Document qa7z8pJM5pvzdbNxR9G4dNXx5
Mobil Oil Corporation
Mary Paxton American Petroleum Institute 1220 L Street, Northwest Washington, DC 20005
April 8, 1993
ENVIRONMENTAL HEALTH AND SAFETY DEPARTMENT P.O. BOX 1001 PRINCETON, NEW JERSEY 08543
Dear Mary,
As agreed at the last Benzene Task Group meeting, I have reworked the table of risk calculations in which I attempted to recapitulate Kenny's numbers for a variety of models. The enclosed t""<r-page Table gives Kenny's results for a full set of models, with restricted assumptions such as L = 5 in cumulative risk situations, all leukemia and AMMLs.
All of the output is based upon the Paustenbach exposure assumptions. In addition, the Table shows my calculations for a representative sub-set of the output. With no exceptions the Crump calculations and the Craig calculations coincide, within a few percentage points of difference. The Table permits a clear appreciation of the influence of model choice and effect of interest, leukemias or AMML.
ILis clear from the Table that the choice of target tumor does not greatly influence the imputed risk, in the any of the models or any of the exposure scenarios ~xplored here. Choice of exposure metric also has minimal influence, except in the case of its combination with multiplicative models. In that circumstance the attributed risks are greater in comparison to analog.eus model-metric combinations. Perhaps the most striking disparities are between the inside non-linear models and all the other outputs.
There are, I think, a number of questions which deserve exploration. Kenny has briefly looked at the relative merit of models and metrics using maximum likelihood estimates as -criteria. There surely is a great deal more to say about that. To begin that discussion, I enclose four (4) graphs of the hazard functions from selected models. In-Crump's parlance, these are...H(O); (unexposed) and H(l); (exposed). The hazard function is of interest because it illustrates the direct play of the exposure metric and risk model upon the background rate
BP-00020041
- 2-
of the target disease. Two exposure scenarios are represented in the graphs; ene is the typical industrial exposure as in Kenny's paper, and the other is an approximately equivalent exposure (in terms of time times concentration), 10 ppm for five years of life. It is obvious in the graphs, I think, why the combination of cumulative exposure metric and multiplicative risk model yields ~igh estimates of attributed risk.
I hope we can get Kenny to reconsider the decision he made to rework his API report a little bit and submit it for publication as an "update of 1984 work. Thousands of man hours and millions of dollars have been devoted to assembling the most accurate possible cancer incidence and exposure data for this Cohort. There will be no better basis for evaluating benzene risks in exposed workers for a long time to come, if ever. The paper needs to provide sufficient information and context that those who would use the risk calculations will not feel compelled to do what the EPA did in 1985: take the geometric mean of a series of model outputs and multiply by a fudge factor.
In a paper which attempts to portray a modern and effective picture of benzene risks, it is important to explain the exposure metrics, the risk models and the combinations among them
from the standpoint of how they work and how that relates to reality. Does choice of model
influence choice of metric, or vice versa? How well do model/metric characteristics reflect biological phenomena, or perhaps Just plain biological common sense? Alternatively, are there some model/metric characteristics which do not make biological sense? Kenny describes the inside and outside variations on non-linear models as being different ways of describing the data mathematically, yet there are :ym: large differences between the two. We most certainly need a better explanation and understanding of that. There is an almost irresistible temptation for anyone reading the paper to look at the risk numbers at the end and then choose model and metric which yield the numbers that suit his purposes best, or are most politically acceptable. The paper should give enough information about metrics and models so there is a rational alternative to that.
-Perhaps you would be kind enough to-send a copy of this to the Benzene Task Group. I have already sent a copy to Kenny in Louisiana.
Best Rei.'ru-dsI,
/I c~,
P. H. Craig
040793.phc Attachment
ec: C. J. DiPerna K. Crump
BP-00020042
Crump's EStimated Additional Lifetime Risks Due to Benzene*
Model
Linear Models
Weighted Additive Linear
kc
pc
kc/pc
Weighted Multiplicative Linear
kc
pc
kc/pc
Cumulative Additive Linear
kc pc kc/pc
Cumulative Multiplicatiye Linear
kc
pc
kc/pc
Non-Linear Moctels
Weighted Additive Outside
kc
pc
kc/pc
Weighted Multiplicative Outside
kc pc kc/pc
Cumulative Additive Outside
kc
pc
kc/pc
Pqe 1 ot2
1 ppb
Contin x lOE-4
0.072
0.078
O.llOO I
q.130
0.005
~.005
0.98
O.Q36
0.014
0.0,5
0.9~
AMMLs
lppm
Contin ll. IOE-4
71 73 0.97 77
99
120
16
17 0.95
36
36
38 0.94
lppm Occup
x IOE-4
15 .
15 1.00
17
18
26
1.7
1.8 0.97
7.5
7.&.1 0.99
3.4
3.5 0.96
1 ppb Con tin
x IOE-4
0.10
0.10
Leukemias
lppm
Contin 11 IOE-4
99 101 0.98
83
0.10 99
0.15 150
0.04
0.04 0.97
0.03
0.02
0.02 0.99
35
36 0.96
32
36
37 0.97
lppm Occup
X lOE-4
19
20 0.97
17
18 0.97
18
18 1.02
31
33 0.94
7.2
7.6 0.95
5.6
5.8 0.97
3.5
3.6 0.97
ttl '"0
I
0 0 0 N 0
+0--
(.N
Non-Linearj Models (con't)
Cumulative
kc
Multiplicative
pc
Outside
kc/pc
0.044
0.044 1.01
WeightA~d
Additive Inside
kc 0.00000086
pc ke/pc
Weighted Multiplicative Inside
Cumulative Additive lnsilde
kc pc kc/pc
kc
pc kc/pc
0.00000097 0.00000120
Cum~.lative
:~~licative
I
kc 0.0000014p
pc
kC/pc
65
65 1.00
0.86
0.97
1.2
1.4
Pauslen~h exposure estimates, risk Jrt,odels by K. Crump (kc~ For all models with cumulative exposure, Lag i= 5 years For models with weighted exposure, k varies from 0.0~ to 0.175 Recapihl~ted calculations of selected models by P. Craig (pc)
Page lofl
10.0
10.0 1.00
0.18
0.20
0.21
0.30
0.14
0.14 1.01
0.00000120
0.00000100
0.00000120
0.00000170
140 29
139 28 1.00 1.02
1.2 0.23
0.24 0.96
1.0 0.22
0.22 1.00
1.2 0.22
0.22 1.01
1.7 0.36
0.35 1.03
Mobil EHSD 04-06-93
ttl ""0
I
0 0 0 N 0
++0----
Weighted Multiplicative Linear Model
l.OOOE-02-r---:---------i.-----------------------.
l.OOOE-03
l.OOOE-04
!
;; ~ l.OOOE-OS
H 1(i), Exposed
HO(i), Unexposed
,../ .................................~-. --
...,.,.......
........-~,.....
......~....- ___..,._.,...,...,.
~-::;.-"
l.OOOE-06 I
Q
! --.----- - II I
.... ..... gVI 0
rr-----c- -
~I I I I I I I I I I I I I I
.,...In
~
~
0
('<"I
V"'
('<"I
5if?
~
Q V"'
V"' V"'
~
~
0r-
.,
r-
0 00
00
o0 -
~
......
Age
ttl '"0
I
0 0 0 N 0 0
.j::o. (Jl
Weighted Multiplicative Linear Model All Leukemias
l.OOOE-02 , - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ,
l.OOOE-03
Hl (i), Exposed ----- HO(i), Unexposed
k = 0.130
l.OOOE-04
~
~
<a ~ l.OOOE-05
..-;,..- ~-~
___ -----,...___----- _.........,....,..
~---
DExpOsure = 10 ppm, Age 20- 25
l.OOOE-06 -f---,~"""T""-r--r-r---r-T"""""~-,.---.--r---,~"""T""-r---r-r---r-T"""""~-i
s -80 ~
~ ~~~~ ~ ~ ~ ~~ ~~~~ ~~~
Age
ttl ""0
I
0 0 0 N 0
+0--
0\
Cumulativ~ Multiplicative Linear Model
l.OOOE-02-,-----:-----~----------------------.
l.OOOE-03
l.OOOE-04
~
i-;; l.OOOE-05
~---
Hl(i). Expo8el)l HO(i). Unexposed
,..,.....................................------
.. ~_..........
.........~ _..,........-.../
...,. .,. ~
,>'""......-;--
l.OOOE-06 -4---,-..,----,.-~=;===r===;===ir==;===;::::==::r===r==:----r-r---r--.,r---r---r~
;zs g 8-!2~ ~ ~ ~ ~0
V")
0
~
In
....-4
N0
1N1"1
0
("'f')
;Q ~ ~ ~ ~
Age
ttl ""0
I
0 0 0 N 0
+0--
-..]
Cumulative Multiplicative Linear Model All Leukemias
l.OOOE-02 - r - . - - - . - - - - - - - - - - - - - - - - - ' - - - - - - - - - - - - - - ,
l.OOOE-03
l.OOOE-04
~
c;j
~ l.OOOE-05
--- Hl(i). Exposed HO(i). Unexposed
L=5
l.OOOE-06 I
0
II
~0 -
Age
Exposure= 10 ppm, Age 20-25
I .--;
-~ 0
-M
~
M
0
~
~ ~
0
~
~ ~
0
~
~ ~
0
~
~ ~
0
~
~ ~
0
~
~ ~
0
~
~ ~
0 0
ttl '"0
I
0 0 0 N 0
+0--
00