sample data.
The treatment of supplies having only negative va uts
reported derives from the treatment of those with negatives and
positives described above. If, for example a VOC was not
observed in two samples from a given system at minimum quanti-
fiable concentrations of 0.1 micrograms/liter and 0.3 micro-
grams/liter, respectively, the system value retained in the
working file was a "negative" 0.2 micrograms/liter. That is,
the VOC was not observed, but if it had been present in both
samples, the maximum possible average concentration for that
supply would have been 0.2 micrograms/liter.
A.2.3 Combining Data from the Federal Surveys
Once the multiple samples for each VOC were averaged to
D-19
-------
obtain a sing-e value for each supply sampled, tables on the
frequency of occurrence of each chemical were prepared for each
survey as presented j.n Sections 2.2.1 and 2.3.1. In addition,
the mean, median, range, and other statistics were computed for
the positive values in each survey.
The next step in developing the national projections was
to combine the results of ail or the surveys together. In
doing this, it was necessary to identify those supplies that
had been sampled in mor* than one survey and compute an average
supply value for each VOC. (The rules for averaging samples
within a survey described in Section A.2.2 applied to computing
averages across surveys.) It should be noted that supplies
sampled in the RViS could not be matched against the other sur-
veys since the RWS locations could only be determined at the
state ana county level, as previously described.
Table A-A presents a list of those systems which were
duplicated across the Federal surveys. When a system was
sampled in two or more surveys, the population used for that
system in the combined survey lile was the one reported in the
most recent survey, represented by the following chronological
order (most re-cent first): GWSS NSP, CWSS, NOMS, and NORS.
D-20
-------
TAbLE A-4
SYSTEMS SAMPLED IN MORE THAN ONE SURVEY
Location
Source
Survey in Which the
System was Sampled
Tuscon, A£
Fresno, CA
Jacksonville, FL
Idaho Falls, ID
Rockroro, XL
Campbellsburg, IN
Soutn Bena, IN
Baton Rouge, LA
Hammond, LA
Lafayette, LA
Senath WO
Webb City, MO
Greenville, MS
Kearney, NE
Lincoln, NE
Albuquerque, NM .
Baldwinsvili^, NY
Dayton, OH
Aliquippa, PA
Sioux Falls, SD
Memphis, TN
San Antonio, TX
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
GWSS, NORS
GWSS, NOMS
NOMS, NORS
GWSS, NORS
GWSS, NOMS
GWSS, CWSS
GWSS, NSP
NSP, NOMS
GWSo, CWSS
CWSS, CWSS
GWSS, CWSS
GWSS, CWSS
NOMS, NORb
GWSS, CWSS
NOMS, NORS
NOMS, NORS
GWSb, CWSS
NSP, NOMS, NORS
GWSS, CWSS
GWSS, NOMS
NOMS, NORS
GWSS, NOMS, NORS
D-2
-------
TABLE A-4 SYSTEMS SAMPLED IN MORE THAN ONE SURVEY (continued)
Location
Provo, UT
Marysville, WA
Spokane , WA
Madison, WI
Poweil, WY
Birmingham, AL
Camden, AR
Little Rock, AR
Phoenix , AZ
Concord , CA
Los Angeles, CA
Oakland , CA
Sacramento, CA
San Diego, CA
San Francisco, CA
Denver , CO
Pueblo, CO
New Haven, CT
Waterbury, CT
Washington, DC
Atlanta, GA
Davenport , 1A
Chicago, 1L
Source
G
G
G
G
U
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
b
S
S
Survey in Which tne
System was Sampled
CWSS
GWSS
GWSb
NSP,
GWSS
NSP,
NOMS
NSP,
NSP,
NOMS
NOMS
NSP,
NSP,
NOMS
NOMS
NSP,
NOMS
NSP,
NOMS
NSP,
NSP,
NSP,
NS±>,
, NOMS
, CWSS
, NSP, NOMS
NOMS
, CWSS
NOMS
, NORS
NOMS
NORS
, NO'.S
, NORS
NOMS
NOMS
, NORS
, NORS
NOMS, NORS
, NORS
NOMS
, NORS
NOMS, NORS
NOMS, NORS
NOMS, NORS
NOMS, NORS
D-22
-------
TABLE A-4 SYSTEMS SAMPLED IN MORE THAN ONb SURVEY (continued)
Location
Fort Wayne, IN
Indianapolis, IN
Whiting, IN
Topeka, KS
]_mii ciri 1 1 o VV
Boston, MA
Lawrence , MA
Baltimore, MD
Portland, ME
Detroit, MI
Grand Rapids, MI
Mount oj.emeni>, MI
St. Paul, MN
Cape Girardeau, MO
Kansas City MO
St, Louis, MO
Jackson, MS
Charlotte, NC
Bismark, ND
Omaha , NE
Manchester, Nh
Elizabeth, NJ
Passaic, NJ
Source
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
Survey in Which the
System was Sampled
NSP,
NSP,
NOMS
NSP,
NSP,
NSP,
NSP,
NSP,
NSP,
NSP,
NSP,
NOMS
NOMS
NOMS
NSP,
NSP,
NSP,
NSP,
CWSS
NSP,
NSP,
NSP,
NSP,
NOMS
NOMS,
, NORS
NOMS,
NOMS
NOMS,
NORS
NOMS,
NOMS
NOMS,
NOMS
, NORS
, NORS
, NORS
NOMS,
NORS
NOMS
NOMS
, NOMS
NOMS
NOMS
NOMS
NOMS,
NORS
NORS
NORS
NORS
NORS
NORb
NORS
D-23
-------
TABLn A-4 SYSTEMS SAMPLED IN MORE THAN ONE SURVEY (continued)
Location
Bufraio, NY
Poughkeepsie, NY
Cincinnati, Oh
Cleveland, OH
Columbus, OH
Toledo, OH
Oklahoma City, OK
Tulsa, OK
Corvailis, OR
Eugene, OR
Portland, OR
Harrisburg, PA
Philadelphia, PA
Pittsburgh, PA
Newport, RI
Providence, RI
Charleston, SO
Huron, SD
Chattanooga , TN
Nashville, TN
Brownsville TX
Dallas, TX
Fort Worth, TX
Source
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
S
Survey in Which the
System was Sampled
NOMS, NORS
CWSS, NOMS
MSP, NORS
NSP, NOMS, NORS
NSP, NOMS NORS
NSP, NOMS
NSP, NOMS, NORb
NSP, NOMS
NSP, CWSS, NOMS, NORS
NSP, CWSS, NOMS
NSP, NOMS
NSP, CWSS
NSP, NORS
NSP, NORS
NOMS, NORS
NSP, NOMS
NSP, NOMS, NORS
NOMS, NORS
NSP, NOMS, NORS
NSP, NOMS, NORS
NOMS, NORS
NOMS, NORS
NSP, NOMS
D-24
-------
TABLE A-4 SYSTEMS SAMPLED IN MORE THAN ONE SURVEY (continued)
Location
Source
Survey in Which the
System was Sampled
Houston, TX
Salt Lake City, UT
Annandale, VA
Richmond, VA
Illwacc, WA
Seattle, WA
Milwaukee, WI
Huntington, WV
Wheeling, WV
S
S
S
S
S
S
S
S
S
NSP, NOMS
NSP, NOMS, NORS
NOMS, NORS
NSP, NOMS
NOMS, NORS
NSP, NORS
NSP, NOMS, NORS
NSP, NOMS, NORS
NOMS, NORS
G - Groundwater.
S » Surface water.
D-25
-------
A.2.4 Testing for Biases in the Data
There was concern that the selection of sample sites in
some of tht surveys was biased towaru VOC contamination,
which, when combined with other random survey data, would bias
the national projections toward a higher estimated trequency
of occurrence and/or mean concentration. Purposeful selection
of sites having a higher tnan average probability ot contami-
nation with VOC's was in fact the design of the nonrandom
portion of the GWSS. It has been suggested that the NORS,
NOMS, and NSP surveys also may have involved a bias towards
systems known or suspected to be contaminated; however, this
could not be confirmed.
Initially, consideration was given to excluding the
GWSS nonrandom data for all VOC's when combining the survey
data because the analysis presented by Westrick et al. (19»3)
showed that 21.2% of the random supplies had at least one VOC
above its quantitation limit, whereas 27.3% of the nonrandom
supplies showed contamination. The nonrandom portion of the
GWSS was reported to have a higher ir^quency 01 occurrence of
VOC's at all concentration levels for both large and small
systems. At higher concentrations, there was a two to four
times higher frequency of occurrence of VOC's in the nonrandom
sample. It is important to nott that these comparisons are
based on the combined results for all VOC's. There was some
concern, however, that sites selected because or suspected
D-26
-------
contamination with one or anotner specific VOC would not
necessarily be biased for other VOC's. Westrick et al. (1983)
did not present a comparison of the random and uonrandom data
on a chemical-by-chemical basis, so it was not clear whether
it was appropriate to exclude the GWSb nonrandora data for each
VOC. It was decided, therefore, that an analysis would be
performed on each VOC to evaluate whether there was a statisti-
cally significant difference in the frequency of occurrence
anu in the mean or the positive values observed in the random
and nonrandoro portions of the GWSS.
To test the difference in trequency ot occurrence, the
results for each "OC at each site were classified as positive
or negative and summarized in a two-way t« ole as shown below:
Portion of
GWSS
Random
Nonrandom
Total
Results
Positive Negative
"11 "12
"21 "22
".1 ".2
Total
"1 .
"2.
n% f
Comparisons of relative frequencies of positive results
in the random and nonrandom segments of the GWSt> were based on
the X2 statistic:
2 X2 » (nnn22 - n12n2l) n_
"1. "2. ".1 ".2
with n's defined in the table above. Under the null hypothesis
D-27
-------
that the relative frequencies tor tne random and nonrandora
segments are equal, the test statistic has an approximate
chx-square distribution with 1 degree of freedom; this distri-
bution was used to compute the P values for each VOC of
interest that was examined in the GWSS.
The mean values of the positive samples in the random
and nonrandora portions of the GWSS wert compared using tne
"t" test, which tests the null hypothesis that the two means
are the same (i.e., w\ «= u2). The statistic lor testing th*.
equality of two population means uj and u2 using independent
samples from each are as follows:
t - (x" - x2) / s2 (1/nj + l/n2)
for n^ and n2 observations with a pooled cariance (s2) of:
s2 - l(n} - 1) s2 - (n2 - 1) s2] / (n} + n2 - z)
where XT and 3c2 are the observed sample means of the two
groups and s2 ana s2 are the corresponding variances. An
underlying assumption xor use of the "t" test xs that the
variables are normally and independently distributed in each
group. The normal distribution is not considered an accept-
able model for the values of the positive samples because of
the positive skewness and large coefficient of variation of
their distribution. Therefore, the tests were based on
natural logarithms of the concentrations to make the normality
assumption more reasonable.
D-28
-------
Table A-5 presents the results of the comparison of th<
frequency of the positives and of the means of the positives
for eacn VOC in the random and nonrandom portion of the GWSb.
The Pt value indicates the significance level for evaluating
the null hypothesis that the means ot the random and nonrandora
sample are equal. This value represents the probability that
the null hypothesis (i.e., the population means are equal) has
been rejected on the basis of the sample means when it is
actually true. For example, tne results for benzene shown in
Table A-5 indicate that there is a 69% probability of being in
error it one were to r«.ject the hypothesis that tne population
means for randomly and nonrandomly selected sites are the same
based on the sampling means of 4.1 m.crograms/liter tor the
nonrandom and 6.2 micrograms/liter for the random sample
observed. Similarly, the PX2 value represents the level or
signiticance tor evaluating the null hypothesis that the fre-
quency of positives are the same. Again, referring to
benzene, tne PX2 value indicates tuat there is a 15% prob-
ability of King in error by rejecting the hypotheses that the
frequency of occurrence of benzene in samples selected ran-
domly and nonrandomly in the GWSS are the same based on the
observed frequencies of 1.7% for the nonrandom and 0.7% for
the ranoom samples.
The critical P value selected tor this analysis was
0.01, implying the acceptance of no more than a 1% probability
of ueing in error by rejecting the null hypothesis ot equal
D-29
-------
TABLE A-5
RESULTS OF COMPARATIVE TESTS FOR RANDOM AND NONRANDOM GWSS DATA
Chemical
Benzene
Carbon
tetrachloride
Chlorobenzene
o-Dichloro-
benzene
m-Dichloro-
benzene
p-Dichloro-
benzene
1,2-Dichioro-
e thane
1,1-Dichioro-
ethylene
cis, trans-1,2-
Dichloro-
ethylene
Tetrachloro-
ethylene
1,1,1-Trichlo-
roe thane
Trichloro-
ethylene
Vinyl chloride
Number Mean
of (x)
Positive (micro-
NR/ Samples'3 Percent grams/
Ra (n) Positiveb liter)
NR
R
NR
R
NR
R
NR
R
NR
R
NR
R
NR
R
NK
R
NR
R
NR
R
NR
R
NR
R
NR
R
8
3
15
15
1
0
2
0
0
0
4
5
7
3
13
9
16
43
34
50
27
61
30
6
1
1.7
0.7
3.2
3.3
0.2
0
0.4
0
0
0
0.9
1.1
1.5
0.7
3.2
2.0
8.0
3.5
9.1
7.5
10.6
5.9
12.9
6.6
1.3
0.2
4.1
6.2
2.2
1.7
2.7
-
2.5
-
_
-
0.77
0.75
3.4
0.68
0.59
1.4
8.5
1.0
4.7
1.5
2.1
1.7
9.0
8.1
4.4
1.1
Standard
Deviation
(s) Range
(micro- (micro-
grams/ grams/ Prc
liter) liter)
?!?
4.2
4.0
-
0.35
-
_
-
0.0*
0.31
2.9
0.23
0.71
2.0
21
0.56
11
4.0
3.3
3.6
22
17
2.9
O.bU-12
0.61-15
0.20-15
0.20-16
-
2.2-2.7
-
0.70-0.90
0.52-1.3
1.1-9.8
0.53-0.95
0.22-3.0
0.22-6.3
0.21-12U
0.21-2.0
0.22-69
0.21-23
0.2U-21
0.20-ltt
0.20-130
0.24-78
1.4-8.4
0.69
0.54
0.71
0.01
0 28
O.Oo
0.04
0.17
0.90
0.22
a NR * nonrandoio; R = random.
b Basud on 4i6 random and 473 nonrandom samples Iron supplies serving 25
more people.
c Based on tht natural logarithms ot the positive concentrations (see tex
d Data may be unr liable uue to the smal, number ot" positive samples
(i.e., <5 for random and/or nonrandom data).
D-30
-------
means or equal frequencies. Using a critical value of 0.01,
it can be assumed that, for those P values which are less than
0.01, the null hypothesis should be rejected (i.e., for a nul
hypothesis u-| - U2, the alternative hypothesis, uj ^ U£, should
be chosen). On the other hand, for values that are greater
than 0.01, it cannot be directly assumed that the nulA hypo-
thesis is true. It is only known that the two valued are not
significantly difierent; it is not known if they are statis-
tically the same. However, assuming the null hypothesis is
testing tor equality betwe-n two values, the P values obtained
from the test can be used as a general guide to determine how
similar the two values are. A higher significance level (e.g.,
P - 0.70) would denote a greater similarity; a lo ;er level
^e.g., P - O.Oi) would denote a greater difference.
In evaluating the null hypothesis based on the P value,
the size of the sample on which the t^st of significance is
made is also important. For small samples, the null hypothesis
is likely to only be rejected if it is very wrong. Conse-
quently, a hypothesis of equality may be accepted when it is
wrong because the test for oeing signiticantly difierent at
the 0.01 level is more stringent. In contrast, with a large
sample small departures from the null hypotnes^s can be deter-
mined to be statistically significant, even though they are
quite unimportant in practice.
As can bt seen from the results shown ^n Table A-5,
D-31
-------
there is an apparent general trend ot higher frequencies 01
occurrence and higher means in the nonrandora samples. How-
ever,, there are a rew cases where the dirferences are statis-
tically significant at the 0.01 level. With respect to means
of positives, only 1,2-dichloroethane is found to have a
statistically significant difference (actually, borderline) in
the means (the nonranuora being higher, as woulu be expected).
For the comparison of frequency of occurrence, statistically
significant aifi.erences wer*. found ror only three 01 the VOC's:
cis/trans-l,2-dichloroethylene, 1,1,1-trichloroethane, and
trichloroethylene. Again, the frequencies were higher in the
nonrandom samples as would be expected. (Note that the small
number ot po i\ .ve samples for several ot the VOC's makes the
analysis of the frequencies of questionable validity.)
Based on these comparisons it was decided that in
preparing the national projections, the GWSS nonrandom data
would be. included except for the four VOC's tound to have
statistically significant differences in the mean of the posi-
tives or frequency of positives.
A.2.5 Establishing a Common Minimum Quantifiable
Concentration for the Combined Survey Data
As noted in the discussion on computing averages for
VOC's in supplies the minimum quantiriable concentration ror
an analytical technique defines the level below which it cannot
D-32
-------
be determined whether the VOC is present and, ir so, at what
concentration. If a common minimum quantifiable concentration
was used across all surveys, then the national projections
would provide estimates of the number of systems at various
ranges of contamination levels above the minimum qaantitiabxe
concentration. Below that concentration, only a total number
of systems would be estimated tor the range or 0 micrograms/
liter to the maximum possible concentration, i.e., just below
the minimum quantifiable concentration. The number of systems
having no contamination and the distribution of systems at
various Bevels below the minimum quantifiable concentration
could not be determined.
In many cases ditfert;nt minimum quantifiable concen-
trations were used in the different surveys and for different
analyses in the same survey. Concequentiy some supplies in
the combined data set were reported as having no measured VOC
(i.e., negative systems) with a maximum posaibte concentration
greater than quantified levels reported for other supplies
analyzed with more sensitive methods.
In some cases, it was reasonable to use the highest
"negative" value as the common minimum quantifiable concentra-
tion above which national estimates of the distribution of sys-
tems containing various VOC levels are made and below which
only a total number of systems is given. Any positive values
below this level were usi.-d as tvidenct that some supplies in
D-33
-------
this iatttr group are contaminated, i.e., all negative values
are not necessarily 0 micrograms/liter. These positive values
were.nof included with positive values above the established
common minimum quantifiable concentration used for selecting
the appropriate model, etc. for estimating national occur-
rence.
Tr> st-veral instances, it happened that the selection
of the highest negative value resulted in many, and sometimes
most, of tht measured positive values tailing below that level,
leaving too few positives above that level for completing the
national projections. Three major alternatives were considered
for handling high nep t ' j alues in these cases. The first
alternative was to establish the common minimum quantifiable
conueuL.j.&ciuu at some lower level that did not exceed substan-
tial numbers ot positive data points, and treat tae high nega-
tive values as though they were positives at their maximum
possible concentration. This would be generally consistent
with the rules for averaging samples described earlier. It was
observed, however that in a j.arge number of cases this woulu
result in a substantial number of the resulting total positive
values btin^ contributed from data where the VOC was not
actually observed; in some cases this contribution would exceed
the number ot actual positive values. This was determined to
be an improper use of the data.
The second alternative was to simply discard the ftigh
D-34
-------
negative data. This was also considered inappropriate since it
meant essentially eliminating valid data because it could not
be made to tit the projection methodology. Furthermore, its
elimination would artificially raise the computed frequency of
posit wet; by lowering the total number of systems sampled.
The third alternative, and the one selected for the
national projections, was to retain the data, but treat them as
though they were negative data below the lower common minmuin
quantifiable concentration selected j.or the combined survey
data. While it is true that VOC's could be present in those
supplies at a higher concentration, the other data sug&et>ted
that the probability is greater that they are not. This alter-
native also avoids the elimination of valid data from ttit
analysis. (It should be noted that, while this third alterna-
tive is used tor the reporting national projections projec-
tions were also calculated using the first alternative to give
the more conservative estimate; these results are noted in tht.
text.)
A.2.6 Model Selection
In developing methods for estimating the numbers ot
drinking water systems and people affected by different pollu-
tant levels, it was necessary to select a statistical modei tor
the distribution of positive concentrations for each data set.
The lotlowing, considerations guided th. selection of a moael:
D-35
-------
0 Tne same model should be applicable to data
from different system size groups for a
given pollutant (to facilitate comparisons
and evaluation of estimation error).
0 A continuous distributional modei should be
used where appropriate (to smooth out random
var ations in relative ^requency among con-
centration intervals).
0 The appropriateness or any continuous model
should be checked through goodness-of-fit
tests.
0 Estimates rrora continuous mouels should
provide upper bounds on the upper tail
of the. observed aistribution (to avoid
underestimating the number of systems
with high-level contamination).
Three types of continuous models were investigated:
0 Statistical distributions: .g , th
lognormal (Aitchison and Brown, 1957,,
ana gamma (Johnson and Kotz, 1970)
distributions
c Transrormations: the Jonnson (1949)
system of transformation to normality
0 Empirical models: fitting the cumula-
tive frequency with a polynomial function
ot contration
The adequacy of different distributional models was tested by
three goodness-of-fit tests: Kolmogorov-Smirnov, Cramer-von
Mises, and Anderson-Darling (Stephens 1974). A model was
considered unsatisfactory if it failed more than one of these
tests at the 0.05 significance xevel. Non^ of the three types
of models worked consistently for different pollutant size-
group data sets.
D-3b
-------
Of the mooeis investigated, th.. delta distribution based
on the lognormal distribution (Aitchison and Brown, 1957)
seemed the most appropriate a priori. Th.s model has been used
successfully in a wide variety of water contamination problems.
It allows tor the positive probability of zero (not detected)
results and the skewed distribution of positive results that
generally characterize the urinking water oata. However, in
some cases this model failed the goodness-of-fit tests for data
from at xeast ont size group. In other cases there were insu±-
ficient positive results to test whether the model was appro-
priate.
Based on the evaluations described above, it tinaliy
was deciced that no continuous model could be identified that
would be useful ror the drinking water data. Therefore, a dis-
crete model had to be employed. The model adopted was the
multinomial distribution (Johnson and Kotz, 1969), in which the
proportion of the distribution in a specified concentration in-
terval is estimated by the ous^rved relative frequency tor that
interval. The intervals used were the ones of interest in the
evaluation (MQC*. MQC-5 micrugr&ms/liter, 5-10 micrograms/
liter, 10 micrograms/liter intervals from 10 to 100 micrograms/
liter, and >100 micrograms/liter).
* MQC is the Minimum Quantifiable Concentration set for the
combined survey data (see Section A.2.5, above).
D-37
-------
Estimating System Size Groupings
Consideration was given to developing estimates from
data grouped by system size because it was thought that con-
tamination might be more likely in larger systems located in
more populous and probably more industrialized areas. The ~
system size groups shown below formed the starting point for
grouping data by system size:
Size Range
Group (number of people served)
1 (very small) l
2 (small) 501-3,300
3 (medium) 3,301-10,000
4 (large) 10,001-100,000
5 (very large) >10u,00u
Frequency tables showing the number of systems sampled and the
numbers* or positive and negative results tor each of tht.se five
size groups were produced for groundwater and surface water
systems for each pollutant. It waa considered desirable to
consolidate these groups as much as possible (consistent with
the data) because relatively tew systems were sampled in some
of the groups
The extent or further consolidation possible was eval-
uated by comparing the relative frequencies of positives in
ditferent groups turough a statistical t st procedure. Group
relative frequencies were compared in the following order, and
the groups were combined when no significant difierence in
D-38
-------
relative frequency of positives was tounu:
1. Groups 1 and 2
2. Groups 4 and 5
3. Group 3 with groups 2 and 2 combined
4. Groups 1, 2, and 3 combined with groups 4 and 5
combined
Step 3 was done oniy ii groups 1 and 2 could be combined as a
result of step 1; step 4 was done only if combinations in pre-
vious steps wery possible. The order of the comparisons was
chosen based on the possibility of a relationship (trend) be-
tween system size and the percentage of positive systems.
In performing the statistical test for equal p'. rcen-
tages of positives in two size groups, the first step was to
form a 2 x 2 summary table as illustrated below for groups 1
and 2:
s
ize Group
1
2
Total
Test
Negative
a
c
a + c
Results
Positive
b
d
b + d
Total
a + b
c + d
n
where
a * the number of negative results xor group 1.
b - the number of positive results for group 1.
a + b = the number of systems sampled in group 1.
n =a+b+c+d, the number of systems sampled
in both groups, etc.
The statistic was:
x2 = (|ad -be I - l/2n) 2n
(a + bJU :r~c) (b r d) (c +
D-39
-------
When this statistic exceeded the critical value 3.84 (the
percentile of the chi-squared distribution with 1 degree or
freedom), the hypothesis ot equal relative trequencies ot
tives in the two groups was rejected, and the groups were not
combined. The chi-squared test lor homogeneity ot percentages
in two populations is discussed by Snedecor and Cochran (1967).
In some cases, the expected number ot positive systems
was too 3u.all for the x2 test to be used (e.g., (a + b)
(b -t- d)/n <5 tor group JL under trie hypothesis ot equal propor-
tions positive). In such cases, the usual alternative proce-
dure, Fisher's Exact Test, was us.;d to test equality of propor-
tions positive. The application of Fisher's test is describee1
in Ostle (1*63).
0 Projections ot National Occurrence
After the final system size groups were selected for a
pollutant, the national projections were computed. First the
proportion of systems in each concentration interval of inter-
est w«« estimated tor each size group by the observed relative
frequency for the sampled systems. The proportions of systems
above the different specified concentrations also were esti-
mated for each size group (again by the observed relative fre-
quency tor sampled systems). For t.xample, it 100 systems weie
sampled in a given size group and three were found in the range
40 50 micrograms/liter, the estimated percentage ot systems in
D-40
-------
that range was 3%; it ten out of the 100 systems sampled were
above 50 micrograms/liter, the estimated percentage of systems
above 40 micrograms/liter was 10%.
The number of systems above each concentration limit tor
a given size group was estimated by multiplying the observed
relative frequency (p^) by the total number ot systems (N^) in
that group. Then the total number of systems or ail sizes above
a given concentration was estimated by the sum of estimates for
the k individual system size groups:
k
m - Ni pi-
i=i
It can be shown based on the multinomial me if that m is an un-
biased estimator of the total number of systems above a spec-
itied concentration (M), and that m has variance
k 2 <
Var(ro) - Nt N1 - nt P* (1-Pj)
i=l NI -1 ni
where n^ is the number of systems sampled out of the NI systems
in group i and P^ is the true percentage of systems above the
specified concentration in group i. Following Cochran (1963),
the term (Pi (l-Pi)/ni in Var(m) was replaced with its unbiased
estimator, Pj.(l - Pj.)Ani - 1), to estimate Var (m) . Then an
approximate 95% confidence interval on M was calculated from
ID ± 1.96 iVar (m)] 1/2.
The accuracy ot this interval improves with increasing sample
size (n^) and is bettern when P^ values are not close to zero.
D-41
-------
When confidence limits obtained from the approximation were
outside feasible M values (i.e., less than zero or greater than
N « N^) , the limits were reset to the nearest feasible value.
Projections for numbers of people exposed to concentra-
tions above specified limits were computed in the manner des-
cribed above, letting N£ represent the number of people served
by systems in the ith system size group.
-------
B.I SLCiION 6
B.I.I System Size Grouping and Selection ot Model for
Estimating National Occurrence of l.I-Dichloroethylene
in Drinking Water Supplies
As indicated in Section A (A 2.6), an initial step in
the estimation process was to group the data based on system
size. Based on results of statistical tests comparing percent-
ages positive in five initial size groups, groundwater data was
collapsed into two groups:
0 <10,000 persons served
0 > 10,000 persons served
(Summary statistics for initi '. .nd final groundwater groups
are given in Taole B-l. Results of the test are shown in
Table B-3.) For surface water, data was collapsed into two
groups:
0 £10,OOU persons served
0 >10,000 persons served
(Summary statistics and test results are given in Tables B-2
and B-3, respectively.) Only one surface water system was
sampled in groups 1-3 tner^fore, these groups weie not com-
bined with 4 and 5 because of insufficient data to test for
dij-ferencts in proportions positive.
For both groundwater and surface water data, the next
step was to fit a delta distribution to each system size
D-43
-------
grouping with suiticitnt positive uata. The oexta distribucion
has cumulative distribution function:
0 , x < 0
P(X 0
with
(logex - )/
F(x) - f(z)dz,
where r(?) ?'? rh standard normal probability density xunction.
The mean and standard deviation of loge-transfonned data were
used to estimate the parameters and ; was estimated by the
observed proportion of n-.gative values. The goooncss-oi-fit
of the delta distribition was tested, and the model was adopted
it no more than one 01 the t r< e tests Bailed for each size
group. The results in Table B-A show that the model failed
for grounawater, and there was insufficient data to test the
model for surface water. Therefore, the multinomial model was
used as tiie basis ior estimation.
D-44
-------
TABLE B-I
FREQUENCY OF POSITIVE SYSTEMS BY SIZE GROUPS
FOR GROUNDVlATER SYSiEMb
System
Initial
(1)
(2)
(3) 3
(4) 10
(5)
Size
Groups
1-500
501-3,300
,301-10,000
,001-100,000
>100,OUO
All
Number
Negative
2/8
199
150
298
38
913
Number
Positive
2
A
4
14
1
25
Number
Sampled
230
203
154
312
39
938
Percent
Positive
0.8/
1.97
2.6u
4.49
2.56
2.67
Final Groups
(1,2,3)
(4,5)
<10 QUO
>10,000
Alx
577
336
913
10
15.
25
587
351
938
1.7U
4.27
2.67
-------
TABLE B-2
FREQUENCY OF POSITIVE SYSTEMS BY SIZE GROUPS
FOR SURFACE WATER SYSTEMS
System
Initial
U)
(2)
(3) 3
(4) 10
(5)
Size
Groups
IOUU
501-3,300
, 301-10, OuO
,001-100,000
MOO, 000
All
Number
Negative
--
1
19
81
101
Number
Positive
--
0
0
2
2
Number
Sampled
0
0
i
19
83
103
Percent
Positive
--
u.oo
0.00
2.4i
1.94
Final Groups
(1,2,3)
(4,5)
<>0,OUO
>10,000
All
1
100
101
0
2
2
1
102
10 J
O.Ou
1.96
1.94
D-46
-------
TABLE B-3
CHI-SQUARED TESTS COMPARING
SYSTEM SIZE GROUP PROPORTIONS POSITIVE
System
Size
Groups
Compared
Groundwater
Surface Water
No. of Test No. of Test
Systems Stat. Decisiona Systems Stat. Decisions8
1 vs. 2
433
Pass
Not done
4 vs. 5
35
Pas*
102
1.2 vs. 3
587
Pass
Not done
1.2,3 s
4,5
93b
4.65
Fail
103
Not done
Critical value tor x2 test (a - 0.05) is 3.84.
Expected number of positives too small for X2 test.
Used Fisher's Exact Test (Ostle, 19i>3).
U.S. Environmental Protection Agency
Region V, L;'\T-Y
2?-...' -',,fi :>. .: - n "..Let
D-47
-------
a o cr to
n o n 2
n n n
f- »- H- II
ft ft ft
*" H- »- 2
o o n e
to to to B
h- h-1 O"
A
< < < 1
to to to
h- >- >- O
C C C Mi
A A A
Co
M-, Ml f *<
O O O CB
n n n rt
ft
0 0 O B
(o
O C O
cr V v Co
it
1- 1- .- B
a re A -o
A A A A
->-»- a
»
O 0 0
Ml M H
CO CO CO
H- (- t-
Oo 00 Oo
333
i- » K
Ml Mi l-h to
M- >- >- 9
o o o a
to to to
333
O O O
A A A
II N II 0>
1
O O C fi
00 t ' ^J A
vc rc a co
er> O" ^*J ft
f*.
B
0>
ft
A
CO
0
Mi
ft
3"
n
M 0
Ml C
09 5
n o.
A C
0>
C rt
0) A
rt it
A
V |/\ V |/V
t-> M ^ |_>
O O O O
« *
o o o o
CO CO
O O CO
H* IX Cn
O V) 00
N> t- -J
C t- O 0
VC O VC VC
O O W 00
CO -sj u>
1 1 1
C/J
i^
CB
rt
A
B
H
Q
A
n
^^N ^O
_ ^,4
O to
n*a rt
^
A
C A
^^ B
09
art W
V-'
z
- 00
1 .
t- 1 vc ^
*- »- NJ
h- ^ *-
1
«J 1 NJ O
^_
O
I*
C
^- ^V x^
(-<-( |_l
33 *i 3
to co o> OD
e c »- c
M^ Ml MM-,
^ P^ f*^
|^. ^* ^.
O O O
A A A
99 C 9
n rt rt
^j
O. O. en o,
09 0) N: o>
rt rt rt
to to 09
M Ml -Ml
O O "1C
n n 09 «
rt a t- n
C A A
to CO CO
rt rt n
VX V^ *~s
C
VC
CT*
er
H-
to
K
^~
(/)
rt
to
rt
D
A
O
^*«
CO
^.
O
CA
rt
09
rt
D
A
O
H-
CO
^*<
O
9
CA
rt
0:
rt
O
A
O
f-
CO
H'
o
3
M- H-
O N
9 A
10
09
«
g
A
rt
A
W
CO
rt
B*
0>
rt
A
CO
to
H
A
CO
rt
C >
0> 3
1 C
** A
H- M
JO C
CT3
1
H
A
(0
rt
0 0
9 *1
M- A
CB «n
A I
CO
o
H
A
to
rt
5*
CA O
B i-
i-B
n o
3 C^
O 0
< n
C-O
^
1
CH
i- P3
SH
f- tA
Or
f.
BE
i
2>
K
o
H
5C
------- |