Quantitative risk analysis for quantal reproductive and developmental effects.

Animal experiments are generally conducted at higher dose levels than anticipated human dose levels in order to elicit otherwise subtle changes in reproduction or developmental effects with relatively few animals. Based on animal data, regulatory strategy generally has been to postulate a no-observed-effect level (NOEL) for toxic effects and to divide this by a safety factor, usually 100, to establish acceptable levels for humans. Various authors have discussed the shortcomings of using NOEL and have suggested the use of an estimable effect level determined from a dose-response curve fitted to bioassay data, e.g., the dose at which 1% of the animals are adversely affected, and employing some form of conservative low dose extrapolation to control risks at lower doses. In this paper, 10 sets of bioassay data on fetal mortality or anomalies were used to compare the estimated upper limits of risk estimated at the NOEL/100 and the lower 95% confidence limit estimate of the dose producing adverse effects in 1% of the embryonic implants or fetuses divided by 100 (LED01/100). The latter quantity is expected to result in a risk (proportion affected) of less than 10(-4) (1 in 10,000). The estimated upper limits of risk associated with the NOEL/100 were from 2 x 10(-4) to 6 x 10(-4) for the 10 data sets investigated.


Introduction
In order to detect potential toxic effects with a limited number of animals, experimental studies typically employ doses that are higher than expected human exposure levels. Thus, high to low dose extrapolation is generally required. The following discussion of extrapolation of developmental toxicity data does not discount the possibility of a biological threshold below which no risk exists.
One approach for setting acceptable levels for developmental toxicity risk has been the use of safety (uncertainty) factors. From a bioassay conducted at several dose levels (generally three or more), a supposedly safe dose for humans is determined by dividing the no-observed-effect level (NOEL) by a safety factor. Lehman and Fitzhugh suggested a safety factor of 100 (1). If the NOEL is taken to be a safe dose for the experimental animals, a safety factor of 10 is applied to allow for potentially higher sensitivities of humans compared to the experimental animals and another factor of 10 to allow for differences in sensitivities among individuals. For irreversible effects, such as death or malformation, an additional safety factor of 10 is suggested (2). Even if a safety factor of 100 is adequate to account for interspecies and intraspecies differences in response, this does not necessarily result in a riskfree dose. The NOEL may not represent a safe dose for *National Center for Toxicological Research, Jefferson, AR 72079. the laboratory animal tested because the power of the experiment may be inadequate to detect subtle toxic effects. That is, the NOEL may have been inadequately established. Also, the safety factors are somewhat arbitrary and may be adequate on the average, but may be inadequate for any particular case. Further, Gaylor (3) and Crump (4) discuss the problem of poorer experiments resulting in higher NOEL's and hence higher allowable levels.
Rather than routinely applying a fixed size safety factor to the NOEL of the critical toxic effect in an animal species to obtain a safe dose, data from the whole dose-response curve should be used in setting acceptable levels for humans that are estimated to be without any appreciable risk. Based upon the upper confidence limit on the risk estimated from the animal dose-response data, the size of the safety factor can be determined to reduce the risk below a certain level for the animal test species (3).
Several mathematical dose-response models have been used for curve fitting of experimental data.
Crump applied modifications of the one-hit and multistage models to determine a benchmark dose to replace the NOEL (4). A benchmark dose was suggested for which the excess risk was estimated not to exceed 1 to 10%, with 95% confidence. Dourson et al. suggested a similar approach (5). For carcinogenic data, Mantel and Bryan (6), Van Ryzin (7), and Farmer et al. (8), suggested estimating the dose that produced an excess tumor incidence of 1% (EDO1) and then using some form of conservative extrapolation from that point for lower doses. Gaylor suggested using safety factors in conjunction with the EDO1 for controlling risks at low doses for any toxic effect (3).
Regardless of the dose-response model used, it will generally be difficult to estimate with precision an excess risk of less than 1 to 10% above the spontaneous background level for quantal responses, e.g., the proportion of implants that are resorbed or the proportion of live fetuses with a type of malformation, from standard developmental toxicity studies.
Kimmel and Gaylor suggest a procedure based on the ED10 for controlling low dose risks (9). A doseresponse curve is fitted to the experimental data on the proportions affected. This curve is used to estimate the dose that produces a low level of risk in the experimental dose range, e.g., the ED10 (the effective dose corresponding to an excess risk of 10%). Then, a lower confidence limit on the dose that produces a 10% risk (the LED10) is obtained. If F represents a safety factor, at a dose of LED1O/F, the risk in the low dose region is estimated to be less than 0.1/F. This procedure is conservative from a safety standpoint when the dose response is curving upward (convex). If a biological threshold exists, the lower limit on risk would be zero if the LED1o/F is less than the threshold dose. This procedure makes greater use of the dose-response data collected in animal studies, whereas the current procedure based on the NOEL makes use of only a single point. The advantage of applying safety factors to the LED10, or some other estimable dose, is that an upper limit on the risk can be estimated, whereas application of a safety factor to the NOEL results in variable levels of risk since the potential risk at the NOEL is not considered.
Most teratological studies are capable of detecting disease incidence of 10% or more. Hence, the NOEL will generally represent a dose at which the risk is less than 10%. It generally is possible with a slight extrapolation to estimate the ED01. Since the EDO1 is generally in the dose range of the NOEL, it is of interest to compare the LEDo1/100 and the NOEL/100. Generally, the EDO1 will not be directly measurable in a typical teratology study and requires some interpolation or extrapolation from the experimental doses. In general, the choice of the model will not have much impact on the estimation of the ED01. Rather than use the NOEL/100 to set acceptable levels, which can vary considerably depending upon the experimental design, a more consistent approach is to divide the lower confidence limit on the EDO1 of a given effect, i.e., the LEDO1, by 100. If the dose response is curving upward in the low-dose region, the risk at the LEDo1/100 is estimated to be less than 10-4 (1 in 10,000). If the LEDo1/100 is less than a threshold dose, if one exists, then the risk is zero.

Methods
Ten sets of dose response data for the proportions of fetuses with reproductive or developmental defects were investigated to compare the NOEL/100 and LEDo1/100 procedures (Table 1). In general, it would be preferable to use data on a litter basis. Since the data were not reported for individual litters, dose-response curves on the proportions of dead or abnormal fetuses per litter were not obtained and quantal data on a fetal basis were used. An exponential polynomial model was fitted to each of the data sets: where P was the proportion of affected fetuses, d was the dose rate, and each b was estimated from the data using the procedure of Howe and Crump (10). These estimated sigmoidal dose-response curves were only used to provide an estimate of the LEDO1. Since the estimated dose response is not used for low dose extrapolation, a biologically based model is not particularly important, and all that is required is a descriptive model in the experimental dose range.
Conservative upper limits on low dose risk estimates are obtained for convex (upward curvature) doseresponse curves by using a risk to dose slope of 0.01/LED01, below the LEDO1, i.e., Estimated risk < . .01 x dose.

LED01
Thus, the risk at the LEDo1/100 is estimated to be less than 10-4 (1 in 10,000). The risk at the NOEL/100 is estimated to be less than 10-4 x NOEL/LEDO0. There is a difference of opinion as to the NOEL from those given by Crump (4) for ethylenethiourea and 2,3,7,8tetrachlorodibenzo-p-dioxin (intestinal anomalies), indicating one of the difficulties of basing procedures on the NOEL.

Results and Discussion
Comparison of the NOEL and the LEDO1 for the 10 data sets is shown in Table 2. The NOEL was a factor of 2 to 6 times the LEDO1, resulting in estimated upper limits of risks at the NOEL/100 to be 2 to 6 times higher than the estimated upper risk of 10-4 at the LEDo1/ 100. Due to the lack of detail of the published data, the issue of litter effects could not be addressed. Ignoring the differences among litters could result in a reduced estimate of the variance. Hence, the LEDO1 values reported are larger than would be obtained when the variation of litters is properly included. Thus, the NOELs for these 10 data sets are actually more than 2 to 6 times greater than the LEDO1 values that would be obtained if interlitter variation were included. The approach presented here for extrapolating risk to low dose levels based upon the LEDO1 takes into account the dose-response data. Therefore, it provides a stronger basis on which to estimate risk for low dose levels. More research is needed for developing appropriate dose-DEVELOPMENTAL QUANTITATIVE RISK ANALYSIS  (14) Ametantrone (15) Linamarin (16) Solanidine (17) Paraxanthine ( 10-4. response models for these types of data. Improved statistical curve-fitting procedures of experimental data can provide better estimates of the LEDO1. For the reproductive and developmental effects examined here, the use of a 100-fold safety factor applied to the NOEL resulted in estimated risks of less than 6 x 10-4. The risk of reproductive or developmental effects is estimated to be less than 10-4 at the LEDo1/100 in the animal test systems. Thus, a lower level of risk can generally be obtained by using the LEDo,/100 rather than the NOEL/100 to set allowable dose levels. If lower levels of potential risk are desired, then larger safety factors must be used.