_ 26453-3
An Industry Approach for the Regulation of
Toxic Pollutants
Appendices
Prepared for
Toxics Integration Project
U.S. Environmental Protection Agency
Prepared by
Putnam, Hayes & Bartlett, 'inc
50 Church Street
Cambridge, MA 02138
15 August 1981
-------
26453
TABLE OF CONTENTS APPENDICES
Appendix A
MATHEMATICAL PROGRAMMING FORMULATION A-l
Introduction A-l
Objective Function A-4
Material Balance Constraints A-7
Factor Input Constraints A-7
Capacity Constraints A-8
Operating Constraints A-8
Pollution Control Constraints . A-10
Health Effect Constraints A-12
Model Operation A-l4
Model Results A-14
Appendix B
EXPOSURE ASSESSMENT B-l
Air Pollution Modeling B-l
Analysis of Human Exposure Through
Drinking Water and Fish Consumption B-7
Ground Water Contamination From
Hazardous Waste B-l2
Appendix C
PROCEDURES FOR ESTIMATING HEALTH RISK C-l
Toxics Integration Program Scoring of
Selected Pollutants for Relative Risk
Prepared by Clement Associates, Inc C-3
Aooendix D
ALTERNATIVE WEIGHTING SYSTEMS
FOR EIGHT HEALTH EFFECTS D-l
Methods for Assigning Weights
to Health Effects in Integrated
Risk-Reduction Models
Prepared by Milton C. We ins te in D-2
-------
MATHEMATICAL PROGRAMMING FORMULATION APPENDIX A
INTRODUCTION
Mathematical programming (MP) is a technique used to
calculate the best use of available resources. Putnam,
Hayes & Bartlett, Inc. (PHB) developed a mathematical
programming model for the chlorinated organic solvents
case study. This model simulates plant operations, pollu-
tion, and the resulting health effects at alternative
levels of pollution control.
Consider an example where a manufacturer has two
products and wants to know how much of each product to make
and the best way to combine his resources to produce each.
In this example, resources might include labor, raw mater-
ials, and equipment. The manufacturer wants to make the
best use of these three resources in the production of the
two products.
An M? program begins with the statement of a goal,
which is expressed in the "objective function." In the
example, the manufacturer might want to maximize operating
cash"flows. The objective function would be established
to calculate the total margin, based on the unit cost of
each resource and its level of usage and the sales of each
product and revenues received. Because the manufacturer
wants to decide on the best sales mix for the two products
and usage of the three resources, the levels of sales and
usages are called "decision variables." This type of prob-
lem is referred to as a "cash flow maximization" problem.
Maximizing the value of the objective function occurs
subject to a series of restrictions, called "constraints."
n
-------
Constraints are established as equations which limit the
range of values for decision variables. One set of con-
straints might specify a maximum sales level (demand) for
each product. Since this example is a cash flow maximization
problem, a maximum sales or production level must be speci-
fied or the solution would be to sell an infinite amount of
profitable products. Another constraint (or set of con-
straints) might limit the usage of the resources. For ex-
ample, there may be only three employees available to pro-
duce the two products. Finally, constraints must be set
to show how much of each resource is required in the pro-
duction of each product so that the plant's internal mate-
rial balance is reflected in the model.
The PHB model for chlorinated organic plants uses a
version of MP called mixed integer programming. Mixed inte-
ger programming allows the model to account for fixed costs
appropriately and to account for any potential economies of
scale. Mixed integer programming allows some decision var-
iables to take on integer values. If a process is operat-
ing, the variable assumes a value of one and the fixed
costs of the process are incurred. When the process is
not operating the variable equals zero and the fixed costs
are avoided.
As in the above example, the PHB model seeks to maxi-
mize the operating cash flows* of the plant subject to a
series of constraints. Production volumes, sales volumes,
and purchases of various factor inputs are all calculated
so as to maximize operating cash flow. Decision variables
fall into one of four groups:
• Level of operation for production processes (such
as chlorination of ethylene),
• Level of sales for chemical products (such as
methyl chloride, vinyl chloride),
• Level of usage for factor inputs (such as labor
and energy), and
• Level of operation for environmental control op-
tions.
* Operating cash flows include revenues less manufacturing
costs and pollution control capital and operating costs.
-------
The selection of decision variable values is subject to six
types of constraints:
• Material balance (yields),
• Factor input (requirements),
• Capacity,
• Operating,
• Pollution control, and
• Health risk.
Each of these constraints is explained below.
Production flows among the specific production process-
es at a plant are defined through yield relationships,
which indicate how much output -of one process (an inter-
mediate product) is necessary as input to the next. These
are often called material balance constraints, because
they insure that eachintermediate product (material) is
produced in an amount consistent with the needs of down-
stream processes.
Factor input requirements are coupled with production
process decision variables to calculate how much of a
particular factor input (labor, power, water, and so forth)
is necessary at a particular process given the production
level of that process. These are also called value added
constraints because they quantify the value of the input
factors used at each process.
Capacity constraints limit the production level at
each process to the capacity of that process. Capacity
constraints can also be"used to force the plant to produce
at levels consistent with expected demand levels.
Operating constraints insure that plant operations are
consistent with a variety of special limitations.
Pollution source constraints are coupled with produc-
tion process and environmental control technology decision
variables to insure that the desired pollution control is
achieved.
A-2
-------
Health risk constraints limit the relative risk of
occurrence of one or more health effects (such as cancer)
or of a weighted group of health effects. This type of
constraint acts to limit pollution levels so as to achieve
a reduced level of health risk. Thus the health risk con-
straints force the model to select the most cost-effective
mix of production and pollution control decisions in order
to reduce the risk of health effects.
In the following pages the objective function of the
model and the various operating constraints, including pol-
lution control and health constraints, are described. Equa-
tions which are the basis for the mathematical programming
model of the selected chlorinated organics are included.
In addition to the equations, a list of the variables used
is included in Exhibit A-l.
OBJECTIVE FUNCTION
The objective function used is a cash flow objective
of the following form.
Maximize
the operating = 1 ?iSi - 1 PjBj - I P^Bfc - £ Km1^
cash flows i j J k m
Where:
or
Bj or
K
'Til
m
= Market price of product
i, factor input j or
material k
= Units of product i sold
38 Units of factor input j
or material k purchased
* Annualized capital cost of
pollution control option m
» Binary decision variable (0=off,
l=on) for pollution control
ootion m
Note that negative signs indicate reductions in cash flow,
while positive signs are used to indicate increases in cash
flow.
A-4
-------
Exhibit A-l
INDICES AND VARIABLES
USED IN THE MODEL EQUATIONS
i Products
j Factor Inputs
k Materials
1 Health Effects
m Pollution Control Option
p Pollutant
r Processes
s Product Mix
Ai,s Fraction of product i
in mix s output
3j Units bought of fac-
tor input j
Bfc Units bought of
material k
Cr Capacity of process r
Dm - Binary decision var-
iable for treatment
option ra
Eifp Emissions of pollu-
tant p from product i
Em,i,p Emissions reduction
of pollutant p from
control option m for
product i
Fj/i Quantity of input j
consumed or produced
per unit of produc-
tion of product i
H Total combined health
risk
Hi
J-i
K
m
N,
Qp,t
Risk of health effect 1 P
(relates emission level
to health risk) of pol- QD/h
lutant p
Total health risk of
effect 1
Quantity of input j
consumed or produced
in a fixed quantity in
process r
Annualized capital cost
of treatment option m
Binary decision variable
for process r
Binary decision variable
which indicates the section
of the utilization curve
a process is operating on
Market price of product i
Market price of input j
Quantity of pollutant p
emitted which has a health
effect
Total quantity of pollut-
ant emitted
The threshhold emission
level
?,t
' then
» Q - Or
P -
A-5
-------
Exhibit A-l
(Continued)
INDICES AND VARIABLES
USED IN THE MODEL EQUATIONS
Rkfi Units of material k
consumed or produced in
the production of i
Sj_ Units sold of product i
Sfc Units sold of material
k
Vs Total quantity of
output at product
mix s
Weighting factor for
health effect 1.
Production volume of i
Production volume of i
produced by process r
Operating level of
treatment option m for
product i
A-6
-------
MATERIAL BALANCE CONSTRAINTS
Material balance constraints match process inputs and
outputs of chemicals. They are generally of the form:
Bk > 0
i
Where: Xj. = Production volume of product i
Rk i ~ On its of material k consumed or
produced per unit of i produced
Sfc = Units of k sold
Bfc = Units of k bought
FACTOR INPUT CONSTRAINTS
These constraints assure that the inputs consumed do
not exceed the sum of those produced or purchased. They
are of the following form:
I Fj^Xi + Jj/rNr + Bj >. 0
Where: ?j,i s Quantity of input j consumed or
produced per unit of i produced
Xj_ » Production of i
Jj r = Quantity of j consumed or
produced in a fixed quantity
when process r is being operated*
Nr = Binary variable which equals
1 when process r is being
operated
Bj - Purchased quantity of input j
j r is generally denominated in dollars for fixed costs.
A-7
-------
CAPACITY CONSTRAINTS
Capacity constraints prevent the model from operating
a process above its rated capacity. Their typical fora is
illustrated below:
I Xifr - CrNr £ 0
1
Where: xi,r s Production of product i from
process r
Cr - Capacity of process r*
Nr = Binary variable which equals 1
when process r is being operated
and 0 when it is not
OPERATING CONSTRAINTS
Operating constraints are used to constrain the model
to simulate actual plant operating behavior. A representa-
tive example of the situations in which these constraints
are used is the equilibrium production of Methyl Chloride,
Methylene Chloride, Chloroform and Carbon Tetrachloride by
the chlorination of methane process. As illustrated in
Exhibit A-2, the output mix is dependent upon the molar ratio
of chlorine to methane in the feed to the reactor. Thus
the ratio of chlorine to methane in the feed to the reactor
determines the fraction by weight of each of the four
products in the output. Note the choice of a product mix
is independent of the choice of total throughput of all
products by weight. Thus the quantity of any product
produced is determined by two choices. These choices result
in a mix percent which when multiplied by the total volume
of material processed equals the quantity of each product
manufactured. This multiplication of two decision variables
is nonlinear and thus not amenable to continuous mathe-
matical programming. It is possible, however, to select a
group of discrete product mixes from which the model must
choose. The model can then select the operating level for
* In the PH3 model, capacities are selected to reflect
anticipated demand levels at the actual plants of inter-
est.
A-8
-------
Exhibit A-2
CHLOROMETHANES BY THERMAL CHLORINATIOH OF METHANE
PRODUCT DISTRIBUTION — HASS AND McBEE
U
z
O
TOO
80
z
O 60
u
LU
O
z
UJ
u
as
•-u
a.
a:
<
O
40
20
0
--*- x J
1 2 34
MCLAX RATIO CF CHLORINE TO METHANE
Source: 4231.
Source: SRI Chemical Economics Handbook.
A-9
-------
the process. The model is thus constructed to treat the
process as a number of separate processes, only one of which
can be operated at any one time.
In Exhibit A-3, seven vertical slices of the product
mix function have been identified. Each of these represents
a fixed product mix alternative (the total number of alter-
natives can be varied). One constraint is required to model
the choice among fixed product mix alternatives and the
choice of a production level. The form of this equation
appears below.
Where: X^ = Quantity of product i manufactured
AifS = Fraction of i in the process output for
product mix s
Vs = Total quantity of product manufactured
using product mix s
(Note: These variables are identified
as Special Ordered Set variables which
require .that only- one variable be non-
zero at any one time. Thus only one
product mix can be active.)
One equation of this form is required for each product the
process can produce.
POLLUTION CONTROL CONSTRAINTS
Pollution treatment options are generally treated as
processes with capacity constraints and coefficients in
the factor input constraints to capture their operating
and fixed costs. As noted previously, an annualized capital
cost for each treatment option appears in the objective
function.
Only one type of constraint is specific to pollution
control. This constraint adds total uncontrolled emissions,
deducts those emissions controlled and sums up total result-
ing emissions in Q. It is of the following form:
£ Bi,p xi - II Em,i/p Xm/i - Qp,t £ 0
m
A-10
-------
Exhibit A-3
CELOROMETHANES BY THERMAL CHLORINATION OF M
PRODUCT DISTRIBUTION — SASS AND McBEE
U
~mt
Q
o
z
o
LU
( ^
V>
Z
LU
U
—
e_
<
O
Source: 423154
Source: SRI Chemical Economics Handbook,
I 2 3 4
MOLAR RATIO OF CHLORJNE TO METHANE
A-li
-------
Where: Ei,o = Emissions of pollutant p per unit of
product i produced
Xi « Quantity of i produced
Erari,p = Emissions reduction of pollutant p
per unit of operation of treatment
option m for product i
*m, i ~ Operating level of treatment option
m for product i
Qp,t = Total quantity of pollutant p emit-
ted. An upper bound on this vari-
able acts as an emissions limit
Note that the operating level of a treatment option and the
production level of a product are treated similarly. This
reflects the assumption that a treatment option can be
operated on a unit of product basis. Thus a treatment
facility can be operated to offset the emissions generated
in the production of each unit of product or at some lesser
level specified in terms of a lower production level.
HEALTH EFFECT CONSTRAINTS
The total risk of each health effect arising from the
pollutants emitted at the plant is calculated using the
following constraint:
HI - I H^pQp
Where: H]_ = Total risk of health effect 1
H]_7p * Health risk of effect 1 from a unit
of pollutant p (this is a constant
which relates emissions level to
health risk)*
Q0 = Quantity of pollutant p emitted
* This constant is developed from the exposure analysis
and health effect analysis.
A-12
-------
By placing an upper bound on HI, the risk from health ef-
fect 1 is limited. Because certain health effects do not
appear until total plant emissions exceed a threshold,* a
special constraint is needed to calculate QD for these
effects. Threshold effects occur only for that quantity
of pollutants that exceeds the threshold. The constraint
which calculates this quantity appears below:
QP,h + Qp,t - Qp s Q'
Where: Qp,t = Total emissions of pollutant p
Qp - Quantity of emissions by which
Qp,t exceeds Q1
Qpfh = Quantity of emissions by which
Qp/t is less than Q1
Q' = Threshold emission quantity
p
Note: 9^/h an<^ ®*> are Designated special ordered set var-
iables. TKis permits only one variable to have a
positive value at any one time; the other is required
to equal zero. Under no circumstances can either
have a negative value.
The final health risk constraint combines the total
health risks of each effect into one health risk number.
The weighting factors which are used in the constraint to
combine health effects are described in Appendix D. The
constraint is formulated as follows:
H - 2 WiHi = 0
Where: H]_ * As defined above.
H * Total combined health risk from
plant.
* Weighting factor for health effect 1.
* Thresholds are determined by comparing plant emissions to
background levels in the plant's environment.
A-13
-------
3y placing an upper bound on K, the combined health risk
from the plant can be reduced.
MODEL OPERATION
The cost effectiveness curves are produced by running
the model with upper bounds on H which reflect a 30 percent,
50 percent, 80 percent, 90 percent, 95 percent and"99 per-
cent reduction from its original uncontrolled value. The
decline in the objective function from its uncontrolled
value is the cost plotted in the curve.
MODEL RESULTS
In addition to the detailed computer output which is
produced by the MP solution program, summary reports are
available which contain the major results of the model.
Two of these reports for the Mississippi River sum plant
appear in Exhibits A-4 and A-5. Both reports are for
1985. Exhibit A-4 contains the results for the base or
uncontrolled case. Exhibit A-5 is a report of the results
for the case in which the total weighted health effect is
reduced to 50 percent of its value in the base case.
In the base case (Exhibit A-4) plant revenues are
approximately $1.2 billion, while gross margin (the objec-
tive function value without annualized capital charges)
equals $555.3 million. Capital charges of $6 million
reflect the annualized capital cost of pollution control
devices installed. This charge reflects the capital cost
of.profitable control options which were installed for
their economic benefit. As can be seen in the profitable
pollution control cost summary, the operating costs of the
control options is exceeded by the recovery credits (the
value of chemicals retrieved by pollution control for sale
or internal use) to result in a $13.5 million profit from
the use of these controls. This profit also exceeds the
combined operating and annualized capital costs of the
devices.
Relative incidence measures of each health effect are
reflected in the health scores for each effect. The weight-
ing factors used to calculate the weighted average health
score are the quality of life weights (discussed in Appen-
dix C) normalized to sum to one. Note that cancer has the
A-14
-------
Exhibit A-4
MISSISSIPPI SUM PLANT - BASE CASE
1IHAI
HAM) IN!) AMU CO!) Hi
ass.av
A.00
F pouurinH CONIKOI. coins
i on IH inn con mm. orir,ATino
RFCOVKRY tlflf-rilTB
Hi: r foiioruiN COWIKOI IWRATINII wurra
PAI'ITftl.
4. A 7 (III)*)
-1H. 13
-13.47
A, 00
lit A» in tu:imr:s
h--
Ln
I ("RAH Kit
: 1 1 VK i i»x 1 1 ; / r v
IIIXICITY
M::MAI TOXICIIY
(inifcii in i era
Hf IIRIII OOICAI TOXir.IlY
IIFAI rn
SCORE!
I 4 • A ,'?
o.o a
0.02
','. . 4 1
t ,")?.
A. 41
7 . 90
0 . 00
MEIIillilHO
FACTOR
O • O'?
0.20
0 . 20
0.10
0 . 0?
0 . 04
0 . 1
F-FRCENT RFIilli;
FROM HAHF I'A
O.OO
0.00
0.00
0 . 00
0 . 00
0 . 00
0 . 00
O . 00
THIN
fii"
;i . on
O i 00
F'lM I III TON Ml I AN I I Tim
TO I At AIR
AW fttlRHIIUin ~ CIIIMICAI H OF r.ONf.fKN
I'lIM FFFI..IIi:.NT
UAIFR IFF! 111 wr - ciiriHinm.a OF CUHCFRN
I-I.-OCFSS WATIR 11 on
-------
Exhibit A-4
(Continued)
MISSISSIPPI SUM PLANT - BAStf CASK
I inn 11 vi in HY
rr.-or.tim ii nri'KiioiiiiimiAnoN of tit m.iiHii
III CHI OIUNAl HIN (IF IIYHSIICAIUKINH
V IKIitCI Hill OKI NAT I ON III l:MIYI FNF.
Vf OXVCIH IIRINA1IWM III I Ilirl Mil
VII I-YKOI YfilH OP I Hi;
i i;oi:i tin vin rim M< i NATION or VIMYI minuim
rimer mi x I>HIYHKOI:III OHINAIHUI or ii-icin IIKKI rnriiir
l-i-MO
211 ,A(>
37A7..IO
4O2.3O
2li27,0'.J
vv . vo
A9.IIA
CAI Ad 1 Y
4A7. l()
241 .AO
.'1'A7..'IO
402.. 10
:v«r:u . 30
vv.vo
it. . oo
III II I/AI1OII
100.00
1 oo . oo
1 0<> . OO
1 00 . 00
»l» . 711
1 00 . 00
V».VA
i Fvn n
cr>
VIHVI CIIIOKIH
"itm IKI in DIII
CAK-HON H1KAi;lll Oli I HI
MI inn nil en t i.i
HFHiri FNI <:IH
i OKIII inn I:NI
1:111
(mi i Hi;)
?476.0tl
A9.IIA
H.U.13
4:*.04
O.A?
3A1..T4
I'-I.IA
i?«:!.o:i
ll'VH.OA
MAI I 8
(HI. %)
43.40
I . 22
4.0A
0.74
0.01
A. :v
0.1.'
21 ,O/
(MM *>
IIVI ..»•»
fi'f . 311
:«).« y
, i/,
(II All 11 X)
2. SO
0.77
0.01
A.V3
7.va
2.AV
n
'ill
: tit*
S'l
tin
J !ifi
1 lit)
II
V
VII
1 1
1 1 1
V
III
"1 1
VI 1
rimiMii
Alii
A IK
AIT.-
•JAM 1 F
HAi 1 1 F
HAS 1 F
UAtllF
MASH
1 MAS II
tltllil T.IlM 1 ANIH:
lll'fif CIK-I 1 Alll'l
iiNiii i:ui
III.
Ill
II I
II 1
III
III
III- 1 MWin HI 1 Mil MI
IIAH AlifillKI'l li
IIII.HUIiAMII.
I IIC 1 Ml liA 1 OR
IIICIMI KAI 1011
1 IIC 1 Nl l-Yil I'll!
Illl Illl I,AI I0||
II" (III I
-------
Bxhlbit A-5
MISSISSIPPI SUM PLANT - 50 PERCENT RISK REDUCTION
DI-IIKATING
nun rosin
IIITAI Rl WNIM B
HITRAUNd (IRIIfif] HAI'-OIM
rAfiiAi ctMMiF.fi
i:.'OA.;s4
{ir.5t.1V!
.30
0.51
r cotiiiriON n
rni i in ION riiNiiau n
K-rrowtMr CKrniifi
MII (•iiiMirioN con i mil oiei
J».13
m i*i
(),'?()
0.10
0.0V
O.OA
O.K!
0 . I fi
MI iniiiru AVI «nr.F in AI in
FROM HfttiC CAKE
47.72
0 . 00
0.00
111. 69
A4.A3
117, 13
•.•(i.av
0 . <><>
5iO.no
fill 1.11 Til *N HIIANTI I flrfi
HHAI AIR
Air: rHifisiniis - cnifHinAi H nr CONCERN
IKin II I I III Nl
UAH: If rmilFNT - Cllt-Hir.AI II III" (IIHCIRU
i i;i»:Kr.n HAIIT< now
si it in ur.sit
Sill III MA.'ill CIIMIJCAI !i III Cniir.tl'H
(II t. MB)
-------
Exhibit A-5
(Continued)
MISSISSIPPI SUM PLANT - 50 PERCENT RISK REDUCTION
ri mil 11 »»ii I. IIY fi-on tin
HIM 1 tifl>
I-MII:FMI n nri'K'icni MimiAi ION or ;n HIAIIMI 4A7,io
IKIICIIi'l III (ill) III* MIA II ON III • MYlihOCAFclUUIS 241. AO
ri;!. ,W
F'l;(li:i Hit V|| PVRIII YB18 IIF (HI! I!!>27 . '.'3
ri (Ul . X) (Mil «)
VINYI rill IIIVI III 217A.2H 43.'.!3 1191.44
VINYI IliFIII- i:ill IIRIHF AV.IIA 1 .'.'.:• !iy.3fl
CAKIUIII n MiAciii imiiii :>:.4|.UA 4.04 30.10
CHI IlliOl 111(11 4.1. OA 0.73 9.;>!i
(IF IIIYI. Clll III--IIM-: 0..4V O.OI O.I 2
lirillYI INK CHI ill* IHF 3A4.34 A.3A 113.110
IKM-ACIII 111(111 IIIYI FNF 24.20 0.42 3.I!S
HiiniiimiiF IIIYI INF o.oo o.oo o.oo
HYiiKiiniH i:ninr2.9n :i:'.vi
(;IIIIII;MI mil MIIII mircMM
I-«IM;I no in A u< ruiii'i HMFK IYF-F ill
I-I.-IICI Mil V fttf nilllil MM It TYI'F III
ri.-in:rim vi AH* INDINI KATIIR
HJOCFHM X AIH INCINhKAIllR
FKiinfiti vii WASIF fii:i:ni»F IANKFIII
1 liOCFHIl II UAU1F IIMIil TUI-'F 1 ANIif II 1
nI n i
i iviicma v UACiiF: iiiiiin:iir*i: IAHHFIII
Cl*lll:l !ill VI HAH IF MHMI Clll(lr IfiNlillll
lldllllliil VII HAHIF IIIIRI CIII;F 1 AMIil M 1
ii*iit:riifi viii (.IAVIII: iitiiiiciiicF IAMMIII
lilCIM.'F 1 ANI'F II 1 (MM 1 li!i>
llll!)Fi;|l|;K IAHH III (MM 1 I'M)
IlilMIIAIMI i:OIIM*IM HI'IMIN!) lilllCMII
l-KIICFUfl II All; HAS Allli'11,1'1 li
I'MU'.t 'till V AM* IIICIIII KftMIH
rlillCKlli VII All* UK) ll'l KAIUK
IMICI MM II I.PAIill IMCIHI I-AI IMH
fldlCFIlH III UAIilF IIICINI I*AI IHN
1 l*lirl li'l V MAS IF Illi:illl I*AI ION
I-I
0.13
0,22
0.17
0.23
0.13
0.14
0,011
O.O2
n;on KB IKI AMU
OV. 72
100.00
100.00
1 00 . 00
0 * OO
0.00
0 , 00
0 . 00
0.00
0 . 00
0 . 00
i , i :••
1 .34
r-i*oiFS!i IK-FAITH
100.0(1
loo.oo
1 0(1 . OO
1 00 . 00
100.00
100.00
100.00
1 OO . 00
100.00
-------
highest score (14.62) followed by other effects (principally
respiratory effects) at 7.90 or renal toxicity at 6.41.
Other health effects have much lower scores.
Indicators of overall pollution to the environment from
this plant are listed under pollution quantities. Air,
water or solid waste emissions which lead to health effects
are listed as "chemicals of concern." Process'water flow
and BOD in the water effluent are listed as supplementary
indications of the plant's environmental burden.
Production levels for each process, the process' capa-
city and the resulting capacity utilization as determined
by the program are listed next. Utilization of less than
100 percent reflects shortages of chemical intermediates
or, in pollution constrained cases, reduced production to
achieve reduced risk levels. All eleven process do not
appear either because they do not appear in the region
(such as processes 1 and 9) or because they are not econom-
ical to operate. Product sales are presented in pounds,
by weight * percent, in dollars, and by percent of total
dollar sales. Product prices in 1985 are also listed.
The control options utilized are described next. The
first type are those with a net cost to the firm either as
a result" of pollution restrictions or of normal plant
operations (such as sanitary landfill for all wastes not
incinerated or placed in secure landfills). In the base
case, only the "2.5 million pounds of waste that is_ the
residual from waste incineration is placed in sanitary
landfills at a net cost to the plant. The percent of
process waste treated of 0.00 percent actually reflects a
percentage of 0.001 percent (the incineration residual)
rounded to zero. The second type of control option is
profitable for the plant and includes gas absorbers for
process vents on process II, air incinerators for process
vents on processes V and VII, and waste incineration (at
99.999 percent rounded to 100.00 percent) for processes
II, III, V, VI, VII and VIII.
In Exhibit A-5, operating gross margin net of capital
charces is 5543.8 million dollars, 50.49 million below the
same'figure for the base case. This reflects the net cost
of adding nonprofitable pollution control devices to achieve
a 50 oercent reduction in health risk. The nonprofitable
option costs include 50.79 million operating cost and a
50.31 million capital charges which are offset by 50.63
million recovery credit.
A-19
-------
The additional control options selected to achieve the
50 percent risk reduction included a type III (95 percent
emission reduction) condenser on the storage vents of
processes III and V, air incinerators on the process vents
of processes VI and X, and secure landfilling of 1.12
million pounds of the hazardous waste from process VII.
All other control options were the same as those selected
in the base case.
A-20
-------
EXPOSURE ASSESSMENT APPENDIX B
This appendix describes the mathematical models used
to estimate population exposure to concentrations of pollut-
ants in the air, surface water and ground water. Whenever
possible/ EPA exposure models and data have been used.
The plan for this appendix is to describe the approach
used in air pollution modeling, water pollution modeling
and modeling of ground water contamination from migration
of hazardous wastes.
AIR POLLUTION MODELING
Two EPA air quality models were used -in th;.s analysis.
The Human Exposure Model (HEM) was used for the chlorinated
solvents plants and the Industrial Source Complex Model
(ISCM) was used for copper smelters. Both of these models
were used to calculate annual average concentrations. The
HEM utilizes a Census Bureau population data base to esti-
mate population exposure. The ISCM was linked to the HEM
so population exposure in the vicinity of copper smelters
could be estimated.
The HEM was developed specifically to model synthetic
organic chemical plants and is designed to determine the
population exposed to various levels of organic chemicals."
* A description of the diffusion equations and auxiliary prc-
arams is'given in Human Exposure to Atmospheric Concentra-
tions of Selected Chemicals, volume I,SPA OAQPS,1980.
-------
The HEM produces two types of output. The first output is
a matrix of concentration levels at points that are the
intersection of the wind directions and concentric circles
at varying distances from the point source. (See Exhibit B-
1. ) The model can calculate concentrations for distances
as great as 80 kilometers from the pollution source.
The second output uses the population database to cal-
culate population exposure. 'Using the longitude and latitude
of the point source and the concentration levels, the
program calculates the population exposed to a pollutant.
The output includes a map of the number of people exposed
to the pollutant within each "square" in the matrix (that
is, in each "square" surrounded by 4 points in the concen-
tration matrix). There is also a table giving a dozen or
more concentration levels, the number of people exposed to
these concentrations, and the dosage received. Exhibit B-2
is an example of this type of output.
For population groups out to 2 kilometers from the
source, the*EEM determines pollutant concentration by taking
the population center at the nearest point in the concen-
tration matrix. Outside of 2 kilometers, the program
interpolates the concentration at the population center
from the surrounding matrix points. The HEM uses 1978
population levels which have been converted to 1985 esti-
mates using Census Bureau growth rates.
The HEM uses a number of standard values for input
parameters to save computing time. Exhibit 3-3 lists the
standard values used i'n the HEM. Exhibit B-4 lists the
input requirements for the HEM. Most of this information
is" available from engineering reports prepared for EPA.
The HEM assumes flat terrain conditions and as indicated
in Exhibit 3-3 can take account of photochemical decay. How-
ever, for the pollutants of interest in this study most of
the decay rates were either unknown or zero.
The HEM has access to meteorological data from 311
U.S. weather (STAR) stations. These data include information
on wind speed, wind direction and meteorological stability
classes. These data are, on average, the mean of about 10
years of weather data collected at STAR stations and are
kept on record at the National Climatic Center. Unless
otherwise specified, the HEM will use data from the STAR
site closest to the specified point source.
B-2
-------
Exhibit B-l
HUMAN EXPOSURE MODE]
population
exposed to
specific
concentration
distance f
plant
rom
wind direction
Ambient concentrations are measured at each intersection of
wind direction vector and concentric circles.
3-3
-------
Exhibit 3-2
Concentration
(aq/m2)
1.00 x 10"4
5.00 x 10~5
2.50 x 10-5
1.00 x 10"5
5.00 x IQ-o
2.50 x 10~6
i.OO x 10"6
5.00 x 10-7
2.50 x 10~7
1.00 x ID'7
5.00 x 10-8
2.50 x 10~8
1.42 x 10-8
EXPOSURE PER POUND OF EMISSION
AT MISSISSIPPI RIVER PLANT
Cumulative
Population Exposed
129
333
558
5656
10,181
11,965
20,607
54,088
212,861
947,264
1,521,227
1,614,931
1,615,386
Cumulative
Dosage
2.16 x ID"2
3.40 x 10-2
4.20 x ID"2
1.23 x 10-1
1.56 x 10-1
1.63 x 10-1
1.74 x 10-1
1.96 x 10-1
2.47 x 10-1
3.52 x 10-1
3.98 x 10-1
4.01 x 10-1
4.01 x 10-1
Source: PHB Analysis.
B-4
-------
Exhibit B-3
HUMAN EXPOSURE MODEL STANDARD VALUES
Wind speed categories: (m/sec)
1.50, 2.46, 4.47, 6.93, 9.61, 12.52
Stability classes:
Daytime: A, B, C, D]_
Nighttime: D2/ E, F
Decay classes:
Time of Dav
Class Description Daytime Nighttime
I Very Reactive 1.0 x 10~2 5.0 x 10~5
II . Reactive 5.0 x 10~3 5 ..0 x iO"5
III Moderately Reactive 5.0 x 10~4 0
IV Unreactive 0 0
Downwind distances:
20 km output: 0.2, 0.3, 0.5, 0.7, 1.0, 2.0, 5.0, 10.0,
15.0, 20.0 km.
80 km output: 0.5, 1.0, 2.0, 5.0, 10.0, 15.0, 20.0, 40.0,
60.0, 80.0 km.
Wind directions:
N, NNE, NE, ENE, E, ESE, SE, SSE, S, S3W, SW, WSW, W,
WNW, NW, NSW
Effective stack heights: (meters)
0.0, 5.0, 10.5, 20.0, 35.0
Source: ?H3 Analysis.
3-5
-------
Exhibit 3-4
INPUTS REQUIRED TO RUN THE
HUMAN EXPOSURE MODEL
Pollutant Name
Facility Identification
• Latitude and Longtitude (degrees, minutes, and
seconds)
• STAR Site
• Ambient Temperature
a Urban or Suburban
• Number of Sources of the Pollutant at the Facility
Source Information
• Emission Type (Process Vent, Storage Vent, Fugitiv?
Emissions, Stack)
• Emission Rate (kg/year)
• Stack Height in Meters
• Building Cross-Sectional Area (height x width)
• Vertical Stack or Nonvertical Stack/Vent?
• Stack Diameter in Meters
• Exit Velocity of Emissions
• Exit Temperature in Degrees Kelvin
Decay Rates - Daytime/Nighttime
" (Reaction Rate Constant)
Source: PKB Analysis.
3-6
-------
The ISCM is a more sophisticated dispersion model used
widely by EPA. A complete description of this model is
provided in EPA's Industrial Source Complex Dispersion
Model User's Guide, published in 1979. The ISCM model was
used for copper smelters because smelters have very tall
stacks and the HEM was not designed to model sources with
tall stacks. The input requirements for the ISCM are
similar to those required for the HEM. The ISCM can take
into account gravitational settling; however, no information
was available on deposition rates for the substances exam-
ined in the smelter study. The ISCM was linked to the
population database of the HEM so the output is produced in
the same format as the HEM's output.
Information on background levels is needed to determine
actual levels of air pollution. For the copper smelrer
study, EPA regional staff provided information on background
levels of sulfur dioxide and total suspended particulates.
Background data on heavy metals were provided by Accurex,
Inc. in a report prepared for this project.* The EPA Office
of Air Quality Planning and Standards provided information
on background levels of substances emitted by chlorinated
solvent plants.
ANALYSIS OF HUMAN EXPOSURE THROUGH
DRINKING WATER AND FISH CONSUMPTION
Water pollutants discharged to the surface water may
cause the contamination of drinking water supplies and
supplies of fish. Also, in the case of volatile organics,
the chemicals discharged into the water will volatilize
into the air resulting in additional air emissions. This
section will present the mathematical formulas used to
determine, for a single source, incremental human exposure
to toxic chemicals through drinking water and consumption
of fish. Exposure from water pollutants that volatilize
into the air were calculated as described in the section on
air exposure.
Estimated Arsenic, Lead, Cadmium, and Mercury Levels Con-
tributed to the Environment 3y Four Nonferrous Smelters:
Accurex Corporation, June 25, 1981.
3-7
-------
Exposure from drinking water has been determined by
identifying all drinking water intakes downstream of each
plant together with the number of people served by these
intakes. The incremental chemical concentrations at these
intakes arising from a plant's operations are then derived.
Finally, it was assumed that each person would consume 2
liters of water from these supplies each day.
EPA computer models and computerised databases were
not as readily available for estimating water pollution
concentrations as for estimating air pollution concentra-
tions. Standard water quality models were used in this
analysis and extensive effort was undertaken to gather the
needed input data. It is important to note the assumptions
underlying the use of these models, specifically:
• Steady-state conditions are assumed, meaning that
the polluting material flows into the waterway at
a constant rate in time.
• Characteristics of the waterway, such as velocity
and river flow- rate, are assumed to be constant
with respect to time and distance from the pollut-
ing source.
• It is assumed that pollutant removal processes
occur in accordance with a first-order reaction.
The water quality model equations used by PHB are dis-
played in Exhibit B-5. The equations developed estimate
exposure from both water and fish consumption. The equation
used for estimating surface water concentrations is adapted
from a recent SPA report.*
By applying appropriate unit risk measures to these
concentrations, one can estimate the increased incidence of a
certain health effect due to pollution from a specified
source. For example, to estimate carcinogenic risk the
Water Quality Criteria Documents give unit risk factors in
Falco et al., "A Screening Procedure for Assessing the
Transport and Degradation of Solid Waste Constituents
in Subsurface and Surface Waters," to be published in
Proceedings of the Society of Environmental Toxicology
and Chemistrv.
B-8
-------
Exhibit B-5
WATER QUALITY MODELING EQUATIONS
V(l-HcS)
(1)
p. . C(xi)j
= 1
jw
(2)
f
' i ' ex? "
^
i-t-kS)
= P
• j • v(-k
Q - K
v(H-kS) 1 - sx?
V(l-fkS) j
for KJ r 0, and
Where:
C(X)
iw
concentration of pollutant j in the surface
water as a function of downstream distance
from the discharge point
E-J w = human exposure to pollutant j from drinking
water
E-; f » human exposure through consumption of fish
• contaminated by pollutant j
i » a drinking water intake downstream of the
source
j • a pollutant for which exposure is being
calculated
Source: PHB Analysis.
B-9
-------
Exhibit 3-5
(continued)
pi = population served by drinking water intake i
Wj = daily mass rate of discharge of pollutant j
from the source
Q = median daily flow rate of the river
V = median velocity of the river
Xj_ = downstream distance from the source to drinking
water intake i
D = distance from the source to the ocean
Pp = population exposed through fish consumption
per unit of distance downstream of the" source
Kj = a first-order reaction constant for principal
removal process for pollutant j. (The predomin-
and removal process is volatilization for most
of the pollutants under study.)
k = sediment/pollutant partition coefficient
S = suspended sediment concentration
Source: ?K3 Analysis.
3-10
-------
units of risk per microgram/Liter (ug/L). Thus, the total
risk would be equal to the risk per ug/L times the number
of people exposed to various concentrations or:
Total Risk = (Ejw "*" EjF^ ' un^t risk factor
It was found that the risk of cancer from eating contaminated
fish was negligible in comparison to the risk associated with
drinking water consumption. Furthermore, subsets of the
population which consume much more than the average intake
of fish were also found to have very low levels of cancer
risk.
For other effects Clement Associates provides risk
factors in units of risk per mg/Xg of body weight. To
convert to these units it has been assumed that an average
person weighs 70 kilograms and consumes 2 liters of water
per day.
Data regarding drinking water intakes and populations
served were obtained from EPA's STORET database. The mass
rate of discharge Wj at typical plants is determined from
engineering reports. These rates can be reduced by appli-
cation of pollution controls. Flow rates for rivers were
obtained from the U.S. Geological Survey {U.S.G.S.) and
river velocities were obtained from the U.S. Army Corps
of Engineers. Distances along the river were estimated
from U.S.G.S. maps and from STORET data. The number of
persons exposed through fish consumption was estimated by
dividing the amount of fish caught for human consumption
downstream of the source by average fish intake per person
(assumed to be 6.5 grams daily in accordance with the
Water Quality Criteria Documents). The amount of fish
caught for human consumption in various regions was ob-
tained from Fishery Statistics of the United States,
1976, publisheoBytheU.S.DepartmentorCommerce.
Reaction constants were obtained from EPA's publication
entitled Water Related Fate of 129 Priority Pollutants
(EPA 440 1479-029). Sediment/pollutant partition coeffi-
cients were derived from actual water partition coefficients
for the pollutants of concern. Data on suspended sediment
concentrations were obtained from the U.S.G.S.
3-11
-------
GROUND WATER CONTAMINATION FROM
HAZARDOUS WASTE
The model used in this analysis to predict ground water
contamination from landfills or lagoons is based upon work
described in a recent EPA paper.* This paper describes a
procedure for rapidly screening many compounds based upon
their physical and chemical properties for potential to
contaminate subsurface waters. A range of estimates of
the movement and persistence of specific compounds is pre-
dicted based upon a variety of environmental conditions
that the compounds may experience at a disposal site.
This model is the most uncertain of the exposure models
used in this study due to the fact that little is known
about che migration of contaminants through ground water.
SPA is currently developing improved models for predic-ing
movement through the ground water. These models may be
available for use next year.
The procedure used in this study involves three steps:
o Estimation of release rates.
• Use of mathematical models to estimate concentra-
tions as a function of distance from the landfill.
• Estimation of population exposure.
The first step assumes that an uncontrolled release from
landfills and lagoons will occur. In the model described
by Falco et al, two categories of contaminants, major and
minor, are designated. The major contaminant is the one
present in the highest concentration. All other contamin-
ants are designated as minor constituents. For the major
contaminant, the concentration in water leaching out of a
disposal site is assumed to be equal to its solubility.
Solubility for all contaminants in our study is included
Falco et al., "A Screening Procedure for Assessing the
Transoort and Degradation of Solid Waste Constituents
in Subsurface and Surface Waters," to be published in
Proceedings of the Society of Environmental Toxicology
and Chemistry.
3-12
-------
in Water Related Environmental Fate of 129 Priority Pollut-
ants. For all minor contaminants, the initial concentration
is assumed to be the equilibrium concentration obtained by
partitioning between sorbed solid phase material and dis-
solved material. This equilibrium concentration is based
on the chemical-specific "octanol water partition coeffi-
cient" (Kow).*
Recent evidence indicates that some materials can be
transported through the ground as free flowing liquids and
thus concentrations could exceed the chemical's solubility
in water. To compensate for this effect it has been assumed
that the initial concentration for all contaminants, not
just the major contaminants, leaching out of a disposal
site is equal to the contaminant's water solubility.
Once the contaminants are released, they are transported
with ground water movement and may eventually pollute drink-
ing water wells. To estimate transport, the procedure as-
sumes steady-state concentrations have been achieved. Fur-
ther, the procedure also assumes that the soil sorptive
capacity has been reached. Given these, assumptions, the
differential equation that defines transport and degradation
in ground water is:
d£ _ -k • C
dx V
Where:
dC
^x" = change in concentration as a function of
distance
C = concentration of the contaminant in
ground water
x » distance from disposal site
V = velocity of ground water movement
k - rate of hydrolysis
* Kow is obtained by mixing the chemical in a container
of two phases, water and octanol. The ratio of the
chemical dissolved in the octanol phase to the chemical
dissolved in the water phase is the Kow.
B-13
-------
The solution to this equation is:
C_ - exp ,-k • *,
Where:
C0 = the initial concentration, i.e., the con-
taminant's water solubility
£_ = the fractional concentration at a distance
Co x from the disposal site.
In PHB's analysis, hydrolysis is assumed to be the only
degradation process. Also, a typical ground water movement
velocity of 10 meters per year is assumed.
Along with hydrolysis, dilution is another important
process for lowering ground water concentrations. It is
assumed that the contaminants will migrate in the direction
of ground water flow. Without site specific information one
does not know the direction of ground water flow nor the mag-
nitude of dilution. Concentration in ground water when di-
lution and degradation are taken into account is as follows:
c * Df9 ' Co ' exP "* ' f
Where:
Dfg = dilution factor for diffusion angle S.
In the model it is assumed that the contaminant will mix
uniformly over a wedge with an angle of & on its face and
'side. The model also assumes that the maximum mixing depth
beneath the surface is 500 meters. Dilution after a certain
distance (x) from the disposal site is proportional to the
volume of the landfill divided by the volume of the wedge
underneath the surface at distance x. (See Figure 3-1. )
S-14
-------
Figure 3-1
DILUTION GEOMETRY
Landfill
The equations which describe the dilution factor are
presented in Exhibit B-6.
Exhibit B-7 indicates the value of the dilution factor
for a given set of input parameters and for various diffusion
angles. The figure' indicates as the angle of diffusion
decreases the concentration increases. In this study ?H3
used a dilution angle of 2.8125-degrees so that ground water
pollution would not be underestimated and because this dilu-
tion angle more closely approximates concentration profiles
given in the literature. Finally, ground water concentra-
tions are estimated only to a distance of 10 kilometers
*rom the disposal site. At the assumed velocity of 10
meters per year, it would take 1000 years to migrate 10
kilometers.
The first two steps provide a procedure for roughly
estimating concentrations "at various distances from the
disposal site. In the analysis it is assumed that the
hazardous waste not incinerated would be disposed of in a
sanitary landfill on site at the plant. The number _ of
oeoole exposed to contaminated ground water is a function
of the number of people who drink well water in the area
o- interest. EPA provided data on the number of people
who drink ground water in each county in which the plants
of interest are located. Assuming this population is
uniforalv distributed within each county the number of
oeoole at varving distances from the plant can be calcu-
lat»d. It is" also assumed that no drinking water weu-ls
would be located within 300 meters of the disposal site.
For a 2.8125 dilution angle, the number of people exposed
to contaminated drinking water is assumed to be:
B-15
-------
Exhibit B-6
DILUTION FACTOR EQUATIONS
W
Df° = w + H*3 - -5I3
for (R < r)
for (r < R < R'/TANS)
— *
Where:
W
r
d
w/d
x
R
rTR3/S3
Sg
R'
for (TANS 1
= volume of hazardous waste from the plant
= — . /~"'w a radius of landfill
V 77d
= depth of landfill (assumed to be 5 meters)
= surface area of landfill
= distance from boundary of landfill
= r + x = distance from center of landfill
= 360/9
« surface area of dilution wedge of angle 3
- wedge volume for diffusion of 9 degrees in
XY and YZ planes
* proportionality constant which depends on
angle of diffusion
R1 » maximum depth of diffusion below surface,
assumed to be 500 meters
R TA2JS = depth of diffusion below surface at dis-
tance R from center of landfill
-i
Source: PH3 Analysis
3-16
-------
Exhibit B-7
DILUTION: ANGLE SENSITIVITY
W = 100,000 M3
d - 5 meters
r = 80 meters
Distance From Landfill Center (m)
22.5 degrees
11.25 degrees
5.525 degrees
2.3125 degress
Source:
-------
££., , 9N
P(R) . 12g f R2 - (300 + r) 2 J for (3 >_ 300 + r)
Where: ?{R) = the number of people drinking well
water at distance R.
P - the average number of people per unit
area who drink well water in the vicin-
ity of the plant.
r = radius of landfill
The exposure is found by multiplying the number of
people exposed at a given distance times the concentration
at that distance.
The method described above is the most uncertain of
the exposure models used in this effort, due to the fact
that little is currently known about movement of contaminants
through ground water. In view of this, it is important to
keep several limitations in mind:
1. The model does not take into account the time it
takes for the contaminant to migrate to the ground.
water (i.e., the model is in steady state).
2. The model considers hydrolysis as the only degra-
dation process. Other processes could be consid-
ered if" significant evidence of occurrence war-
ranted.
3. The model assumes uniform mixing in a wedged
shape volume. Dilution of the contaminant as a
function of distance is proportional to the vol-
ume of the wedge.
4. The model assumes that the contaminant is trans-
ported as a dissolved solute in water. Recent
evidence has indicated that in some situations
solvents can be transported as a free flowing
liquid. Thus, concentrations in ground water may
exceed their water solubility.
5. The model is not applicable for transport of
trace metals.
6. Site specific information on hydrogeology, loca-
tion of wells relative to disposal sites and so
forth are not included in the model.
B-18
-------
Finally, PHB has assumed zero risk when hazardous waste
is disposed of in a secure landfill.
ESTIMATION OF MAXIMUM INDIVIDUAL RISK
The methods described in this appendix have been used to
estimate aggregate exposure levels. When these exposure
levels are combined with information on dose-response rela-
tionships, total population risk can be estimated.
Regulatory authorities are often concerned with indi-
vidual risk levels as well as aggregate risk levels. PHB
has estimated maximum individual risk levels for a large
chlorinated solvent plant which is located in the most
densely populated area in one of the three regions. For
this analysis only cancer risk was considered.
The approach adopted examined the exposure pattern
(i e the"distribution of concentrations) of each pollutant
emitte^ '^om the plant. An example air pollution exposure
pattern for one pound of pollutant was provided in Exhibit
3-2. R1'sks from exposure to water • pollution were ignored
in this"approach because the risks are very small compared
to air pollution and they are incurred by different popula-
tion subgroups.
By applying the CAG unit risk factors to the air expo-
su-e pattern, the risk to each population subgroup can be
calculated. For the particular plant examined, six carcin-
oaenic pollutants are emitted and the risks were summed to
comoute" the total risk to each population subgroup._ The
results for uncontrolled plant operations are shown in the
figure below.
Figure B-2
MAXIMUM INDIVIDUAL RISK
Annual Individual
Persons Cancer Risk
96 10-4 to 10~5
1490
10~5 to 10~6
3-19
-------
PROCEDURES FOR ESTIMATING
HEALTH RISK APPENDIX C
?HB asked Clement Associates, Inc. to develop an ap-
proach for estimating the risks associated with exposure
to pollutants which are released during copper smelting and
in the manufacture of chlorinated solvents. For each of
these pollutants the following eight health effects have
been evaluated:
• Carcinogenicity,
• Teratogenicity,
• Reproductive tcxicity,
• Mutagenicity,
• Hepatotoxicity,
• Renal toxicity,
• Neurobehavioral toxicity, and
• Toxic effects on other organ systems.
In implementing their approach Clement has used crude
linear dose-response models to estimate incidence. These
models have not gained acceptance for low dose extrapolation
for effects other than cancer. In the absence of better un-
derstanding of dose-response relationships at low dose
levels, it cannot be claimed that these models provide an
absolute measure of risk. Review and further development
-------
of this technique by EPA should improve the predictive
value of these models.
The next section of this appendix describes the Clement
approach in detail. Following this section is an example
of the application of the approach to a specific pollutant.
Clement has prepared a short paper on each pollutant which
summarizes the basis for estimating health risks. Because
of the volume of this material only one example is included
in this appendix. A copy of this material may be obtained
directly from Clement Associates or from EPA.
It should be noted that whenever CAG unit risk factors
were available, they were used by PHB in place of the
Clement estimates. In almost all cases the CAG unit risk
factors and the Clement estimates are comparable. Further,
in the case of respiratory effects associated with exposure
to sulfur dioxide SPA supplied a risk function which was
used rather than Clement's estimate.
The Clement estimates when combined with the exposure
estimates described in Appendix B provide an overall measure
of health risk to the exposed human population. For example,
if the probability of occurrence of health effect i from
pollutant j at concentration C^ is denoted by the Clement
score SCCfcl^j, then the overall measure of risk of health
effect i from pollutant j is defined as follows:
I pj(cV • s«Vij
Where Pj(C^) "is the population exposed to each concentration
of pollutant j.
Summing over all pollutants — over the j subscript —
provides an estimate of the overall measures of risk of
health effect i. This can be defined as:
I I PtC) • StC)
C-2
-------
TOXICS INTEGRATION PROGRAM SCORING OF
SELECTED POLLUTANTS FOR RELATIVE RISK
Prepared by:
Clement Associates, Inc.
1010 Wisconsin Avenue, N.W.
Washington, D.C. 20007
June 26, 1981
C-3
-------
INTRODUCTION
At the request of Putnam, Hayes & Bartlett, Clement
Associates has evaluated 41 water and/or air pollutants for
carcinogenicity, teratogenicity, reproductive toxicity, mu-
tagenicity, hepatotoxicity, renal toxicity, neurobehavioral
toxicity and toxic effects on other organ systems. The
chemicals evaluated are shown below.
CHEMICALS SCORED FOR TOXIC EFFECTS
Arsenic
Benzene
Bis(2-ethylhexyl) phthalate
N-Butyl chloride
Cadmium
Carbon tetrachloride
Chloroform
2-Chlorophenol
Chromium
Di-n-butyl phthalate
1,1,Dichloroe thane
2,4-Dichlorophenol
1,3-Dichloropropene
Ethane
Ethyl chloride
Ethylene
Ethylene dichloride
Fluorene
Hexachlorobenzene
Hexachlorobutadiene
Hexachloroethane
Lead
Mercury
Methanol
Methyl chloride
Methyl Ether
Methylene chloride
Monochloroace tylene
Nickel
Phenol
Propane
Sulfur dioxide
Tetrachloroethylene
Total Suspended Particulates
Trans/cis-dichloroethylene
1,1,1-Trichloroethane
1,1,2-Trichloroethane
Trichloroethylene
2,4,6-Trichloro?henol
Vinyl chloride
Vinylidene chloride
Using a semiquantitative method for relative risk ranking,
Clement assessed the strength of the available evidence for
each of the 41 chemicals and estimated the relative likeli-
hood that each chemical is a human toxicant. Where suffi-
cient data existed, an estimate was also made of the proba-
bility that a given toxic effect will occur in exposed
humans per unit dose of exposure. It. should be emphasized
that this procedure is intended only for the purpose of rel-
ative risk ranking of the 41 chemicals, and cannot be used
to estimate how many members of the exposed population are
at risk or to estimate the risk to an exposed individual.
C-4
-------
OUTLINE OF METHOD
1. For each pollutant, Clement provides a score (S) that
represents the probability that a given hazard will
occur in exposed human populations.
2. The effects that are scored are carcinogenicity, tera-
togenicity, reproductive toxicity, mutagenicity, hepa-
totoxicity, renal toxicity, neurobehavioral toxicity,
and effects in other organ systems. A score (S) Is
assigned for each of these forms of toxicity for each
pollutant.
3. Clement did not use data from the primary literature
to score these pollutants. Scoring was based solely
on toxicity data provided in EPA and NIOSH criteria
documents, in IARC and NAS reviews, and in a few other
readily available secondary sources. Thus, the scor-
ing system accommodates the fact that secondary litera-
ture often does not contain complete information, espe-
cially on dose—response relations. All conclusions
regarding toxicity and risks are thus necessarily
limited.
4. The score S is a product of two measures of risk:
T = probability that the pollutant is toxic to humans,
based on inferences from animal data or on direct
measures of human toxicity.
P = probability of occurrence of the toxic effect in
exposed humans, assuming that the agent is a human
toxicant.
5. The score T is a function of the strength of scientif-
ic evidence that a given pollutant is capable of pro-
ducing toxic effects in humans. It is derived by
considering evidence from human studies, animal stud-
ies, and, in some cases, from in vitro studies. It
is also based on knowledge regarding the predictive
power of animal and in vitro tests. For toxic effects
other than carcinogenicity, when the value of T has
a maximum value of 0.95, since even very strong animal
data may not be 100% predictive for humans for these
effects. It is not a measure of the absolute probabil-
ity of human hazard. No such absolute measure can be
defined.
C-5
-------
The score P takes one of three possible forms, depend-
ing on the nature of the hazard being scored:
a) For carcinogens:
P = risk per unit dose of carcinogen, and it is
taken directly from EPA assessments of car-
cinogenic risk, times X (actual human exposure-
in mg/kg of body weight)
b) For pollutants exhibiting other forms of toxicity
and yielding dichotomous responses:
P » I/MED • X
where
I = observed incidence of effects above the con-
trol incidence at the minimum effective dose
(MED)
X = actual human exposure in mg/kg of body weight
It will be assumed that this measure of risk ap-
plies everywhere on the dose-response curve for
the pollutant. Because such an assumption is
needed, it cannot be claimed that ? is an abso-
lute measure of risk per unit dose; however, it
is likely that application of this approach to
all the pollutants being scored will lead to a
ranking that places the pollutants in the order
of their relative risks. No methods have been
developed for treating dichotomous dose-response
data other than carcinogenicity for purposes of
estimating risk per unit dose at low doses.
Presumably, the method used to treat carcinoge-
nicity could also be adapted to other dichotoraous
toxicity data. However, to do so would have
required a significant development effort, and
could not have been accomplished within the con-
straints imposed on this contract, not the least
of which was the restriction on use of primary
sources. The measure I/MED is a rough first
approximation of the risk per unit dose of low
doses, and at least permits a relative risk rank-
ing.
C-6
-------
c) For toxic agents giving rise to a graded response,
? is estimated as follows:
1) It is assumed that if humans are exposed at
the NOEL or MED for test animals, close to
the entire population will be affected. Tra-
ditionally, toxicologists have assumed .that
the general population is likely to be more
sensitive to toxic agents than a given test
animal population. Thus, test animal NOELs
are considered insufficiently low to protect
human health, and these NOELs are divided by
safety factors (see below) to yield accept-
able levels of human exposure. Although it
is far from certain that most members of
the human population will be affected at a
test animal NOEL, such an assumption can be
used in the present situation because the
purpose is only to estimate relative risks.
Thus, all compounds rated will be assumed
to produce adverse effects in the same frac-
tion of an exposed human population (i.e.,
near 100%), when human exposure is at the
test animal NOEL.-
The use of MED is confined to situa-
tions in which an experimental NOEL is not
available. If the MED is truly a "minimum"
effect dose, then it should approximate the
NOEL.
2) It is assumed that if humans are exposed at
specified fractions of the NOEL or MED for
test animals, no significant number of per-
sons will be affected. The denominator (f)
of those fractions depends on the source of
data from which the NOEL or MED is taken:
a) If the NOEL or MED is derived from chron-
ic studies, f = 100
b) If the NOEL or MED is derived from sub-
chronic studies, f « 1,000
c) If the NOEL or MED is derived from con-
trolled studies of human exposure, f = 10
C-7
-------
These values of f represent the traditional safety
factors used in toxicology to assign acceptable
human exposure levels for toxic agents (NAS 1980) .
Although their application does not ensure that
the derived exposure level is below a threshold
for the entire human population, experience has
shown that there is a high probability that this
is the case. Use of such an assumption in the
present scheme provides consistency with current
methods of toxicologicai science used to assign
acceptable intakes of agents giving rise to graded
toxic effects.
3) It is further assumed that at dose levels between
the NOEL (or MED) and the NOEL/f, the probability
that an exposed population is above a threshold
is directly proportion to the dose.
Thus, ? for pollutants causing graded toxic
responses is given as follows:
where
X - the actual human exposure in rag/kg of
body weight
NOTE: If NOELs are not available, MEDs will be
used in the above formula. It should be noted
that this formulation leads not to relative risk
per unit dose, but to the relative risk (i.e.,
the actual exposure (X) is included in the above
formula). It again must be emphasized that this
method yields, at best, a relative risk ranking,
and cannot be used to estimate actual risk.
If certain toxic effects (e.g., respiratory effects)
were thought to be related to route of exposure, then
toxicity data reflecting only the relevant route of
exposure were used to score a pollutant. Otherwise,
all routes of exposure were considered relevant to
risk assessment.
C-8
-------
8. Where both acute and chronic effects of a given pollu-
tant exist in a single organ system, the scoring was
based on the chronic effects because they are considered
to be more relevant to human exposure to environmental
pollutants.
DISCUSSION
Elimination of Certain
Compounds from Scoring
Health effects scoring was not done for ethane, ethy-
lene, fluorene, and monochloroacetylene because it is un-
likely that any of these compounds could be present in the
environment in quantities that could adversely affect human
health or environmental quality.
Ethane and ethylene are hydrocarbon gases that for all
practical purposes are biologically inert (Patty 1963).
These two compounds cause no known systemic toxic effects
until they reach sufficient concentrations to exclude oxy-
gen, in which case the signs and symptoms of intoxication
are those of oxygen deprivation. It is implausible that
either compound could be present in the open environment at
levels sufficient to exclude oxygen.
Fluorene is a compound whose toxicity is not well
studied. The 1979 NIOSH Registry of Toxic Effects of Chemi-
cal Substances reports an oral LD$Q in rats of 5,000 mg/kg.
Fluorene is probably narcotic at higher vapor concentrations
but its systemic toxicity is expected to be low. It is un-
likely to pose any significant threat of impairment at the
concentration that might be expected in the open environment.
Monochloroacetylene is not listed in chemical diction-
aries or in any standard toxicology resources. The 40th
edition of The Handbook of Chemistry and Physics describes
it as an "unstable, spontaneously inflammable gas." It
appears that if monochloroacetylene were released into the
environment it would be immediately transformed into some
other substance or substances.
C-9
-------
Limitations in Scoring
The development of a biologically meaningful procedure
for estimating the probability of adverse human health ef-
fects (i.e., the risk) due "to exposure to chemicals is
beset with difficulties and scientific uncertainties. Data
obtained from experimental studies in animals and from epid-
emiological investigations in humans are used as evidence
in describing the nature and extent of chemically induced
adverse health effects. Using such data, it is not possible
simply to categorize chemicals as either toxic or nontoxic.
It is most unfortunate that toxicity is not recognized as a
concept rather than a finite event easily measured. Accord-
ingly, risk and, therefore, safety are relative concepts.
Procedures for risk assessment and safety evaluation of po-
tentially toxic chemicals are confounded by poorly defined
toxicological principles. Moreover, content'and quality of
tcxicological information is often quite variable, and the
interpretation of a study's results is often equivocal.
As with all approaches of the type used here, the
method developed by Clement has inherent weaknesses and un-
certainties, which include:
1. Lack of quantitative data for some of the pollu-
tants that were scored
2. Inability to extrapolate the results of in vitro
mutagenicity studies to humans
3. Utilization of arbitrary (although traditionally
used) safety factors for effects having graded
responses
4. Unknown but likely interspecies differences in
absorption, metabolism, pharmacokinetics, and ex-
cretion.
In addition, the system penalizes those compounds that
have been the most extensively studied relative to compounds
for which little or no health effects data exist (i.e., un-
tested is not "safer"). These weaknesses and uncertainties
are discussed in detail.
An integral part of this project was to provide a quan-
titative measure of the relative probability of occurrence
of toxic effects in exposed humans per unit dose cf exposure.
C-10
-------
Because of the time and resource limitations for this
project, it was not possible to identify, acquire, and cri-
tically evaluate primary literature resources for each of
the pollutants. Therefore, the toxicity profiles were pri-
marily based on data reported in the IARC monographs, HAS
reviews, EPA water quality criteria documents and NIOSH
criteria documents. The extent and validity of the data
reported in these secondary sources were a major limitation
in assigning ?-values for each of the toxic endpoints eval-
uated. The qualitative nature of the data from the secon-
dary sources posed a major problem. In many cases detailed
quantitative data on the compounds of interest were not
available. An additional problem posed by the use of
secondary literature sources was their emphasis on a pre-
dominant effect or route of exposure. The EPA water crite-
ria documents emphasize carcinogenicity and acute toxic
effects. The NIO*SH criteria documents emphasize the inha-
lation and dermal routes of exposure, and primarily report
on effects to the principle target organs.
In Clement's method for determining the strength of
evidence for a given toxic endpoinn, those compounds that
have been extensively studied are penalized when they are
compared to compounds for which there is a paucity of data.
In the absence of empirical data for a given toxic endpoint
for a compound, Clement has arbitrarily elected not to
apply some minimum scoring factor for the effect and has
applied a score of 0 when an effect has not been studied.
However, Clement is not assuming that the compound does not
produce that effect. Rather, Clement assumes that there is
too much uncertainty to estimate a scoring factor in the
absence of empirical data that would be biologically meaning-
ful and scientifically justifiable.
Ideally, the probability of exposure to multiple toxic
compounds requires a consideration of their interacting ef-
fects. Concurrent exposures may alter the rates of absorp-
tion, metabolism, or excretion of one or more of the inter-
acting chemicals. The biological effects resulting from
multiple exposure may be altered, with responses equal to,
greater than, or less than the sum of effects of the indi-
vidual chemicals. While Clement is aware that potentiation,
interactive effects, the state-of-the-art is not sufficient-
ly developed to allow for this. Very few studies have been
C-ll
-------
performed to investigate interactive effects in low-dose,
chronic experiments.
The use of safety factors is a crude form of extrapola-
tion. The no-observed-effect level (NOEL) is divided by
some safety factor, e.g., 100, to arrive at an acceptable
exposure level in humans. Application of a safety factor
assumes that the factor used reduces the risk to a negligible
level. The shortcomings of safety factors are obvious.
They are chosen arbitrarily, and do not reflect the sensi-
tivity of the measurement used to derive the NOEL nor the
relevance of the animal data to the biological response in
humans. In addition, since the NOEL is by definition a
likely subthreshold dose level, safety factors are not
applicable for agents producing nonthreshold effects. Math-
ematical models exist for estimating risks associated with
dichotomous responses, such as cancer. Because graded
responses are difficult to quantify, safety factors are
applied in determining risks associated with certain types
of biological responses. While the safety factor approach
has several substantial limitations, it is the traditional
method used by regulatory agencies to- account for possible
inter-species differences in susceptibility, for uncer-
tanties inherent in animal testing due to biological vari-
ability, and for the limited number of test animals that
can practically be used (NAS 1980). No single safety factor
margin has been designated as a universal standard for all
chemicals. With experience a reasonable range of safety
factors has emerged in regulatory toxicology and these
safety factors has emerged in regulatory toxicology and
these"safety factors were the ones chosen by Clement. How-
ever, as pointed out above, the choice of such safety
factors is without a systematic scientific basis and in-
troduces uncertainty as to the true risk associated with
exposure to a given chemical.
Uncertainties exist in both the low-dose extrapolation
procedures used by Clement, and in the extensions of extra-
polated results to humans. The extension of dose-response
data obtained from animals to humans poses two different
problems: the estimation of species sensitivity and the
estimation of appropriate "scaling11 factors (conversion
factors for extrapolation of animal data to humans, such
as body weight, surface area, metabolic rate, and life
expectancy). There is a paucity of data on the concordance
between the doses effective in humans and doses effective
in the most sensitive animal species. Clement has assumed
C-12
-------
that for all toxic endpoints, humans are the most sensitive
animal species, although empirically this may not be the
case. In the absence of empirical data we have arbitrarily
applied safety factors to the data.
An additional source of uncertainty is interspecies
and intraspecies differences in the absorption, metabolism,
pharmacokinetics, and excretion of chemicals. These dif-
ferences may be both quantitative and qualitative and may
be influenced by the exposure conditions, dose level, and
the presence of one or more interactive chemicals. Intra-
species differences have been extensively documented in the
literature. In addition, there may be special subpopula-
tions with increased sensitivity for a particular effect
due to exposure to a given chemical (NAS 1930).
The importance of interspecies differences was evident
in methanol, one of the chemicals studied. There are sig-
nificant qualitative differences in the metabolism and
quantitative differences in the excretion kinetics of meth-
anol given to nonprimate versus primate species. Because
of these differences nonprimates are not considered to be
an appropriate model for the toxic effects of methanol in
humans. It is not known whether this is the case for the
other chemicals evaluated by Clement. In this absence of
empirical data, such differences must be considered an
imporcant uncertainty factor.
One of the major uncertainties in this project was the
extension of the results of mutagenicity tests to humans.
Mutagenicity tests include, but are not limited to, assays
for induction or repair of DNA damage; mutagenesis in bac-
teria, yeast, or Drosophila melanoqaster; mutagenesis in
mammalian somatic cells; mutagenesis in mammalian germinal
cells; and tests for neoplastic transformation of mammalian
cells in culture. Evidence is also acquired from studies
in whole animals and from data showing damage to human
chromosomes. The validity of extrapolating such data to
predict human mutagenic risk is not established, and this
adds further uncertainty to the risk assessment procedure.
Clement has been unable to devise a procedure than
allows for the extension of results obtained in in vitro
svstems to humans. For compounds that have been tested for
mutagenic activity in an in vivo mammalian test system
C-13
-------
(e.g., heritable translocation assay, dominant lethal test,
sperm abnormality test), P-values have been calculated when
quantitative data are available in the secondary literature.
However, for compounds assayed only in an in vitro system,
Clement cannot, at this time, calculate the probability of
a mutagenic effect in exposed humans per unit dose of expos-
ure.
CONCLUSION
Given the above limitations, the claims made for the
scoring system must be limited. Thus, it yields at best a
relative measure of risk for the 41 compounds scored. It
cannot be used to estimate actual risk.
Nevertheless, while there are weaknesses in the system,
the underlying concept represents an ideal goal for assess-
ment of risks from exposure to toxic substances. Further
development work should allow imDroveraents in raanv areas.
C-14
-------
REFERENCES
INTERNATIONAL AGENCY FOR RESEARCH ON CANCER (IARC). 1980.
An evaluation of chemicals and industrial processes asso-
ciated with cancer in humans based on human and animal
data: IARC Monographs, Volumes 1-20. Cancer Res. 40:1-12
NATIONAL ACADEMY OF SCIENCES (NAS). 1980. Drinking Water
and Health. Washington, D. C.
PATTY, F. A. 1963. Industrial Hygiene and Toxicology.
.2nd ed. Interscience Publishers, New York
C-15
-------
CHEMICAL: ARSENIC
Arsenic is a human carcinogen causing skin cancer in
persons exposed to inorganic arsenic through drugs, drink-
ing water, and pesticides. Lung cancer has been shown to
result from exposure to airborne arsenic compounds in
pesticide manufacture in smelter workers. Cancer of the
liver has also been associated with occupational arsenic
exposure.
The evidence that arsenic is rautagenic is equivocal
but strongly suggestive. Inorganic arsenic compounds cause
terata and other adverse reproductive effects in animals
and the evidence is strongly suggestive that it causes
similar effects in humans.
Arsenic also causes noncancerous skin lesions and is
neurotoxic.
C-16
-------
SUMMARY OF EVIDENCE FOR CARCINOGENICITY
Arsenic (As) in the drinking watar has been correlated
with skin cancer in a village in Taiwan; the incidence of
cancer at water concentrations less than 0.29 ppm (0.008 ing/
kg/day) was 2.6 per 1,000 residents, 21.4 at 0.60 ppm or
less (0.017 mg/kg/day) (Tseng 1977, as reported bv USEPA
1980).
IARC (1980) has concluded that there is sufficient
evidence that inorganic arsenic compounds are skin and lung
carcinogens in humans.
The value for ? 15.94 x 10~3 (ug/m3)-l] j_s taken from
an assessment of risk for respiratory cancer in workers
exposed to arsenic atmospheres; the calculation is based on
epidemiology studies (Clement 1980).
T = 1.0
P « 27.8 (isgAg/day)"1 • X
C-17
-------
SUMMARY 0? EVIDENCE FOR MUTAGENICITY
There is substantial evidence that exposure to arsenic
compounds either in an occupational setting or via drug
products is correlated with an increased incidence of
chromosomal aberrations in humans (IARC 1980, USEPA 1980).
However, in no case was a quantitative dose-response rela-
tionship shown. Many of these reports also showed the
presence of confounding factors such as cigarette smoking,
X-ray exposure, other metals, and treatment with other
drugs.
In bacterial systems, sodium arsenite caused mutations
in Sscherichia coli WP2 and Bacillus subtilis, but not in
Salmonella (Ames test) (Nishioka 1975, Kada et al., 1980,
Lofroth 197S; as reported by IARC 1980). Arsenate was
also negative in the Ames test and positive in the B. sub-
tilis system (Lofroth 1978, .Kada et al. 1980; as reported
by IARC 1980). Both arsenite and arsenate induced chromo-
somal aberrations in cultured mammalian cells including
human lymphocytes (Oppenhim 1965, Paton 1972; as reported
by IARC'1980)".
Administration of 10 or 100 mg/liter of sodium arsenite
in drinking water in rats caused a slight increase in chromo-
somal aberrations in mouse bone marrow cells, but arsenic
was not active in dominant lethal assays in mice at levels
of 100 mg/liter in drinking water or at oral doses of 0.25,
0.5, and 1 mg/kg (Sram 1976, Gencik et al. 1977; as reported
by IARC 1980).
The evidence for mutagenesis of arsenic is equivocal
but strongly suggestive. The T-value for this effect is
0.55. P is calculated assuming an incidence of 1 percent
(0.01) at an MED of 10 ing/liter of sodium arsenite in drink-
ing water, equivalent to 1.37 mgAg/day as elemental arsenic.
T = 0.55
? = 7.30 x 10~3
C-18
-------
SUMMARY OF EVIDENCE FOR TERATOGENICITY
An epidemiology study indicated that there was an
increased incidence of multiple malformations among children
born to women who worked at a smelter compared to women who
lived in the area, but did not work at the smelter. Arsenic,
however, was only one of several "potentially genotoxic"
agents present and no exposure levels were given (Nordstrom
1978 a, b, c, and d; as reported by IARC 1980).
In animal studies, various arsenic compounds have been
shown to be teratogenic in a number of different species by
various routes (IARC 1980, USEPA 1980). The following
table is compiled from reports in IARC 1980 and USEPA 1980.
Dose
Soecies Route
Compound
Author
Chicken Egg Unspecified Unspecified Farm 1977
injection
Golden IV or I? Sodium arsenate 15-25
hamster
Fenti 1977
Mouse
.Mouse
Mouse
Rat
Rat
I?
IP
IP
IP
I?
Sodium
Sodium
Sodium
Sodium
Sodium
arsenate
arsenate
arsenate
arsenate
arsenate
45
10-12
40
20-40
45
Hood 1972
Hood 1972
Hood 1978
3eaudoin
1974
Burk 1977
In neither secondary reference were incidences of mal-
formations reported, nor was there a NOEL for teratogenesis
given in any of the studies. For the purpose of calculating
?, the MED is 10 mgAg of sodium arsenite, assuming an in-
cidence of malformations of 10 percent at this dose.
T = 0.35
? = 1.73 x lO-2 (mgAg/dayr1 • X
C-19
-------
SUMMARY OF EVIDENCE FOR REPRODUCTIVE TOXICITY
Fetal mice given 10-12 mg/kg of arsenate while in
utero showed a significant (p<0.05) increase in mortality.
Sodium arsenate at 40 ng/kg-given to the dam intraper-
itoneally caused fetal death and a reduction in fetal body
weight in mice (Hood et al. 1977, as reported by USEPA 1980)".
Oral administration of 10-40 mg/kg of sodium arsenate
on days 9, 10, or 11 of pregnancy caused an increased number
of resorptions in ICR mice (Matsumoto et al. 1973, as reported
by IARC 1980). No incidence data were given. For the purpose
of calculating P, a 10 percent increase in the incidence of
resorptions at 10 mg/kg of sodium arsenate was assumed.
T = 0.55
P = 2.48 x 10~2
C-20
-------
SUMMARY OF EVIDENCE FOR HEPATOTOXICITY
Hepatomegaly was reported in "the majority of patients"
who consumed arsenic-contaminated soy sauce, but both liver
function tests and biopsies revealed "few abnormalities"
(Mizuta et al. 1956, as reported by USEPA 1980). Infants
fed formula made with arsenic-contaminated powdered milk
also showed hepatomegaly, but no further data were oresented
(USEPA 1980).
Because of the limitations of these reports and in the
absence of further evidence, arsenic cannot be considered
hepatoxic.
C-21
-------
SUMMARY OF EVIDENCE FOR RENAL TOXICITY
There were no studies available for review from which
to assess the renal toxicity of arsenic.
C-22
-------
SUMMARY OF EFFECTS ON OTHER ORGAN SYSTEMS
Chronic exposure to arsenic has been associated with
progressive changes in the skin, which may ultimately result
in skin cancer. The sequelae begin with melanosis and
progress to keratoses, which have been called precancerous
(USEPA 1980). No specific data were given, but the effects
were observed in the Taiwanese people (Tseng et al. 1963,
as reported by USEPA 1980); thus ? cannot be established.
1.0
C-23
-------
SUMMARY OF EVIDENCE FOR NEUROBEHAVIORAL TOXICITY
Various authors have reported peripheral neuropathy in
humans occupational!/ exposed to inorganic arsenic; the
symptoms and signs are for a progressive polyneuropathy
involving both motor and sensory nerves and particularly
affecting distal extremities and myelinated long-axon neu-
rons (USEPA 1980) .
Quantitative data for exposure are limited. Peripheral
neuropathies were reported in 44/220 patients poisoned with
arsenic-contaminated soy sauce; the total dose was estimated
to be 60 mg (0.86 mgAg/ 70-kg man) (Mizuta et al. 1956, as
reported by USEPA 1980). Polyneuropathies were reported in
37/74 patients ingesting arsenic trioxide or arsenic sulfide
at 0.05 or 0.15 mgAg/cay (Tay and Shea 1975, as reported
bv USEPA 1980). Japanese infants fed arsenic-contaminated
formula showed CNS damage (epilepsy, mental retardation),
hearing and visual pathologic effects, and brain wave (EEC)
abnormalities at a total intake of 90-140 mg (18-28 mgAg)
(USEPA 1980). The calculation of P uses f = 10 and MED =
0.05
1.0
X - 0.005 .
0.045
C-24
-------
REFERENCES
CLEMENT ASSOCIATES, INC. 1980. Assessment of Human Respi-
ratory Cancer Risk Due to Arsenic Exposure in the Workplace.
September 5, 1980.
INTERNATIONAL AGENCY FOR RESEARCH ON CANCER (IARC). 1980.
IARC Monographs on the Evaluation of Carcinogenic Risk of
Chemicals to Humans. Vol. 23: Some Metals and Metallic
Compounds. World Health Organization, Lyon, France.
U.S. ENVIRONMENTAL PROTECTION AGENCY (USEPA). 1980. Ambi-
ent Water Quality Criteria for Arsenic. Office of Water
Regulations and Standards, Criteria and Standards Division,
Washington, D.C. October 1980. EPA 440/5-80-021.
C-25
-------
ALTERNATIVE WEIGHTING SYSTEMS
FOR SIGHT HEALTH EFFECTS APPENDIX D
Having developed relative risk measures for each of
eight health effects, it is possible to define total risk
as a weighted sum of these measures. Although it is impos-
sible to objectively determine a unique set of weights", it
is interesting to explore the sensitivity.of results to a
range of weighting systems. To assist PHB in performing
such a sensitivity analysis, Dr. Milton Weinstein of the
Harvard School of Public Health proposed a methodology for
deriving alternative weighting systems reflecting one or
more of the following factors:
e Years of life lost per case
• Quality of life lost per case
• Direct costs of medical treatment per case
• Indirect costs (earnings lost due to morbidity
and premature death) per case
Dr. Weinstein's methodology is presented in his paper en-
titled Method for Assigning Weights to Health Effects in
Integrated Risk Reduction Models"datedMay28,1981.The
first section of this appendix will present this paper in
its entirety. The second section of the appendix will
discuss FSB's implementation of the methodology outlined in
Dr. Weinstein1s paper.
-------
METHODS FOR ASSIGNING WEIGHTS
TO HEALTH EFFECTS IN INTEGRATED
RISK-REDUCTION MODELS
Milton C. Weinstein, Ph.D.
May 28, 1981
D-2
-------
1. INTRODUCTION
1.1. Objectives
We are given a set of scores, Sj_, which will serve as
proxies for the increased incidence rates for several toxic
effects of concern (i = 1, 2, ..., n). We need not assume
that these scores have any quantitative significance in
absolute terms, but we must assume that they do provide
meaningful estimates of the relative increases in incidence
rates for the several toxic effects. Thus, in a linear
model, any positive multiple of the vector of scores (Si_)
is functionally equivalent to the original set.
The task at hand is to develop a meaningful objective
function, f(S]_, £2, • •., Sn), that reflects the overall
health impact on the society of this multivariate perturba-
tion of prevailing disease incidence rates. Assuming that
impact is, as a first-order approximation, proportional to
incidence (i.e./ that the impact per case is constant),
the problem reduces to that of assigning a set of weights ,
wj_, to each of the toxic effects, such that the function
f(Si, S2, ..., Sn) = ^ wisi
i=l
becomes the measure of societal health impact. As for the
scores themselves, the weights need not have any absolute
meaning; we only require that the ratios, wj_/wj , reflect
the relative health impacts of effects i and j. Thus,
without loss of generality, and to minimize the illusion
that the weights have any absolute significance, it may be
desirable to normalize them to sum to unity after having
developed them in admittedly arbitrary units.
The weights are a means to an end — that of exploring
the efficiency frontier for pollution reduction — and not
an end in themselves. They enable one to compare quantities
that are inherently incommensurable except by invoking
painfully restrictive assumptions about the social cost of
illness/ In economic theory, such assumptions are often
made by postulating the existence of willingness-to-pay
values, the sums of which provide the sought-for numeraire.
In practice, the problem is multidimensional, and no measure
that is both objective and comprehensive .exists. We can
D-3
-------
achieve objectivity only by excluding from consideration
those dimensions of health (e.g., pain) that are inherently
subjective; we can be comprehensive only by introducing
value judgments into the process (e.g., by assigning quality-
of-life weights to time spent in various disease states).
Hence, perhaps more useful than a unique set of weights
would be an efficient strategy for exploring the efficiency
frontier in the model, by varying the weights across the n-
simplex of possible values. For example, much could be
learned by allocating the weights to the vertices
(1, 0, 0, ...), (0, 1, 0,...), etc., or to the midpoints
of the edges (0.5, 0.5, 0, 0, ...), (0.5, 0, 0.5, 0, ...),
etc., or to the centroid (1/n, 1/n, 1/n, ...) of the
n-simplex. However, given limited resources with which
to experiment with the model, a small subset of plausible
and more-or-less defensible weight combinations might be
useful. The objective, then, is to develop such weights
that reflect, in each of several defensible senses of the
term, the social health impact of toxic effects in question.
1.2. Ideal Versus Practical
Assessment of Weights
Ideally, the data entering into the weighting process
(i.e., the toxic effects scores) would be expressed in
terms of specific human diseases, and would be disaggregated
in the form of age-specific incidence rates, with attention
to the latency period between exposure and effect. With
disease-, age-, and time-specific data of this kind, the
weighting system could be similarly disaggregated, with a
separate weight, wj_jt, assigned to a case of disease i in a
person of age j in year t. This would make possible the
use of disease-specific models of the cost and health
consequences of increased incidence, and would add credibil-
ity to the weights.
Unfortunately, the ideal is not possible under the
present circumstances. The level of scientific knowledge
does not provide a basis for estimating objectively the
disease-age-time-specific increases in incidence. I would
argue that this could be done subjectively by expert
toxicologlsts, but perhaps the scope and objectives of this
study do not justify that level of effort. Perhaps, on
another occasion, such an exercise might be attempted; in
that event, the more theoretically satisfying method of
assessing weights (which is referred to below, somewhat
D-4
-------
hyperbolically, as the "ideal") might be attempted as
well. Such an approach is sketched out below, in addition
to the "practical" approach recommended for the immediate
task, not for theoretical reasons, but to accentuate the
limitations of the admittedly blunt instruments with which
we are operating at present.
2. VALUED ATTRIBUTES
Any approach to developing a set of weights must
recognize the existence of at least the following conse-
quences of an episode of illness:
I. Health care resource costs
II. Loss of life expectancy
III. Impaired quality of life (including disability)
Health care resource costs are the costs of detection,
treatment and rehabilitation for the disease in question.
They are coirraonly, and uncontroversially, measured in dol-
lars. Loss of life expectancy requires no explanation,
except to point out that its only natural unit of measurement
is years. Impaired quality of life encompasses the dis-
ability, pain, suffering, anxiety, and other personal ef-
fects of illness. To these we might also add effects on
the quality of the life of others, such as family and
friends. There is no natural unit of measurement for these
effects.
Two conceptual frameworks have been applied to make
commensurate these consequences. In benefit-cost analysis,
one seeks economic equivalents, in dollars, for all econ-
omically measurable attributes. These are conventionally
classified as direct costs, indirect costs, and intangible
costs. The direct costs are the health care resource costs
(category I). The indirect costs, using a human capital
approach, consist of the value of productivity lost due to
premature mortality (category II) and due to morbidity
(category III). The aspects of death and disease included
in categories II and III, but not captured by the economic
measure, are the intangible costs; except for a few enter-
prising attempts to assess willingness to pay to avoid
these effects, these are usually excluded from analysis.
D-5
-------
Another approach, founded in multiattribute utility
theory, is to seek a common unit of health effect (categories
II and III). This is usually done by assigning weights
(between 0 and 1) to each possible health state (1 = well,
0 * dead), such that the overall measure of loss is expressed
in "quality-adjusted" years of life expectancy. (Such a
linear weighting places rather rigid restrictions on the
form of the implicit utility function.) Some researchers
have developed scales for a wide range of health status
levels. The commensuration of direct cost, in dollars, and
quality-adjusted life years is more troublesome, but could
be accomplished by exploring a range of values per quality-
ad justed" year of life or, alternatively, by ignoring the
direct costs altogether and focusing on health effects per
se.
In developing a set of weights, it seems .reasonable,
therefore, to "consider at least the following implicit
utility functions:
1. Weight diseases in proportion to loss of life ex-
pectancy per case;
2. Weight diseases in proportion to loss of quality-
adjusted life expectancy per case, suitably
(though subjectively) defined;
3. Weight diseases in proportion to direct health
care cost per case;
4. Weight diseases in proportion to total economic
cost per case (direct plus indirect);
5. Weight diseases in proportion to Ac + A.ALE,
where Ac is direct cost, ALE is loss of life
expectancy per case, and A is a weight (i.e., a
shadow price) that is allowed to vary;
6. Weight diseases in proportion to Ac + AAQALS,
where AQALE is loss of quality-adjusted life ex-
pectancy per case, and \ is a weight (i.e., a shadow
price) that is allowed to vary.
D-6
-------
The weights, for each condition, might be inferred from
existing cost-effectiveness studies of treatment or screen-
ing in the corresponding diseases (e.g. , dialysis for
renal disease, amniocentesis for genetic birth defects,
surgery for cancer of the pancreas) or from willingness-
to-pay studies in life-threatening or chronic conditions.
My recommendation for now would be to assign weights based
on rules 1, 2, 3, and 4, since these are easily understood,
although a more theoretically appealing scheme would be
based on rule 6.
3. ASSESSMENT OF WEIGHTS:
THE "IDEAL"
What I shall call the "ideal" methodology (which is
really far from that but, by comparison with "the proposed
"practical" approach, might seem so) takes as input estimates
(or proxies) of disease-age-time-specific incidence rates.
Although the level of detail can be overdone, given the
paucity of objective evidence,"I do believe that scientific
experts can at least bound the range of specific diseases
that might occur and, if pressed, could' quantify their
beliefs in the form of subjective probabilities.
The salient characteristic of this "ideal" method
would be its reliance on simulated cohort analyses for each
disease entity. Either a probability tree model or a Markov
state-transition model would be used to simulate the course
of events following the incidence of disease. There are
numerous examples in the literature of both kinds of models
applied to cost-effectiveness analyses of medical interven-
tions. Consider the data that would be required in order
to use such a model to estimate the main attributes of
outcome:
I. Resource Cost. The model would require estimates
of the probabilities that various medical interventions
would be applied at various stages of the illness, of
the probabilities of consequences of those interven-
tions (including test results and subsequent induced
treatment costs), and of the unit costs of each
intervention (e.g., hospital days, drugs, physician
time, institutional stay). By averaging out the prob-
ability tree, taking care to discount downstream costs
to present value, an expected cost per episode of ill-
ness could be generated.
D-7
-------
II. Life Expectancy. Age-specific survival rates ob-
tained by extrapolating from the medical literature
could be used to project life expectancy under various
treatment options. Data on utilization patterns could
then be used to average out across treatment options
to yield a life expectancy. Comparison with life-table
values yields the net loss in life expectancy. Future
years of life lost may be discounted to present value
(see Weinstein in New Enaland Journal of Medicine,
1977), if desired.
IIA. Economic Cost of Life Expectancy. Survival
curves generated by the life-table method can be
combined with data (Cooper and Rice, 1976) on mean
age-specific earnings, and suitably discounted, to
generate an estimate of the expected, present-value,
loss of productivity.
III. Quality-Adjusted Life Expectancy. A probability
tree or Markov chain approach could be used (and both
have been used) to project expected numbers of years
spent in each of several defined health states.
Weights mav be assigned to these states either by
relying on" available health states indexes (e.g.,
Kaplan, Bush, and Patrick in Health Services Research,
1976), or bv subjectively assigning weights. Quality-
adjusted years of life lost may be discounted to
present value, if desired.
'IIIA. Economic Cost of Morbidity. If desired, only
those health states associated with inability to work
mav be considered, and these may be valued according
to"foregone earnings and discounted to present value.
Th^s method is not easy to implement. Subjectivity
and expert opinion will be required. There are special
problems in assigning values to the teratologic effects,
considering that some health states may be considered worse
than death. (A loss of more than one quality-adjusted year
oer vear spent in that state would be conceivable. ) Problems
of "how to assicr. values to spontaneous or elective abortions,
t-o -:-^°rtilitv" or other reproductive effects, do not fall
^ea*-lv~w
-------
4. ASSESSMENT OF WEIGHTS: A PRACTICAL METHODOLOGY
Finally, consider the following, extremely crude meth-
odology for use in the present study. The only advantage
of this methodology is that it can be implemented in a few
days' time. Moreover, since the toxic effects scores will
be expressed in terms of broad disease categories, so will
the weights. It is not clear that weights derived by this
methodology have any substantial advantage over merely
exploring the range of possible weight vectors, as if they
were chosen at random. On the other hand, at least the
conceptual basis for the methodology is sound, even if the
data base for implementing it is woefully imprecise.
The methodology will be described in the following
sections: 1) the required data base; 2) methods for
estimating the major components of impact (direct cost per
case, life lost per case, quality-of-life lost per case,
and economic equivalents of the latter two); 3) methods for
combining the components into a single weight, wj_, for
each toxic effect.
4.1. Data Base
The data on impacts by broad disease category may be
obtained from Cooper, 3. S., and Rice, D. P., The economic
cost of illness revisited, Social Security Bulletin 39: 21-
36, 1976. Since only the relative weights are ofconcern,
it is of no consequence that these data are from 1972.
In order to apply these broad categories of illness to
the toxic effects of concern, some arbitrary translations
are needed. For example, the following may be defensible:
1. carcinogenicity > neoplasms
2. teratogenicity r» congenital anomalies
3. reproductive toxicity * complications of pregnancy
and childbirth
4. hepatotoxicity * diseases of the digestive
system
5. renal toxicity > diseases of the genito-
urinary system
D-9
-------
6. neurotoxicity * diseases of the nervous
system and sense organs
7. behavioral toxicity =* mental disorders
8. other organ toxicity ^diseases of the respiratory
system
Since we are concerned only with a proxy for the cost per
case, we need not be concerned that these disease categories
do not match cleanly (e.g., reproductive toxicity includes
some "diseases of the genitourinary system"; teratogenicity
includes some "mental disorders," etc.). The implicit
assumption is that the cost per case can be approximated by
the cost per case in the corresponding disease category.
If desired, weights based on two or more disease categories
can be averaged to arrive at a weight for a particular
toxic effect. However, since this is not a major obstacle,
the rest of this discussion assumes that a one-to-one
correspondence between toxic effects and disease categories,
as described above, will be followed.
Having defined the categories of disease for purposes
of accessing the NCHS (National Center for Health Statistics)
data base, consider the data required for the assessment of
weights. All data elements are" subscripted by i, where i
indexes the disease categories.
a. Prevalence and Incidence Data
i. Pj_ = prevalence of disease category i
ii. Ii = incidence (per year) of disease
category i
iii. Ni = mean duration (years) of diseases in
category i
iv. Mj. = case fatality rate (per year) in
category i
v. DI « deaths (per year) in category i
Anv three of the five (i)-(v) are sufficient to
determine the others, by virtue of the following
identities that hold in the steady state (and
which, in this crude procedure, may be assumed to
hold):
D-10
-------
Di » ?£ x Mi
Deaths (Dj_) are available from Cooper and Rice
(1976). Therefore we need two from among (Pj_, Ij_,
Nj_, and MI). If it is possible to get prevalence
and incidence data from NCHS or other sources, then
that is sufficient. If it is possible to get one
but not the other, then an independent (subjective)
estimate of either duration (Nj_) or case-fatality
rate (Mj_) is required. If neither prevalence nor
incidence is available, then both Nj_ and Mj_ will
have to be assessed subjectively. I recommend that
an epidemiologist be consulted for the subjective
estimates, if needed.
b. Direct Cost Data
Ci = direct costs per year (U.S.) for
category i
c. Mortality Data
Yi = years of life lost (U.S.) owing to
deaths in one year in category i
YCi = earnings lost 'U.S.) owing to deaths
in one year in category i
d. Morbidity Data
Zi = person-years of work lost (U.S.) per
year owing to disease category i
ZCi = earnings lost (U.S.) per year owing
to disability in category i
e. Quality-of-Life Data (Subjective)
Qi = loss of quality-adjusted life for each
person-year lost from work
Q.; = loss of quality-adjusted life for each
year of disease not lost from work
D-ll
-------
Data on direct costs, mortality, and morbidity are obtainable
from Cooper and Rice. The quality of life data (Q, Q1 )
reflect subjective judgments and must be assessed indepen-
dently. These reflect the extent to which years lost from
work captures the adverse effects on quality of life (Q1),
and the severity of the major adverse effects (Q).
4.2. Calculation.of Components
of Impact
4.2.1. Direct Cost Weight
The weight for direct cost is calculated as
wCi = NI* . Ci/?i,
where Nj_* is the present value of a unit annuity of Nj_
years. This, has the effect of discounting the stream of
costs to present value, as of the time of incidence. Since
all costs* may be thought of as real dollars (i'.e., no
assumed increase in unit cost over time), a real discount
rate of 4-6 percent seems reasonable.
4.2.2. Mortality Weight
The weight for life years lost is calculated as
This is tantamount to:
wYj_
where the first factor is the number of years lost per
death, and the second factor is the case fatality rate for
the disease category.
Alternatively, life-years may be discounted to present
value, in which case
D-12
-------
wYi
where (Yj_/D^)* is the present value of a unit annuity of
Yi/Di
4.2.3. Cost-of-Mortality
Weight
A weight which values life years lost in terms of
foregone earnings is calculated as
4.2.4. Morbidity Weight (wZjJ
The weight for person-years lost from work is calcu-
lated as
wZi - Ni* (Zi/Pi),
where Nj_* is the present value of an annuity of N^ years.
4.2.5. Cost-of -Morbidity
Weight (wZCj_)
A weight which values disability from work in terms of
foregone earnings is calculated as
wZCi = Ni* (ZCi/?i)
4.2.6. Quality-of-Life-Lost
Weight
A weight that reflects the subjective impact on quality
of life is calculated as
wQi « Ni* [QiZi + Q[(?i - ZiJl/Pi
= (Qi) (wZi) + Q^CNi* - wZi)
D-13
-------
4.3. Pooling the Component
Weights
Alternative procedures for pooling the weight compo-
nents were described in Section 2. To review, these are
as follows:
a. wj_ = wYj_ (life expectancy)
b. wj_ = wQj_ + wYj_ (quality-adjusted life expectancy)
c. wj. = wCj, (direct cost)
d. Wj_ = wCj_ + wYCj_ •(• wSCj_ (economic cost)
e. wj_ = wCj_ + X wY^ (economic cost, assigning value
to life other than human capital)
f. Wj_ = wCj_ T /.(wQj_ + wYj_) (economic cost, assigning
value to morbidity and mortality other than
human capital)
5. RECOMMENDATION
My suggestion would be to use (a), (b), and (d) to
develop three sets of weights to supplement the arbitrary
sets of the form (1, 0, 0, ...), (0, 1, 0, ...), etc., and
(1/n, 1/n, ...).
6. APPENDIX ON DISEASE CATEGORIES
Cooper and Rice, and the NCHS, use ICDA codes to define
the broad disease categories. Other data bases, such as
those of insurance companies, may use similar coding. The
codes are:
neoplasms (ICDA 140-239)
mental disorders (ICDA 290-315)
diseases of the nervous system and sense organs (ICDA
320-389)
diseases of the respiratory system (ICDA 460-519}
diseases of the digestive system (ICDA 520-577)
diseases of the genitourinary system (ICDA^ 530-629)
comolications of pregnancy, etc. (ICDA 630-673)
congenital anomalies (ICDA 740-759)
D-14
-------
IMPLEMENTATION OF METHODOLOGY
This section will discuss the derivation of alternative
weighting systems which rank the eight health effects con-
sidered in the analysis according to economic cost, years
of life lost, and quality of life lost per case. All weights
presented here are defined according to Dr. Weinstein's paper
as follows:
Weight
Direct Cost Weight
Cost of Mortality
Weight
Cost of Morbidity
Weight
Mortality Weight
Morbidity Weight
Quality of Life
Weight
Quality Adjusted
Life Expectancy
Weight
Variable
Name
wC
wYC
wZC
wY
wQALE
Interpretation
The present value of all
medical treatment costs
per case
The present value of fore-
gone earnings due to
mortality per case
The present value of fore-
gone earnings due to
morbidity per case
The number of years of
life lost per case
The number of years lost
from work per case
The subjective impact on
quality of life, measured
in equivalent years of
life lost
The sum of wYj_ and vQj_
Exhibit D-l shows the derivation of the weights and Exhibits
D-2 and D-3 translate these weights into alternative indices.
EPA staff were consulted on the selection of a weight-
ing scheme. Because of resource limitations, ?HB wished to
D-15
-------
Exhibit D-l
DERIVATION OF WEIGHTS FOR BASIC DISEASE CATEGORIES
Coinpl Icatlons Dlucnuoa uf Dlaeauei of Dlaeaauu Dlacaaea of
Congenital of t'i c'jiiiiiicy Digestive f!en Hour I nary of Nervous Respiratory
I
M
a\
C|- direct coata per year for
category I (In Millions)9
Y|- years of life loat owing to
deutha In year (In thotigandol9
YC'i-oai nlngs loat owing to deaths
(In lull lions discounted at 4t)9 12,633
11" pet non-yeara of work loat due to
disability (In thouaando)9
ZC i-eainlinia lost owing to disability
(In Millions)9
11* Incidence par year
l)|- deaths per year
Mj» moan duration (yra)
M,» caaa fatality rate (t)3
Pj- prevalenca of dlaeaao
M| « « 4%
o,10
n(10
Direct Coat Weight
Coat of Mortality Height
Coat of Morbidity Height
Total economic Cont Weight
Mortality Weight
Morbidity Weight
Oua'.lly of Life Height
Duality Adjusted Life Expectancy
Height 6.237
* lixoluileu dental services.
'* r.ncludus eye glanues.
Neo^laama
3,672
5, 101
9 12,63]
to
115
Ity
U62
1,020,61)1)'
352,800
2. 12
u. e
2, 100,000*
2.06
.6
.3
»J,TJO
$12,378 •
$846
$17,022
5.506
.11
.651
Anomalies
381
942
1,284
26
2311
250, ODD1
15,050
1B.08
.3
4,500,000
13.2
.6
.1
$1,1111
55,136
$690
$6,952
3.76U
.10
1. J7
t Oil idli I rid
2,607
IB
DO
48
245
4,161,000S 3,
700
.13*
.1
540,930' 21,
.13
.75
.375
$627
?19
?5U
$704
. OO'J
.01
.053
Uyntom
5,519*
1,402
1,781
299
2,606
320,000s
75,004
6.42
.4
361,000s
5. B
.3
.1
$1,499
$1,139
$708
$1,346
.422
.09
.590
System
4,471
390
736
164
1,249
1,514,000s
27.215
3.8*
.5
5,603.000s
3.6
.3
.1
$2.832
$486
S791
$4,109
.258
.11
. 1H2
Uystem
4,051«*
476
1,060
482
3,944
663,000s 5,
16,644
10. 72
.2
7,111,000s SI,
0.9
.3
.1
$5,070
$1,599
$4,936
$11,605
.710
.73
1.036
Syatem
5,931
1,934
3,434
U40
7,089
528,000s
111,596
9.3*
.2
407,000s
7.9
.2
.1
$911
$621
$1,009
$2,621
.350
.15
.1105
5.138
.062
1.02
.640
1.754
1.155
6
7
8
9
III
Calculated baned on 1981 ratio of deaths to new cusos o« reported by American Cancer .Society
(Ainiirlcnn Cancer Society, 1981 Fact and figures, p. U).
Calculated as Pj T I(.
Calculated aa l>, t P..
National Center for Health Statistics - Number of cancer patients discharged from hospitals In 1972.
National Center for Health Stallatlca - Health Interview Surveys.
Hat limited bailed on convcroat Ion with I'atrlcla Adams of National Center for Health Statistics.
Soui I:IM March of Dimes.
Msl I MI,111:1) b.inod on Information from March of Dimes.
iiuuiri!: Cooper fc It Ice.
Uiiurci.-: Hllli.n C. Heliiiiteln.
-------
Exhibit D-2
ALTERNATIVE INDICES FOR BASIC DISEASE CATEGORIES
Quality
Adjusted
1. Neoplasms
2. Congenital Anomalies
3, Diseases of Nervous
System
4. Diseases of
Genitourinary System
5. Diseases of Digestive
System
6. Diseases of Respiratory
System
7. Complications of
Pregnancy 5 Childbirth .01
••^•^••^•^•B
1.00
Economic
Costs
.37
.15
.25
.09
.07
•v
.06
i .01
Years of
Life Lost
.50
.34
.07
.03
.04
.03
^
Quality of
Life Lost •
.13
.28
.21
.08
.12
.16
.01
Life
Expectancv
.39
.32
."
.04
.06
.07
.
1.00
1.00
1.00
D-17
-------
Exhibit D-3
ALTERNATIVE INDICES FOR EIGHT HEALTH EFFECTS*
Cancer (1)
Mutagenicity (2)
Teratogenicity (2)
Neurological
Disorder (3)
Reproductive
Toxicity
[(2) + (7)] * 2
Kidney Disease {4}
Hepatotoxicity (5)
Other (6)
Economic
Costs
.30
.12
.12
.20
.07
.07
.06 -
.05
Years of
Life Lost
.33
.22
.22
.04
.11
.02
.03
.02
Quality of
Life Lost
.09
.20
.20
.15
.10
.06
.09
.12
Quality
Adjusted
Life
Expectancy
.26
.22
.22
.07
.11
.03
.04
.05
1.00
1.00
1.00
1.00
Numbers in parentheses correspond to basic disease categories i:
Exhibit D-2.
D-18
-------
select one scheme on which most of the analyses were to be
conducted.* The SPA particioants reached agreement on
these points:
• An acceptable scheme should include several or
all of the health effects. This excludes weight-
ing schemes considering only a single health
effect such as cancer.
• Although an equal weight scheme is more attrac-
tive than a single health effect scheme, it
would mask differences in relative importance
among health effects that most people intuitively
feel are significant. An unequal weight scheme
is preferable to an equal weight scheme.
• Among the unequal weight schemes, all had some
attractive characteristics, but the EPA partici-
pants generally agreed that weighting solely on
the basis of ecoraomic. costs of the diseases or
years of life lost missed some important charac-
teristics of the difference among the health ef-
fects. The quality of life lost scheme was more
attractive than the others, although it was in-
herently more subjective. The participants recog-
nized that all the weighting schemes were based
on limited data. Choosing the quality .of life
lost measures would acknowledge the subjective
nature of the exercise. It would minimize the
possibility of a reader assuming (incocrectly)
that the economic cost and years of life lost
weights had more inherent validity than they de-
serve, simply because they are more data-oriented.
• Sensitivity analysis to test the effects of alter-
native weighting schemes on the study's results
was deemed important by the EPA participants.
Thus, PH3 used the quality of life lost weights as the
basis for most analysis and performed sensitivity analysis
with equal weight and an average of the economic cost and
vears of life lost schemes.
* The effect of alternate health weighting schemes was
tested, as reported in Chapter 4.
D-19
-------
The principal sources of data for this analysis were:
• The Economic Cost of Illness Revisited by Bar-
bara S. Cooper and Dorothy P. Rice, Social Secur-
ity Bulletin 39:21-36, 1976. This article pro-
vided aggregated 1972 data on direct costs, years
of life lost, earnings lost, deaths, and person
years of work lost for each of the disease cate-
gories. For purposes of.this analysis, the costs
of dental services and eyeglasses have been ex-
cluded from the derivation of direct cost per
case.
• The National Center for Health Statistics (NCHS)
which provided information on prevalence and inci-
dence for all disease categories except cancer and
congenital anomalies. These data were obtained
from health interview surveys conducted periodi-
cally by NCHS. Where data were gathered for years
other than 1972, the figures have been adjusted
for population growth. NCHS also provided an es-
timate of the number of cancer patients discharged
from hospitals in 1972 which has been used as a
proxy for prevalence of cancer in the absence of
better data.
• Cancer Facts and Figures, 1981 published by the
American Cancer Society. This publication provid-
ed data on estimated incidence of cancer and
cancer deaths in 1981. The ratio of incidence
to deaths has been applied to 1972 deaths (from
Cooper and Rice) to yield an estimate of 1972
incidence.
« The March of Dimes organization, which provided
estimates of incidence and prevalence of congeni-
tal anomalies.
« Milton C. Weinstein provided the quality of life
weights shown in Exhibit D-l.
D-20
------- |