home edit2 whole TSD November 2002 PDF format - OEHHA

calculated by maximizing the likelihood function of the data.
US EPA uses an updated version (GLOBAL86, Howe et al., 1986) of the computer program
GLOBAL79 developed by Crump and Watson (1979) to calculate the point estimate and the 95%
upper confidence limit of the extra risk A(d). Upper 95% upper confidence limits on the extra risk and
lower 95% confidence limits on the dose producing a given risk are determined from a 95% upper
confidence limit q *
1 on a parameter q 1 . When q 1 ≠ 0, at low doses the extra risk A(d) has
approximately the form A(d) = q * 1 * d. This term is a 95% upper confidence limit on the extra risk and
*
R / q 1 is an approximate 95% lower confidence limit on the dose producing an extra risk of R. The
*
*
upper limit of q 1 is calculated by increasing q 1 to a value q 1 such that when the log-likelihood is
remaximized subject to this fixed value q * for the linear coefficient, the resulting maximum value of the
log-likelihood L 1 satisfies the equation 2(L 0 - L 1 ) = 2.70554, where L 0 is the maximum value of the loglikelihood
function and 2.70554 is the cumulative 90% point of the chi-square distribution with one
degree of freedom, corresponding to a 95% upper limit (one-sided). This method of calculating the
upper confidence limit for the extra risk A(d) is a modification of the Crump (1980) model. The upper
*
confidence limit for the extra risk calculated at very low doses is always linear with dose. The slope q 1
is taken as an upper bound of the potency of the chemical in inducing cancer at low doses.
In fitting the dose-response model, the number of terms in the polynomial is chosen equal to k (where 1
< k < 6). The algorithm for selecting the number of stages in the model selects the value of k that
minimizes q 1 * , while providing an adequate model fit (p ≥ 0.01). If the model does not sufficiently fit
the data, data from the highest dose are deleted and the model is refitted to the remaining data. This
process is continued until an acceptable fit to the remaining data is accomplished. For purposes of
determining if a fit is acceptable, the chi-square
x 2 =
h
∑
2
( Xi − NiPi
)
NP i i( − Pi)
i=
1
1
is calculated, where N i is the number of animals in the i th dose group, x i is the number of animals in the
i th dose group with a tumor response, P i is the probability of a response in the i th dose group estimated
by fitting the multistage procedure to the data, and h is the number of remaining groups. The fit is
unacceptable when chi-square (Χ 2 ) is larger than the cumulative 99% point of the chi-square
distribution with f degrees of freedom, where f is the number of dose groups minus the number of nonzero
multistage coefficients.
US EPA separates tumor incidence data according to organ sites or tumor types. The incidence of
benign and malignant tumors is combined whenever scientifically defensible. US EPA considers this
incidence combination scientifically defensible unless the benign tumors are not considered to have the
potential to progress to the associated malignancies of the same histogenic origin. The primary
comparison in carcinogenicity evaluation is tumor response in dosed animals as compared to
contemporary matched control animals. However, US EPA states that historical control data could be
used along with concurrent control data in the evaluation of carcinogenic responses, and notes that for
the evaluation of rare tumors, even small tumor responses may be significant compared to historical
9