v>EPA
United States
Environmental Protection Agency
Office of Chemical Safety and
Pollution Prevention
Final Risk Evaluation for
n-Methylpyrrolidone
Benchmark Dose Modeling Supplemental File
CASRN: 872-50-4
December 2020
-------
Table of Contents
1 INTRODUCTION 10
2 BENCHMARK DOSE MODELING OF EFFECTS FOR POST-IMPLANTATION LOSSES
AND RESORPTIONS 11
2.1 Resorptions: Results for Saillenfait et al. (2002) using Cmax 20
2.2 Resorptions: Results for Saillenfait et al. (2002) using AUC 25
2.3 Resorptions: Results for Saillenfait et al. (2003) using Cmax 30
2.4 Resorptions: Results for Saillenfait et al. (2003) using AUC 35
2.5 Post-implantation Losses: Results for Saillenfait et al. (2002) using Cmax 40
2.6 Post-implantation Losses: Results for Saillenfait et al. (2002) using AUC 43
2.7 Post-implantation Losses: Results for Saillenfait et al. (2003) using Cmax 46
2.8 Post-implantation Losses: Results for Saillenfait et al. (2003) using AUC 49
2.9 Post-implantation Losses: Results for Saillenfait et al. (2003; 2002) combined using Cmax 52
2.10 Post-implantation Losses: Results for Saillenfait et al. (2003; 2002) combined using AUC 55
3 BENCHMARK DOSE MODELING OF FETAL AND PUP BODY WEIGHT CHANGES ...58
3.1 Results for Saillenfait et al. (2003) using AUC 62
3.2 Results for Saillenfait et al. (2002) using AUC 65
3.3 Results for DuPont, 1990 using AUC 69
4 BENCHMARK DOSE MODELING OF MALE FERTILITY, FEMALE FECUNDITY,
LITTER SIZE AND PUP DEATH IN EXXON, 1991 72
4.1 Summary of BMD Modeling for Exxon, 1991 Data 76
4.2 Results of BMD Modeling of P2 Male and Female Fertility Indices (Exxon, 1991) 77
4.2.1 P2/F2A Male Fertility (Males Unsuccessful/Males Used; Exxon Appendix AF) 79
4.2.2 P2/F2B Male Fertility (Males Unsuccessful/Males Used; Exxon Appendix AG) 82
4.2.3 P2/F2A Female Fecundity (Females Unsuccessful/Females Mated; Exxon Appendix AF). 85
4.2.4 P2/F2B Female Fecundity (Females Unsuccessful/Females Mated; Exxon Appendix AG) 88
4.3 Results of BMD Modeling of P2 Litter (Exxon (1991a)) 92
4.3.1 P2/F2A Litter Size - 50 g Rat (Exxon Appendix AJ, "Total Pups Born") 94
4.3.2 P2/F2B Litter Size - 50 g Rat (Exxon Appendix AK, "Total Pups Born") 98
4.3.3 P2/F2A Litter Size - GD 6-21 Rat (Exxon Appendix AJ, "Total Pups Born") 102
4.3.4 P2/F2B Litter Size - GD 6-21 Rat (Exxon Appendix AK, "Total Pups Born") 106
4.4 Results of BMD Modeling of P2 Pup Death (Exxon (1991a)) 110
4.4.1 P2/F2A Pups Dead at Day 0 (Stillborn Day 0/Total Pups Born; Exxon 1991 Appendix AJ)
Ill
4.4.2 P2/F2B Pups Dead at Day 0 (Stillborn Day 0/Total Pups Born; Exxon 1991 Appendix AK)
117
4.4.3 P2/F2A Pups Dead by Day 4 (Dead by Day 4/Total Pups Born; Exxon Appendix AJ) 118
4.4.4 P2/F2B Pups Dead by Day 4 (Dead by Day 4/Total Pups Born; Exxon Appendix AK).... 119
5 BENCHMARK DOSE MODELING OF FETAL AND PUP BODY WEIGHT, PUP DEATH,
STILLBIRTHS, AND ABSOLUTE TESTES WEIGHT IN NMP PRODUCERS GROUP 1999A,B
120
5.1 Overall BMD Modeling Approach for NMP Producers Group 1999a,b Data 122
5.2 PBPK Analysis for NMP Producers Group (1999a, b) 126
Page 2 of244
-------
5.3 Comparison of PODs for Critical Effects and for Effects Reported in the NMP Producers
Group Studies 128
5.4 Results for Benchmark Dose Modeling of Absolute Testes Weight in PO Male Wistar Rats
(NMP Producers Group (1999b)) 133
5.5 Results for BMD Modeling for Reduced Fetal and Pup Body Weight for Sprague-Dawley Rats
(NMP Producers Group (1999a)) 136
5.5.1 Sprague-Dawley Rat F2B Fetal Body Weight at PND1 (Females) 136
5.5.2 Sprague-Dawley Rat F2B Fetal Body Weight at PND1 (Males) 137
5.5.3 Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Females) 138
5.5.4 Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Males) 144
5.5.5 Sprague-Dawley Rat F2B Pup Body Weight at PND21 (Females) 145
5.5.6 Sprague-Dawley Rat F2B Pup Body Weight at PND21 (Males) 148
5.6 Results for BMD Modeling for Reduced Fetal and Pup Body Weight for Wistar Rats (NMP
Producers Group (1999b)) 149
5.6.1 Wistar Rat F1A Fetal Body Weight at PND1 (Females) 149
5.6.2 Wistar Rat F1A Fetal Body Weight at PND1 (Males) 152
5.6.3 Wistar Rat F1A Pup Body Weight at PND7 (Females) 160
5.6.4 Wistar Rat F1A Pup Body Weight at PND7 (Males) 163
5.6.5 Wistar Rat F1A Pup Body Weight at PND21 (Females) 164
5.6.6 Wistar Rat F1A Pup Body Weight at PND21 (Males) 167
5.7 Results for BMD Modeling for Stillbirths, and PND4 and PND21 Pup Deaths in Sprague-
Dawley Rats (NMP Producers Group (1999a)) 173
5.7.1 Sprague-Dawley Rat F1A stillborn/total delivered (NMP Producers Group (1999a)) 173
5.7.2 Sprague-Dawley Rat F2B Pup death at PND4/total delivered (NMP Producers Group
(1999a)) 178
5.7.3 Sprague-Dawley Rat F2B Pup death at PND21/PND4 post-cull (NMP Producers Group
(1999a)) 182
5.8 Results for BMD Modeling for Stillbirths, and PND4 and PND21 Pup Deaths in Wistar Rats
(NMP Producers Group (1999b)) 190
5.8.1 Wistar Rat F1A stillborn/total delivered (NMP Producers Group (1999b)) 190
5.8.2 Wistar Rat F1A Pup death at PND4/total delivered (NMP Producers Group (1999b)) 204
5.8.3 Wistar Rat FIB stillborn/total delivered (NMP Producers Group (1999b)) 212
5.8.4 Wistar Rat F2B Pup death at PND4/total delivered (NMP Producers Group (1999b)) 216
5.8.5 Wistar Rat F2B Pup death at PND21/PND4 post-cull (NMP Producers Group (1999b)).. 220
6 REFERENCES 228
APPENDICES 230
Appendix A Analysis of Continuous Response Summary Data Subject to Litter Effects 230
Appendix B Tests for Differences and Trends in Saillenfait et al. (2003; 2002) Post-Implantation
Dose-Response Data 232
Page 3 of244
-------
List of Figures
Figure 2.1-1 Plot of Response by Dose, with Fitted Curve for Selected Hill Model for Resorptions
(Mean % per Litter) in Rats Exposed to NMP via Gavage (Saillenfait et al. (2003; 2002))
23
Figure 2.2-1 Plot of Response by Dose, with Fitted Curve for Selected Polynomial Degree 4 Model for
Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage (Saillenfait et al.
(2003; 2002)) 28
Figure 2.3-1 Plot of Response by Dose, with Fitted Curve for Selected Linear Model for Resorptions
(Mean % per Litter) in Rats Exposed to NMP via inhalation (Saillenfait et al. (2003)) .. 33
Figure 2.4-1 Plot of Response by Dose, with Fitted Curve for Polynomial Degree 3 Model for
Resorptions (Mean % per Litter) in Rats Exposed to NMP via inhalation (Saillenfait et al.
(2003)) 38
Figure 2.5-1 Plot of Response by Dose, with Fitted Curve for Selected Log-Probit Model for Post-
implantation Loss in Rats Exposed to NMP via Gavage (Saillenfait et al. (2003; 2002))41
Figure 2.6-1 Plot of Response by Dose, with Fitted Curve for Selected Log-Probit Model for Post-
implantation Loss in Rats Exposed to NMP via Gavage (Saillenfait et al. (2003; 2002))44
Figure 2.7-1 Post-Implantation Loss (Incidence) vs. Cmax (Saillenfait et al. (2003)) - Log-Logistic Model
with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the
BMDL 47
Figure 2.8-1 Post-Implantation Loss (Incidence) vs. AUC (Saillenfait et al. (2003)) - Log-Logistic
Model with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for
the BMDL 50
Figure 2.9-1 Plot of Response by Dose, with Fitted Curve for Selected Log-Probit Model for Post-
implantation Loss in Rats Exposed to NMP via Gavage or Inhalation (Saillenfait et al.
(2003; 2002)) 53
Figure 2.10-1 Plot of Response by Dose, with Fitted Curve for Selected Log-Probit Model for Post-
implantation Loss in Rats Exposed to NMP via Gavage or Inhalation (Saillenfait et al.
2003; 2002)) 56
Figure 3.1-1 Plot of Mean Response by Dose, with Fitted Curve for Selected Exponential 3 Model for
Fetal Body Weight in Rats Exposed to NMP via Inhalation (Saillenfait et al. (2003)).... 62
Figure 3.2-1 Plot of Mean Response by Dose, with Fitted Curve for Selected Exponential 3 Model for
Fetal Body Weight in Rats Exposed to NMP via Gavage (Saillenfait et al. (2002)) 66
Figure 3.3-1 Plot of Mean Response by Dose, with Fitted Curve for Selected Exponential 3 Model for
Fetal Body Weight in Rats Exposed to NMP via Inhalation (DuPont (1990)) 70
Figure 5.4-1 Plot of Mean Response by Dose, with Fitted Curve for Frequentist Exponential 4 Model for
Absolute Testes Weight in Male Wistar Rats Exposed to NMP via Oral Gavage (NMP
Producers Group (1999b)) 133
Figure 5.5-1 Plot of Mean Response by Dose, with Fitted Curve for Linear Model with Constant
Variance for Sprague-Dawley Rat F2B Fetal Body Weight at PND1 (Females) 136
Figure 5.5-2 Plot of Mean Response by Dose, with Fitted Curve for Linear Model with Constant
Variance for Sprague-Dawley Rat F2B Fetal Body Weight at PND1 (Males) 137
Figure 5.5-3 Plot of Mean Response by Dose, with Fitted Curve for Exponential 3 Model with Constant
Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Females) 138
Figure 5.5-4 Plot of Mean Response by Dose, with Fitted Curve for Exponential 3 Model with Constant
Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Females) 141
Figure 5.5-5 Plot of Mean Response by Dose, with Fitted Curve for Exponential 3 Model with Constant
Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Females) 142
Page 4 of 244
-------
Figure 5.5-6 Plot of Mean Response by Dose, with Fitted Curve for Linear Model for Sprague-Dawley
Rat F2B Pup Body Weight at PND7 (Males) 144
Figure 5.5-7 Plot of Mean Response by Dose, with Fitted Curve for Exponential 4 Model with Constant
Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND21 (Females) 145
Figure 5.5-8 Plot of Mean Response by Dose, with Fitted Curve for Linear Model with Constant
Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND21 (Males) 148
Figure 5.6-1 Plot of Mean Response by Dose, with Fitted Curve for Exponential 3 Model with Constant
Variance for Wistar Rat F1A Fetal Body Weight at PND1 (Females) 149
Figure 5.6-2 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with Constant
Variance for Wistar Rat F1A Fetal Body Weight at PND1 (Males) 152
Figure 5.6-3 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with Constant
Variance for Wistar Rat F1A Fetal Body Weight at PND1 (Males) 155
Figure 5.6-4 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with Constant
Variance for Wistar Rat F1A Fetal Body Weight at PND1 (Males) 157
Figure 5.6-5 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with Non-
constant Variance for Wistar Rat F1A Pup Body Weight at PND7 (Females) 160
Figure 5.6-6 Plot of Mean Response by Dose, with Fitted Curve for Lines Model with Constant
Variance for Wistar Rat F1A Pup Body Weight at PND7 (Females) 163
Figure 5.6-7 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with Constant
Variance for Wistar Rat F1A Pup Body Weight at PND21 (Females) 164
Figure 5.6-8 Plot of Mean Response by Dose, with Fitted Curve for Polynomial Degree 3 Model with
Constant Variance for Wistar Rat F1A Pup Body Weight at PND21 (Males) 167
Figure 5.6-9 Plot of Mean Response by Dose, with Fitted Curve for Polynomial Degree 3 Model with
Constant Variance for Wistar Rat F1A Pup Body Weight at PND21 (Males) 170
Figure 5.6-10 Plot of Mean Response by Dose, with Fitted Curve for Polynomial Degree 3 Model with
Constant Variance for Wistar Rat F1A Pup Body Weight at PND21 (Males) 171
Figure 5.7-1 Plot of NLogistic (no LSC; ICC estimated) model for AUC (hr mg/L) versus Sprague-
Dawley Rat F1A stillborn/total delivered 176
Figure 5.7-2 Plot of NLogistic (no LSC; ICC estimated) model for Cmax(mg/L) versus Sprague-Dawley
RatFIA stillborn/total delivered 177
Figure 5.7-3 Plot of NLogistic (no LSC; ICC estimated) model for AUC (hr mg/L) versus Sprague-
Dawley Rat F2B Pup Death at PND4/Total Delivered 181
Figure 5.7-4 Plot of NLogistic model (LSC = LD1 dam weight; ICC estimated) for AUC (hr mg/L)
versus Sprague-Dawley Rat F2B Pup Death at PND21/PND4 Live Post-cull 185
Figure 5.8-1 Plot of NLogistic (LSC = LD1 dam weight; ICC estimated) model for AUC (hr mg/L)
versus Wistar Rat F1A Stillborn/Total Delivered 193
Figure 5.8-2 Plot of NLogistic (LSC = LD1 dam weight; ICC estimated) model for Cmax (mg/L) versus
Wistar RatFIA Stillborn/Total Delivered 198
Figure 5.8-3 Plot of NCTR model (LSC = LD1 dam weight; ICC estimated) for AUC (hr mg/L) versus
Wistar Rat F1A Pup Death at PND4/Total Delivered 207
Figure 5.8-4 Plot of NCTR model (LSC = LD1 dam weight; ICC estimated) for AUC (hr mg/L) versus
Wistar Rat FIB stillborn/total delivered 214
Figure 5.8-5 Plot of NCTR model (LSC = LD1 dam weight; ICC estimated) for Cmax (mg/L) versus
Wistar Rat FIB stillborn/total delivered 215
Figure 5.8-6 Plot of NCTR model (LSC = LD1 dam body weight; ICC estimated) for AUC (hr mg/L)
versus Wistar Rat F2B Pup Death at PND4/Total Delivered 219
Figure 5.8-7 Plot of NLogistic model (no LSC; ICC estimated) for AUC (hr mg/L) versus Wistar Rat
F2B Pup Death at PND21/Live PND4 Post-cull 223
Page 5 of244
-------
List of Tables
Table 2-1 Resorptions (Mean % per litter) Data selected for Dose-Response Modeling for NMP 12
Table 2-2 Post-implantation Loss Data Selected for Dose-Response Modeling for NMP 13
Table 2-3 BMD and BMDL Derivations from the Variance (SD) Sensitivity Analysis of Saillenfait et al.
(2003; 2002) Resorption Data, with Corresponding NOAELs 17
Table 2-4. Summary of PODs identified for Cmax and AUC Dose Metrics for Post-Implantation Loss and
Resorptions 18
Table 2-5 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using Cmax as the Dose Metric (Saillenfait et al. (2003; 2002)) 20
Table 2-6 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using Cmax as the Dose Metric (Saillenfait et al. (2003; 2002)) 21
Table 2-7 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using Cmax as the Dose Metric (Saillenfait et al. (2003; 2002)) 22
Table 2-8 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using AUC as the Dose Metric (Saillenfait et al. (2003; 2002)) 25
Table 2-9 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using AUC as the Dose Metric (Saillenfait et al. (2003; 2002)) 26
Table 2-10 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using AUC as the Dose Metric (Saillenfait et al. (2003; 2002)) 27
Table 2-11 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using Cmax as the Dose Metric (Saillenfait et al. (2003)) 30
Table 2-12 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using Cmax as the Dose Metric (Saillenfait et al. (2003)) 31
Table 2-13 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using Cmax as the Dose Metric (Saillenfait et al. (2003)) 32
Table 2-14 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using AUC as the Dose Metric (Saillenfait et al. (2003)) 35
Table 2-15 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using AUC as the Dose Metric (Saillenfait et al. (2003)) 36
Table 2-16 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage
Using AUC as the Dose Metric (Saillenfait et al. (2003)) 37
Table 2-17 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in Rats
Exposed to NMP via Gavage Using Cmax as the Dose Metric (Saillenfait et al. (2003;
2002)) 40
Table 2-18 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in Rats
Exposed to NMP via Gavage Using AUC as the Dose Metric (Saillenfait et al. (2003;
2002)) 43
Table 2-19 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in Rats
Exposed to NMP via Inhalation Using Cmax as the Dose Metric (Saillenfait et al. (2003))
46
Table 2-20 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in Rats
Exposed to NMP via Inhalation Using AUC as the Dose Metric (Saillenfait et al. (2003))
49
Table 2-21 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in Rats
Exposed to NMP via Gavage or Inhalation Using Cmax as the Dose Metric (Saillenfait et
al. (2003; 2002)) 52
Page 6 of 244
-------
Table 2-22 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in Rats
Exposed to NMP via Gavage or Inhalation Using AUC as the Dose Metric (Saillenfait et
al. (2003; 2002)) 55
Table 3-1 Fetal Body Weight Data Selected for Dose-Response Modeling for NMP 60
Table 3-2. Summary of Recommended BMD and BMDL Values for Fetal Weight 61
Table 3-3. Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Inhalation Using
Daily Average AUC as the Dose Metric (Saillenfait et al. (2003)) 62
Table 3-4 BMD and BMDL Estimates from the Sensitivity Analysis of Fetal Body Weights (Saillenfait
et al. (2002)) 65
Table 3-5 Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Gavage Using Daily
Average AUC as the Dose Metric (Saillenfait et al. (2002)); Observed SD case 66
Table 3-6 Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Gavage Using Daily
Average AUC as the Dose Metric (Saillenfait et al. (2002)); Minimume SD Case 68
Table 3-7 Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Gavage Using Daily
Average AUC as the Dose Metric (Saillenfait et al. (2002)); Maximum SD Case 68
Table 3-8 Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Inhalation using Daily
Average AUC as the Dose Metric (DuPont (1990)) 69
Table 4-1 PBPK-predicted average blood concentrations (Cavg, mg/L) in juvenile rats 74
Table 4-2 PBPK-predicted average blood concentrations (Cavg, mg/L) during gestation for P2/F2A... 74
Table 4-3 PBPK-predicted average blood concentrations (Cavg, mg/L) during gestation for P2/F2B ... 74
Table 4-4 BMD Modeling Summary for Exxon (1991b) 76
Table 4-5 Model Predictions for Reduced Male Fertility in P2/F2A Male Rats (Exxon (1991b)) 79
Table 4-6 Model Predictions for Reduced Male Fertility in P2/F2B Male Rats (Exxon (1991b)) 82
Table 4-7 Model Predictions for Reduced Fecundity in P2/F2A Female Rats (Exxon (1991b)) 85
Table 4-8 Model Predictions for Reduced Fecundity in P2/F2B Female Rats (Exxon (1991b)) 88
Table 4-9 Model Predictions for Litter Size in P2/F2A Rats Based on Post-weaning Exposure (Exxon
(1991b)) 94
Table 4-10 Model Predictions for Litter Size in P2/F2B Rats Based on Post-weaning Exposure (Exxon
(1991b)) 98
Table 4-11 Model Predictions for Litter Size in P2/F2A Rats Based on Gestational Exposure (Exxon
(1991b)) 102
Table 4-12 Model Predictions for Litter Size in P2/F2B Rats Based on Gestational Exposure (Exxon
(1991b)) 106
Table 4-13 Model Predictions for Pup Death at Day 0 in P2/F2A Rats (Exxon (1991b)) Ill
Table 4-14 Model Predictions for Pup Death at Day 0 in P2/F2B Rats (Exxon (1991b)) 117
Table 4-15 Model Predictions for Pup Death at Day 4 in P2/F2A Rats (Exxon (1991b)) 118
Table 4-16 Model Predictions for Pup Death at Day 4 in P2/F2B Rats (Exxon (1991b)) 119
Table 5-1 Description of Endpoints from NMP Producers Group Studies (1999a, b) that were used for
BMD Modeling 121
Table 5-2 BMDsPct and BMDLsivt derivations from the variance (SD) sensitivity analysis of body and
organ weight data, with corresponding NOAELs 124
Table 5-3 Acute PODs: Comparison of PODs for critical effects and for effects reported in the NMP
Producers Group Studies (1999a, b) 129
Table 5-4 Chronic PODs: Comparison of PODs for critical effects and for effects reported in the NMP
Producers Group Studies (1999a, b) 130
Table 5-5 Model Predictions for AUC (hr mg/L) versus Wistar Rat Absolute Testes Weight (P0 Adult
Males) (NMP Producers Group (1999b)) 133
Page 7 of244
-------
Table 5-6 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Fetal Body Weight at
PND1 (Females) Using Daily Average AUC as the Dose Metric 136
Table 5-7 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Fetal Body Weight at
PND1 (Males) Using Daily Average AUC as the Dose Metric 137
Table 5-8 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body Weight at
PND7 (Females) Using Daily Average AUC as the Dose Metric 138
Table 5-9 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body Weight at
PND7 (Females) Using Daily Average AUC as the Dose Metric 140
Table 5-10 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body Weight at
PND7 (Females) Using Daily Average AUC as the Dose Metric 141
Table 5-11 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body Weight at
PND7 (Males) Using Daily Average AUC as the Dose Metric 144
Table 5-12 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body Weight at
PND21 (Females) Using Daily Average AUC as the Dose Metric 145
Table 5-13 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body Weight at
PND21 (Males) Using Daily Average AUC as the Dose Metric 148
Table 5-14 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Fetal Body Weight at PND1
(Females) Using Daily Average AUC as the Dose Metric 149
Table 5-15 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Fetal Body Weight at PND1
(Males) Using Daily Average AUC as the Dose Metric 152
Table 5-16 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Fetal Body Weight at PND1
(Males) Using Daily Average AUC as the Dose Metric 154
Table 5-17 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Fetal Body Weight at PND1
(Males) Using Daily Average AUC as the Dose Metric. All SDs set to Maximum SD
Across the Group 157
Table 5-18 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at PND7
(Females) Using Daily Average AUC as the Dose Metric 160
Table 5-19 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at PND7
(Males) Using Daily Average AUC as the Dose Metric 163
Table 5-20 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at PND21
(Females) Using Daily Average AUC as the Dose Metric 164
Table 5-21 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at PND21
(Males) Using Daily Average AUC as the Dose Metric 167
Table 5-22 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at PND21
(Males) Using Daily Average AUC as the Dose Metric 169
Table 5-23 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at PND21
(Males) Using Daily Average AUC as the Dose Metric 170
Table 5-24 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Sprague-Dawley Rat
F1A stillborn/total delivered (NMP Producers Group (1999a)); BMR = 1% extra risk. 175
Table 5-25 Summary of BMDS nested modeling results for Cmax (mg/L) versus Sprague-Dawley Rat
F1A stillborn/total delivered (NMP Producers Group (1999a)); BMR = 1% extra risk. 176
Table 5-26 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Sprague-Dawley Rat
F2B Pup death at PND4 /total delivered (NMP Producers Group (1999a)); BMR = 1%
extra risk 180
Table 5-27 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Sprague-Dawley Rat
F2B Pup death at PND21/PND4 post-cull (NMP Producers Group (1999a)) 184
Table 5-28 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat F1A
stillborn/total delivered (NMP Producers Group (1999b)); BMR= 1% extra risk 192
Page 8 of244
-------
Table 5-29 Summary of BMDS nesting modeling results for Cmax (mg/L) versus Wistar Rat F1A
stillborn/total delivered (NMP Producers Group (1999b)); BMR = 1% extra risk 198
Table 5-30 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat F1A Pup
death at PND4/total delivered (NMP Producers Group (1999b)); BMR = 1% extra risk.
206
Table 5-31 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat FIB
stillborn/total delivered (NMP Producers Group (1999b)); BMR = 1% extra risk 214
Table 5-32 Summary of BMDS nested modeling results for Cmax (mg/L) versus Wistar Rat FIB
stillborn/total delivered (NMP Producers Group (1999b)); BMR = 1% extra risk 215
Table 5-33 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat F2B Pup
death at PND4/total delivered (NMP Producers Group (1999b)); BMR = 1% extra risk.
218
Table 5-34 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat F2B Pup
death at PND21 /PND4 post-cull (NMP Producers Group (1999b)); BMR = 1% extra
risk 222
Page 9 of 244
-------
1 INTRODUCTION
This supplemental file describes benchmark dose (BMD) modeling approaches and results for all critical
endpoints considered in the derivation of points of departure (PODs) for NMP. Reduced male fertility,
reduced female fecundity, and reduced fetal body weights were all identified as sensitive reproductive
and developmental endpoints associated with repeated dose exposures and were evaluated as the
potential basis for chronic PODs. Post-implantation loss (resorptions and fetal mortality) and resorptions
were identified as sensitive developmental endpoints that are relevant for single dose exposures and
were evaluated as the potential basis for acute PODs.
In addition to the critical endpoints identified in the NMP risk evaluation, EPA performed dose-response
analysis on several additional reproductive and developmental endpoints, including absolute testes
weight, pup body weights, pup mortality, and stillbirth. These additional endpoints provide supporting
evidence for POD selection, but contain uncertainties (e.g., around exposure levels, or relevant exposure
durations) that make them less suitable as the quantitative basis for PODs. For example, the relevance of
stillbirths and pup mortality for acute versus chronic exposures is unclear. Stillbirths and pup mortality
have been reported following repeated exposures throughout gestation, but could conceivably result
from single exposures.
BMD modeling for post-implantation loss (resorptions and fetal mortality) and resorptions (Sections
2.1-2.10), fetal and pup body weight changes (Sections 3.1-3.3 and 5.5-5.6), male fertility and female
fecundity (Sections 4.2-4.3), and absolute testes weight (Section 5.4) was performed using USEPA's
BMD Software package version 3.1.1 (BMDS 3.1.1, released 07/31/2019), 3.1.2 (BMDS 3.1.2, released
11/ 8/2019) or 3.2 (BMDS 3.2, released 08/20/2020). Choice of BMD software was dictated by software
availability at the time of BMD modeling for each endpoint. As each BMDS release provides updates,
fixes, and enhancements to BMDS version 3, EPA chose to use the most up-to-date BMDS version
available when conducting BMD modeling.1 BMD modeling for stillbirths and pup death (Sections 4.4
and 5.7-5.8) was performed using USEPA's BMD Software package version 2.7 (BMDS 2.7, released
08/18/2017). The pup death and stillbirth endpoints were analyzed using BMDS 2.7 because it contains
a larger suite of nested dichotomous models compared to BMDS version 3, and nested dichotomous
models are preferred for these endpoints because they contain an intra-litter correlation coefficient for
the assessment of litter-specific responses. All BMD modeling was conducted in a manner consistent
with BMD technical guidance (U.S. EPA (2012)).
A peer-reviewed rat PBPK model for NMP (Poet et al. (2010)) modified by EPA (as described in
Appendix I of the final NMP risk evaluation) was used to describe dose-response data for each endpoint
in terms of internal doses (blood concentrations) in exposed rats. PODs based on internal doses in rats
can be compared to blood concentrations in people predicted by human PBPK models for each condition
of use. Internal dose metrics calculated with the rat PBPK model are in units of either AUC (hr mg/L)
for chronic exposures or peak blood concentration (Cmax, mg/L) for acute exposures.
1 For a complete history of BMDS Version 3 software updates see: https://www.epa.gov/bmds/benchmark-dose-software-
bmds-version-3 -release-history
Page 10 of 244
-------
2 Benchmark Dose Modeling of Effects for Post-implantation Losses
and Resorptions
The Saillenfait et al. (2003; 2002). Becci et al. (1982) and Sitarek et al. (2012) studies were selected for
dose-response analysis of resorptions and post-implantation loss (resorptions and fetal mortality). Data
available from the Sitarek et al. (2012) study did not allow for the analysis of post-implantation loss, as
only fetal mortality data was reported. Fetal mortality is considered a less sensitive endpoint than the
combined endpoint of post-implantation loss, which incorporates resorptions and fetal mortality. In the
Sitarek et al. (2012) study, the mean percent dead fetuses across litters was significantly increased only
in the highest dose group. Furthermore, the number of live pups in the highest exposure group was also
significantly lower, and there were dam deaths and total litter loss in the highest exposure group.
Benchmark dose (BMD) analysis of fetal mortality as a continuous response was not conducted for this
data set, as study data were not consistent with this approach (e.g., the mean and standard deviation was
zero for some dose groups) (see Table 2-1). Thus, a NOAEL of 265 mg/L (based on Cmax) was chosen as
a POD for the Sitarek et al. (2012) study. Similarly, the dose-response data for resorptions in the Becci
et al. (1982) dermal study was not amenable to BMD modeling, and a NOAEL of 662 mg/L (based on
Cmax) was chosen as a POD.
BMD modeling of resorptions and post-implantation loss (resorptions and fetal mortality) endpoints was
performed for the Saillenfait et al. oral (2002) and inhalation (2003) studies using USEPA's BMD
Software package version 3.1.2 (BMDS 3.1.2), in a manner consistent with BMD technical guidance
(U.S. EPA (2012)). Dichotomous models were used to fit post-implantation loss incidence data and
continuous models were used to fit dose-response data for mean number of resorptions. A BMR of 1%
relative deviation (post-implantation loss) or 1% absolute deviation (resorptions) was used to address the
relative severity of these endpoints (U.S. EPA (2012)). The peak NMP in maternal blood (Cmax) and
average area under the curve (AUC) blood concentration of NMP were used as dose metrics for these
endpoints. The doses and response data used for the modeling post-implantation losses and resorptions
are presented in Table 2-1 and Table 2-2, respectively. Model options and standard dichotomous and
continuous BMDS 3.1.2 models applied to the post-implantation loss and the resorption endpoints are
listed below.
Standard Dichotomous BMDS 3.1.2 Models Applied to Post-Implantation Loss Endpoint
Gamma-restricted (Gam)
Log-Logistic-restricted (Lnl)
Multistage-restricted (Mst); from degree = 1 to degree = # dose groups - 1
Weibull-restricted (Wei)
Dichotomous Hill-unrestricted (Dhl)
Logistic (Log)
Log-Probit-unrestricted (Lnp)
• Probit (Pro)
Model Options Used for Dichotomous Response Modeling of Post-Implantation Loss
• Risk Type: Extra Risk
• Benchmark Response (BMR): 0.01 (1%)
• Confidence Level: 0.95
Background: Estimated
Standard Continuous BMDS 3.1.2 Models Applied to Resorptions
• Exponential 2 (Exp2)-restricted
• Exponential 3 (Exp3)-restricted
Page 11 of 244
-------
• Exponential 4 (Exp4)-restricted
• Exponential 5 (Exp5)-restricted
• Hill (Hil)-restricted
• Polynomial Degree 4 (Ply4)-restricted
• Polynomial Degree 3 (Ply3)-restricted
• Polynomial Degree 2 (Ply2)-restricted
• Power (Pow)-restricted
• Linear (Lin)
Model Options Used for Continuous Response
• Benchmark Response (BMR): 1% Absolute Deviation
• Response Distribution-Variance Assumptions
o Normal Distribution-Constant Variance
o Normal Distribution-Non-Constant Variance
o Lognormal Distribution, which assumes Constant Variance (if normal distribution models
do not fit means)
• Confidence Level: 0.95
• Background: Estimated
Table 2-1 Resorptions (Mean % per litter) Data selected for Dose-
Response Modeling for NMP
Reference and Endpoint
Cmax
(mg/L)
AUC
(hr mg/L)
Number of
Litters
Mean ± SD
Saillenfait et al. (2002)
Resorptions
0
0
21
4.1 ±6.1
120
1,145
22
8.9 ± 21.2
250
2,504
24
4.5 ±6.6
531
5,673
25
9.4 ± 8.9
831
9,228
25
91 ± 16
Saillenfait et al. (2003)
Resorptions
0
0
24
2.7 ±3.7
15
156.2
20
4.3 ±4.1
30
318.3
20
9.9 ±22.3
62
665.5
25
7 ±9.4
Sitarek et al. (2012)
Fetal Mortality
0
0
22
0.18 ±0.85
76
902
24
0±0
265
3,168
20
0.13 ±0.34
669
8,245
15
0.8 ± 1.1
Page 12 of 244
-------
Table 2-2 Post-implantation Loss Data Selected for Dose-Response Modeling for NMP
Reference
and
Endpoint
Cmax
(mg/
L)
AUC
(hr
mg/
L)
Litters
w /
Implants
Mean
Implants
Total
Implants
Live
Litters
Mean
Live
Fetuses
Total
Live
Fetuses
Total
Lost
Fetuses
Proportion
Lost
Fetuses
Design
Effect
RS-
Implants"
RS-
Lossa
Saillenfait
et al.
(2002)
Post-
implant-
ation loss
0
0
21
13.3
279.3
21
12.7
266.7
12.6
0.0451
2.0812
134.20
6.0541
120
1145
22
13.6
299.2
21
13.1
275.1
24.1
0.0805
2.5498
117.34
9.4516
250
2504
24
13.3
319.2
24
12.7
304.8
14.4
0.0451
2.0812
153.37
6.9190
531
5673
25
14
350
25
12.4
310
40
0.1143
2.8824
121.42
13.877
831
9228
25
13.8
345
8
2.4
19.2
325.8
0.9443
6.0479
57.044
53.870
Saillenfait
et al.
(2003)
Post-
implant-
ation loss
0
0
24
14.3
343.2
24
13.9
333.6
9.6
0.0280
1.7605
194.94
5.4529
15
156.2
20
13.4
268
20
12.6
252
16
0.0597
2.2958
116.73
6.9692
30
318.3
20
14.1
282
19
14
266
16
0.0567
2.2552
125.04
7.0946
62
665.5
25
12.9
322.5
25
12
300
22.5
0.0698
2.424
133.01
9.2798
Combined
Saillenfait
et al.
(2003:
2002)
0b
0b
21
13.3
279.3
21
12.7
266.7
12.6
0.0451
2.0812
134.20
6.0541
0b
0 b
24
14.3
343.2
24
13.9
333.6
9.6
0.0280
1.7605
194.94
5.4529
15
156.5
20
13.4
268
20
12.6
252
16
0.0597
2.2958
116.73
6.9692
30
319
20
14.1
282
19
14
266
16
0.0567
2.2552
125.04
7.0946
Post-
implant-
ation loss
62
660.8
25
12.9
322.5
25
12
300
22.5
0.0698
2.424
133.01
9.2798
120
1145
22
13.6
299.2
21
13.1
275.1
24.1
0.0805
2.5498
117.34
9.4516
250
2504
24
13.3
319.2
24
12.7
304.8
14.4
0.0451
2.0812
153.37
6.9190
531
5673
25
14
350
25
12.4
310
40
0.1143
2.8824
121.42
13.877
831
9228
25
13.8
345
8
2.4
19.2
325.8
0.9443
6.0479
57.044
53.870
Data highlighted in gray was used for dose-response modeling for NMP.
a The Rao-Scott transformation (RS) entails dividing the total numbers of implantations and post-implantation loss by a design effect to approximate the true variance
in the clustered data.
b Calculating the design effects separately for the control groups from each study is preferred as it captures possible differences between the groups.
Page 13 of 244
-------
Dose-response results from the Saillenfait et al. oral (2002) and inhalation (2003) studies were modeled
separately and combined for the Cmax and AUC dose metrics for the post-implantation loss dichotomous
endpoint. The BMD analyses combining the oral and inhalation results are recommended for this
endpoint, and this recommendation is supported by the following considerations:
• Saillenfait et al. (2003) reported that "mean numbers of implantation sites and of live fetuses and
the incidences of non-live implants and resorptions were comparable across groups" up to and
including their highest-exposure group, for which EPA's PBPK model estimates a 62 mg/L
(C max ) internal dose (Table 2-1). Saillenfait et al. also point out that their findings are in
agreement with the absence of teratogenic effects found in previous studies on the developmental
toxic potential from similar inhalation exposures to NMP.
• A deviance test indicates no significant difference between dose-response relationships in the
two Saillenfait et al. oral and inhalation studies, from combined and separate study results for
doses at or below 530 mg/L Cmax internal dose. Appendix B provides additional technical details
on the statistical approach. Technically the statistical approach assumed that the dose-response in
the region analyzed is sufficiently flat or otherwise linear for each study so that it can be
approximated by a linear regression. Then the slopes and intercepts could be equal or unequal for
the two Saillenfait studies. A useful null hypothesis is that both the slope and intercept are equal.
This approach avoided complications of dependence on selecting a nonlinear model and
technical issues of statistics with constrained parameter spaces (compared to Stiteler et al.
(1993)). The assumption of a dose response curve with a flat or approximately linear portion at
low doses is supported by graphical analysis and by tests for nonlinear trend (discussed further in
the following bullet). The regression approach assumed that the response variable has a binomial
distribution. The deviance test suggests that the data are consistent with equal intercepts and
equal slopes in the dose range evaluated, which includes doses above the 62 mg/L Cmax high dose
blood concentration estimated for the Saillenfait et al. (2003) inhalation study. The analyses (as
well as the trend analysis - next bullet) used Rao-Scott adjusted incidence values to account for
possible litter effects.
• For the inhalation data, test for a trend in the relationship of incidence to Cmax internal doses
(further details provided in Appendix B) did not provide substantial evidence of an effect, thus
did not support separate modeling. Modeling the combined data allows for an effect in the
inhalation study, but attention is needed to the possibility of different dose-response curves for
the inhalation study versus the oral gavage study. The trend analysis provided by the EPITOOLS
software also provide a test of nonlinearity, which does not suggest any deviation from linearity
at the lowest doses, providing some support for the deviance test (previous bullet) as a test for a
difference in the dose-response relationship.
• Close similarity of strain, breed, source and housing helps alleviate uncertainties associated with
combining control and test rat dose-response data from the two studies - The Saillenfait et al.
(2003; 2002) oral and inhalation studies were conducted in the same laboratory within a year of
each other, using the same strain of rats from the same source (Sprague-Dawley rats supplied by
IFFA CREDO Breeding Laboratories, Saint-Germain-surl' Arbresle, France). Control rats of the
oral study were not gavaged, making them more comparable to the inhalation study controls.
Body weight on day 0, body weight gain and food consumption during the treatment period were
nearly identical for control rats of both studies.
• Confidence in the PBPK estimates of internal doses helps alleviate uncertainties associated with
combining control and test rat dose-response data from the two studies - EPA has confidence in
Page 14 of 244
-------
the NMP PBPK model used for this purpose as it has been thoroughly vetted through multiple
reviews (further discussion of the NMP PBPK model is provided in Appendix I of the final NMP
risk evaluation). An advantage that can come from use of a PBPK model with an appropriate
internal dose metric is that it allows one to combine dose-response data from studies with
different designs, such as inhalation studies with different daily exposure durations, oral
exposure by gavage versus drinking water, and exposures by more than one route of exposure.
Evaluation of whether the dose metric is appropriate is accomplished first by plotting the results
of the health-effects studies together, using the PBPK-predicted dose metric as the measure of
dose, and evaluating the overall congruence of the sets of results.2 Statistical tests for consistency
of dose-response relationships, as described in the above bullets, can then be performed for a
rigorous analysis.
• Adding inhalation dose groups to the oral study increases confidence in the modeling results,
particularly in the low dose region - Use of the post-implantation loss endpoint data from the
Saillenfait et al. (2003) inhalation study alone is not recommended given the lack of a statistical
or pharmacokinetic evidence for a dose-response trend. Use of the Saillenfait et al. (2002) oral
study alone is not recommended for this endpoint given the lack of data in the low dose region of
interest. The combination of two dose-response studies presumes that the data, including the
endpoint incidence in control animals, are derived from the same overall population distribution
(i.e., the distribution of incidence versus exposure that would occur in the entire population of
pregnant Sprague-Dawley rats), with differences only occurring because each study provides
data on a different sample from that distribution. Given this assumption, the data for the two
control groups can be combined, to provide a better estimate of the true response incidence
among unexposed animals. While EPA recognizes the uncertainties associated with combining
data from two studies, EPA does not think uncertainties outweigh the benefits associated with
the increased statistical power that comes from combining the studies, which allows EPA to
more confidently estimate low dose specific response levels (i.e., the BMD and BMDL) for the
post-implantation loss endpoint.
Analysis of Post-Implantation Loss as a Dichotomous Response:
Increases in post-implantation losses/implantations (Saillenfait et al. (2003; 2002)). which accounts for
both resorptions and fetal/pup death, is evaluated as a dichotomous endpoint. To perform this analysis,
incidences of post-implantation loss from the reported litter means3 were modeled with standard BMDS
3.1.2 dichotomous models after adjusting for litter effects using a Rao-Scott transformation. Normally,
individual animal data are necessary in order to account for intralitter correlation present in nested
developmental toxicity data (i.e., the observation that pups from one litter are more likely to respond
alike one another compared to pups from another litter). But in this situation, study authors were unable
to provide litter level data and instead an approximate approach was used. Briefly, the numbers of total
implantations and total fetal loss (dead fetuses plus resorptions) were scaled by a design effect in order
to approximate the true variance of the clustered data. This transformation is called the Rao-Scott
transformation and has been shown to reasonably approximate the variance due to clustering and
intralitter correlation in developmental toxicity data (Fox et al. (2016)). Details of the Rao-Scott
transformation are shown in Table 2-2.
As discussed above, a two-sided test for trend indicates no significant trend in the Rao-Scott transformed
Saillenfait et al. (2003) response data with increasing inhalation dose (Appendix B). Consequently, the
2 A previous example of such an analysis was performed by Sasso et al. (2013) for chloroform-induced renal toxicity.
3 Total post-implantation loss was calculated as follows: (mean implantations per litter x total litters) - mean live fetuses per
litter x litter) = total number of post-implantation losses.
Page 15 of 244
-------
BMDLs derived from this dataset alone (Sections 2.7 and 2.8a) are not recommended for use and are not
presented in the summary of BMD and BMDL results (Table 2-4).
The analysis of the eight dose groups associated with the combined dose response data from the two
Saillenfait et al. (2003; 2002) studies presents a unique situation for the Multistage model that requires
careful consideration. The default number of Multistage model degrees run in BMDS 3.1.2 is n-1, where
n is the number of dose-groups in the dataset. Thus, in this case, the 1st degree through 7th degree
Multistage models were run. Consideration needs to be given as to whether that many Multistage
degrees are necessary and appropriate for the dataset being evaluated. Of the Multistage models, the 7th
degree Multistage provides an adequate fit to the data that is similar to the model fit achieved by some
non-Multistage models, but its BMDL estimate is nearly four-fold lower (Table 2-21 and Table 2-22
The Multistage degree 7 BMDL is lower because it contains several extra parameters (Peta coefficients
for degrees 1 through 6). These parameters contribute to the BMDL estimation but are restricted at the 0
boundary criteria for the purposes of the maximum likelihood, BMD estimation. Thus, while the BMD
estimates (377 mg/L Cmax) of the 7th degree Multistage model are similar to adequately fitting non-
Multistage models (423-472 mg/L Cmax), its BMDL estimates are nearly four-fold lower (113 mg/L Cmax
versus 364-437 mg/L Cmax for non-Multistage models). Hence, it appears that the extra parameters in the
higher degree Multistage models are solely driving the derivation of the lower BMDLs for these models.
In situations where BMDLs vary substantially (i.e., by greater than three-fold), EPA BMD Technical
Guidance (U.S. EPA (2012)) states that "expert statistical judgment may help at this point to judge
whether model uncertainty is too great to rely on some or all of the results." In this case, given that trend
tests of the combined dataset indicate a lack of linear dose-response trend in the low dose region up to
and including 531 mg/L Cmax, EPA's judgment is that the Multistage 7 model is not appropriate for the
derivation of a BMDL from this dataset, despite its adequate statistical fit (p-value > 0.1) to the data.
Because BMDLs from the remaining adequately fitting models are sufficiently close, the BMDL is
derived from the model with the lowest AIC (U.S. EPA (2012)). which is the Log-Probit model (see
Sections 2.9 and 2.10).
Analysis of Resorptions as a Continuous Response:
Summary statistics available from Saillenfait et al. (2003; 2002) do not allow for the preferred approach
of evaluating of resorptions as dichotomous responses. Hence mean percent resorptions per litter
reported by Saillenfait et al. (2003; 2002) were evaluated as continuous responses. As with the fetal
weight data discussed in Section 3, because the Saillenfait et al. (2003; 2002) resorption datasets were
obtained from a nested design, with fetus nested within litter, it is preferable to analyze the individual
fetal data in order to incorporate variability across fetuses. However, fetal data were not available for
this study; thus, the means and standard deviations (SDs) of litter mean percent resorption as well as
number of litters in each dose group were modeled (see Method 2 in Appendix A for details).
Standard models gave adequate results for all endpoints, and thus non-standard models were not
considered. Also, since adequate fits to the means were obtained using normal distribution models,
lognormal models were not applied.
The variances for resorptions from Saillenfait et al. (2003; 2002) could not be fit using either the
constant or nonconstant variance models available in BMDS. Therefore, a sensitivity analysis using the
original, minimum and maximum SDs for the dataset, was conducted to determine the influence of the
variances on the resorption results. Briefly, from the results of the modeling using the observed SDs, a
model was selected from the models that fit the means adequately, assuming constant variance. Then the
data were modeled by replacing the SDs in all the groups by the minimum SD across the groups,
assuming constant variance and only fitting models that fit the means adequately for the observed SD
Page 16 of 244
-------
case, and a model was selected from these fits. This step was repeated with the SDs in all the groups
replaced by the maximum SD across the groups. The results of the sensitivity analysis are summarized
in Table 2-3 and the BMD modeling details are presented in Sections 2.1-2.4. For three datasets
(Sections 2.1, 2.2 and 2.4), the lowest BMDL from an adequately fitting model would typically be
recommended because the selected BMDLs for each of the three variance cases did not differ greatly
(i.e., BMDLs varied by less than three-fold). However, due to uncertainty caused by the lack of model
fits (Test 4 P-value <0.1) when SDs were set to the minimum SDs of the group, these BMDLs were
compared to the NOAEL for the endpoint and the lowest of these BMDL and NOAEL values is
recommended as the POD. For the other dataset (Section 2.3), the sensitivity analysis indicated the
selected BMDLs for the three variance cases differ greatly (i.e., BMDLs differed by more than three-
fold). Thus, EPA does not regard the available model results as acceptable and hence the NOAEL for
the endpoint is recommended as the POD. Table 2-3 summarizes the results for this variance sensitivity
analysis, with the recommended POD values highlighted in gray and shown in bold font. Note that the
"free-standing" Cmax and AUC NOAELs from Saillenfait et al. (2003) are not bolded and not
recommended due to the existence of higher Cmax and AUC NOAELs from Saillenfait et al. (2002).
Table 2-3 BMD and BMDL Derivations from the Variance (SD) Sensitivity Analysis of Saillenfait
et
Section
Response
Dose
Metric
SD Case a
Selected
Model
Test 4
P-value
BMDiad
BMDLiad
Cmax
(mg/L)
Observed
Hill
0.389
535
511
2.1
Resorption
(Mean %)
Saillenfait et al.
(2002)
Minimum
Hillb
0.015
535
522
Maximum
Hill
0.696
535
502
NOAEL
~
—
—
250
AUC
Observed
Hill
0.389
5,719
5,462
2.2
(hr
mg/L)
Minimum
Power b
0.014
5,797
5,298
Maximum
Polv 4
0.417
4,307
3,222
NOAEL
—
—
—
2504
Cmax
Observed
Linear
0.251
14.078
6.30
2.3
Resorption
Minimum
Hill b
0.00874
14.5
13.7
(mg/L)
Maximum
Linear
0.668
14.077
4.31
(Mean %)
NOAELc
—
—
—
62
Saillenfait et al.
AUC
Observed
Linear
0.248
151
67.9
2.4
(2003)
(hr
mg/L)
Minimum
Exd5 b
0.00874
152
83.7
Maximum
Polv 3
0.6664
151
46.4
NOAELc
—
—
—
666
a The lowest BMDL from an adequately fitting model is selected and bolded if all BMDLs are reasonably close (i.e.,
withing threefold) and the BMDL is lower than the NOAEL. Otherwise, the NOAEL is selected and bolded.
bNo model adequately fit the dataset means (Test 4 p-value <0.1); results for the model with the lowest AIC are shown.
0 The "free-standing" Cm,,-, and AUC NOAELs from Saillenfait et al. (2003) are not bolded and are not recommended
for use as PODs due to the existence of higher Cm,,-, and AUC NOAELs from Saillenfait et al. (2002) studv.
For each dataset-specific BMD analysis, a single preferred model was chosen from the standard set of
models and modeling options listed above. The modeling restrictions and the model selection criteria
facilitated in BMDS 3.1.2 and defined in the BMDS 3.1.2 User Guide were applied in accordance with
EPA BMD Technical Guidance (U.S. EPA (2012)). Briefly, for each dataset, BMDS models with
standard restrictions were fitted to the data using the maximum likelihood method. For dichotomous
Page 17 of 244
-------
models, if the BMDLs from adequately fitting models (p-value < 0.1) were sufficiently close (within a
threefold range), the model with the lowest AIC was selected as the best-fitting model, and its BMDL
was used as the POD. Per BMD Technical Guidance "This criterion is intended to help arrive at a single
BMDL value in an objective, reproducible manner." If the BMDLs are not sufficiently close (not within
a threefold range), it was determined that the BMDLs were substantially model-dependent; thus, the
BMDL from the adequately fitting model with the lowest BMDL was used as the POD. From
continuous models applied to the resorptions endpoint, model fit was assessed by a series of tests as
follows. For each model, first the homogeneity of the variances was tested using a likelihood ratio test
(BMDS Test 2). If Test 2 was not rejected (%2 p-value > 0.05), the model was fitted to the data assuming
constant variance. If Test 2 was rejected (%2 p-value < 0.05), the variance was modeled as a power
function of the mean, and the variance model was tested for adequacy of fit using a likelihood ratio test
(BMDS Test 3). For fitting models using either constant variance or modeled variance, models for the
mean response were tested for adequacy of fit using a likelihood ratio test (BMDS Test 4, with yl p-
value < 0.10 indicating inadequate fit). From among the models that yielded an adequate fit, the model
for POD determination was selected using the same procedure as for the dichotomous models. For both
the dichotomous and continuous model analyses, other factors were also used to assess the model fit,
such as scaled residuals, visual fit, and adequacy of fit in the low-dose region and in the vicinity of the
BMR.
Comparisons of model fits obtained for post-implantation losses and resorptions are provided in Table
2-5 through Table 2-22. The best-fitting models, based on the criteria described above, are bolded and
highlighted in gray. For each of the best fitting models in Sections 2.1-2.10, subsequent tables and
figures show the model version number, model form, benchmark dose calculation, parameter estimates
and estimated values.
PODs identified based on the best fit models for post-implantation loss and resorptions for the
Saillenfait et al. (2003; 2002) studies are summarized in Table 2-4.
Table 2-4. Summary of PODs identified for Cmax and AUC Dose Metrics for Post-Implantation
Loss and Resorptions
Section
Response
Dose
Metric
Selected
Model a
BMDier
or NOAEL b
BMDLier or
NOAEL b
2.5
Post-Impl antati on
Losses/Implants
(Saillenfait et al. (2002))
Cmax
(mg/L)
Log-Probit
474
437
2.6
AUC
(hr mg/L)
Log-Probit
5,010
4,592
2.9
Post-Impl antati on
Losses/Implants
(Saillenfait et al. (2003;
2002) combined)
Cmax
(mg/L)
Log-Probit
470
437
2.10
AUC
(hr mg/L)
Log-Probit
4,990
4,590
2.1
Resorption (Mean %)
(Saillenfait et al. (2002))
Cmax
(mg/L)
NOAEL c
—
250
2.2
AUC
(hr mg/L)
NOAEL c
—
2,500
Page 18 of 244
-------
a Since standard models gave adequate results for all endpoints, non-standard models were not considered. Since fits
to the means of the mean % resorption data were obtained using normal distribution models, lognormal models
were not applied.
bBMD and BMDL values are for BMR of 1% Extra Risk (1ER) for post-implantation losses/Implants and NOAELs
for mean % resorptions (see Table 2-3).
0 The NOAEL for this dataset is recommended for use over the BMDL values derived for this endpoint (see Table
2-3).
Page 19 of 244
-------
2.1 Resorptions: Results for Saillenfait et al. (2002) using Cmax
Table 2-5 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using Cmax as the Dose Metric (Saillenfait et al. (2002))
3MR = 1% Absolute Deviation (AD)
Modela
Goodness of fit
BMD
(mg/L)
BMDL
(mg/L)
BMDU
(mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.007044
947.069
393.2644
323.4513
4479.5071
Only Exponential 3, Hill and
Power models provided an
adequate fit (Test 4 p-value >
0.10). Of these, the Hill model
was selected based on lowest
AIC.
Exponential 3
0.379373
938.906
548.3582
424.5308
723.8003
Exponential 4
<0.0001
1084.78
51.9053
46.8503
58.0743
Exponential 5
0.000125
953.690
481.7744
0
521.0350
Hill b
0.389268
938.855
535.1995
511.3336
704.9748
Polynomial 4°
0.007024
947.054
387.6039
351.9091
Infinity
Polynomial 3°
<0.0001
968.580
300.8686
277.3312
Infinity
Polynomial 2°
<0.0001
1011.52
183.1171
167.1364
Infinity
Power
0.388006
938.861
541.5904
459.9201
574.7369
Linear
<0.0001
1073.71
42.9383
38.2120
49.0001
aNo variance model fit this dataset using reported SD values (BMDS Test 2 and 3 p-values < 0.0001). Results for constant
variance model are shown.
b Model selection based on lowest AIC from adequately fitting models.
Page 20 of 244
-------
Table 2-6 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using Cmax as the Dose Metric (Saillenfait et al. (2002))
3MR = 1% Absolute Deviation (AD); minimum SD among groups used for all groups in analysis
Modela
Goodness of fit
BMD
BMDL
BMDU
Basis for model selection
Test 4
P-value
AIC
(mg/L)
(mg/L)
(mg/L)
Exponential 2
<0.0001
803.718
393.2456
353.6334
437.6397
No model adequately fit the mean
response data.
Exponential 3
0.012984
766.749
548.3653
483.8053
636.6293
Exponential 4
<0.0001
1054.98
51.8658
47.6549
56.8808
Exponential 5
0.003606
768.533
538.1709
509.8833
603.9132
Hill
0.014526
766.525
535.1998
521.5652
561.5738
Polynomial 4°
<0.0001
808.344
387.6039
376.5096
Infinity
Polynomial 3°
<0.0001
867.713
300.8687
290.5825
Infinity
Polynomial 2°
<0.0001
950.354
183.1170
173.1190
Infinity
Power
0.014322
766.553
541.5890
498.5351
605.4818
Linear
<0.0001
1041.71
42.9380
38.7557
48.1304
aNo model adequately fit the means of this dataset using the 6.1 minimum SD value for all dose groups (BMDS Test 4 <
0.1). No variance model fit this dataset using reported SD values (BMDS Test 2 and 3 p-values < 0.0001). Results for
constant variance model are shown.
Page 21 of 244
-------
Table 2-7 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using Cmax as the Dose Metric (Saillenfait et al. (2002))
3MR = 1% Absolute Deviation (AD); maximum SD among groups used for all groups in analysis
Modela
Goodness of fit
BMD
BMDL
BMDU
Basis for model selection
Test 4
P-value
AIC
(mg/L)
(mg/L)
(mg/L)
Exponential 2
0.189295
1052.3296
393.2780
290.2496
539.9314
The Hill model was
selected based on lowest
AIC among adequately
fitting models (Test 4 P-
Exponential 3
0.689247
1050.3022
548.3492
363.5049
732.9579
value > 0.1).
Exponential 4
<0.0001
1128.5692
51.8672
46.2820
58.9588
Exponential 5
0.394674
1052.2824
537.0627
420.4268
708.2826
Hill b
0.696127
1050.2823
535.1428
502.2433
705.4570
Polynomial 4°
0.232517
1051.1411
387.6039
302.4135
Infinity
Polynomial 3°
0.004892
1060.4678
300.8687
251.0415
Infinity
Polynomial 2°
<0.0001
1082.9218
183.0936
155.4972
Infinity
Power
0.694961
1050.2857
544.5289
420.2064
671.7613
Linear
<0.0001
1120.6978
42.93835
37.2979
50.5875
a Results for constant variance model are shown. SD set to maximum value of 21.2 for all dose groups.
b Model selection based on lowest AIC from adequately fitting models.
Page 22 of 244
-------
120
-20
Cmax{mg/L)
Figure 2.1-1 Plot of Response by Dose, with Fitted Curve for Selected Hill Model for Resorptions
(Mean % per Litter) in Rats Exposed to NMP via Gavage (Saillenfait et al. (2002))
BMR = 1% AD; all SDs set to the maximum SD across the groups
Estimated Probability
Response at BMD
O Data
BMD
BMDL
USER INPUT
Info
Model
frequentist Hill vl.l
Dose-Response Model
M[dose] = g + v*doseAn/(kAn + doseAn)
Variance Model
Var[i] = alpha
Model Options
BMR Type
Abs. Dev.
BMRF
1
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Cmax (mg/L)
Independent Variable
Mean% Resorptions per litter
Total # of Observations
5
Adverse Direction
Automatic
Page 23 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
535.1428123
BMDL
502.243266
BMDU
705.4569526
AIC
1050.282358
Test 4 P-value
0.696126859
D.O.F.
2
Model Parameters
# of Parameters
5
Variable
Estimate
8
5.813844994
V
85.79623561
k
631.9163554
n
Bounded
alpha
432.9057039
Goodness of Fit
Dose
Size
Observed
Mean
Estimated SD
Calc'd SD
Observed
SD
Scaled Residual
0
21
4.1
20.80638613
21.2
21.2
-0.377471819
120
22
8.9
20.80638613
21.2
21.2
0.695716689
250
24
4.5
20.80638613
21.2
21.2
-0.309353259
531
25
9.4
20.80638613
21.2
21.2
-0.000237076
831
25
91
20.80638613
21.2
21.2
0.001359032
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-520.7789555
6
1053.557911
A2
-520.7786645
10
1061.557329
A3
-520.7789555
6
1053.557911
fitted
-521.1411788
4
1050.282358
R
-598.5686115
2
1201.137223
"ests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
155.5798939
8
<0.0001
2
0.000581864
4
0.999999958
3
0.000581864
4
0.999999958
4
0.724446735
2
0.696126859
* Includes additive constant of -107.51581. This constant was not included in the LL
derivation prior to BMDS 3.0.
Page 24 of 244
-------
2.2 Resorptions: Results for Saillenfait et al. (2002) using AUC
Table 2-8 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using AUC as the Dose Metric (Saillenfait et al. (2002))
3MR = 1% Absolute Deviation (AD)
Goodness of fit
BMD
(hr mg/L)
BMDL
(hr mg/L)
BMDU
(hr mg/L)
Modela
Test 4
P-value
AIC
Basis for model selection
Exponential 2
0.008986
946.544
4043.1086
3295.7539
5010.8819
Only Exponential 3, Hill and
Power models provided an
adequate fit (Test 4 p-value >
0.10). Of these, the Hill model
Exponential 3
0.379432
938.906
5875.3609
4450.0672
8103.4619
was selected based on lowest
AIC.
Exponential 4
<0.0001
1075.55
9
559.6636
507.9306
622.4397
Exponential 5
0.000124
953.701
5263.5918
0
5585.7252
Hill"
0.389268
938.855
5718.8813
5462.4791
7838.6124
Polynomial 4°
0.040523
942.962
4306.9967
3852.5679
Infinity
Polynomial 3°
<0.0001
961.304
3339.8312
3068.5732
Infinity
Polynomial 2°
<0.0001
1000.87
5
2027.7227
1855.6345
Infinity
Power
0.388006
938.861
5797.0548
4853.5551
6326.2892
Linear
<0.0001
1067.16
2
472.3225
422.3903
535.6331
aNo variance model fit this dataset using reported SD values (BMDS Test 2 and 3 /^-values < 0.0001). Results for constant
variance model are shown.
b Model selection based on lowest AIC from adequately fitting models.
Page 25 of 244
-------
Table 2-9 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using AUC as the Dose Metric (Saillenfait et al. (2002))
3MR = 1% Absolute Deviation (AD); minimum SD among groups used for all groups in analysis
Goodness of fit
BMD
(hr mg/L)
BMDL
(hr mg/L)
BMDU
(hr mg/L)
Modela
Test 4
P-value
AIC
Basis for model selection
Exponential 2
<0.0001
801.920
4042.9257
3615.8068
4532.9584
No model adequately fit the
mean response data.
Exponential 3
0.012993
766.748
5875.3642
5127.2682
6661.9635
Exponential 4
<0.0001
1043.697
559.4295
515.9626
610.3890
Exponential 5
0.003623
768.525
5734.7899
5504.3446
6241.6250
Hill
0.003623
768.525
5718.9466
5571.8781
6000.1162
Polynomial 4°
<0.0001
794.383
4306.9964
4175.4460
Infinity
Polynomial 3°
<0.0001
849.688
3339.8310
3231.0905
Infinity
Polynomial 2°
<0.0001
933.375
2027.7128
1925.1043
Infinity
Power
0.014322
766.553
5797.0518
5298.1645
6543.8546
Linear
<0.0001
1033.026
472.3283
428.5655
526.0557
aNo model adequately fit the means of this dataset using the 6.1 minimum SD value for all dose groups (BMDS Test 4 <
0.1). No variance model fit this dataset using reported SD values (BMDS Test 2 and 3 p-values < 0.0001). Results for
constant variance model are shown.
Page 26 of 244
-------
Table 2-10 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using AUC as the Dose Metric (Saillenfait et al. (2002))
Goodness of fit
BMD
(hr mg/L)
BMDL
(hr mg/L)
BMDU
(hr mg/L)
Modela
Test 4
P-value
AIC
Basis for model selection
Exponential 2
0.207145
1052.116
4043.0558
2952.3908
5709.1001
Exponential 2, 3, Hill,
Power, and Polynomial 4
models provided an
adequate fit (Test 4 p-value
Exponential 3
0.689288
1050.302
5875.7504
3757.6576
8151.6558
>0.10). Of these, the
Polynomial 4 model was
selected based on lowest
AIC.
Exponential 4
<0.0001
1124.251
560.0255
496.0853
640.6819
Exponential 5
0.015662
1057.398
5198.5546
3645.3150
5974.0792
Hill
0.695791
1050.283
5710.8575
5360.7253
7835.6455
Polynomial 4° b'c
0.417028
1049.477
4306.9964
3221.5295
Infinity
Polynomial 3°
0.020166
1057.206
3339.8268
2748.7368
Infinity
Polynomial 2°
<0.0001
1076.304
2027.7670
1720.5274
Infinity
Power
0.695257
1050.284
5797.0501
4400.0709
7721.2377
Linear
<0.0001
1116.355
472.29675
412.2081
552.9808
a Results for constant variance model are shown. SD set to maximum value of 21.2 for all dose groups.
b Model selection based on lowest AIC from adequately fitting models.
0 Scaled residuals for selected Poly 4 model for 0, 1144, 2503, 5674 and 9231 lir mg/L were 0.1170, 1.183, 0.1094, -1.543,
and 0.2195, respectively.
Page 27 of 244
-------
120
-20
AUC (hr mg/L)
Figure 2.2-1 Plot of Response by Dose, with Fitted Curve for Selected Polynomial Degree 4 Model
for Resorptions (Mean % per Litter) in Rats Exposed to NMP via Gavage (Saillenfait et al. (2002))
BMR = 1% AD for the BMD and 0.95 lower confident limit for the BMDL; all SDs set to the maximum
SD across the groups
Estimated Probability
Response at BMD
O Data
BMD
BMDL
USER INPUT
Info
Model
frequentist Polynomial degree 4 vl.l
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha
Model Options
BMR Type
Abs. Dev.
BMRF
1
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
AUC (hr mg/L)
Independent Variable
Mean % Resorptions per litter
Total # of Observations
5
Adverse Direction
Automatic
MODEL RESULTS
Page 28 of 244
-------
Benchmark Dose
BMD
4306.996436
BMDL
3221.529462
BMDU
Infinity
AIC
1049.477283
Test 4 P-value
0.417027552
D.O.F.
4
Model Parameters
# of Parameters
6
Variable
Estimate
8
3.561301059
bl
Bounded
b2
Bounded
b3
Bounded
b4
Bounded
alpha
444.8896007
Goodness of Fit
Dose
Size
Observed
Mean
Estimated
SD
Calc'd SD
Observed
SD
Scaled Residual
0
21
4.1
21.09240623
21.2
21.2
0.117038741
1144
22
8.9
21.09240623
21.2
21.2
1.182653022
2503
24
4.5
21.09240623
21.2
21.2
0.109405041
5674
25
9.4
21.09240623
21.2
21.2
-1.543357153
9231
25
91
21.09240623
21.2
21.2
0.219466033
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-520.7789555
6
1053.557911
A2
-520.7786645
10
1061.557329
A3
-520.7789555
6
1053.557911
fitted
-522.7386417
2
1049.477283
R
-598.5686115
2
1201.137223
* Includes additive constant of -107.51581. This constant was not included in the LL
derivation prior to BMDS 3.0.
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
155.5798939
8
<0.0001
2
0.000581864
4
0.999999958
3
0.000581864
4
0.999999958
4
3.919372507
4
0.417027552
Page 29 of 244
-------
2.3 Resorptions: Results for Saillenfait et al. (2003) using Cmax
Table 2-11 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using Cmax as the Dose Metric (Saillenfait et al. (2003))
BMR = 1% Absolute Deviation (AD)
Modela
Goodness of fit
BMD
(mg/L)
BMDL
(mg/L)
BMDU
(mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.213813
699.611
21.11962
12.83895
Infinity
The Linear model was selected
based on lowest AIC.
Exponential 3
0.213813
699.611
21.12014
12.83898
Infinity
Exponential 4
0.209689
700.099
2.947766
0.432368c
Infinity
Exponential 5
NA
701.204
13.80411
0
31.64662
Hill
NA
701.204
14.48207
5.265227
41.21115
Polynomial 4°
--
--
--
--
--
Polynomial 3°
0.250824
699.291
13.97376
6.300609
Infinity
Polynomial 2°
0.250824
699.291
13.9729
6.30061
Infinity
Power
0.250824
699.291
13.97397
6.300571
Infinity
Linear b
0.250824
699.291
13.97378
6.300621
Infinity
a No variance model fit this dataset using reported SD values (BMDS Test 2 and 3 p-values < 0.0001). Results for constant
variance model are shown.
b Model selection based on lowest AIC from adequately fitting models.
0 Model is considered to be of questionable relevance as the estimated BMDL is substantially > 10 below lowest dose.
Page 30 of 244
-------
Table 2-12 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using Cmax as the Dose Metric (Saillenfait et al. (2003))
3MR = 1% Absolute Deviation (AD); minimum SD among groups used for all groups in analysis
Goodness of fit
BMD
(mg/L)
BMDL
(mg/L)
BMDU
(mg/L)
Modela
Test 4
P-value
AIC
Basis for model selection
Exponential 2
<0.0001
515.375
21.12172
16.46037
34.91679
No model adequately fit the mean
response data.
Exponential 3
<0.0001
515.375
21.12172
16.46037
34.91678
Exponential 4
<0.0001
504.624
2.948015
1.542205
6.039347
Exponential 5
NA
498.236
13.73017
8.170257
24.87779
Hill
0.008742
496.236
14.501
13.66569
22.98806
Polynomial 4°
--
--
--
--
--
Polynomial 3°
<0.0001
512.813
13.97417
9.791243
24.39673
Polynomial 2°
<0.0001
512.813
13.97452
9.791353
24.39666
Power
<0.0001
512.813
13.97453
9.791287
24.39600
Linear
<0.0001
512.813
13.97459
9.791277
24.39678
a No model adequately fit the means of this dataset using the 3.7 minimum SD value for all dose groups (BMDS Test 4 <
0.1). No variance model fit this dataset using reported SD values (BMDS Test 2 and 3 p-values < 0.0001). Results for
constant variance model are shown.
Page 31 of 244
-------
Table 2-13 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using Cmax as the Dose Metric (Saillenfait et al. (2003))
3MR = 1% Absolute Deviation (AD); maximum SD among groups used for all groups in analysis
Modela
Goodness of fit
BMD
(mg/L)
BMDL
(mg/L)
BMDU
(mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.637063
807.997
21.11877
10.41853
Infinity
The Linear model was selected
based on lowest AIC.
Exponential 3
0.637063
807.997
21.11949
10.41858
Infinity
Exponential 4
0.498954
809.552
2.947133
0.32408 c
Infinity
Exponential 5
NA
811.292
13.61688
0
Infinity
Hill
NA
811.291
14.50418
4.648527
Infinity
Polynomial 4°
--
--
--
--
--
Polynomial 3°
0.667837
807.902
13.97068
4.310606
Infinity
Polynomial 2°
0.667837
807.902
13.95906
4.3106
Infinity
Power
0.667837
807.902
13.97127
4.332837
Infinity
Linear b
0.667837
807.902
13.96877
4.310579 d
Infinity
a Results for constant variance model are shown. SD set to maximum value of 22.3 for all dose groups.
b Model selection based on lowest AIC from adequately fitting models.
0 Model is considered to be of questionable relevance as the estimated BMDL is substantially > 10 below lowest dose.
d BMDLIad selection for this dataset (bolded) based on lowest BMDL from selected models for each SD approach (Tables
a-c) for this BMR type. Selected models in bold; scaled residuals for selected Linear model for doses 0, 15, 30, and 62
mg/L were -0.2719, -0.1407, 0.7839 and -0.3091, respectively.
Page 32 of 244
-------
25
Estimated Probability
Response at BMD
O Data
BMD
BMDL
-10
Cmax (mg/L)
Figure 2.3-1 Plot of Response by Dose, with Fitted Curve for Selected Linear Model for
Resorptions (Mean % per Litter) in Rats Exposed to NMP via inhalation (Saillenfait et al. (2003))
BMR = 1% AD; all SDs set to the maximum SD across the groups
USER INPUT
Info
Model
frequentist Linear vl. 1
Dose-Response Model
M[dose] = g + b 1 *dose
Variance Model
Var[i] = alpha
Model Options
BMR Type
Abs. Dev.
BMRF
1
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Cmax (mg/L)
Independent Variable
Mean% Resorptions per litter
Total # of Observations
4
Adverse Direction
Automatic
Page 33 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
13.96877456
BMDL
4.310579039
BMDU
Infinity
AIC
807.9022344
Test 4 P-value
0.667837175
D.O.F.
2
Model Parameters
# of Parameters
4
Variable
Estimate
8
3.914947346
betal
0.071588238
alpha
479.2687721
Goodness of Fit
Dose
Size
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled Residual
0
24
2.7
21.892208
22.3
22.3
-0.271877653
15
20
4.3
21.892208
22.3
22.3
-0.140701988
30
20
9.9
21.892208
22.3
22.3
0.783904443
62
25
7
21.892208
22.3
22.3
-0.309109551
Likelihooc
s of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-400.5474063
5
811.094813
A2
-400.5468877
8
817.093775
A3
-400.5474063
5
811.094813
fitted
-400.9511172
3
807.902234
R
-401.2235829
2
806.447166
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
1.353390341
6
0.96863225
2
0.001037257
3
0.99999112
3
0.001037257
3
0.99999112
4
0.807421769
2
0.66783718
Page 34 of 244
-------
2.4 Resorptions: Results for Saillenfait et al. (2003) using AUC
Table 2-14 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using AUC as the Dose Metric (Saillenfait et al. (2003))
3MR = 1% Absolute Deviation (AD)
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr mg/L)
BMDU
(hr
mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.211980
699.6287
227.9498
137.8811
Infinity
The Linear model was selected
based on lowest AIC.
Exponential 3
0.211980
699.628
227.9498
137.8820
Infinity
Exponential 4
0.212197
700.082
30.916
4.641077c
Infinity
Exponential 5
NA
701.204
151.6068
0
509.9996
Hill
NA
701.205
149.9484
38.47086
519.3482
Polynomial 4°
--
--
--
--
--
Polynomial 3°
0.248246
699.312
151.1512
67.92561
Infinity
Polynomial 2°
0.248246
699.312
151.2107
67.92571
Infinity
Power
0.248246
699.312
151.2246
67.92488
Infinity
Linear b
0.248246
699.312
151.1247
67.92584
Infinity
aNo variance model fit this dataset using reported SD values (BMDS Test 2 and 3 p-values < 0.0001). Results for constant
variance model are shown.
b Model selection based on lowest AIC from adequately fitting models.
0 Model is considered to be of questionable relevance as the estimated BMDL is substantially > 10 below lowest dose.
Page 35 of 244
-------
Table 2-15 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using AUC as the Dose Metric (Saillenfait et al. (2003))
3MR = 1% Absolute Deviation (AD); minimum SD among groups used for all groups in analysis
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr
mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
<0.0001
515.511
227.8383
177.0871
379.0394
No model adequately fit the mean
response data.
Exponential 3
<0.0001
515.511
227.8254
177.0872
379.0402
Exponential 4
0.000102
504.469
30.91629
16.28533
63.04971
Exponential 5
0.008743
496.236
151.6093
83.69496
274.3896
Hill
0.008744
496.236
150.9994
140.9521
243.9417
Polynomial 4°
--
--
--
--
--
Polynomial 3°
<0.0001
512.980
151.1449
105.6824
265.1396
Polynomial 2°
<0.0001
512.980
151.1439
105.6817
265.1420
Power
<0.0001
512.980
151.1268
105.6816
265.1342
Linear
<0.0001
512.980
151.1255
105.6819
265.1440
aNo model adequately fit the means of this dataset using the 3.7 minimum SD value for all dose groups (BMDS Test 4 <
0.1). No variance model fit this dataset using reported SD values (BMDS Test 2 and 3 p-values < 0.0001). Results for
constant variance model are shown.
Page 36 of 244
-------
Table 2-16 Model Predictions for Resorptions (Mean % per Litter) in Rats Exposed to NMP via
Gavage Using AUC as the Dose Metric (Saillenfait et al. (2003))
BMR = 1% Absol
ute Deviation (AD); maximum SD among grou
ps used for all groups in analysis
Modela
Goodness of fit
BMD
BMDL
BMDU
Basis for model selection
Test 4
P-value
AIC
(hr mg/L)
(hr mg/L)
(hr mg/L)
Exponential 2
0.635442
808.002
227.8262
111.7975
Infinity
The Polynomial 3 model
was selected based on
lowest AIC.
Exponential 3
0.635442
808.002
227.8273
111.7974
Infinity
Exponential 4
0.501532
809.547
30.91004
3.47685 c
Infinity
Exponential 5
NA
811.291
151.6049
0
Infinity
Hill
NA
811.293
147.7947
0
Infinity
Polynomial 4°
--
--
--
--
--
Polynomial 3° b
0.665804
807.908
151.1354
46.43517 d
Infinity
Polynomial 2°
0.665803
807.908
151.1375
46.43597
Infinity
Power
0.665803
807.908
151.1461
46.62033
Infinity
Linear
0.665803
807.908
151.1459
46.43533
Infinity
a Results for constant variance model are shown. SD set to maximum value of 22.3 for all dose groups.
b Model selection based on lowest AIC from adequately fitting models.
0 Model is considered to be of questionable relevance as the estimated BMDL is substantially > 10 below lowest dose.
dBMDLlAD selection for this dataset (bolded) based on lowest BMDL from selected models for each SD approach (Tables
a-c) for this BMR type. Scaled residuals for selected Poly 3 model for doses 0, 156.5, 319 and 660.8 lir mg/L were -
0.2781, -0.1382, 0.7866 and -0.3075, respectively.
Page 37 of 244
-------
25
Estimated Probability
Response at BMD
O Data
BMD
BMDL
-10
AUC (hr mg/L)
Figure 2.4-1 Plot of Response by Dose, with Fitted Curve for Polynomial Degree 3 Model for
Resorptions (Mean % per Litter) in Rats Exposed to NMP via inhalation (Saillenfait et al. (2003))
BMR = 1% AD; all SDs set to the maximum SD across the groups
USER INPUT
Info
Model
frequentist Polynomial degree 3 vl.l
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha
Model Options
BMR Type
Abs. Dev.
BMRF
1
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
AUC (hr mg/L)
Independent Variable
Mean% Resorptions per Litter
Total # of Observations
4
Adverse Direction
Automatic
Page 38 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
151.1354168
BMDL
46.43517436
BMDU
Infinity
AIC
807.9083341
Test 4 P-value
0.665803491
D.O.F.
2
Model Parameters
# of Parameters
4
Variable
Estimate
8
3.942998221
bl
0.006616582
b2
Bounded
b3
Bounded
alpha
479.3001679
Goodness of Fit
Dose
Size
Observed
Mean
Estimated
SD
Calc'd
SD
Observe
d SD
Scaled Residual
0
24
2.7
21.8929251
22.3
22.3
-0.278145692
156.5
20
4.3
21.8929251
22.3
22.3
-0.138192477
319
20
9.9
21.8929251
22.3
22.3
0.786644232
660.8
25
7
21.8929251
22.3
22.3
-0.307481493
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-400.5474063
5
811.094813
A2
-400.5468877
8
817.093775
A3
-400.5474063
5
811.094813
fitted
-400.954167
3
807.9083341
R
-401.2235829
2
806.447166
* Includes additive constant of -81.78553. This constant was not included in the LL derivation prior
to BMDS 3.0.
"ests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
1.353390341
6
0.96863225
2
0.001037257
3
0.99999112
3
0.001037257
3
0.99999112
4
0.813521422
2
0.665803491
Page 39 of 244
-------
2.5 Post-implantation Losses: Results for Saillenfait et al. (2002) using Cmax
Table 2-17 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in
Rats Exposed to NMP via Gavage Using Cmax as the Dose Metric (Saillenfait et al. (2002))
3MR =1% Relative Deviation (RD)
Model
Goodness of fit
BMD
(mg/L)
BMDL
(mg/L)
BMDU
(mg/L)
Basis for model selection
P-value
AIC
Dichotomous
Hill
0.1572321
292.17137
455.2478
410.9323
521.50995
The Log-Probit model is
selected based on lowest
AIC.
Gamma
0.0026842
302.78107
370.8095
351.7727
390.48307
Log-Logistic
0.3677639
290.17137
455.2251
410.9016
521.51172
Multistage 4°
<0.0001
315.70804
237.824
156.5285
252.46991
Multistage 3°
<0.0001
334.71559
171.2772
123.605
184.79851
Multistage 2°
<0.0001
361.93988
92.36062
71.96177
103.27055
Multistage 1°
(Quantal
<0.0001
405.75225
17.6278
14.49097
21.907659
Weibull
0.3666823
290.17729
426.3443
365.4222
519.08535
Logistic
<0.0001
339.80554
86.68156
65.90755
114.83487
Log-Probita
0.367832
290.171
473.6389
437.3743
523.85736
Probit
<0.0001
351.11489
68.46759
52.41983
90.924229
a Scaled residuals for selected Log-Probit model for doses 0, 120, 250, 531 and 831 mg/L were -0.5201, 1.192, -0.5560,
6.428E-08 and -6.292E-07, respectively.
Page 40 of 244
-------
^—Estimated Probability
^—Response at BMD
O Data
BMD
— BMDL
Figure 2.5-1 Plot of Response by Dose, with Fitted Curve for Selected Log-Probit Model for Post-
implantation Loss in Rats Exposed to NMP via Gavage (Saillenfait et al. (2002))
BMR = 1% RD
USER INPUT
Info
Model
frequentist Log-Probit vl. 1
Dose-Response Model
P[dose] = g+(l-g) * CumNorm(a+b*Log(Dose))
Model Options
Risk Type
Extra Risk
BMR
0.01
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
Cmax (mg/L)
Independent Variable
Post-Implantation Loss
Total # of Observations
5
MODEL RESULTS
Benchmark Dose
BMD
473.6389478
BMDL
437.3742789
BMDU
523.8573611
AIC
290.171
P-value
0.367832005
D.O.F.
2
Chi2
2.000257907
Cmax(mg/L)
Page 41 of 244
-------
Model Parameters
# of Parameters
4
Variable
Estimate
8
0.055381721
a
-44.95639542
b
6.919962006
(
joodness of Fit
Dose
Estimated Probability
Expected
Observed
Size
Scaled Residual
0
0.055381721
7.432241273
6.05414704
134.2003
-0.520105
120
0.055381721
6.498562598
9.45162137
117.3413
1.1918875
250
0.055381721
8.493990027
6.91902519
153.3717
-0.556015
531
0.114285712
13.87709518
13.8770954
121.4246
6.428E-08
831
0.944347845
53.86959138
53.8695903
57.04423
-6.29E-07
Analysis of Deviance
Model
Log Likelihood
# of Parameters
Deviance
Test d.f.
P-value
Full Model
-141.1497769
5
-
-
-
Fitted Model
-142.0855
3
1.87144618
2
0.3923021
Reduced Model
-251.1748556
1
220.050157
4
<0.0001
Page 42 of 244
-------
2.6 Post-implantation Losses: Results for Saillenfait et al. (2002) using
AUC
Table 2-18 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in
Rats Exposed to NMP via Gavage Using AUC as the Dose Metric (Saillenfait et al. (2002))
3MR =1% Relative Deviation (RD)
Modela
Goodness of fit
BMD
(hr mg/L)
BMDL
(hr mg/L)
BMDU
(hr mg/L)
Basis for model selection
P-value
AIC
Dichotomous
Hill
0.157272
292.17101
5035.17
4290.223
Infinity
The Log-Probit model is
selected based on lowest
AIC.
Gamma
0.0117941
298.74127
25
4025.695
3809.819
4246.349
Log-Logistic
0.3677636
290.17138
4799.243
4292.557
5568.1149
Multistage 4°
0.0001122
310.60220
2589.069
1673.568
2750.7374
Multistage 3°
<0.0001
329.10985
1854.91
1328.066
2002.4376
Multistage 2°
<0.0001
356.41697
991.3199
767.9789
1109.1373
Multistage 1°
(Quantal
<0.0001
400.78140
184.3376
151.5097
229.07485
Weibull
0.3667049
290.17731
4468.994
3782.108
5537.3985
Logistic
<0.0001
335.28734
892.2392
682.6192
1177.6672
Log-Probita
0.3678321
290.171
5010.495
4592.073
5591.3344
Probit
<0.0001
346.05448
706.5968
544.8604
933.0203
a Scaled residuals for selected Log-Probit model for doses 0, 1144, 2503, 5674 and 9231 lir mg/L were -0.5201, 1.192, -
0.5560, 1.958E-05 and -1.140E-05, respectively.
Page 43 of 244
-------
^—Estimated Probability
^—Response at BMD
O Data
BMD
BMDL
Figure 2.6-1 Plot of Response by Dose, with Fitted Curve for Selected Log-Probit Model for Post-
implantation Loss in Rats Exposed to NMP via Gavage (Saillenfait et al. (2002))
BMR = 1% RD
USER INPUT
Info
Model
frequentist Log-Probit vl. 1
Dose-Response Model
P[dose] = g+(l-g) * CumNorm(a+b*Log(Dose))
Model Options
Risk Type
Extra Risk
BMR
0.01
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
AUC (hr mg/L)
Independent Variable
Post-Implantation Loss
Total # of Observations
5
MODEL RESULTS
Benchmark Dose
BMD
5010.494649
BMDL
4592.073168
BMDU
5591.334418
AIC
290.171
P-value
0.367832091
D.O.F.
2
Chi2
2.000257437
i
10
J 0-9
g 0.8
> 0.7
-------
Model Parameters
# of Parameters
4
Variable
Estimate
8
0.055381734
a
-56.59546142
b
6.370145171
Goodness of Fit
Dose
Estimated Probability
Expected
Observed
Size
Scaled Residual
0
0.055381734
7.432243129
6.05414704
134.2003
-0.520105
1145
0.055381734
6.498564221
9.45162137
117.3413
1.1918867
2504
0.055381734
8.493992148
6.91902519
153.3717
-0.556016
5673
0.114285641
13.87708647
13.8770954
121.4246
2.548E-06
9228
0.944348009
53.86960073
53.8695903
57.04423
-6.03E-06
Analysis of Deviance
Model
Log Likelihood
# of Parameters
Deviance
Test d.f.
P-value
Full Model
-141.1497769
5
-
-
-
Fitted Model
-142.0855
3
1.87144618
2
0.3923021
Reduced
Model
-251.1748556
1
220.050157
4
<0.0001
Page 45 of 244
-------
2.7 Post-implantation Losses: Results for Saillenfait et al. (2003) using Cmax
Table 2-19 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in
Rats Exposed to NMP via Inhalation Using Cmax as the Dose Metric (Saillenfait et al. (2003))
3MR =1% Relative Deviation (RD)
Modela
Goodness of fit
BMD
BMDL
BMDU
Basis for model
P-value
AIC
(mg/L)
(mg/L)
(mg/L)
selection
Dichotomous
Hill
0.7370213
230.47434
2.175642
0
85.665865
The Log-Logistic model
is selected based on
lowest AIC.
Gamma
0.6213002
229.26394
13.68791
6.54628
Infinity
Log-Logistic a
0.6269402
229.24610
13.37256
6.275856
Infinity
Multistage 3°
0.6213117
229.26394
13.68903
6.545976
Infinity
Multistage 2°
0.6213117
229.26394
13.68903
6.545497
Infinity
Multistage 1°
(Quantal
0.6213133
229.26394
13.68924
6.545934
Infinity
Weibull
0.6213117
229.26394
13.68904
6.546292
Infinity
Logistic
0.5383909
229.56539
19.87295
12.25974
Infinity
Log-Probit
0.774706
230.44280
0.192445
0
Infinity
Probit
0.5484738
229.52492
18.94939
11.37275
Infinity
a Scaled residuals for selected Log-Logistic model for doses 0, 15, 30 and 62 mg/L were -0.4312, 0.8051, 0.08788 and -
0.3033, respectively.
Page 46 of 244
-------
0.2
Estimated Probability
Response at BMD
O Data
O BMD
BMDL
0.02 T
0
0 10 20 30 40 50 60
Cmax (mg/L)
Figure 2.7-1 Post-Implantation Loss (Incidence) vs. Cmax (Saillenfait et al. (2003)) - Log-Logistic
Model with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
BMR = 1% RD
USER INPUT
Info
Model
frequentist Log-Logistic vl. 1
Dose-Response Model
P[dose] = g+(l-g)/[l+exp(-a-b*Log(dose))]
Model Options
Risk Type
Extra Risk
BMR
0.01
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
Cmax (mg/L)
Independent Variable
Post-Implantation Loss
Total # of Observations
4
MODEL RESULTS
Benchmark Dose
BMD
13.37255834
BMDL
6.275856421
BMDU
Infinity
AIC
229.2461007
P-value
0.626940156
D.O.F.
2
Chi2
0.933808377
0.18
0.16
£ 0.14
0.12
^ 0.08
£ 0.06
0.04
Page 47 of 244
-------
Model Parameters
# of Parameters
4
Variable
Estimate
8
0.033531344
a
-7.188324572
b
Bounded
Goodness of Fit
Dose
Estimated Probability
Expected
Observed
Size
Scaled Residual
0
0.033531344
6.536681453
5.45293497
194.9424
-0.431177
15
0.044359046
5.178190759
6.96916961
116.7336
0.8051078
30
0.054946823
6.870674643
7.09459598
125.0423
0.0878755
62
0.076768071
10.21092153
9.27976787
133.01
-0.303273
Analysis of Deviance
Model
Log Likelihood
# of Parameters
Deviance
Test d.f.
P-value
Full Model
-112.1799236
4
-
-
-
Fitted Model
-112.6230503
2
0.88625345
2
0.6420258
Reduced Model
-114.0109142
1
3.66198111
3
0.3003533
Page 48 of 244
-------
2.8 Post-implantation Losses: Results for Saillenfait et al. (2003) using
AUC
Table 2-20 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in
Rats Exposed to NMP via Inhalation Using AUC as the Dose Metric (Saillenfait et al. (2003))
3MR =1% Relative Deviation (RD)
Modela
Goodness of fit
BMD
BMDL
BMDU
Basis for model
P-value
AIC
(hr mg/L)
(hr mg/L)
(hr mg/L)
selection
Dichotomous
Hill
0.735317008
230.4759003
22.2922271
0
921.8672575
The Log-Logistic
model is selected
based on lowest
Gamma
0.611906143
229.2931847
147.6424963
70.26024961
Infinity
AIC.
Log-Logistica
0.617393517
229.275429
144.2205252
67.33913159
Infinity
Multistage 3°
0.611906294
229.2931847
147.641914
70.25803339
Infinity
Multistage 2°
0.611906294
229.2931847
147.6420727
70.25444987
Infinity
Multistage 1°
(Quantal
0.611916694
229.2931847
147.6569874
70.25519203
Infinity
Weibull
0.611906298
229.2931847
147.6419509
70.25999005
Infinity
Logistic
0.531714218
229.590634
214.0066138
131.5210445
Infinity
Log-Probit
0.773895796
230.4434248
1.807927221
0
Infinity
Probit
0.541437038
229.5508735
204.1141883
122.0150256
Infinity
a Scaled residuals for selected Log-Logistic model for doses 0, 156.5, 319 and 660.8 mg/L were -0.445, 0.816, 0.0951 and -
0.3031, respectively.
Page 49 of 244
-------
0.2
0.18
Estimated Probability
Response at BMD
O Data
BMD
BMDL
0
0 100 200 300 400 500 600
AUC (hr mg/L)
Figure 2.8-1 Post-Implantation Loss (Incidence) vs. AUC (Saillenfait et al. (2003)) - Log-Logistic
Model with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
BMR = 1% RD
USER INPUT
Info
Model
frequentist Log-Logistic vl. 1
Dose-Response Model
P[dose] = g+(l-g)/[l+exp(-a-b*Log(dose))]
Model Options
Risk Type
Extra Risk
BMR
0.01
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
AUC (hr mg/L)
Independent Variable
Post-Implantation Loss
Total # of Observations
4
MODEL RESULTS
Benchmark Dose
BMD
144.2205252
BMDL
67.33913159
BMDU
Infinity
AIC
229.275429
P-value
0.617393517
D.O.F.
2
Chi2
0.964497336
Page 50 of 244
-------
Model Parameters
# of Parameters
4
Variable
Estimate
8
0.033730478
a
-9.566463403
b
Bounded
Goodness of Fit
Dose
Estimated
Probability
Expected
Observed
Size
Scaled Residual
0
0.033730478
6.575501106
5.45293497
194.9424
-0.445347
156.5
0.044187105
5.158119485
6.96916961
116.7336
0.8156396
319
0.054802059
6.852573038
7.09459598
125.0423
0.0950973
660.8
0.076763207
10.21027468
9.27976787
133.01
-0.303071
Analysis of Deviance
Model
Log Likelihood
# of Parameters
Deviance
Test
d.f.
P Value
Full Model
-112.1799236
4
-
-
-
Fitted Model
-112.6377145
2
0.915581768
2
0.632679766
Reduced
Model
-114.0109142
1
3.66198111
3
0.3003533
Page 51 of 244
-------
2.9 Post-implantation Losses: Results for Saillenfait et al. (2003; 2002)
combined using C max
Table 2-21 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in
Rats Exposed to NMP via Gavage or Inhalation Using Cmax as the Dose Metric (Saillenfait et al.
(2003: 2002))
3MR =1% Relative Deviation (RD)
Modela
Goodness of fit
BMD
(mg/L)
BMDL
(mg/L)
BMDU
(mg/L)
Basis for model
selection
P-value
AIC
Dichotomous Hill
0.2853
520.3039
453.0362
409.9375
507.9537
BMDL estimates from
adequately fitting
models are sufficiently
close (within 3-fold).
Per EPA BMD
Technical Guidance
endpoints (U.S. EPA
(2012)). the Loe-Probit
model is selected based
on it resulting in the
lowest AIC from among
appropriate and
adequately fitting
models.
Gamma
0.00901
530.8042
370.8499
351.8593
390.4129
Log-Logistic
0.4135
518.3039
452.829
409.7989
507.9488
Multistage Degree 7b
0.3437
518.0253
377.3410
113.1514
393.1504
Multistage Degree 6
0.09912
522.4614
338.3081
142.9029
353.8167
Multistage Degree 5
0.00771
530.7404
292.7263
160.1728
307.7792
Multistage Degree 4
0.0001
543.8711
238.1524
153.6198
252.7783
Multistage Degree 3
<0.0001
563.1458
171.9726
122.7886
185.5231
Multistage Degree 2
<0.0001
590.8305
93.54928
71.63636
104.6315
Multistage Degree 1
<0.0001
631.9849
17.7798
14.59179
22.1200
Weibull
0.4129
518.3090
422.892
364.4164
500.5706
Logistic
<0.0001
576.8098
58.3057
49.52224
69.1827
Log-Probit
0.4136
518.3036
471.6574
436.584
514.4843
Probit
<0.0001
586.5139
47.8555
40.97493
56.2891
a Scaled residuals for selected Log-Probit model for doses 0, 15, 30, 62, 120, 250, 531, and 831 mg/L were -1.43, 0.35,
0.21, 0.89, 1.36, -0.41, -0.003, and 0.003, respectively.
b The analysis of the eight dose groups associated with the combined dose response data from the two Saillenfait et al.
studies (2003; 2002) dresents a uniaue situation for the Multistage model that reauires consideration. The default number
of Multistage model degrees runinBMDS 3.1.2 is n-1, where nis the number of dose-groups in the dataset. Thus, in this
case, the 1st degree through 7th degree Multistage models were run. Consideration needs to be given as to whether that
many Multistage degrees are necessary and appropriate for the dataset being evaluated. Of the Multistage models, the 7th
degree Multistage provides an adequate fit to the data that is similar to the model fit achieved by some non-Multistage
models, but its BMDL estimate is nearly 4-fold lower. The Multistage degree 7 BMDL is lower because it contains
several extra parameters ((teta coefficients for degrees 1 through 6). These parameters contribute to the BMDL estimation
but are restricted at the 0 boundary criteria for the purposes of the maximum likelihood, BMD estimation. Thus, while the
BMD estimates (377 mg/L Cmax) of the 7th degree Multistage model are similar to adequately fitting non-Multistage
models (423-472 mg/L Cmax). its BMDL estimates are nearly 4-fold lower (113 mg/L Cmax versus 364-437 mg/L Cmax for
non-Multistage models). Hence, it appears that the extra parameters in the higher degree Multistage models are solely
driving the derivation of the lower BMDLs for these models. In situations where BMDLs vary substantially (i.e., >3-fold),
EPA BMD Technical Guidance (U.S. EPA (2012)) states that "c\Dcrt statistical iudement mav liclo at this doint to iudee
whether model uncertainty is too great to rely on some or all of the results." In this case, given that trend tests of the
combined dataset indicate a lack of linear dose-response trend in the low dose region up to and including 531 mg/L Cmax,
EPA's judgment is that the Multistage 7 model is not appropriate for the derivation of a BMDL from this dataset, despite
its adequate statistical fit (p-value > 0.1) to the data.
Page 52 of 244
-------
Estimated Probability
Response at BMD
O Data
BMD
BMDL
Dose
Figure 2.9-1 Plot of Response by Dose, with Fitted Curve for Selected Log-Probit Model for Post-
implantation Loss in Rats Exposed to NMP via Gavage or Inhalation (Saillenfait et al. (2003;
2002))
BMR = 1% RD; Dose shown is Cmax in units of mg/L
USER INPUT
Info
Model
frequentist Log-Probit vl. 1
Dose-Response Model
P[dose] = g+(l-g) * CumNorm(a+b*Log(Dose))
Model Options
Risk Type
Extra Risk
BMR
0.01
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
Cmax (mg/L)
Independent Variable
Post-Implantation Loss
Total # of Observations
8
MODEL RESULTS
Benchmark Dose
BMD
471.6573999
BMDL
436.5840183
BMDU
514.484334
AIC
518.3035838
P-value
0.413581098
D.O.F.
5
Chi2
5.018874228
Page 53 of 244
-------
Mode
Parameters
# of Parameters
3
Variable
Estimate
8
0.052551843
a
-44.61801666
b
6.869709488
Goodness of Fit
Dose
Estimated
Probability
Expected
Observed
Size
Scaled
Residual
0
0.052551843
17.29705468
11.50708201
329.1426844
-1.430252
15
0.052551843
6.134565349
6.96916961
116.7336
0.3461872
30
0.052551843
6.57120091
7.09459598
125.0423
0.2097633
62
0.052551843
6.989920959
9.27976787
133.01
0.8898004
120
0.052551843
6.166501106
9.45162137
117.3413
1.3591085
250
0.052551843
8.059966818
6.91902519
153.3717
-0.412875
531
0.11436183
13.88633778
13.8770954
121.4246
-0.002635
831
0.944242254
53.86356803
53.8695903
57.04423
0.003475
Analysis of Deviance
Model
Log Likelihood
# of Parameters
Deviance
Test
d.f.
P Value
Full Model
-253.6691293
8
-
-
-
Fitted Model
-256.1517919
3
4.965325251
5
0.420126602
Reduced
Model
-382.8277672
1
258.317276
8
<0.0001
Page 54 of 244
-------
2.10 Post-implantation Losses: Results for Saillenfait et al. (2003; 2002)
combined using AUC
Table 2-22 Model Predictions for Post-implantation Losses (Resorptions and Fetal Mortality) in
Rats Exposed to NMP via Gavage or Inhalation Using AUC as the Dose Metric (Saillenfait et al.
(2003: 2002))
3MR =1% Relative Deviation (RD)
Model
Goodness of fit
P-value
AIC
BMD
(hr mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Dichotomous Hill
0.2853866
520.3036
4981.221
4279.555
Infinity
Gamma
0.0306728
526.7632
4025.814
3810.557
4244.9365
Log-Logistic
0.4134005
518.3039
4771.014
4283.133
5408.5029
Multistage Degree 7
0.4748693
516.7767
4153.835
1001.527
4333.9282
Multistage Degree 6
0.2155793
519.7480
3710.335
1364.709
3884.9435
Multistage Degree 5
0.0280884
526.6318
3198.668
1634.376
3366.2372
Multistage Degree 4
0.000571
538.7165114
2591.817
1622.292
2753.2022
Multistage Degree 3
<0.0001
557.4245223
1860.889
1309.782
2008.5784
The Log-Probit
model is
selected based
on it resulting in
the lowest AIC
from among
appropriate and
adequately
fitting models.
Multistage Degree 2
<0.0001
585.0614255
1001.789
762.5566
1120.9533
Multistage Degree 1
<0.0001
626.9736975
185.8796
152.5732
231.13825
Weibull
0.4128887
518.3089987
4430.582
3768.154
5323.8529
Logistic
<0.0001
571.0790154
621.6337
528.0541
737.29755
Log-Probit
0.4136068
518.3036
4988.572
4585.262
5483.0131
Probit
<0.0001
580.4845
509.6028
436.2995
599.34703
a Scaled residuals for selected Log-Probit model for doses 0, 156.5, 319, 660.8, 1144, 2503, 5674, and 9231 mg/L were -
1.43, 0.35, 0.21, 0.89, 1.36, -0.41, 1.7E-6, and -4.7E-6, respectively.
b The analysis of the eight dose groups associated with the combined dose response data from the two Saillenfait et al.
studies (2003: 2002) presents a unique situation for the Multistage model that requires consideration. The default number
of Multistage model degrees run in BMDS 3.1.2 is n-1, where n is the number of dose-groups in the dataset. Thus, in this
case, the 1st degree through 7th degree Multistage models were run. Consideration needs to be given as to whether that
many Multistage degrees are necessary and appropriate for the dataset being evaluated. Of the Multistage models, the 6th
and 7th degree Multistage models provide an adequate fit to the data that is similar to the model fit achieved by some non-
Multistage models, but BMDL estimates are 3- to 4-fold lower. The Multistage degree 6 and 7 BMDLs are lower because
they contain several extra parameters ((teta coefficients for degrees 1 through 6). These parameters contribute to the
BMDL estimation but are restricted at the 0 boundary criteria for the purposes of the maximum likelihood, BMD
estimation. Thus, while the BMD estimates (3710 - 4154 hr mg/L AUC) of the 6th and 7th degree Multistage models are
similar to adequately fitting non-Multistage models (4431 - 4989 hr mg/L AUC), BMDL estimates for Multistage models
are 3- to 4-fold lower than non-Multistage models (1002 - 1365 hr mg/L AUC versus 3768 - 4585 mg/L AUC for non-
Multistage models). Hence, it appears that the extra parameters in the higher degree Multistage models are solely driving
the derivation of the lower BMDLs for these models. In situations where BMDLs vary substantially (i.e., >3-fold), EPA
BMD Technical Guidance (U.S. EPA (2012)) states that "expert statistical judgment may help at this point to judge
whether model uncertainty is too great to rely on some or all of the results." In this case, given that trend tests of the
combined dataset indicate a lack of linear dose-response trend in the low dose region, EPA's judgment is that the
Multistage 6 and 7 models are not appropriate for the derivation of a BMDL from this dataset, despite the models
adequate statistical fit (p-value > 0.1) to the data.
Page 55 of 244
-------
1
0.9
0.8
0.7
| 0.6
I 0.5
CO
£ °-4
0.3
0.2
0.
ft
Estimated Probability
Response at BMD
O Data
BMD
BMDL
1000
2000
3000
6000
7000
8000 9000
4000 5000
Dose
Figure 2.10-1 Plot of Response by Dose, with Fitted Curve for Selected Log-Probit Model for Post-
implantation Loss in Rats Exposed to NMP via Gavage or Inhalation (Saillenfait et al. 2003;
2002))
BMR = 1% Relative Deviation; Dose shown is AUC in units of hr mg/L
USER INPUT
Info
Model
frequentist Log-Probit vl. 1
Dose-Response Model
P[dose] = g+(l-g) * CumNorm(a+b*Log(Dose))
Model Options
Risk Type
Extra Risk
BMR
0.01
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
AUC (mg/L)
Independent Variable
Post-Implantation Loss
Total # of Observations
8
MODEL RESULTS
Benchmark Dose
BMD
4988.571582
BMDL
4585.262231
BMDU
5483.013135
AIC
518.3035647
P-value
0.413606752
D.O.F.
5
Chi2
5.018663223
Page 56 of 244
-------
Model
'arameters
# of Parameters
8
Variable
Estimate
8
0.05255397
a
-56.20158759
b
6.327168701
Goodness of Fit
Dose
Estimated
Probability
Expected
Observed
Size
Scaled
Residual
0
0.05255397
17.29775491
11.50708201
329.1426844
-1.430398
156.5
0.05255397
6.134813689
6.96916961
116.7336
0.3460775
319
0.05255397
6.571466926
7.09459598
125.0423
0.2096527
660.8
0.05255397
6.990203926
9.27976787
133.01
0.8896734
1144
0.05255397
6.16675074
9.45162137
117.3413
1.3589793
2503
0.05255397
8.060293103
6.91902519
153.3717
-0.412986
5674
0.114285666
13.87708957
13.8770954
121.4246
1.666E-06
9231
0.944347968
53.86959837
53.8695903
57.04423
-4.67E-06
Page 57 of 244
-------
3 Benchmark Dose Modeling of Fetal and Pup Body Weight Changes
BMD modeling for fetal and pup body weight changes was performed using USEPA's BMD Software
package version 3.1.2 (BMDS 3.1.2), in a manner consistent with BMD technical guidance (U.S. EPA
(2012)).
The DuPont (1990). Becci et al. (1982). Saillenfait et al. (2002). and Saillenfait et al. (2003) studies
were selected for dose-response analysis. Individual fetal and pup data were not available for these
studies. Thus, the reported litter means and standard deviations (SDs) applying to the litter level data
were modeled. The data tables in the source reports were not explicit about types of means presented for
pup weight, however, the methods section of Saillenfait et al. (2003; 2002) indicated that analyses were
performed on a per litter basis supporting modeling in this manner. Further details on the analysis
method are provided in Appendix A.
The dose-response data for fetal weight change reported in the dermal study conducted by Becci et al.
(1982) was not amenable to BMD modeling as mean body weight increased gradually from the control
to the middle dose group and then decreased significantly at the high dose group (see Table 3-1). This
dose-response pattern is essentially equivalent to one where only the highest dose has a response and
thus the model estimates of the parameters and BMDs would not be reliable. Hence the NOAEL was
used to derive a POD from the Becci et al. (1982) study.
EPA considered combing data from the Saillenfait et al. (2002) oral and Saillenfait et al. (2003)
inhalation studies to provide a more extensive characterization of the dose-response curve across
exposure routes. However, the Saillenfait et al. (2003) inhalation study observed a statistically
significant decrease in fetal body weights at an internal dose that corresponds to an oral dose lower than
the NOAEL in the Saillenfait et al. (2002) oral study. This implies that fetal body weights were more
sensitive to inhalation exposures and this was not fully accounted for in the PBPK model. Therefore,
datasets from the two studies were not combined for this endpoint.
Benchmark dose modeling was conducted using U.S. EPA BMD Software version 3.1.2 (BMDS 3.1.2)
in accordance with EPA BMD Technical Guidance (U.S. EPA (2012)). Mean fetal and pup body weight
was evaluated with standard continuous response models available in BMDS 3.1.2. Standard continuous
models and model options used for evaluating mean fetal and pup body weight are listed below. Since
adequate model fits to the mean were achieved for continuous models in all cases for the standard model
suite, no non-standard modeling was conducted.
Standard Continuous BMDS 3.1.2 Models Applied to Mean Fetal Body Weight
• Exponential 2 (Exp2)-restricted
• Exponential 3 (Exp3)-restricted
• Exponential 4 (Exp4)-restricted
• Exponential 5 (Exp5)-restricted
• Hill (Hil)-restricted
• Polynomial Degree 4 (Ply4)-restricted
• Polynomial Degree 3 (Ply3)-restricted
• Polynomial Degree 2 (Ply2)-restricted
• Power (Pow)-restricted
• Linear (Lin)
Page 58 of 244
-------
Model Options Used for Continuous Response
• Benchmark Response (BMR): 5% Relative Deviation for Fetal Body Weight
• Response Distribution-Variance Assumptions
o Normal Distribution-Constant Variance
o Normal Distribution-Non-Constant Variance
o Lognormal Distribution, which assumes Constant Variance (if normal distribution models do not
fit means)
• Confidence Level: 0.95
• Background: Estimated
A BMR of 5% relative deviation (RD) from control mean was applied in modeling pup body weight
changes under the assumption that it represents a minimal biologically significant response. In adults, a
10% decrease in body weight in animals is generally recognized as a biologically significant response
associated with identifying a maximum tolerated dose. During development, however, identification of a
smaller (5%) decrease in body weight is consistent with the assumptions that development represents a
susceptible lifestage and that the developing animal is more adversely affected by a decrease in body
weight than the adult. In humans, reduced birth weight is associated with numerous adverse health
outcomes, including increased risk of infant mortality as well as heart disease and type II diabetes in
adults (Barker (2007; Reyes and Manalich (2005)). The selection of a 5% BMR is additionally
supported by data from Kavlock et al. (1995). which found that a BMR of 5% RD for fetal weight
reduction was statistically similar to several other BMR measurements as well as to statistically-derived
NOAEL values. For these reasons, a BMR of 5% RD was selected for decreased pup weight.
Daily AUC for NMP in blood, averaged over the exposure period until the day of measurement (e.g.,
GD 6-20 for Becci et al. (1982) or GD 5-21 for Saillenfait et al. (2002)). was used as the dose metric for
this endpoint. The doses and response data from Saillenfait et al. (2003; 2002) and DuPont (1990) used
for BMD modeling are presented in Table 3-1.
Page 59 of 244
-------
Table 3-1 Fetal Body Weight Data Selecl
ed for Dose-Response Modeling for NMP
Reference
Dose
AUC (hr mg/L)
Number of
litters
Fetal body weight (g)
Mean ± SD
Saillenfait et al.
(2003)
0
24
5.671 ±0.37
156.2
20
5.623 ±0.36
318.3
19
5.469 ±0.25
665.5
25
5.393 ±0.45
Saillenfait et al.
(2002)
0
21
5.73 ±0.5
1145
21
5.59 ±0.22
2504
24
5.18 ±0.35
5673
25
4.02 ±0.21
9228
8
3.01 ±0.39
DuPont (1990)
0
39
7.48 ±0.701
51
16
7.03 ±0.705
268
15
7.13 ±0.695
633
22
6.66 ±0.616
Becci et al. (1982)
0
24
3.45 ±0.20
561
22
3.49 ±0.24
2052
23
3.54 ±0.29
7986
22
2.83 ±0.39
For each dataset-specific BMD analysis, a single preferred model was chosen from the standard set of
models and modeling options listed above. The modeling restrictions and the model selection criteria
facilitated in BMDS 3.1.2 and defined in the BMDS 3.1.2 User Guide were applied in accordance with
EPA BMD Technical Guidance (U.S. EPA (2012)). Briefly, for each dataset, BMDS models with
standard restrictions were fitted to the data using the maximum likelihood method. For continuous
models applied to the fetal weight endpoint, model fit was assessed by a series of tests as follows. For
each model, first the homogeneity of the variances was tested using a likelihood ratio test (BMDS Test
2). If Test 2 was not rejected (%2 p-value > 0.05), the model was fitted to the data assuming constant
variance. If Test 2 was rejected (%2 p-value < 0.05), the variance was modeled as a power function of the
mean, and the variance model was tested for adequacy of fit using a likelihood ratio test (BMDS Test 3).
For fitting models using either constant variance or modeled variance, models for the mean response
were tested for adequacy of fit using a likelihood ratio test (BMDS Test 4, with yl p-value <0.10
indicating inadequate fit). Additional factors were also used to assess the model fit, such as scaled
residuals, visual fit, and adequacy of fit in the low-dose region and in the vicinity of the BMR.
With respect to the continuous model distribution-variance modeling options, responses were first assumed
to be normally distributed with constant variance across dose groups. If no model achieved adequate fit to
response means (BMDS Test 4 p>0.1) and response variances (BMDS Test 2 p>0.05) under those
Page 60 of 244
-------
assumptions, models that assume normal distribution with non-constant variance, variance modeled as a
power function of the dose group mean were considered (U.S. EPA (2012)). If no normal distribution
model achieved adequate fit to response means under the non-constant variance assumption (BMDS Test 3
p>0.05), models that assume lognormal distribution with constant variance were considered and the same
approach for evaluating model fit for mean and variance used for the normal distribution data was applied.
A comparison of model fits obtained for each data set of fetal/pup body weight changes is provided in
each section. The best-fitting models, based on the criteria described above, are indicated in bold. For
each of the best fitting models in Sections 3.1-3.3, subsequent tables and figures show the model version
number, model form, benchmark dose calculation, parameter estimates and estimated values.
PODs identified for fetal body weight in each of the studies evaluated here are summarized in Table 3-2.
Table 3-2. Summary of Recommended BMP and BMDL Values for Fetal Weight.
Section
Response
Selected Model a
BMDspct
(hr mg/L)
BMDLspct
(hr mg/L)
3.1
Saillenfait et al. (2003)
Exp 3
654
414
3.2
Saillenfait et al. (2002)
Exp 3 b
1400
981
3.3
DuPont (1990)
Exp 3
315
223
N/A
Becci et al. (1982)
No model recommended.
NOAEL = 2,052
N/A
N/A
a Since standard models gave adequate results for all endpoints, non-standard models were not considered. Since fits to the
means were obtained using normal distribution models, lognormal models were not applied.
b For Saillenfait et al. (2002), the BMD and BMDL reported are from modeling the data with all the SDs eaual to the
maximum SD across the groups.
Page 61 of 244
-------
3.1 Results for Saillenfait et al. (2003) using AUC
Individual fetal data were not available for the Saillenfait et al. (2003) inhalation study. Thus, the
reported litter means and standard deviations applying to the litter level data were modeled. The tables
in the source report were not explicit about types of means presented for pup weight, however, the
paper's methods section indicated that analyses were performed on a per litter basis supporting modeling
in this manner. Additional details on the analysis method are provided in Appendix A (Method 2).
Table 3-3. Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Inhalation Using
Daily Average AUC as the Dose Metric (Saillenfait et al. (2003))
Model a'b
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.733
78.008
654
414
1543
Exponential model 3 was
selected based on lowest
AIC among adequately
fitting models (Test 4 P-
value > 0.1).
Exponential 3
0.733
78.008
654
414
1543
Exponential 4
0.431
80.008
654
215
1543
Polynomial 3°
0.726
78.028
657
422
1528
Polynomial 2°
0.726
78.028
657
422
1528
Power
0.726
78.028
657
422
1528
Linear
0.726
78.028
657
422
1528
a Constant variance case presented (BMDS Test 2 /j-value = 0.074), selected model in bold; scaled residuals for
selected model for doses 0, 158, 323 and 668 hr mg/L were 0.08, 0.329, -0.68 and 0.22, respectively.
b Exponential 5 and Hill models were not fit to the dataset because these models are overparameterized according to
model selection criteria (i.e., same number of parameters as dose groups).
5.9
5.8
!§> 5.7
(V
1 5.6
¦o
o
cq 5.5
re
2 5.4
S 5.3
2
5.2
5.1
") •
Estimated Probability
Response at BMD
O Data
BMD
BMDL
100
200
300 400
AUC (hr mg/L)
500
600
Figure 3.1-1 Plot of Mean Response by Dose, with Fitted Curve for Selected Exponential 3 Model
for Fetal Body Weight in Rats Exposed to NMP via Inhalation (Saillenfait et al. (2003))
BMR = 5% Relative Deviation
Page 62 of 244
-------
USER INPUT
Info
Model
frequentist Exponential degree 3 vl.l
Dose-Response Model
M[dose] = a * exp(±l * (b * dose)Ad)
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
654.2564991
BMDL
414.2823399
BMDU
1543.192782
AIC
78.00786013
Test 4 P-value
0.732911552
D.O.F.
2
Model Parameters
# of Parameters
4
Variable
Estimate
a
5.665131549
b
7.83994E-05
d
Bounded
log-alpha
-2.019605929
Page 63 of 244
-------
Goodness of Fit
Dose
Size
Observed Mean
Estimated SD
Calc'd
SD
Observed SD
Scaled
Residual
0
24
5.671
0.36429075
0.37
0.37
0.078918885
156
20
5.623
0.36429075
0.36
0.36
0.329255934
318
19
5.469
0.36429075
0.25
0.25
-0.676169981
666
25
5.393
0.36429075
0.45
0.45
0.217779171
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-35.69319981
5
81.3863996
A2
-32.2216643
8
80.4433286
A3
-35.69319981
5
81.3863996
fitted
-36.00393006
3
78.00786013
R
-39.97467922
2
83.9493584
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
15.50602984
6
0.01666578
2
6.943071035
3
0.0737346
3
6.943071035
3
0.0737346
4
0.621460501
2
0.732911552
Page 64 of 244
-------
3.2 Results for Saillenfait et al. (2002) using AUC
Individual fetal data were not available for the Saillenfait et al. (2002) oral study. Thus, the reported
litter means and standard deviations applying to the litter level data were modeled. The tables in the
source report were not explicit about types of means presented for pup weight, however, the paper's
methods section indicated that analyses were performed on a per litter basis supporting modeling in this
manner. Additional details on the analysis method are provided in Appendix A (Method 2).
Mean fetal body weight data reported in Saillenfait et al. (2002) was amenable to BMD modeling,
however, neither constant nor non-constant variance models fit the variances adequately (i.e., the p-
value was <0.05 for Tests 2 and 3). To address the lack of fit of the variance models, a sensitivity
analysis was conducted to determine the influence of the variances on the results. The variances change
haphazardly with dose, with no discernible pattern, so the data were modeled as follows. First, assuming
constant variance, models that adequately fit the means were selected (i.e., Hill and Exponential models
3 and 5; see Table 3-5). Then, assuming constant variance, the data were modeled by replacing the SDs
across all dose groups with the minimum SD observed across all dose groups (Table 3-6). This step was
then repeated by replacing the SDs across all dose groups with the maximum SD observed across all
dose groups (Table 3-7). Finally, the BMDLs were compared for the models selected across the three
cases. BMDLs across the three scenarios did not differ greatly (i.e., by more than threefold), so the
lowest BMDL was selected for use as the POD for this endpoint. The lowest BMDL came from the
maximum SD analysis (Table 3-7). The selected BMD and BMDL are 1402 and 981 hr mg/L,
respectively.
Table 3-4 BMD and BMDL Estimates from the Sensitivity Analysis of Fetal Body Weights
Saillenfait et al. (2002))
Standard
Deviation Case
Selected
Model
Test 4
P-value
BMD
(hr mg/L)
BMDL
(hr mg/L)
Observed
Exp 3
0.386
1400
1100
Minimum
Hill
0.872
1680
1400
Maximum a
Exp 3
0.641
1400
981
a The standard deviation case with the lowest BMDL is bolded and highlighted
in gray.
Page 65 of 244
-------
Table 3-5 Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Gavage Using
Daily Average AUC as the Dose Metric (Saillenfait et al. (2002)); Observed SD case
3MR = 5% Relative Deviation (RD)
Model a'b
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr
mg/L)
Basis for Model Selection
Test 4
P-value
AIC
Exponential 2
0.001
85.305
768
713
831
Only exponential models 3
and 5 and the Hill model
provided an adequate fit to
the means (Test 4 p-value >
0.10). Of these, exponential
model 3 was selected based
on lowest AIC.
Exponential 3
0.3856
72.505
1402
1105
1736
Exponential 4
0.001
85.305
768
713
831
Exponential 5
0.849
72.635
1661
1227
2143
Hill
0.921
72.608
1683
1236
2161
Polynomial 4°
0.029
77.649
1027
897
1259
Polynomial 3°
0.029
77.649
1027
897
1259
Polynomial 2°
0.029
77.649
1027
897
1259
Power
0.068
75.981
1198
922
1518
Linear
0.051
76.363
940
890.2856
998
a Constant variance case presented (BMDS Test 2 p-value < 0.001), selected model in bold; scaled residuals for selected
model for doses 0, 1144, 2503, 5674 and 9231 hr mg/L were -0.54, 0.55, 0.43, -0.79 and 0.69, respectively.
b Model selection was conducted in the context of addressing lack of variance fit and thus ignores the inadequate fit of the
constant variance model.
6.5
6
£ 5.5
op
'•
"O
O
CO 4.5
ra
a>
u_ 4
c
OJ
-------
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
5
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
1402.377226
BMDL
1104.917894
BMDU
1735.983131
AIC
72.5047725
Test 4 P-value
0.385541575
D.O.F.
2
Model Parameters
# of Parameters
4
Variable
Estimate
a
5.76964136
b
8.16174E-05
d
1.370304417
log-alpha
-2.186350781
Goodness of Fit
Dose
Size
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
21
5.73
0.335
0.50
0.50
-0.54
1145
21
5.59
0.335
0.22
0.22
0.55
2504
24
5.18
0.335
0.35
0.35
0.43
5673
25
4.02
0.335
0.21
0.21
-0.80
9228
8
3.01
0.335
0.39
0.39
0.69
Page 67 of 244
-------
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
Al
-31.29928
6
74.59856
A2
-19.79763928
10
59.5952786
A3
-31.29928
6
74.59856
fitted
-32.25238625
4
72.5047725
R
-133.0258433
2
270.051687
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
226.456408
8
<0.001
2
23.00328146
4
<0.001
3
23.00328146
4
<0.001
4
1.906212488
2
0.385541575
Table 3-6 Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Gavage Using
Daily Average AUC as the Dose Metric (Saillenfait et al. (2002)); Minimume SD Case.
3MR = 5% RD; minimum SD among groups used for all groups in analysis
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr
mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 3
0.085
-20.250
1402
1212
1607
The Hill model was
selected based on
lowest AIC.
Exponential 5
0.757
-23.094
1662
1389
1952
Hill
0.872
-23.163
1683
1407
1967
a Constant variance case presented, selected model in bold; only models that provided adequate fit in the observed
SD case were modeled; scaled residuals for selected model for doses 0, 1144, 2503, 5674 and 9231 lir mg/L were
-0.06, 0.12, -0.08, 0.03, and -0.02, respectively.
Table 3-7 Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Gavage Using
Daily Average AUC as the Dose Metric (Saillenfait et al. (2002)); Maximum SD Case.
Modela
Goodness of fit
BMD
(hr mg/L)
BMDL
(hr
mg/L)
BMDU
(hr
mg/L)
Basis for model
selection
Test 4 P-value
AIC
Exponential 3
0.641
147.465
1402
981
1900
Exponential
model 3 was
selected based on
lowest AIC.
Exponential 5
0.897
148.593
1662
1050
2392
Hill
0.946
148.581
1683
1052
2395
a Constant variance case presented, selected model in bold; only models that provided adequate fit in the observed
SD case were modeled; scaled residuals for selected model for doses 0, 1144, 2503, 5674 and 9231 hr mg/L were
-0.06, 0.12, -0.08, 0.03, and -0.02, respectively.
Page 68 of 244
-------
3.3 Results for DuPont, 1990 using AUC
For the DuPont (1990) inhalation study, individual fetal data were not available, but the means and sizes
of the individual litters were. Thus, in addition to modeling the means and standard deviations (SDs) of
litter means, an alternative analysis was attempted in which SD values were adjusted to represent a pup-
based (not litter based) model of fetal body weight. Additional details of this alternative analysis are
provided in Appendix A (Method 1). This analysis should ostensibly yield approximately similar results
as the analysis of the means and SDs of the litter means, provided the variability in the litter weight is
not excessively high. However, in the alternative analysis, neither the constant nor the non-constant
variance models fit the variances adequately (Test 2 and 3 p-value < 0.05), and none of the models fit
the means adequately (Test 4 p-value < 0.10). By contrast, when modeling using the litter level means
and SDs, both variance models fit adequately, and many models fit the means adequately. Modeling
results using the litter level means and SDs are shown below. The BMDLs per model differed only
slightly between the two analyses. Thus, the results from the modeling of means of litter means were
used for DuPont (1990). Exponential model 5 or the Hill model were not fit to the dataset because these
models are overparameterized (same number of parameters as dose groups). Also, the residual of the
low dose group was rather high (-1.72) for all the models, including the selected model. The response at
this dose group was low and appeared to be outside the pattern of the other three groups. Thus, it was
considered an outlier and so was deemed not sufficiently significant to reject the model fit. The selected
BMD and BMDL are 315 and 223 (hr mg/L) respectively.
Table 3-8 Model Predictions for Fetal Body Weights in Rats Exposed to NMP by Inhalation using
Daily Average AUC as the Dose Metric (DuPont (1990))
3MR = 5% Relative Deviation (RD)
Model a'b'c
Goodness of fit
BMD
(hr mg/L)
BMDL
(hr mg/L)
BMDU
(hr mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.139
196.355
315
223
528
Exponential model 3 was
selected based on lowest AIC.
(Exponential model 4 had a
reported BMDL of zero, but
this model was excluded
because it did not fit the data
adequately, Test 4 p-value <
0.10.).
Exponential 3
0.139
196.355
315
223
528
Exponential 4
0.047
196.355
315
0
528
Polynomial 3°
0.138
196.377
323
234
572
Polynomial 2°
0.138
196.377
323
234
555
Power
0.138
196.377
323
234
594
Linear
0.138
196.377
323
234
532
a Non-constant variance case presented (Test 2 p-value = 0.905), selected model in bold; scaled residuals for selected
model for doses 0, 51, 268, and 633 hr mg/L were 0.88, -1.72, 0.35, and 0, respectively.
b Scaled residuals of the low dose group were high (1.72) for all the models, including the selected Exponential 3 model.
The response at the low dose group was low and appeared to be outside the pattern of the other three dose groups. Thus,
the low dose group was considered an outlier and the high scaled residual was deemed not sufficiently significant to
reject the model fit.
Exponential 5 and Hill models were not fit to the dataset because these models are overparameterized according to
model selection criteria (i.e., same number of parameters as dose groups).
Page 69 of 244
-------
Estimated Probability
Response at BMD
O Data
BMD
BMDL
Dose
Figure 3.3-1 Plot of Mean Response by Dose, with Fitted Curve for Selected Exponential 3 Model
for Fetal Body Weight in Rats Exposed to NMP via Inhalation (DuPont (1990))
BMR = 5% RD; Daily Average AUC as Dose Shown in hr mg/L
USER INPUT
Model
frequentist Power vl. 1
Dataset Name
NMP: fetal weight in rats
Dose-Response Model
M[dose] = a * exp(±l * (b * dose)Ad)
Variance Model
Var[i] = alpha * mean[i] A rho
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Non-Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
314.8047273
BMDL
223.1325027
BMDU
528.274145
AIC
196.3549556
Test 4 P-value
0.139323996
D.O.F.
2
Page 70 of 244
-------
Model Parameters
# of Parameters
4
Variable
Estimate
a
7.383675776
b
0.000162937
d
Bounded
log(alpha)
-0.76880135
Goodness of Fit
Dose
Size
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
39
7.48
0.68085857
0.701
0.701
0.883508874
51
16
7.03
0.68085857
0.705
0.705
-1.71884906
268
15
7.13
0.68085857
0.695
0.695
0.351596022
633
22
6.66
0.68085857
0.616
0.616
-0.00059944
Likelihoods of Interest
Model
Log Likelihood
# of Parameters
AIC
A1
-93.20652463
5
196.413049
A2
-92.92594586
8
201.851892
A3
-92.97423292
6
196.413049
fitted
-95.16107509
4
196.354956
R
-103.0646149
2
210.12923
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
20.27733814
6
0.00247147
2
0.561157542
3
0.90526397
3
0.561157542
2
0.90526397
4
3.941906304
2
0.139324
Page 71 of 244
-------
4 Benchmark Dose Modeling of Male Fertility, Female Fecundity, Litter
Size and Pup Death in Exxon, 1991
BMD modeling for reduced male fertility, female fecundity, and reduced litter size described in a 2-
generation reproductive study in rats exposed through diet (Exxon (1991b)) was performed using
USEPA's BMD Software package version 3.1.1 (BMDS 3.1.1) or 2.7 (BMDS 2.7) in a manner
consistent with Benchmark Dose Technical Guidance.
In the Exxon (1991b) study, two generations of both sexes were dosed daily for at least ten weeks prior
to mating and throughout the mating period. Target doses for the exposed groups were 50, 160 and 500
mg/kg-day. Individual litter data reported in Appendices to the Exxon (1991b) report were used for the
determination of dichotomous response incidence and continuous response means and standard
deviations modeled in this report.
The strongest dose-responses for reproductive effects in the Exxon (1991b) study were observed for
reduced Male Fertility Index and Female Fecundity Index in the first (P2/F2A; Table 73 of the Exxon
report) and second (P2/F2B; Table 74 of the Exxon report) litters of the P2 (F1A) 2nd generation parents.
Overall BMD Modeling Approach for Exxon 1991 Data
Benchmark dose software version 3.1.1 (BMDS 3.1.1) was used to analyze male fertility, female
fecundity and litter size. The pup death endpoint was analyzed using BMDS 2.7 because it contains the
larger suite of nested dichotomous models.4 Nested dichotomous models are preferred for this endpoint
because they contain an intra-litter correlation coefficient for the assessment of litter-specific responses.
Only BMDS models that use likelihood optimization and profile likelihood-based confidence intervals
were used in this analysis. All continuous models applied assume normal response distribution. Also, the
benchmark response levels and dose metrics for the analysis are:
1. Fertility and Fecundity for P2/F2A and P2F2B parental rats - estimate BMDs for 10% extra
risk using PBPK estimates of average daily blood concentrations for young (50 g) rat as doses
(four datasets), plus a sensitivity analysis using average daily blood concentrations for 250 g, 350
g and 450 g rats.
2. Litter Size for P2/F2A and P2 F2B - estimate BMDs for 1 SD change from control mean using
PBPK estimates of average daily blood concentrations for young (50 g) rat and GD 6-21 dams as
doses (four datasets)
3. Pup death for P2 F2A and P2 F2B - estimate BMDs for death at Day 0 and by day 4 for 10%.
5% and 1% extra risk using PBPK estimates of average daily blood concentrations for GD 6-21
dam as doses (four datasets)
Standard and non-standard forms of these models- (defined for each endpoint below) were run
separately in BMDS 3.1.1, but EPA model selection procedures (U.S. EPA (2012)) were applied only to
the results of the standard model runs when adequate fit was achieved with any standard model. Since
adequate model fits were obtained in all cases for the standard model suites, no non-standard modeling
results are shown or discussed in this report.
4 BMDS 3.1.1 contains the same NLogistic model, which is preferred because it has received the more extensive QA testing
and is deemed to be the most reliable nested model, but NCTR and RaiVR models are provided as alternatives in this report.
5 The set of standard models are identified in accordance with EPA BMD technical guidance (U.S. EPA (2012)) and are the
default models in BMDS 3.1.1. Non-standard models are the remaining (non-default) models available in BMDS 3.1.1.
Page 72 of 244
-------
Model Restrictions and Model Selection
Restrictions for BMDS 3.1.1 models are defined in the BMDS 3.1.1 User Guide and are applied in
accordance with EPA BMD Technical Guidance (U.S. EPA (2012)). For each BMD analysis, a single
preferred model was chosen from among the preferred standard set of models (noting instances where
consideration of non-standard models may be justified) in accordance with EPA BMD Technical Guidance
(U.S. EPA (2012)). For continuous responses, dose group response standard deviation (SD) was modeled
assuming constant variance across dose groups. If adequate fit (p>0.1) was not achieved for this variance
model a non-constant variance assumption that models SD as a power function of the mean was applied
(U.S. EPA (2012)). Nested dichotomous models were run two ways, with intra-litter correlation (ILC)
coefficients estimated and with ILC coefficients assumed to be zero. Because potential litter-specific
covariates (LSCs) such as dam BW are affected by dose, no appropriate LSC could be determined and
LSCs were not estimated in the BMDS nested dichotomous model runs.
PBPK Analysis for Exxon 1991 Data
Details of the PBPK models for rats and humans are provided in Appendix I of the NMP Risk
Evaluation. The models were developed to describe dosimetry in adult females during pregnancy and so
were slightly adapted to estimate dosimetry in juvenile (post-weaning) rats and adult men.
Because NMP has a relatively short half-life in both rats and humans, exposures only need to be
simulated for several days to a week to determine the internal dosimetry from a consistent exposure
pattern, such as occurs in an animal bioassay or in the workplace (5 day/week). Therefore, adult human
single-day or workplace exposures outside of pregnancy were assumed to be adequately represented by
running the model for the first day or week of pregnancy, when physiological changes are minimal.
Also, physiological differences between men and women were assumed to have minimal impact on the
predicted dosimetry, except that a male-specific body weight (BW) and hand surface area (SA) were
used to estimate dosimetry in men. Changing the BW also affects cardiac output, respiration, and
metabolism, which all scale as BW°75 in the model. Exposures were simulated for a single day
(residential consumer use) or a week (workplace, with 5 d/w exposure) and the average daily area-
under-the-curve (AUC) blood concentration6 was calculated.
For the rat, where pregnancy only lasts 21 days, the model code was modified to allow a user-specified
day for the start of gestation (GSTART), so results for non-pregnant animals could be obtained; i.e.,
with time < GSTART. As for humans, physiological differences between males and females were
assumed to not significantly impact internal dosimetry, hence the non-pregnant female model was used
to simulate male dosimetry. Simulations for post-weaning juvenile animals in the Exxon (1991b)
bioassay were conducted by setting the (initial) BW to 50 g (and for comparison, 250 g, 350 g and 450
g). Because metabolism is scaled as BW°75 in the rats (as well as humans) the internal dose decreases as
BW decreases, so using this BW yields the lowest estimated internal dose for post-weaning rats
(weaning presumed to occur at about this BW). Using this BW in dose-response analysis for fertility and
fecundity provides a lower bound on the internal dose that could give rise to those effects, since they
could result from toxicity at any point in development or during maturity. Target exposure levels (50,
160, and 500 mg/kg/d) were used as exposure levels, exposure was simulated for one-week to go beyond
any initial accumulation and the average blood concentration (Cavg) in the last day of exposure used as
6 Since the 24-hour AUC can vary from day to day, in particular for workplace scenarios, a time-averaged AUC is computed
as AUCavg = AUC(averaging time)*(24 h)/(averaging time), where "averaging time" is typically a week. The average blood
concentration is simply Cavg = AUC(averaging time)/(averaging time). Hence Cavg = AUCavg/(24 h).
Page 73 of 244
-------
internal dose. Food consumption was assumed to occur 12 h/d, at a constant rate over the 12 h to match
the target exposure. Results are given in Table 4-1.
Table 4-1 PBPK-predicted average blood concentrations (Cavg, mg/L) in juvenile rats
Exposure rate
Cavg
Cavg
Cavg
Cavg
(mg/kg/d)
(50 g rat)
(250 g rat)
(350 g rat)
(450 g rat)
0
0
0
0
0
50
13.9
21.1
23.1
24.6
160
48.4
75.2
82.6
88.6
500
181.4
292.6
324.0
349.8
The existing PBPK model does not describe lactational dosimetry, hence the analysis did not include
exposure during that period.
Since effects on litter size and pup viability could result from exposure during gestation, for these
endpoints Cavg in the rat dam over gestation days (GDs) 6-21 days of gestation was estimated. For
simulation of gestation, group-specific mean BW on GD 0 from Table 53 (P2/F2A) and Table 56
(P2/F2B) of the Exxon (1991b) report were used to set the initial BW of the animals. The gestational
BW gain simulated by the model depended on the number of fetuses (NUMFET), an input parameter.
Since group-specific BW values were also given on GD 20 (Tables 53 and 56 of the Exxon report), a
nominal NUMFET was selected for each group to match, as closely as possible, the GD 20 BW value,
though the NUMFET did not necessarily match the average number actually born. This choice was
made since the BW impacts the internal dose, so it was considered most important to match the BW
increase. The dose rates for each exposure group were calculated as the average of measured doses for
days 6-20 from Tables 67 (P2/F2A) and 69 (P2/F2B) of the Exxon (1991b) report. The resulting internal
doses are given in Table 4-2 and 4-3.
Table 4-2 PBPK-predicted average blood concentrations (Cavg, mg/L) during gestation for
P2/F2A
GD 0 BW
(kg)
GD 6-20
Predicted
GD 6-21
Exposure rate
(mg/kg/d)
GD 20 BW (kg)
(# fetuses simulated)
Cavg
(mg/L)
0.3243
52.475
0.4505 (17)
26.12
0.3054
166.75
0.4394 (19)
92.55
0.2815
494.1
0.3872 (14)
326.1
Table 4-3 PBPK-predicted average blood concentrations (Cavg, mg/L) during gestation for
P2/F2B
GD 6-20
Predicted
GD 6-21
GD 0 BW (kg)
Exposure rate
GD 20 BW (kg)
Cavg
(mg/kg/d)
(# fetuses simulated)
(mg/L)
0.3706
49.350
0.5075 (18)
25.25
0.3536
156.70
0.4935 (19)
89.03
0.3187
466.63
0.4188 (12)
311.9
Page 74 of 244
-------
For human workplace and residential exposures, input parameters were specified in Excel spreadsheets.
For workplace exposures, estimated air concentrations were assumed to be constant over each period of
use, but the air concentration, liquid concentration (weight fraction), and duration of use varied between
scenarios. Internal average blood concentrations for varying levels of protective equipment (face mask
and/or gloves with varying protection factors (PFs)) were estimated assuming a five-day work week in
which the exposure was repeated each day followed by two days without exposure. Residential
applications were assumed to occur for a single day and air-concentration time-courses estimated for
each application, along with liquid weight fraction and dermal contact duration specific to each use
scenario. These inputs were read by a model script from Excel spreadsheets. For the analysis of potential
for effect on male fertility, BW and hand surface area (SA) were set to male-specific values. For the
analysis of potential for gestational effect, BW and SA were set to female-specific values. Residential
application evaluated exposure for both adult and teenage women. Model results are written back to the
Excel spreadsheet from which exposure inputs were obtained.
Since human internal doses are calculated as 24-h average AUC values, these must be divided by 24 h
before comparison to Cavg BMD(L) values, or the Cavg BMD(L) values multiplied by 24 h, prior to MOE
calculation.
Page 75 of 244
-------
4.1 Summary of BMD Modeling for Exxon, 1991 Data
Table 4-4 BMP Modeling Summary for Exxon (1991b)
Sec.
Response
Basis for
Internal Dose
Calculations
Selected
Modelb
BMR
BMDC
(mg/L)
BMDLC
(mg/L)
BMDUC
(mg/L)
BMDd
24hr AUC
(h mg/L)
BMDL"
24hr AUC
(h mg/L)
4.2.1
P2/F2A Male Rat Fertility
Young rat (50 g)
Log-Logistic
10% ER
20.5
10.9
81.7
492
262
4.2.2
P2/F2B Male Rat Fertility
Young rat (50 g)1
Log-Logistic
10% ER
14.2
7.64
65.1
341
183
4.2.3
P2/F2A Female Rat Fecundity
Young rat (50 g)
Log-Logistic
10% ER
35.9
16.7
179
862
401
4.2.4
P2/F2B Female Rat Fecundity
Young rat (50 g)
Log-Logistic
10% ER
17.5
8.40
58.4
420
202
4.3.1
P2/F2A Litter Size
Young rat (50 g)
Polynomial 3
1 SD
203
151
715
4872
3624
4.3.2
P2/F2B Litter Size
Young rat (50 g)
Linear
1 SD
153
99.6
332
3672
2390
4.3.3
P2/F2A Litter Sizee
Dam (GD 6-21)
Polynomial 3
1 SD
364
274
1280
8736
6576
4.3.4
P2/F2B Litter Size6
Dam (GD 6-21)
Linear
1 SD
265
172
575
6360
4128
4.4.1
P2/F2A Pup Death at Day 0
(stillborn)
Dam (GD 6-21)
NLogistic - ILC
5% ER
327
205
NC
7848
4920
1% ER
281
49.3
NC
6744
1183
4.4.2
P2/F2B Pup Death at Day 0
(stillborn)
Dam (GD 6-21)
No Model
Selected
5% ER
NA
NA
NA
NA
NA
1% ER
NA
NA
NA
NA
NA
4.4.3
P2/F2A Pup Death by Day 4
Dam (GD 6-21)
No Model
Selected
5% ER
NA
NA
NA
NA
NA
1% ER
NA
NA
NA
NA
NA
4.4.4
P2/F2B Pup Death by Day 4
Dam (GD 6-21)
No Model
Selected
5% ER
NA
NA
NA
NA
NA
1% ER
NA
NA
NA
NA
NA
a BMDL estimates from the selected model (Log-Logistic) for this most sensitive endpoint using internal doses based on 250 g, 350 g and 450 g rats, were 12.1, 13.4
and 14.4 mg/L, respectively.
b As described in Section 4.1, BMDs were derived from the standard set of models as defined in the EPA BMD technical guidance and as identified inBMDS 3.1.1 as
defaults. Since the standard approach gave adequate results for all endpoints, non-standard models were not considered for BMD derivations.
0 BMD, BMDL and BMDU values are in terms of average concentration over 24 hrs and are reported to more than 3 significant figures in the tables in Section 4.2, 4.3
and 4.4. This lias been done to facilitate QC (i.e., replication of the results to a higher number of significant figures gives greater assurance that QA model runs have
been performed using the same modeling options).
d Adjusted BMD and BMDL are in terms of 24-hour AUC blood concentration. These units are directly comparable with BMDLs previously calculated for the NMP
risk evaluation.
e Effects on litter size during gestation are of interest for acute exposure and would therefore be most appropriately evaluated based on maximum concentrations as
opposed to 24 lir average or AUC concentrations shown here.
NC = not calculated; NA = not applicable
Page 76 of 244
-------
4.2 Results of BMD Modeling of P2 Male and Female Fertility Indices
(Exxon, 1991)
The strongest dose-responses for reproductive effects in the Exxon (1991b) study were observed for
reduced Male Fertility Index and Female Fecundity Index in the first (P2/F2A; Table 73 of the Exxon
report) and second (P2/F2B; Table 74 of the Exxon report) litters of the P2 (F1A) 2nd generation parents.
Incidence data for these effects were obtained from Appendices AF (P2/F2A parents) and AG (P2/F2B
parents) of the Exxon (1991b) report. Because BMDS models dichotomous data using dose-response
curves that are increasing in dose-response, the results reported in Appendices AF and AG in terms of
successful impregnations were inverted to obtain incidence data in terms of "number of males
unsuccessful at impregnating any female" per "number of males used for mating" (Males Unsuccessful/
Males Used) and "number of females that did not get pregnant" per "number of females sperm positive
(confirmed mated or confirmed pregnant)" (Females Unsuccessful/Females Mated). These ratios were
derived slightly differently from the Male Fertility and Female Fecundity indices shown in Tables 73
and 74 of the Exxon (1991b) report in that a confirmed pregnancy was counted as "sperm positive"
regardless of whether the mating was "confirmed" (cases where this occurred are identified with
footnotes in the tabular results of this Section).
Because of the existing uncertainty regarding the lifestage "window of toxicity," and the possibility that
reproductive effects of concern could have been associated with early life exposures, the BMD analyses
of potential reproductive effects were performed using PBPK estimates of internal doses that assume an
early lifestage rat body weight of 50 g. A sensitivity analysis was performed on the P2/F2B Male Rat
Fertility to determine the impact of the body weight assumption. As indicated in Footnote 1 of the table
in Section 4.3, BMDL estimates for this most sensitive endpoint increased by less than 2-fold for body
weight assumptions at or below 450 g. The following standard and non-standard dichotomous models
and general modeling options were used to fit fertility incidence data.
Standard Dichotomous Models Applied to Fertility and Fecundity Responses:
• Gamma-restricted
• Log-Logistic-restricted
• Multistage-restricted; from degree = 1 to degree = # dose groups - 1
• Weibull-restricted
• Dichotomous Hill-unrestricted
• Logistic
• Log-Probit-unrestricted
• Probit
Non-Standard Dichotomous Models Applied to Fertility and Fecundity Responses:
• Dichotomous Hill-restricted
• Log-Probit-restricted
• Gamma-unrestricted
• Log-Logistic-unrestricted
• Multistage-unrestricted
• Weibull-unrestricted
Page 77 of 244
-------
General Model Options Used for Fertility and Fecundity Dichotomous Responses:
• Benchmark Response (BMR): 0.1 (10%) Extra Risk
• Confidence Level: 0.95
• Background: Estimated
Page 78 of 244
-------
4.2.1 P2/F2A Male Fertility (Males Unsuccessful/Males Used; Exxon Appendix AF)
mg/L Blood - 50 g Rat
N
Incidence
0
29
2
13.9
29
8
48.4
29
8
181.4
30
16
Table 4-5 Model Predictions for Reduced Male Fertility in P2/F2A Male Rats (Exxon (1991b))
Standard
Models
Restriction b
10% Extra Risk
(mg/L blood - 50 g
Rat)
P-value
AIC
BMDS
Recommends
BMDS Recommendation
Notes
BM
D
BMD
L
BMDU
Gamma
Restricted
28.82
54
18.06
77
106.50
62
0.221224
4
131.36474
26
Viable -
Alternate
Log-
Logistic a
Restricted
20.47
39
10.93
76
81.732
23
0.267407
3
130.87451
55
Recommended
Basis: Lowest BMDL In a > 3-
Fold BMDL Range
Lowest AIC
Multistage
Degree 3
Restricted
28.82
54
18.06
78
109.51
57
0.221224
131.36474
26
Viable -
Alternate
Multistage
Degree 2
Restricted
28.82
54
18.06
75
91.607
10
0.221224
1
131.36474
26
Viable -
Alternate
Multistage
Degree 1
Restricted
28.82
53
18.06
76
56.969
40
0.221223
8
131.36474
26
Viable -
Alternate
Weibull
Restricted
28.82
54
18.06
76
115.14
04
0.221223
9
131.36474
26
Viable -
Alternate
Dichotom
ous Hill
Unrestricted
4.245
66
0.000
24
41.015
37
0.309315
6
131.38255
36
Questionable
BMD/BMDL ratio > 2
BMD/BMDL ratio > 3
BMD 3x lower than lowest non-
zero dose
BMDL lOx lower than lowest
non-zero dose
Logistic
NA
51.42
08
38.19
85
79.828
21
0.162073
5
132.33267
84
Viable -
Alternate
Log-Probit
Unrestricted
4.642
11
0.000
37
37.710
69
0.294224
6
131.45311
68
Questionable
BMD/BMDL ratio > 20
BMD/BMDL ratio > 3
BMD 3x lower than lowest non-
zero dose
BMDL lOx lower than lowest
non-zero dose
Probit
NA
48.86
14
36.41
63
77.278
41
0.166761
4
132.24053
29
Viable -
Alternate
a Selected Model (Gray); residuals for doses 0, 13.9, 48.4, and 181.4 were -0.811610042, 1.353899534, -0.296031585 and -
0.242023672, respectively
bRestrictions defined in the BMDS 3.1.1 User Guide; NA = Not Applicable
Page 79 of 244
-------
BMDS 3.1.1 Standard Model Plots for P2/F2A Male Rat Fertility (Males
Unsuccessful/Males Used) vs NMP Blood Concentration - 50 g Rat (Exxon, 1991;
Appendix AF)
-Frequentist Gamma Estimated
Probability
Frequentist Log-Logistic Estimated
Probability
Frequentist Multistage Degree 3
Estimated Probability
-Frequentist Multistage Degree 2
Estimated Probability
¦Frequentist Multistage Degree 1
Estimated Probability
-Frequentist Weibull Estimated
Probability
-Frequentist Dichotomous Hill
Estimated Probability
-Frequentist Logistic Estimated
Probability
Selected Model - Log-Logistic (Restricted) - Extra Risk, BMR = 0.1
USER INPUT
Info
Model
Log-Logistic vl.O
Dataset Name
P2F2A Male Fertility
Model Options
Risk Type
Extra Risk
BMR
0.1
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
[Dosel
Independent Variable
[Incidence]
Total # of Observations
4
MODEL RESULTS
Benchmark Dose
BMD
20.4738478
BMDL
10.93759459
BMDU
81.7322316
AIC
130.8745155
P-value
0.267407255
D.O.F.
2
Chi2
2.637964966
Page 80 of 244
-------
Model Parameters
# of Parameters
3
Variable
Estimate
B
0.117496501
a
-5.216372932
b
Bounded
Goodness of Fit
Dose
Estimated
Probability
Expected
Observed
Size
Scaled
Residual
0
0.117496501
3.407398541
2
29
-0.81161
13.9
0.17939856
5.202558252
8
29
1.3538995
48.4
0.301079065
8.731292894
8
29
-0.296032
181.4
0.555291468
16.65874405
16
30
-0.242024
Analysis of
deviance
Model
Log
Likelihood
#of
Parameters
Deviance
Test
d.f.
P Value
Full Model
-62.1675397
4
-
-
-
Fitted Model
-63.43725776
2
2.53943612
2
0.2809108
Reduced Model
-70.51432209
1
16.6935648
3
0.0008171
P2/F2A Male Rat Fertility (Males Unsuccessful/Males Used) vs NMP Blood
Concentration - 50 g Rat (Exxon, 1991; Appendix AF) - Log-Logistic Model with
BMR of 10% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the
BMDL
^—Estimated Probability
Response at BMD
O Data
BMD
BMDL
80 100
Dose
Page 81 of 244
-------
4.2.2 P2/F2B Male Fertility (Males Unsuccessful/Males Used; Exxon Appendix AG)
mg/L Blood - 50 g Rat
N
Incidence
0
30
5
13.9
29
9
48.4
30
12
181.4
29
19
Table 4-6 Model Predictions for Reduced Male Fertility in P2/F2B Male Rats (Exxon (1991b))
Standard
Models
Restriction b
10% Extra Risk
(mg/L blood - 50 g Rat)
P-
value
AIC
BMDS
Recommends
BMDS Recommendation Notes
BM
D
BMD
L
BMDU
Gamma
Restricted
21.46
13
13.74
89
76.52064
0.666
6306
145.51839
72
Viable -
Alternate
Log-
Logistic a
Restricted
14.21
25
7.638
24
65.11825
0.824
8283
145.08067
89
Recommended
Basis: Lowest BMDL In a > 3-
Fold BMDL Range
Lowest AIC
Multistage
Degree 3
Restricted
21.46
13
13.74
89
87.34237
0.666
6306
145.51839
72
Viable -
Alternate
Multistage
Degree 2
Restricted
21.46
13
13.74
87
75.00523
0.666
6309
145.51839
72
Viable -
Alternate
Multistage
Degree 1
Restricted
21.46
13
13.74
88
40.46712
0.666
6306
145.51839
72
Viable -
Alternate
Weibull
Restricted
21.46
13
13.74
89
80.30469
0.666
6306
145.51839
72
Viable -
Alternate
Dichotomo
us Hill
Unrestricted
8.677
17
0.171
04
60.82728
0.656
4479
146.89849
18
Questionable
BMD/BMDL ratio > 20
BMDL lOx lower than lowest
non-zero dose
Logistic
NA
36.72
71
27.09
45
56.56066
0.442
6321
146.39715
35
Viable -
Alternate
Log-Probit
Unrestricted
9.269
62
0.241
78
59.56593
0.616
1031
146.95220
17
Questionable
BMD/BMDL ratio > 20
BMDL lOx lower than lowest
non-zero dose
Probit
NA
35.70
14
26.71
57
55.32779
0.453
3689
146.34376
72
Viable -
Alternate
a Selected Model (Gray); residuals for doses 0, 13.9, 48.4 and 181.4 were -0.300662226, 0.518709072, -0.122358174 and -
0.103594189, respectively.
b Restrictions defined in the BMDS 3.1.1 User Guide; NA = Not Applicable
Page 82 of 244
-------
BMDS 3.1.1 Stan (laid Model Plots for P2/F2B Male Rat Fertility (Males
Unsuccessful/Males Used; Appendix AG) vs NMP Blood Concentration - 50 g Rat
(Exxon, 1991)
Frequentist Gamma Estimated
Probability
Frequentist Log-Logistic Estimated
Probability
Frequentist Multistage Degree 3
Estimated Probability
^^—Frequentist Multistage Degree 2
Estimated Probability
¦^^—Frequentist Multistage Degree 1
Estimated Probability
^^—Frequentist Weibull Estimated
Probability
¦^^"Frequentist Dichotomous Hill
Estimated Probability
^^—Frequentist Logistic Estimated
Probability
^^—Frequentist Log-Probit Estimated
Probability
Selected Model - Log-Logistic (Restricted) - Extra Risk, BMR =0.1
USER INPUT
Info
Model
Log-Logistic vl.O
Dataset Name
P2F2B Male Fertility
Model Options
Risk Type
Extra Risk
BMR
0.1
Confidence Level
0.95
Background
Estimated
MODEL RESULTS
Benchmark Dose
BMD
14.21245366
BMDL
7.638241538
BMDU
65.11824629
AIC
145.0806789
P-value
0.824828266
D.O.F.
2
Chi2
0.385160154
Page 83 of 244
-------
Model Parameters
# of Parameters
3
Variable
Estimate
B
0.188119322
a
-4.851343176
b
Bounded
Goodness of Fit
Dose
Estimated
Probability
Expected
Observed
Size
Scaled
Residual
0
0.188119322
5.643579645
5
30
-0.300662
13.9
0.267697459
7.763226311
9
29
0.5187091
48.4
0.410991312
12.32973936
12
30
-0.122358
181.4
0.664257058
19.26345469
19
29
-0.103594
Analysis of Deviance
Log
#of
Test
Model
Likelihood
Parameters
Deviance
d.f.
P-value
Full Model
-70.35048621
4
-
-
-
Fitted Model
-70.54033943
2
0.37970644
2
0.8270805
Reduced Model
-78.43743444
1
16.1738965
3
0.0010446
P2/F2B Male Rat Fertility (Males Unsuccessful/Males Used) vs NMP Blood
Concenti'ation - 50 g Rat (Exxon, 1991; Appendix AG) - Log-Logistic Model
with BMR of 10% Extra Risk for the BMD and 0.95 Lower Confidence Limit
for the BMDL
0.8
0.7
«> 0.6
C/2
n
8. 0.5
C/3
-------
4.2.3 P2/F2A Female Fecundity (Females Unsuccessful/Females Mated; Exxon Appendix
AF)
mg/L Blood - 50 g Rat
N
Incidence
0
29 a
2
13.9
29 b
6
48.4
28
7
181.4
23
9
a Includes 1 presumed mating (JAB 149 with JAB273) that was not
"Confirmed" but resulted in pregnancy of JAB273
b Includes 1 presumed mating (JAB008 with JAB 105) that was not
"Confirmed" but resulted in pregnancy of JAB 105
Table 4-7 Model Predictions for Reduced Fecundity in P2/F2A Female Rats (Exxon (1991b))
Standard
Models
Restriction b
10% Extra Risk
(mg/L blood - 50 g
Rat)
P-value
AIC
BMDS
Recommends
BMDS Recommendation Notes
BM
D
BMD
L
BMDU
Gamma
Restricted
44.96
90
24.27
97
166.87
43
0.410732
8
112.25409
63
Viable -
Alternate
Log-
Logistic a
Restricted
35.85
00
16.70
86
178.83
94
0.464483
7
111.95596
85
Recommended
Basis: Lowest AIC
Multistage
Degree 3
Restricted
44.96
9
24.27
93
152.75
87
0.410732
9
112.25409
63
Viable -
Alternate
Multistage
Degree 2
Restricted
44.96
90
24.27
97
145.56
55
0.410732
8
112.25409
63
Viable -
Alternate
Multistage
Degree 1
Restricted
44.96
90
24.27
94
139.99
63
0.410732
9
112.25409
63
Viable -
Alternate
Weibull
Restricted
44.96
90
24.27
97
176.62
68
0.410732
8
112.25409
63
Viable -
Alternate
Dichotomo
us Hill
Unrestricted
6.584
76
0
78.866
85
NA
114.50099
14
Unusable
BMD computation failed; lower
limit includes 0 BMDL not
estimated
d.f.=0 (Goodness of fit test
cannot be calculated)
Logistic
NA
72.81
42
49.22
49
179.07
43
0.311254
6
112.97438
42
Viable -
Alternate
Log-Probit
Unrestricted
7.047
68
0
74.365
06
0.736000
8
112.51903
46
Unusable
BMD computation failed; lower
limit includes 0 BMDL not
estimated
Probit
NA
69.29
99
46.38
35
174.67
04
0.320756
4
112.89541
63
Viable -
Alternate
a Selected Model (Gray); residuals for doses 0, 13.9, 48.4 and 181.4 were -0.754747582, 0.857664083, 0.263750831 and -
0.398574381, respectively.
b Restrictions defined in the BMDS 3.1.1 User Guide; NA = Not Applicable
Page 85 of 244
-------
BMDS 3.1.1 Standard Model Plots for P2/F2A Female Rat Fecundity (Females
Unsuccessful/Females Mated) vs NMP Blood Concentration - 50g Rat(Exxon, 1991;
Appendix AF) ^—Frequentist Gamma Estimated
Probability
Frequentist Log-Logistic Estimated
Probability
Frequentist Multistage Degree 3
Estimated Probability
Frequentist Multistage Degree 2
Estimated Probability
^^—Frequentist Multistage Degree 1
Estimated Probability
^^—Frequentist Weibull Estimated
Probability
Frequentist Dichotomous Hill
Estimated Probability
^^—Frequentist Logistic Estimated
Probability
Frequentist Log-Probit Estimated
Probability
Frequentist Probit Estimated
Probability
O Data
Selected Model - Log-Logistic - Extra Risk, BMR = 0.1
USER INPUT
Info
Model
Log-Logistic vl.O
Dataset Name
P2F2A Female Fecundity
Model Options
Risk Type
Extra Risk
BMR
0.1
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
mg/L Blood 50 g Rat
Independent Variable
Females Unsuccessful
Total # of Observations
4
MODEL RESULTS
Benchmark
Jose
BMD
35.85003887
BMDL
16.70857886
BMDU
178.8394143
AIC
111.9559685
P-value
0.464483699
D.O.F.
2
Chi2
1.53365763
Page 86 of 244
-------
Model Parameters
# of Parameters
3
Variable
Estimate
B
0.11340654
a
-5.776569229
b
Bounded
Goodness of Fit
Dose
Estimated Probability
Expected
Observed
Size
Scaled Residual
0
0.11340654
3.288789653
2
29
-0.754748
13.9
0.150024089
4.350698589
6
29
0.8576641
48.4
0.22905425
6.41351901
7
28
0.2637508
181.4
0.432477945
9.946992746
9
23
-0.398574
Analysis of Deviance
Model
Log Likelihood
# of Parameters
Deviance
Test d.f.
P Value
Full Model
-53.20227182
4
-
-
-
Fitted Model
-53.97798425
2
1.55142486
2
0.4603757
Reduced Model
-57.45827043
1
8.51199723
3
0.0365346
1
0.9
0.8
0.7
£ 0.6
s
& 0.5
C«
-------
4.2.4 P2/F2B Female Fecundity (Females Unsuccessful/Females Mated; Exxon Appendix
AG)
mg/L Blood - 50 g Rat
N
Incidence
0
27
2
13.9
29 a
9
48.4
28
10
181.4
21 b
11
a Includes 2 presumed matings (JAB194 with JAB279; JAB201 with
JAB293) not "Confirmed" but resulting in pregnancies
bIncludes 1 presumed mating (JAB022 with JAB134) that was not
"Confirmed" but resulted in pregnancy of JAB 134
Table 4-8 Model Predictions for Reduced Fecundity in P2/F2B Female Rats (Exxon (1991b))
Standard
Models
Restriction b
10% Extra Risk
(mg/L blood - 50 g
Rat)
P-
value
AIC
BMDS
Recommends
BMDS Recommendation
Notes
BMD
BMDL
BMDU
Gamma
Restricted
27.75
96
15.948
1
82.142
00
0.134
9299
123.9885415
Viable - Alternate
Log-
Logistic a
Restricted
17.45
28
8.3958
6
58.448
82
0.192
5123
123.0293723
Recommended
Basis: Lowest AIC
Multistage
Degree 3
Restricted
27.75
98
15.948
2
97.117
40
0.134
9306
123.9885415
Viable - Alternate
Multistage
Degree 2
Restricted
27.75
98
15.948
2
87.010
75
0.134
9306
123.9885415
Viable - Alternate
Multistage
Degree 1
Restricted
27.76
19
15.948
3
68.871
17
0.134
946
123.9885416
Viable - Alternate
Weibull
Restricted
27.76
00
15.948
3
84.747
89
0.134
9318
123.9885415
Viable - Alternate
Dichotomo
us Hill
Unrestricted
1.071
72
0
18.132
80
NA
123.9261336
Unusable
BMD computation failed;
lower limit includes 0
BMDL not estimated
BMD lOx lower than lowest
non-zero dose
d.f.=0 (Goodness of fit test
cannot be calculated)
Logistic
NA
49.48
25
34.009
0
100.18
99
0.089
0178
125.2278017
Questionable
Goodness of fit p-value <0.1
Log-Probit
Unrestricted
1.359
20
0
18.120
44
0.660
4573
121.9394443
Unusable
BMD computation failed;
lower limit includes 0
BMDL not estimated
BMD lOx lower than lowest
non-zero dose
Probit
NA
47.44
59
32.803
8
97.343
69
0.091
8383
125.1319918
Questionable
Goodness of fit p-value <0.1
a Selected Model (Gray); residuals for doses 0, 13.9, 48.4 and 181.4 were -0.976071189, 1.341257654, 0.170425804 and -
0.717257235, respectively.
b Restrictions defined in the BMDS 3.1.1 User Guide; NA = Not Applicable
Page 88 of 244
-------
BMDS 3.1.1 Standard Model Plots for P2/F2B Female Rat Fecundity (Females
Unsuccessful/Females Mated) vs NMP Blood Concentration - 50 g Rat (Exxon, 1991;
Appendix AG)
^^—Frequentist Gamma Estimated
Probability
¦ Frequentist Log-Logistic Estimated
Probability
Frequentist Multistage Degree 3
Estimated Probability
^^—Frequentist Multistage Degree 2
Estimated Probability
^^—Frequentist Multistage Degree 1
Estimated Probability
^^—Frequentist Weibull Estimated
Probability
^^—Frequentist Dichotomous Hill
Estimated Probability
^^—Frequentist Logistic Estimated
Robability
^^—Frequentist Log-Probit Estimated
Probability
^^—Frequentist Pi'obit Estimated
Probability
O Data
Selected Model - Log-Logistic (Restricted) - Extra Risk, BMR = 0.1
USER INPUT
Info
Model
Log-Logistic vl.O
Dataset Name
P2F2B Female Fecundity
Model Options
Risk Type
Extra Risk
BMR
0.1
Confidence Level
0.95
Background
Estimated
Model Data
Dependent Variable
[Dosel
Independent Variable
[Incidence!
Total # of Observations
4
Page 89 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
17.45276136
BMDL
8.395858147
BMDU
58.44881649
AIC
123.0293723
P-value
0.192512349
D.O.F.
2
Chi2
3.295189957
Model Parameters
# of Parameters
3
Variable
Estimate
8
0.139072629
a
-5.056722458
b
Bounded
Goodness of Fit
Dose
Estimated
Probability
Expected
Observed
Size
Scaled
Residual
0
0.139072629
3.754960985
2
27
-0.976071
13.9
0.209064738
6.062877397
9
29
1.3412577
48.4
0.341865741
9.572240753
10
28
0.1704258
181.4
0.600472417
12.60992076
11
21
-0.717257
Analysis o
' Deviance
Model
Log Likelihood
# of Parameters
Deviance
Test d.f.
P Value
Full Model
-57.87277378
4
-
-
-
Fitted Model
-59.51468613
2
3.2838247
2
0.1936094
Reduced Model
-64.55874867
1
13.3719498
3
0.0038975
Page 90 of 244
-------
P2/F2B Female Rat Fecundity (Females Unsuccessful/Females Mated) vs NMP
Blood Concentration - 50 g Rat (Exxon, 1991; Appendix AG) - Log-Logistic Model
with BMR of 10% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the
BMDL
1
0.9
0.8
Estimated Probability
Response at BMD
O Data
BMD
BMDL
Dose
Page 91 of 244
-------
4.3 Results of BMD Modeling of P2 Litter (Exxon (1991a))
The next most sensitive dose-related reproductive effect noted in the Exxon (1991b) study, other than
the reduction in male fertility and female fecundity, was the reduction in litter size, which was most
pronounced for the first (F2A) and 2nd (F2B) P2 rat litters. However, the Exxon (1991b) study also
reported a dose-related increase in pup death by postnatal day 4 that was also most pronounced in the
F2A and F2B litters of the P2 parental rats. Thus, the extent to which the reduction in litter size is due to
reproductive effects on the parents or gestational effects on the fetus is not clear, and the Exxon (1991b)
reproductive study design does not allow for a definitive investigation of that question (e.g., the number
of implantations and resorptions were not identified). For these reasons, the litter size reduction effect
was analyzed three ways:
• Model litter size means and SD (live and stillborn pups) using BMDS continuous models
against estimates of internal doses to young (50 g) parental rats (Sections 4.3.1 and 4.3.2).
• Model litter size means and SD (live and stillborn pups) using BMDS continuous models
against estimates of internal doses to P2 maternal rats during GD 6-21 (Sections 4.3.3 and
4.3.4).
• Model pup death at day 0 (stillborn) and by postnatal day 4 per total pups born as incidence
data using BMDS nested dichotomous models against estimates of internal doses to P2
maternal rats during GD 6-21 (Section 4.4).
Individual litter data that allows for the calculation of dose-specific means and standard deviations for
litter size are available in Appendix AJ (for P2/F2A litters) and AK (for P2/FB litters) of the Exxon
(1991b) report.
Standard and nonstandard continuous models (defined below) were used to fit litter size data. BMDs
were estimated for 1 SD change from control mean. Internal doses used for BMD modeling were based
on PBPK estimates of average daily blood concentrations for young (50 g) rat and GD 6-21 dams.
Standard Continuous Models Applied to Litter Size Response:
• Exponential 2-restricted
• Exponential 3-restricted
• Exponential 4-restricted
• Exponential 5-restricted
• Hill-restricted
• Polynomial Degree 3-restricted
• Polynomial Degree 2-restricted
• Power-restricted
• Linear
Non-Standard Continuous Models Applied to Litter Size Response:
• Hill-unrestricted
• Polynomial Degree 3-unrestricted
• Polynomial Degree 2-unrestricted
• Power-unrestricted
Page 92 of 244
-------
General Model Options Used for Litter Size Continuous Response:
• Benchmark Response (BMR): 1 Standard Deviation (SD) Change from Control Mean
• Confidence Level: 0.95
• Background: Estimated
Page 93 of 244
-------
4.3.1 P2/F2A Litter Size - 50 g Rat
(Exxon Appendix A J, "Total Pups Born")
mg/L Blood - 50 g Rat
N
Mean
SD
0
27
15.2592593
3.558225
13.9
23
13.2608696
4.937955
48.4
21
14.9047619
3.871754
181.4
14
11.6428571
3.272429
Table 4-9 Model Predictions for Litter Size in P2/F2A Rats Based on Post-weaning Exposure
(Exxon (1991b))
Standard
Models
Restriction b
BMR = 1 Standard
Deviation (mg/L
blood - 50 g Rat)
P-
value
AIC
BMDS
Recommends
BMDS Recommendation Notes
BM
D
BMD
L
BMDU
Exponential
2 (CV)
Restricted
264.
277
140.4
44
1032.840
0.1317
861
483.41059
57
Viable -
Alternate
BMD higher than maximum dose
Exponential
3 (CV)
Restricted
190.
060
149.0
59
788.7670
0.0625
955
484.82469
12
Questionable
Goodness of fit p-value <0.1
BMD higher than maximum dose
Exponential
4 (CV)
Restricted
264.
120
140.4
42
1032.835
0.1317
865
483.41059
02
Viable -
Alternate
BMD higher than maximum dose
Exponential
5 (CV)
Restricted
190.
171
149.0
60
788.7498
NA
486.82469
61
Questionable
BMD higher than maximum dose
d.f.=0 (Goodness of fit test
cannot be calculated)
Hill (CV)
Restricted
999
9
0
Infinity
0.0625
977
484.82463
33
Unusable
BMD computation failed
BMD not estimated
BMDL not estimated
Goodness of fit p-value <0.1
Polynomial
Degree 3
(CV)a
Restricted
202.
696
150.6
74
714.9564
0.1718
518
482.87969
17
Recommended
Basis: Lowest AIC
BMD higher than maximum dose
Polynomial
Degree 2
(CV)
Restricted
214.
035
148.9
14
757.4027
0.1605
273
483.01602
8
Viable -
Alternate
BMD higher than maximum dose
Power (CV)
Restricted
183.
783
182.1
12
698.8191
0.0625
983
484.82461
5
Questionable
Goodness of fit p-value <0.1
BMD higher than maximum dose
BMDL higher than maximum
dose
Linear (CV)
NA
248.
915
145.0
61
875.6812
0.1364
343
483.34127
Viable -
Alternate
BMD higher than maximum dose
a Selected Model (Gray); Constant variance case presented (Test 2 p-value = 0.24158); scaled residuals for doses 0, 13.9, 48.4
and 181.4 were 0.958706516, -1.509731959, 0.501737513 and -0.010801354, respectively.
b Restrictions defined in the BMDS 3.1.1 User Guide; NA = Not Applicable; CV = Constant Variance Model; NCV = Non-
Constant Variance Model.
Page 94 of 244
-------
BMDS3.1.1 Standard Model Plots for P2/F2A Litter Size (Exxon, 1991; Appendix
AJ, "Total Pups Born") vs NMP Blood Concentration-50 g Rat
50
150
^—Frequentist Exponential Degree 2
Estimated Probability
—Frequentist Exponential Degree 3
Estimated Probability
Frequentist Exponential Degree 4
Estimated Probability
^—Frequentist Exponential Degree 5
Estimated Pr obability
^—Frequentist Hill Estimated Probability
^—Frequentist Polyiromial Degree 3
Estimated Probability
^—Frequentist Polynomial Degree 2
Estimated Probability
^—Frequentist Power Estimated
Probability
^—Frequentist Linear Estimated
Probability
O Data
Selected
USER INPUTS
100
Dose
Model - Polynomial Degree 3 (Restricted) - Extra Risk, BMR = 1 SD
Info
Model
Polynomial degree 3 vl.l
DatasetName
P2F2A Litter Size
Dose-Response Model
M[dose] = g • bl*dose + b2*doseA2 + ...
Model Options
BMR Type
Std. Dev.
BMRF
1
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Model Data
Dependent Variable
[Dose]
Independent Variable
[Response]
Total # of Observations
85
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
202.6960934
BMDL
150.6744181
BMDU
714.956421
AIC
482.8796917
Test 4 P-value
0.171851757
D.O.F.
2
Page 95 of 244
-------
Mode
Parameters
# of Parameters
5
Variable
Estimate
8
14.52128961
bl
Bounded
b2
Bounded
b3
-4.80285E-07
alpha
15.99813687
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Median
Estimated
SD
Calc'd
SD
Observed
DS
Scaled
Residual
0
27
14.52128961
15.2592593
15.2592593
3.9997671
3.558225
3.558225
0.958706516
13.9
23
14.51999975
13.2608696
13.2608696
3.9997671
4.937955
4.937955
-1.50973196
48.4
21
14.466835
14.9047619
14.9047619
3.9997671
3.871754
3.871754
0.501737513
181.4
14
11.6544036
11.6428571
11.6428571
3.9997671
3.272429
3.272429
-0.01080135
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-236.6787228
5
483.357446
A2
-234.583299
8
485.166598
A3
-236.6787228
5
483.357446
fitted
-238.4398459
3
482.879692
R
-241.3113542
2
486.622708
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
13.45611034
6
0.03633832
2
4.190847665
3
0.24157981
3
4.190847665
3
0.24157981
4
3.522246101
2
0.17185176
Page 96 of 244
-------
18
16
14
12
Iu
a 10
o
a,
-------
4.3.2 P2/F2B Litter Size - 50 g Rat
Exxon Appendix AK, "Total Pups Born")
mg/L Blood - 50 g Rat
N
Mean
SD
0
25
15.24
2.947881
13.9
20
14.35
3.422449
48.4
18
14.39
3.972536
181.4
9
11
3.708099
Table 4-10 Model Predictions for Litter Size in P2/F2B Rats Based on Post-weaning Exposure
(Exxon (1991b))
Standard
Models
Restriction b
BMR = 1 Standard
Deviation (mg/L blood
- 50 g Rat)
P-value
AIC
BMDS
Recommends
BMDS Recommendation
Notes
BMD
BMDL
BMDU
Exponenti
al 2 (CV)
Restricted
151.2
11
90.014
4
358.880
7
0.710819
6
385.22188
7
Viable -
Alternate
Exponenti
al 3 (CV)
Restricted
156.9
52
90.562
6
352.685
4
0.435551
2
387.14718
89
Viable -
Alternate
Exponenti
al 4 (CV)
Restricted
151.1
78
90.014
5
358.868
5
0.710823
3
385.22187
65
Viable -
Alternate
Exponenti
al 5 (CV)
Restricted
156.9
62
50.816
4
352.691
NA
389.14720
32
Viable -
Alternate
BMD/BMDL ratio > 3
d.f.=0 (Goodness of fit test
cannot be calculated)
Hill (CV)
Restricted
79.46
42
51.861
2
Infinity
NA
389.31785
9
Questionable
d.f.=0 (Goodness of fit test
cannot be calculated)
Polynomia
1 Degree 3
(CV)
Restricted
162.7
87
100.26
4
324.548
3
0.478185
6
387.04221
2
Viable -
Alternate
Polynomia
1 Degree 2
(CV)
Restricted
159.7
31
100.10
2
326.253
1
0.467703
9
387.06660
93
Viable -
Alternate
Power
(CV)
Restricted
157.0
00
99.763
0
329.895
1
0.446602
9
387.11847
29
Viable -
Alternate
Linear
(CV)a
NA
153.2
31
99.615
8
331.517
7
0.740097
5
385.14116
03
Recommende
d
Basis: Lowest AIC
a Selected Model (Gray); Constant variance case presented (Test 2 p-value = 0.60824); scaled residuals for doses 0, 13.9, 48.4
and 181.4 were 0.209483207, -0.589116734, 0.445351928 and -0.100787718, respectively.
bRestrictions defined in the BMDS 3.1.1 User Guide; NA = Not Applicable; CV= Constant Variance Model; NCV = Non-
Constant Variance Model.
Page 98 of 244
-------
BMDS 3.1.1 Standard Model Plots for P2/F2B Litter Size (Exxon, 1991; Appendix
AK, "Total Pups Born") vs NMP Blood Concentration-50g Rat
FrequeiltisT Exponential Degree 2
Estimated Probability
Frequentist Exponential Degree 3
Estimated Pr obability
Frequentist Exponential Degree 4
Estimated Probability
^^—Frequentist Exponential Degree 5
Estimated Probability
•^^—Frequentist Hill Estimated Probability
^^—Frequentist Polynomial Degree 3
Estimated Probability
^^—Frequentist Polynomial Degree 2
Estimated Piobability
^^—Frequentist Power Estimated
Probability
^^—Frequentist Linear Estimated
Probability
O Data
USER INPUT
Info
Model
Linear vl.l
Dataset Name
P2F2B Litter Size
User notes
[Add user notes here]
Dose-Response Model
M[dose] = g + bl*dose
Model Options
BMR Type
Std. Dev.
BMRF
1
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Model Data
Dependent Variable
[Dose]
Independent Variable
[Response]
Total # of Observations
72
Adverse Direction
Automatic
18
6
4
2
0
0 50 100 150
Dose
Selected Model - Linear - Extra Risk, BMR = 1 SD
Page 99 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
153.2308251
BMDL
99.6158179
BMDU
331.5176516
AIC
385.1411603
Test 4 P-value
0.740097541
D.O.F.
2
Model Parameters
# of Parameters
3
Variable
Estimate
8
15.09893919
betal
-0.02197258
alpha
11.33585663
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimate
dSD
Calc'd
SD
Observ
ed SD
Scaled
Residual
0
25
15.098939
19
15.24
15.24
3.366876
39
2.9478
81
2.9478
81
0.2094832
07
13.9
20
14.793520
33
14.35
14.35
3.366876
39
3.4224
49
3.4224
49
0.5891167
48.4
18
14.035466
34
14.38888
89
14.38888
89
3.366876
39
3.9725
36
3.9725
36
0.4453519
28
181.
4
9
11.113113
26
11
11
3.366876
39
3.7080
99
3.7080
99
0.1007877
Likelihoods of Interest
Model
Log
Likelihood*
#of
Parameters
AIC
A1
-189.2696069
5
388.539214
A2
-188.354168
8
392.708336
A3
-189.2696069
5
388.539214
fitted
-189.5705801
3
385.14116
R
-194.2508792
2
392.501758
Tests of Interest
Test
-2 *Log(Likelihood
Ratio)
Test df
p-value
1
11.79342232
6
0.06673919
2
1.830877708
3
0.60823876
3
1.830877708
3
0.60823876
4
0.601946577
2
0.74009754
Page 100 of 244
-------
18
P2/F2B Litter Size (Exxon, 1991; Appendix AK, "Total Pups Born") vs NMP Blood
Concentration-50 g Rat - Linear Model with BMR of 1 Std. Dev. for the BMD and
0.95 Lower Confidence Limit for the BMDL
^—Estimated Probability
Response at BMD
O Data
BMD
BMDL
80 100
Dose
Page 101 of 244
-------
4.3.3 P2/F2A Litter Size - GD 6-21 Rat (
xxon Appendix A J, "Total Pups Born")
mg/L Blood - GD 6-21 Rat
N
Mean
SD
0
27
15.2592593
3.558225
26.1207
23
13.2608696
4.937955
92.5466
21
14.9047619
3.871754
326.1056
14
11.6428571
3.272429
Table 4-11 Model Predictions for Litter Size in P2/F2A Rats Based on Gestational Exposure
Exxon (1991b))
Standard
Models
Restriction b
BMR = 1 Standard
Deviation
(mg/L Blood - GD 6-
21 Rat)
P-value
AIC
BMDS
Recommends
BMDS Recommendation Notes
BM
D
BMD
L
BMDU
Exponenti
al 2 (CV)
Restricted
479.8
77
254.4
30
1919.1
52
0.126001
7
483.50036
47
Viable -
Alternate
BMD higher than maximum dose
Exponenti
al 3 (CV)
Restricted
341.0
70
272.8
16
1398.6
51
0.062593
9
484.82473
34
Questionable
Goodness of fit p-value <0.1
BMD higher than maximum dose
Exponenti
al 4 (CV)
Restricted
479.8
45
254.4
27
1919.0
11
0.041809
485.50036
47
Viable -
Alternate
Goodness of fit p-value <0.1
BMD higher than maximum dose
Exponenti
al 5 (CV)
Restricted
335.9
07
105.7
78
369.62
51
NA
486.82461
64
Questionable
BMD/BMDL ratio > 3
BMD higher than maximum dose
d.f.=0 (Goodness of fit test
cannot be calculated)
Hill (CV)
Restricted
9999
0
Infinity
NA
486.82461
56
Unusable
BMD computation failed
BMD not estimated
BMDL not estimated
d.f.=0 (Goodness of fit test
cannot be calculated)
Polynomi
al Degree
3 (CV)a
Restricted
364.3
94
273.7
96
1275.7
35
0.170808
482.89187
58
Recommended
Basis: Lowest AIC
BMD higher than maximum
dose
Polynomia
1 Degree 2
(CV)
Restricted
384.9
61
270.0
21
1364.6
28
0.157874
4
483.04935
69
Viable -
Alternate
BMD higher than maximum dose
Power
(CV)
Restricted
329.9
08
275.4
82
1240.3
89
0.062598
3
484.82461
5
Questionable
Goodness of fit p-value <0.1
BMD higher than maximum dose
Linear
(CV)
NA
450.8
59
261.8
83
1618.6
56
0.130882
7
483.42435
33
Viable -
Alternate
BMD higher than maximum dose
a Selected Model (Gray); Constant variance case presented (Test 2 p-value = 0.24158); scaled residuals for dosesO, 26.1207,
92.5466 and 326.1056were 0.954993534, -1.512767309, 0.511175014 and -0.013313118, respectively.
b Restrictions defined in the BMDS 3.1.1 User Guide; NA = Not Applicable; CV = Constant Variance Model; NCV = Non-
Constant Variance Model.
Page 102 of 244
-------
BMDS 3.1.1 Standard Model Plots for P2/F2A Litter Size (Exxon, 1991; Appendix
AJ, "Total Pups Born") vs NMP Blood Concentration - GD 6-21 Rat
Frequentist Exponential Degree 2
Estimated Probability
^^—Frequentist Exponential Degree 3
Estimated Probability
Frequentist Exponential Degree 4
Estimated Probability
^^—Frequentist Exponential Degree 5
Estimated Probability
^^—Frequentist Hill Estimated Probability
^^—Frequentist Polynomial Degree 3
Estimated Probability
^^—Frequentist Polynomial Degree 2
Estimated Probability
^^—Frequentist Power Estimated
Probability
^^—Frequentist Linear Estimated
Probability
O Data
Selected Model - Polynomial Degree 3 (Restricted) - Extra Risk, BMR = 1
USER INPUT
Info
Model
Polynomial degree 3 vl.l
Dataset Name
P2F2A Litter Size GD 6-21
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Model Options
BMR Type
Std. Dev.
BMRF
1
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Model Data
Dependent Variable
[Dosel
Independent Variable
[Response]
Total # of Observations
85
Adverse Direction
Automatic
18
6
4
2
0
0 50 100 150 200 250 300
Page 103 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
364.3935627
BMDL
273.7956247
BMDU
1275.734624
AIC
482.8918758
Test 4 P-value
0.170808016
D.O.F.
2
Model Parameters
# of Parameters
5
Variable
Estimate
8
14.52409502
bl
Bounded
b2
Bounded
b3
-8.26711E-08
alpha
16.00042971
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observe
d Mean
Estimate
dSD
Calc'd
SD
Observe
d SD
Scaled
Residual
0
27
14.5240950
2
15.2592
593
15.2592
593
4.00005
371
3.55822
5
3.55822
5
0.954993
534
26.1207
23
14.5226216
6
13.2608
696
13.2608
696
4.00005
371
4.93795
5
4.93795
5
1.512767
92.5466
21
14.4585657
8
14.9047
619
14.9047
619
4.00005
371
3.87175
4
3.87175
4
0.511175
014
326.1056
14
11.6570896
6
11.6428
571
11.6428
571
4.00005
371
3.27242
9
3.27242
9
0.013313
Likelihoot
s of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-236.6787228
5
483.357446
A2
-234.583299
8
485.166598
A3
-236.6787228
5
483.357446
fitted
-238.4459379
3
482.891876
R
-241.3113542
2
486.622708
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test df
p-value
1
13.45611034
6
0.03633832
2
4.190847665
3
0.24157981
3
4.190847665
3
0.24157981
4
3.534430134
2
0.17080802
Page 104 of 244
-------
P2/F2A Litter Size (Exxon, 1991; Appendix AJ, "Total Pups Born") vs NMP
Blood Concentration- GD 6-21 Rat - Polynomial Degree 3 Model with BMRof
1 Std. Dev. for the BMD and 0.95 Lower Confidence Limit for the BMDL
18
16
14
12
10
8
6
4
2
0
O T
C)
IL
^—Estimated Probability
Response at BMD
O Data
BMD
BMDL
50
100
150
200
Dose
250
300
350
400
Page 105 of 244
-------
4.3.4 P2/F2B Litter Size - GD 6-21 Rat (Exxon Appendix AK, "Total Pups Born")
mg/L Blood - GD 6-21 Rat
N
Mean
SD
0
25
15.24
2.947881
25.25
20
14.35
3.422449
89.03
18
14.39
3.972536
311.9
9
11
3.708099
Table 4-12 Model Predictions for Litter Size in P2/F2B Rats Based on Gestational Exposure
(Exxon (1991b))
BMR = 1 Standard
Standard
Models
Restriction b
Deviation
(mg/L Blood - GD 6-21
Rat)
P-value
AIC
BMDS
Recommends
BMDS Recommendation
Notes
BMD
BMDL
BMDU
Exponential
2 (CV)
Restricted
262.3
67
156.20
9
625.5100
0.6820873
385.30440
9
Viable -
Alternate
Exponential
3 (CV)
Restricted
273.9
39
157.87
8
606.7505
0.4253036
387.17482
76
Viable -
Alternate
Exponential
4 (CV)
Restricted
262.3
75
156.20
8
625.4980
0.6820873
385.30440
9
Viable -
Alternate
Exponential
5 (CV)
Restricted
273.9
09
157.87
6
606.7426
NA
389.17482
74
Questionable
d.f.=0 (Goodness of fit test
cannot be calculated)
Hill (CV)
Restricted
111.0
61
95.288
1
Infinity
NA
389.31790
07
Questionable
d.f.=0 (Goodness of fit test
cannot be calculated)
Polynomial
Degree 3
(CV)
Restricted
281.8
42
173.62
8
556.2398
0.4745885
387.05048
62
Viable -
Alternate
Polynomial
Degree 2
(CV)
Restricted
276.8
75
173.24
1
560.2511
0.4606428
387.08354
61
Viable -
Alternate
Power (CV)
Restricted
273.9
07
172.50
2
568.1038
0.4351554
387.14823
81
Viable -
Alternate
Linear
(CV)a
NA
264.7
04
171.88
3
574.9049
0.717494
385.20319
5
Recommende
d
Basis: Lowest AIC
a Selected Model (Gray); Constant variance case presented (Test 2 p-value = 0.60824); scaled residuals for selected model for
doses 0, 25.25, 89.0333, and 311.8896 were 0.180266075, -0.593822034, 0.507945167 and -0.133410146, respectively.
b Restrictions defined in the BMDS 3.1.1 User Guide; NA = Not Applicable; CV =
Constant Variance Model; NCV = Non-
Constant Variance Model
Page 106 of 244
-------
BMDS 3.1.1 Standard Model Plots for P2/F2B Litter Size (Exxon, 1991; Appendix
AK, "Total Pups Born") vs NMP Blood Concentration- GD 6-21 Rat
^^—Frequentist Exponential Degree 2
Estimated Probability
Frequentist Exponential Degree 3
Estimated Probability
Frequentist Exponential Degree 4
Estimated Piobability
Frequentist Exponential Degree 5
Estimated Probability
^^—Frequentist Hill Estimated Piobability
^^—Frequentist Polynomial Degree 3
Estimated Probability
^^—Frequentist Polynomial Degree 2
Estimated Piobability
^^—Frequentist Power Estimated
Probability
^^—Frequentist Linear Estimated
Probability
O Data
USER INPUT
Info
Model
Linear vl.l
Dataset Name
P2F2B Litter Size GD 6-21
Dose-Response Model
M[dose] = g + b 1 *dose
Model Options
BMR Type
Std. Dev.
BMRJF
1
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Model Data
Dependent Variable
Dose!
Independent Variable
Response]
Total # of Observations
72
Adverse Direction
Automatic
50
100
150
Dose
200
250
300
Selected Model -Linear - Extra Risk, BMR = 1 SD
Page 107 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
264.7037947
BMDL
171.8830314
BMDU
574.9048606
AIC
385.203195
Test 4 P-value
0.717494025
D.O.F.
2
Model Parameters
# of Parameters
3
Variable
Estimate
g
15.11856069
betal
-
0.012724921
alpha
11.34568072
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observ
ed SD
Scaled
Residual
0
25
15.118560
69
15.24
15.24
3.368335
01
2.9478
81
2.94788
1
0.1802660
75
25.25
20
14.797256
43
14.35
14.35
3.368335
01
3.4224
49
3.42244
9
-0.593822
89.0333
18
13.985618
94
14.38888
89
14.38888
89
3.368335
01
3.9725
36
3.97253
6
0.5079451
67
311.889
6
9
11.149790
02
11
11
3.368335
01
3.7080
99
3.70809
9
-0.133410
Likelihoods of Interest
Model
Log Likelihood*
#of
Parameters
AIC
A1
-189.2696069
5
388.539214
A2
-188.354168
8
392.708336
A3
-189.2696069
5
388.539214
fitted
-189.6015975
3
385.203195
R
-194.2508792
2
392.501758
Tests of Interest
Test
-2 *Log(Likelihood
Ratio)
Test df
p-value
1
11.79342232
6
0.06673919
2
1.830877708
3
0.60823876
3
1.830877708
3
0.60823876
4
0.663981316
2
0.71749403
Page 108 of 244
-------
P2/F2B Litter Size (Exxon, 1991; Appendix AK, "Total Pups Born") vs NMP Blood
Concentr ation - GD 6-21 Rat - Linear Model with BMRof 1 SD for the BMD and
0.95 Lower Confidence Limit for the BMDL
18
Estimated Probability
Response at BMD
O Data
BMD
BMDL
Page 109 of 244
-------
4.4 Results of BMD Modeling of P2 Pup Death (Exxon (1991a))
Nested dichotomous models were applied to fit pup death for the P2/F2A and P2/F2B litters. Nested
dichotomous models are preferred for this endpoint because they contain an intra-litter correlation
coefficient for the assessment of litter-specific responses. Details regarding pup death at day 0 (stillborn)
and by day 4 are available in Appendix AJ (for P2/F2A litters) and AK (for P2/FB litters) of the Exxon
(1991b) report.
The pup death endpoint was analyzed using BMDS 2.7 because it contains the larger suite of nested
dichotomous models. To assess intra-litter correlations (ILC) BMDS nested dichotomous models were
run two ways, with ILC coefficients estimated and with ILC coefficients assumed to be zero. Because
potential litter-specific covariates (LSCs) such as dam BW are affected by dose, LSCs were not assessed
in the BMDS nested dichotomous model runs. The following nested dichotomous models and general
modeling options were used to the pup death incidence data.
Nested Dichotomous Models Applied to Pup Death Response7:
• NLogistic - Nested Logistic model with ILC coefficients assumed to be 0
• NLogistic-ILC - Nested Logistic model with ILC coefficients estimated
• NCTR - National Center for Toxicological Research model with ILC coefficients assumed to be
0
• NCTR-ILC - NCTR model with ILC coefficients estimated
• RaiVR - Rai and Van Ryzin model with ILC coefficients assumed to be 0
• RaiVR-ILC - Rai and Van Ryzin model with ILC coefficients estimated
General Model Options Used for Pup Death Nested Dichotomous Response:
• Benchmark Response (BMR): 10% (not shown in report), 5% and 1% Extra Risk
• Confidence Level: 0.95
• Background: Estimated
7 As indicated in the tables in 2.6, the NLogistic model is generally preferred because it has received the more extensive QA
testing, but the NCTR and RaiVR models are provided as alternative models.
Page 110 of 244
-------
4.4.1 P2/F2A Pups Dead at Day 0 (Stillborn Day O/Total Pups Born; Exxon 1991 Appendix
AJ)
Control
26.1207 avg. mg/L blood
GD 6-21
92.5466 avg. mg/L blood
GD 6-21
326.1056 avg. mg/L blood
GD 6-21
Dam
N
Stillborn
Dam
N
Stillborn
Dam
N
Stillborn
Dam
N
Stillborn
JAB248
12
0
JAB029
17
0
JAB302
15
0
JAB325
13
0
JAB026
16
0
JAB032
17
0
JAB038
14
1
JAB327
12
0
JAB251
14
0
JAB279
14
2
JAB 110
15
0
JAB041
13
8
JAB097
15
0
JAB 104
13
1
JAB305
16
1
JAB135
7
0
JAB254
9
0
JAB282
13
0
JAB 113
20
1
JAB 136
4
0
JAB 100
18
2
JAB285
16
1
JAB 116
22
1
JAB045
14
0
JAB257
17
1
JAB288
17
0
JAB311
16
0
JAB050
12
0
JAB260
18
0
JAB035
14
1
JAB 121
9
0
JAB336
11
0
JAB263
15
0
JAB 107
19
0
JAB319
15
0
JAB329
11
0
JAB266
15
0
JAB292
1
1
JAB322
14
0
JAB330
8
2
JAB269
18
1
JAB295
7
0
JAB320
3
0
JAB046
14
0
JAB 10
18
1
JAB347
16
0
JAB306
13
0
JAB328
14
0
JAB270
18
0
JAB298
5
0
JAB313
17
1
JAB 134
16
1
JAB273
15
0
JAB348
19
1
JAB323
14
0
JAB341
14
1
JAB252
16
0
JAB293
5
0
JAB310
15
1
JAB028
18
1
JAB037
14
1
JAB 117
14
0
JAB275
18
0
JAB349
16
0
JAB040
20
0
JAB255
16
0
JAB278
16
1
JAB309
14
1
JAB264
15
0
JAB 105
14
0
JAB039
16
0
JAB267
17
0
JAB297
15
0
JAB317
14
0
JAB262
17
0
JAB 106
17
0
JAB 112
17
0
JAB 102
17
3
JAB281
6
0
JAB246
2
1
JAB290
14
0
JAB256
10
0
JAB098
15
0
JAB249
15
0
JAB253
18
0
Table 4-13 Model Predictions for Pup Death at Day 0 in P2/F2A Rats (Exxon (1991b))
Preferre
d
Models a
5% Extra Risk
1% Extra Risk
P-value
AIC
BMDS
Recommends b
BMDS Recommendation
Notes
BMD
BMDL
BMD
BMDL
NLogistic
326.34
240.809
280.408
50.7883
0.0007
334.364
Questionable
BMD/BMDL ratio > 3
Goodness of fit p-value <0.1
NLogisti
c-ILC
327.095
205.186
281.145
49.3219
0.1017
313.315
Recommended
Basis: Lowest AIC
BMD/BMDL ratio > 3 for 1%
Extra Risk
Alternative Models
NCTR
326.327
271.939
282.34
235.284
0
332.364
Questionable
Goodness of fit p-value <0.1
NCTR-
ILC
327.114
0.63378
5
327.114
0.63378
5
0.1103
311.315
Questionable
BMD/BMDL ratio > 20
RaiVR
281.131
234.276
281.131
234.276
0
332.364
Questionable
Goodness of fit p-value <0.1
RaiVR-
ILC
327.118
0.63378
5
280.539
0.47224
4
0.0867
311.315
Questionable
BMD/BMDL ratio > 20
a NLogistic is preferred because it is the more rigorously tested nested model. All nested models were restricted. Restrictions
are defined in the BMDS 3.1.1 User Guide; ILC = Intra-litter Correlation Coefficients estimated; Because potential litter-
specific covariates (LSCs) such as dam BW are affected by dose, LSCs were not estimated.
b Selected Model (Gray); the average scaled residual for dose group nearest the BMD05 and BMD01 were -0.3523 and -
0.3523, respectively.
Page 111 of244
-------
Selected Model Results- NLogistic- ILC, BMR = 0.01 and 0.05 Extra Risk
NLogistic Model. (Version: 2.20; Date: 04/27/2015)
Input Data File: C:/Users/jgift/BMDS2704/Data/NMP/P2F2A Dead Day 0/nln_P2F2A Day 0
DeathsNln-BMRO 1-Restrict-noLSC .(d)
BMDS Model Run
The probability function is:
Prob. = alpha + thetal*Rij + [1 - alpha - thetal*Rij]/
[ 1 +exp(-beta-theta2* Rij -rho* log(Dose))],
where Rij is the litter specific covariate.
Restrict Power rho >= 1.
Total number of observations = 85
Total number of records with missing values = 0
Total number of parameters in model = 9
Total number of specified parameters = 2
Maximum number of iterations = 500
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Number of Bootstrap Iterations per run: 1000
Bootstrap Seed: 1564538600
User specifies the following parameters:
thetal = 0
theta2 = 0
Tue Jul 30 22:03:20 2019
Default Initial Parameter Values
alpha = 0.02553
beta= -66.0821
thetal = 0 Specified
theta2 = 0 Specified
rho = 10.9041
phil = 0.0392728
phi2 = 0
phi3 = 0
phi4 = 0.310565
Parameter Estimates
Variable Estimate
alpha 0.02553
beta -66.0821
rho 10.9041
phil 0.0392728
0.00468854
0.792172
0.0311563
Std. Err.
NA
Page 112 of 244
-------
phi2 0 Bounded
phi3 0 Bounded
phi4 0.310565 NA
Log-likelihood: -151.658 AIC: 313.315
Litter Data
Lit.-Spec. Litter Scaled
Dose Cov. Est. Prob. Size Expected Observed Residual
0.0000
2.0000
0.026
2
0.051
1
4.1730
0.0000
9.0000
0.026
9
0.230
0
-0.4236
0.0000
10.0000
0.026
10
0.255
0
-0.4400
0.0000
12.0000
0.026
12
0.306
0
-0.4686
0.0000
14.0000
0.026
14
0.357
0
-0.4928
0.0000
15.0000
0.026
15
0.383
0
-0.5036
0.0000
15.0000
0.026
15
0.383
0
-0.5036
0.0000
15.0000
0.026
15
0.383
0
-0.5036
0.0000
15.0000
0.026
15
0.383
0
-0.5036
0.0000
15.0000
0.026
15
0.383
0
-0.5036
0.0000
15.0000
0.026
15
0.383
0
-0.5036
0.0000
15.0000
0.026
15
0.383
0
-0.5036
0.0000
16.0000
0.026
16
0.408
0
-0.5136
0.0000
16.0000
0.026
16
0.408
0
-0.5136
0.0000
16.0000
0.026
16
0.408
0
-0.5136
0.0000
17.0000
0.026
17
0.434
0
-0.5230
0.0000
17.0000
0.026
17
0.434
0
-0.5230
0.0000
17.0000
0.026
17
0.434
1
0.6820
0.0000
17.0000
0.026
17
0.434
3
3.0920
0.0000
18.0000
0.026
18
0.460
0
-0.5318
0.0000
18.0000
0.026
18
0.460
1
0.6254
0.0000
18.0000
0.026
18
0.460
1
0.6254
0.0000
18.0000
0.026
18
0.460
0
-0.5318
0.0000
18.0000
0.026
18
0.460
0
-0.5318
0.0000
18.0000
0.026
18
0.460
2
1.7826
0.0000
18.0000
0.026
18
0.460
1
0.6254
0.0000
18.0000
0.026
18
0.460
0
-0.5318
26.1207
1.0000
0.026
1
0.026
1
6.1782
26.1207
5.0000
0.026
5
0.128
0
-0.3619
26.1207
5.0000
0.026
5
0.128
0
-0.3619
26.1207
6.0000
0.026
6
0.153
0
-0.3965
26.1207
7.0000
0.026
7
0.179
0
-0.4282
26.1207
13.0000
0.026
13
0.332
1
1.1748
26.1207
13.0000
0.026
13
0.332
0
-0.5836
26.1207
14.0000
0.026
14
0.357
0
-0.6056
26.1207
14.0000
0.026
14
0.357
2
2.7833
26.1207
14.0000
0.026
14
0.357
0
-0.6056
26.1207
14.0000
0.026
14
0.357
1
1.0888
26.1207
14.0000
0.026
14
0.357
1
1.0888
26.1207
15.0000
0.026
15
0.383
0
-0.6269
Page 113 of 244
-------
26.1207
16.0000
0.026
16
0.408
1
0.9376
26.1207
16.0000
0.026
16
0.408
0
-0.6474
26.1207
16.0000
0.026
16
0.408
0
-0.6474
26.1207
16.0000
0.026
16
0.408
1
0.9376
26.1207
17.0000
0.026
17
0.434
0
-0.6674
26.1207
17.0000
0.026
17
0.434
0
-0.6674
26.1207
17.0000
0.026
17
0.434
0
-0.6674
26.1207
17.0000
0.026
17
0.434
0
-0.6674
26.1207
19.0000
0.026
19
0.485
1
0.7490
26.1207
19.0000
0.026
19
0.485
0
-0.7055
92.5466
3.0000
0.026
3
0.077
0
-0.2804
92.5466
9.0000
0.026
9
0.230
0
-0.4856
92.5466
13.0000
0.026
13
0.332
0
-0.5836
92.5466
14.0000
0.026
14
0.357
0
-0.6056
92.5466
14.0000
0.026
14
0.357
1
1.0888
92.5466
14.0000
0.026
14
0.357
0
-0.6056
92.5466
14.0000
0.026
14
0.357
1
1.0888
92.5466
14.0000
0.026
14
0.357
0
-0.6056
92.5466
14.0000
0.026
14
0.357
0
-0.6056
92.5466
15.0000
0.026
15
0.383
0
-0.6269
92.5466
15.0000
0.026
15
0.383
0
-0.6269
92.5466
15.0000
0.026
15
0.383
0
-0.6269
92.5466
15.0000
0.026
15
0.383
1
1.0101
92.5466
16.0000
0.026
16
0.408
0
-0.6474
92.5466
16.0000
0.026
16
0.408
1
0.9376
92.5466
16.0000
0.026
16
0.408
0
-0.6474
92.5466
17.0000
0.026
17
0.434
1
0.8703
92.5466
17.0000
0.026
17
0.434
0
-0.6674
92.5466
20.0000
0.026
20
0.511
1
0.6938
92.5466
20.0000
0.026
20
0.511
0
-0.7239
92.5466
22.0000
0.026
22
0.562
1
0.5925
326.1056
4.0000
0.073
4
0.291
0
-0.4031
326.1056
7.0000
0.073
7
0.509
0
-0.4379
326.1056
8.0000
0.073
8
0.582
2
1.0835
326.1056
11.0000
0.073
11
0.800
0
-0.4585
326.1056
11.0000
0.073
11
0.800
0
-0.4585
326.1056
12.0000
0.073
12
0.873
0
-0.4617
326.1056
12.0000
0.073
12
0.873
0
-0.4617
326.1056
13.0000
0.073
13
0.946
8
3.4649
326.1056
13.0000
0.073
13
0.946
0
-0.4645
326.1056
14.0000
0.073
14
1.018
1
-0.0085
326.1056
14.0000
0.073
14
1.018
0
-0.4669
326.1056
14.0000
0.073
14
1.018
0
-0.4669
326.1056
14.0000
0.073
14
1.018
0
-0.4669
326.1056
16.0000
0.073
16
1.164
1
-0.0663
Scaled Residual(s) for Dose Group Nearest the BMD
Minimum scaled residual for dose group nearest the BMD = -0.4669
Minimum ABS(scaled residual) for dose group nearest the BMD = 0.0085
Page 114 of 244
-------
Average scaled residual for dose group nearest the BMD = -0.3523
Average ABS(scaled residual) for dose group nearest the BMD = 0.3523
Maximum scaled residual for dose group nearest the BMD = -0.0085
Maximum ABS(scaled residual) for dose group nearest the BMD = 0.4669
Number of litters used for scaled residual for dose group nearest the BMD = 4
Observed Chi-square = 120.2685
Bootstrapping Results
Number of Bootstrap Iterations per run: 1000
Bootstrap Chi-square Percentiles
Bootstrap
Run P-value 50th 90th 95th 99th
1 0.1020 80.1651 120.8799 132.3672 165.0942
2 0.0930 81.2319 117.9970 132.3763 160.2242
3 0.1050 81.1876 121.5273 137.2496 166.6223
Combined 0.1000 80.9778 120.2642 133.6763 165.0942
The results for three separate runs are shown. If the estimated p-values are sufficiently
stable (do not vary considerably from run to run), then then number of iterations is
considered adequate. The p-value that should be reported is the one that combines
the results of the three runs. If sufficient stability is not evident (and especially
if the p-values are close to the critical level for determining adequate fit, e.g., 0.05),
then the user should consider increasing the number of iterations per run.
To calculate the BMD and BMDL, the litter specific covariate is fixed
at the mean litter specific covariate of all the data: 14.035294
Benchmark Dose Computation
Specified effects = 0.01,0.05
Risk Type = Extra risk
Confidence level = 0.95
BMDs = 281.145,327.095
BMDLs = 49.3219,205.186
Page 115 of 244
-------
Selected Model Plots- NLogistic- ILC, BMR = 0.01 and 0.05 Extra Risk
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
dose
22:03 07/30 2019
Nested Logistic Model, with BMR of 5% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
dose
Page 116 of 244
-------
4.4.2 P2/F2B Pups Dead at Day 0 (Stillborn Day O/Total Pups Born; Exxon 1991 Appendix
AK)
Control
25.25 avg. mg/L blood
GD 6-21
89.03 avg. mg/L blood
GD 6-21
311.9 avg. mg/L blood GD
6-21
Dam
N
Stillborn
Dam
N
Stillborn
Dam
N
Stillbor
n
Dam
N
Stillbor
n
JAB245
18
3
JAB029
15
0
JAB 3 02
19
0
JAB327
14
0
JAB248
14
0
JAB032
15
0
JAB038
14
1
JAB045
15
0
JAB026
16
0
JAB279
14
0
JAB 110
15
0
JAB339
4
0
JAB251
12
0
JAB 104
18
7
JAB305
15
0
JAB329
14
13
JAB097
18
0
JAB288
15
0
JAB 113
16
0
JAB330
13
0
JAB254
8
0
JAB035
15
0
JAB 116
5
0
JAB343D
10
0
JAB 100
16
0
JAB 107
6
0
JAB308
6
0
JAB337
8
0
JAB257
16
2
JAB292
12
1
JAB311
17
0
JAB328
13
0
JAB260
18
0
JAB295
7
0
JAB 121
13
0
JAB 134
8
5
JAB266
11
0
JAB347
15
0
JAB 127
14
1
JAB269
14
0
JAB348
19
0
JAB 130
17
0
JAB 101
15
0
JAB293
19
1
JAB319
18
0
JAB270
20
0
JAB037
15
0
JAB 3 20
17
0
JAB273
18
0
JAB349
16
0
JAB313
11
0
JAB252
11
1
JAB278
11
0
JAB040
18
1
JAB028
16
0
JAB 105
18
0
JAB309
15
0
JAB275
15
0
JAB289
15
1
JAB039
11
0
JAB255
20
0
JAB297
13
0
JAB 112
18
0
JAB264
14
0
JAB 106
16
0
JAB262
16
1
JAB290
13
0
JAB 102
17
1
JAB256
14
0
JAB098
11
1
JAB249
16
0
JAB253
17
0
Table 4-14 Model Predictions for Pup Death at Day 0 in P2/F2B Rats (Exxon (1991b))
Standard
Models a
5% Extra Risk
1% Extra Risk
P-value
AIC
BMDS
Recommends b
BMDS Recommendation Notes
BMD
BMDL
BMD
BMDL
NLogistic
327.408
275.906
285.459
73.5614
0
246.193
Questionable
BMD/BMDL ratio > 3
Goodness of fit p-value <0.1
NLogistic
-ILC
CF
CF
CF
CF
CF
209.115
Unusable
BMD computation fail; Lower
limit includes 0
Alternative Models
NCTR
327.13
0.88668
9
285.638
0.23745
6
0
244.193
Questionable
BMD/BMDL ratio > 20
Goodness of fit p-value <0.1
NCTR-
ILC
324.07
0.65928
9
283.317
0.19183
3
0.256,
0.224
206.511
Questionable
BMD/BMDL ratio > 20
RaiVR
327.208
0.88668
9
285.513
0.51411
5
0
244.193
Questionable
BMD/BMDL ratio > 20
Goodness of fit p-value <0.1
RaiVR-
ILC
324.124
0.65928
9
283.199
0.51702
1
0.2407
206.511
Questionable
BMD/BMDL ratio > 20
a NLogistic is preferred because it is the more rigorously tested nested model. All nested models were restricted. Restrictions
are defined in the BMDS 3.1.1 User Guide; ILC = Intra-litter Correlation Coefficients estimated; Because potential litter-
specific covariates (LSCs) such as dam BW are affected by dose, LSCs were not estimated.
b No model selected as all models were questionable or unusable.
Page 117 of 244
-------
.4.3 P2/F2A Pups
Dead by Day 4 (Dead by Day 4/Total Pups Born; Exxon Appendix A J)
Control
26.1207 avg. mg/L
blood GD6-21
92.5466 avg. mg/L
blood GD6-21
326.1056 avg. mg/L blood
GD6-21
Dam
N
Dead
by Day
4
Dam
N
Dead
by Day
4
Dam
N
Dead
by Day
4
Dam
N
Dead by Day
4*
JAB248
12
0
JAB029
17
4
JAB 3 02
15
0
JAB325
13
9
JAB026
16
0
JAB032
17
0
JAB038
14
1
JAB327
12
12
JAB251
14
0
JAB279
14
3
JAB 110
15
1
JAB041
13
13
JAB097
15
0
JAB 104
13
1
JAB305
16
1
JAB 135
7
0
JAB254
9
0
JAB282
13
5
JAB 113
20
1
JAB 136
4
0
JAB 100
18
2
JAB285
16
1
JAB 116
22
1
JAB045
14
2
JAB257
17
1
JAB288
17
0
JAB311
16
0
JAB050
12
12
JAB260
18
3
JAB035
14
1
JAB 121
9
0
JAB336
11
11
JAB263
15
2
JAB 107
19
2
JAB319
15
0
JAB329
11
1
JAB266
15
0
JAB292
1
1
JAB 3 22
14
2
JAB330
8
8
JAB269
18
1
JAB295
7
0
JAB 3 20
3
0
JAB046
14
0
JAB 10
18
1
JAB347
16
0
JAB 3 06
13
0
JAB328
14
14
JAB270
18
0
JAB298
5
0
JAB313
17
1
JAB 134
16
16
JAB273
15
0
JAB348
19
3
JAB323
14
1
JAB341
14
14
JAB252
16
2
JAB293
5
0
JAB310
15
1
JAB028
18
3
JAB037
14
1
JAB 117
14
0
JAB275
18
5
JAB349
16
0
JAB040
20
2
JAB255
16
2
JAB278
16
3
JAB309
14
1
JAB264
15
0
JAB 105
14
0
JAB039
16
2
JAB267
17
1
JAB297
15
1
JAB317
14
0
JAB262
17
0
JAB 106
17
0
JAB 112
17
0
JAB 102
17
10
JAB281
6
3
JAB246
2
2
JAB290
14
0
JAB256
10
0
JAB098
15
1
JAB249
15
0
JAB253
18
0
Table 4-15 Model Predictions for Pup Death at Day 4 in P2/F2A Rats (Exxon (1991b))
Standard
Models a
5% Extra Risk
1% Extra Risk
P-value
AIC
BMDS
Recommends b
BMDS Recommendation Notes
BMD
BMDL
BMD
BMDL
NLogistic
253.849
136.252
226.386
91.5542
0
771.038
Questionable
Goodness of fit p-value <0.1
NLogistic
-ILC
257.878
132.515
231.394
88.2173
0.0317
608.697
Questionable
Goodness of fit p-value <0.1
Alternative Models
NCTR
261.47
217.891
232.338
193.615
0
769.038
Questionable
Goodness of fit p-value <0.1
NCTR-
ILC
267.663
223.052
240.654
200.545
0.0307,
0.0303
606.697
Questionable
Goodness of fit p-value <0.1
RaiVR
261.996
218.33
233.057
194.214
0
769.038
Questionable
Goodness of fit p-value <0.1
RaiVR-
ILC
267.488
222.907
240.412
200.344
0.0333,
0.034
606.697
Questionable
Goodness of fit p-value <0.1
a NLogistic is preferred because it is the more rigorously tested nested model. All nested models were restricted. Restrictions
are defined in the BMDS 3.1.1 User Guide; ILC = Intra-litter Correlation Coefficients estimated; because potential litter-
specific covariates (LSCs) such as dam BW are affected by dose, LSCs were not estimated.
b No model selected as all models were questionable or unusable.
Page 118 of 244
-------
4.4.4 P2/F2B Pups Dead by Day 4 (Dead by Day 4/Total Pups Born; Exxon Appendix AK)
Control
25.25 avg. mg/L blood
GD6-21
89.03 avg. mg/L blood
GD6-21
311.9 avg. mg/L blood
GD6-21
Dead
Dead
Dead
Dead
Dam
N
by Day
4
Dam
N
by Day
4
Dam
N
by Day
4
Dam
N
by Day
4
JAB245
18
18
JAB029
15
0
JAB302
19
1
JAB327
14
14
JAB248
14
0
JAB032
15
0
JAB038
14
1
JAB045
15
2
JAB026
16
0
JAB279
14
0
JAB 110
15
1
JAB339
4
4
JAB251
12
0
JAB 104
18
7
JAB305
15
0
JAB329
14
14
JAB097
18
0
JAB288
15
0
JAB 113
16
0
JAB330
13
13
JAB254
8
0
JAB035
15
0
JAB 116
5
0
JAB343D
10
10
JAB 100
16
0
JAB 107
6
0
JAB308
6
1
JAB337
8
8
JAB257
16
10
JAB292
12
1
JAB311
17
1
JAB328
13
13
JAB260
18
4
JAB295
7
1
JAB 121
13
1
JAB 134
8
8
JAB266
11
0
JAB347
15
0
JAB 127
14
1
JAB269
14
0
JAB348
19
0
JAB 130
17
1
JAB 101
15
0
JAB293
19
2
JAB319
18
0
JAB270
20
0
JAB037
15
2
JAB320
17
0
JAB273
18
2
JAB349
16
0
JAB313
11
0
JAB252
11
1
JAB278
11
1
JAB040
18
1
JAB028
16
2
JAB 105
18
2
JAB309
15
0
JAB275
15
1
JAB289
15
6
JAB039
11
0
JAB255
20
1
JAB297
13
0
JAB 112
18
0
JAB264
14
0
JAB 106
16
0
JAB262
16
3
JAB290
13
1
JAB 102
17
2
JAB256
14
0
JAB098
11
3
JAB249
16
0
JAB253
17
3
Table 4-16 Model Predictions for Pup Death at Day 4 in P2/F2B Rats (Exxon (1991b))
Standard
Models3
5% Extra Risk
1% Extra Risk
P-value
AIC
BMDS
Recommends b
BMDS Recommendation Notes
BMD
BMDL
BMD
BMDL
NLogistic
229.655
126.176
206.373
92.1515
0
637.258
Questionable
BMD/BMDL ratio > 3
Goodness of fit p-value <0.1
NLogistic
-ILC
229.334
114.81
209.236
85.9385
0.065,
0.053
468.948
Questionable
Goodness of fit p-value <0.1
Alternative Models
NCTR
243.777
203.148
218.255
181.88
0
635.258
Questionable
Goodness of fit p-value <0.1
NCTR-
ILC
250.449
208.707
228.766
190.639
0.0623,
0.0687
466.948
Questionable
Goodness of fit p-value <0.1
RaiVR
243.156
202.63
217.451
181.209
0
635.258
Questionable
Goodness of fit p-value <0.1
RaiVR-
ILC
250.449
208.707
228.766
190.639
0.059,
0.0603
466.948
Questionable
Goodness of fit p-value <0.1
a NLogistic is preferred because it is the more rigorously tested nested model. All nested models were restricted. Restrictions
are defined in the BMDS 3.1.1 User Guide; ILC = Intra-litter Correlation Coefficients estimated; Because potential litter-
specific covariates (LSCs) such as dam BW are affected by dose, LSCs were not estimated.
b No model selected as all models were questionable or unusable
Page 119 of 244
-------
5 Benchmark Dose Modeling of Fetal and Pup Body Weight, Pup Death,
Stillbirths, and Absolute Testes Weight in NMP Producers Group
1999a,b
BMD modeling for reduced fetal and pup bodyweight, increase pup death and stillbirths, and increased
absolute testes weight described in two-generation reproductive studies in Sprague-Dawley rats (NMP
Producers Group (1999a)) and Wistar rats (NMP Producers Group (1999b)) exposed to NMP through
diet was performed using USEPA BMD Software package versions 2.7 (BMDS 2.7) or 3.2 (BMDS 3.2)
in a manner consistent with Benchmark Dose Technical Guidance (U.S. EPA (2012)).
In both NMP Producers Group studies (NMP Producers Group (1999a. b)), male and female rats were
exposed to NMP through diet for two generations (prior to mating through gestation, lactation, weaning,
etc). Each parental generation produced two litters (A and B). In both studies, initial doses were 0, 50,
160 and 500 mg/kg-day and the high dose was reduced from 500 mg/kg-day to 350 mg/kg-day after the
F1A litter due to a high level of mortality in dams exposed to 500 mg/kg-day. F1A litters were exposed
to 500 mg/kg-day; FIB, F2A, and F2B litters were exposed to 350 mg/kg-day. The number of pregnant
dams in each dose group was 20-25 in all of the rat strain and generation combinations except for the
500 mg/kg-day dose group, which had a range of 5-13 pregnant dams across the rat strain and generation
combinations.
Due to uncertainties, several of the endpoints (i.e., pup death, stillbirth, and absolute testes weight)
significantly affected by NMP exposure in these studies were not the critical endpoints identified as the
focus of dose-response analysis in the risk evaluation. For example, stillbirths were observed following
repeated exposure to NMP throughout gestation; however, it is unknown whether stillbirths are the
result of a single dose at a critical stage of development or are the result of repeated exposure to NMP.
Thus, there is uncertainty around whether stillbirths should be considered most relevant for acute or
chronic exposures. EPA performed BMD modeling on these additional reproductive and developmental
endpoints (including pup death, stillbirth, and absolute testes weight) to provide information on a
broader set of endpoints in support of POD selection.
In both NMP Producers Group studies (1999a. b), individual animal data was available for stillbirth and
pup survival through PND4 and PND21 in both litters of both generations. However, pups were culled
on PND4, so PND21 survival should not be compared to PND1 or pre-cull PND4 numbers. Individual
animal data was not available for the fetal and pup body weight endpoints for either study, and therefore
summary statistics for fetal and pup body weights from PND1-PND21 in both litters of both generations
were used for BMD modeling. Additional details regarding modeled endpoints are provided in Table
5-1.
Page 120 of 244
-------
Table 5-1 Description of Endpoints from NMP Producers Group Studies (1999a, b) that were used
for BMP Modeling
Species &
Reference
Endpoint Description
Endpoints Modeled
Litter
Sprague-
Dawley
Rats
(NMP
Individual animal data on stillbirth and pup
survival through PND4 and PND21 in both litters
of both generations; note pups are culled on PND4,
so PND21 survival should not be compared to
PND1 or pre-cull PND4 numbers
Percent stillborn
F1A
Survival to PND4
F2B
Survival to PND21
F2B
Producers
Summary statistics for fetal and pup body weights
on PND1-PND21 in both litters of both
generations.
Fetal body weight PND1
F2B
Group
(1999a))
Pup body weight PND7
F2B
Pup body weight
PND21
F2B
Wistar
Rats
(NMP
Producers
Grouo
(1999b))
Individual animal data on stillbirth and pup
survival through PND4 and PND21 in both litters
of both generations; note pups are culled on PND4,
so PND21 survival should not be compared to
PND1 or pre-cull PND4 numbers
Percent stillborn
F1A,
FIB
Survival to PND4
F1A,
F2B
Survival to PND21
F2B
Summary statistics for fetal and pup body weights
on PND1-PND21 in both litters of both
generations.
Fetal body weight PND1
F1A
Pup body weight PND7
F1A
Pup body weight
PND21
F1A
Absolute testes weights
Absolute testes weights
PO
adult
males
Page 121 of 244
-------
5.1 Overall BMD Modeling Approach for NMP Producers Group 1999a,b
Data
Benchmark dose software was used and EPA BMD Technical Guidance (U.S. EPA (2012)) followed for
the analysis of all endpoints. All endpoints were evaluated with preferred nested dichotomous models
available in BMDS 2.7.0.4 and preferred continuous response models available in BMDS 3.28 using
standard, restricted modeling options (listed below). No non-standard, unrestricted modeling results are
shown or discussed in this section as they either were not needed to achieve adequate model fits or did
not improve upon inadequate standard, restricted model fits.
Standard Nested Dichotomous BMDS 3.1.2 Models Applied to Stillbirth. PND4 and PND21 Pup Death
Endpoints
• Nested Logistic (Nln)-restricted
• NCTR (Nct)-restricted
Model Options Used for Nested Dichotomous Response Modeling of Pup Death Endpoints
• Risk Type: Extra Risk
• Benchmark Response (BMR): 0.01 (1%), 0.05 (5%)
• Confidence Level: 0.95
• Background: Estimate
• Litter Specific Covariate (LSC): Dam weight at Lactation Day 1 (LND1)
Standard Continuous BMDS 3.2 Models Applied to Fetal and Pup Body Weight and Absolute Testes
Weight Endpoints
• Exponential 2 (Exp2)-restricted
• Exponential 3 (Exp3)-restricted
• Exponential 4 (Exp4)-restricted
• Exponential 5 (Exp5)-restricted
• Hill (Hil)-restricted
• Polynomial Degree 4 (Ply4)-restricted
• Polynomial Degree 3 (Ply3)-restricted
• Polynomial Degree 2 (Ply2)-restricted
• Power (Pow)-restricted
• Linear (Lin)
Model Options Used for Continuous Response
• Benchmark Response (BMR): 5% Relative Deviation for Fetal Body Weight and 1% Absolute
Deviation for Resorption
• Response Distribution-Variance Assumptions
• Normal Distribution-Constant Variance
• Normal Distribution-Non-Constant Variance
• Lognormal Distribution-Constant Variance (if normal distribution models do not fit means)
• Confidence Level: 0.95
8 The nested dichotomous (pup death) modeling was performed using the nested logistic and NCTR models contained in
BMDS 2.7.0.4 and the continuous response (body weight) modeling was performed using the standard (default) BMDS 3.2
continuous response models.
Page 122 of 244
-------
• Background: Estimated
Model Restrictions and Model Selection
Each nested dichotomous model analysis of the stillborn and pup death endpoints was performed four
ways, with intra-litter correlation coefficients (ICCs) and LSC estimated, with only LSC estimated, with
only ICCs estimated and with no ICC or LSC estimations. For both the nested dichotomous and
continuous response analyses, each dataset-specific BMD analysis, a single preferred model was chosen
from the standard set of models and modeling options listed above. The modeling restrictions and the
model selection criteria facilitated in BMDS and defined in the BMDS 3.2 User Guide were applied in
accordance with EPA BMD Technical Guidance (U.S. EPA (2012)). Briefly, for each dataset, BMDS
models with standard restrictions were fit to the data using the maximum likelihood method. For nested
dichotomous models applied to pup death endpoints, if the BMDLs from adequately fitting models (P-
value <0.1) were sufficiently close (within a threefold range), the model with the lowest AIC was
selected as the best-fitting model, and its BMDL was used as the POD. Per BMD Technical Guidance
"This criterion is intended to help arrive at a single BMDL value in an objective, reproducible manner."
If the BMDLs are not sufficiently close (not within a threefold range), it was determined that the
BMDLs were substantially model-dependent; thus, the BMDL from the adequately fitting model with
the lowest BMDL was used as the POD.
For continuous models applied to the body and testes weight endpoints, model fit was assessed by a
series of tests as follows. For each model, first the homogeneity of the variances was tested using a
likelihood ratio test (BMDS Test 2). If Test 2 was not rejected (%2 p-value > 0.05), the model was fit to
the data assuming constant variance. If Test 2 was rejected (%2 p-value < 0.05), the variance was
modeled as a power function of the mean, and the variance model was tested for adequacy of fit using a
likelihood ratio test (BMDS Test 3). For fitting models using either constant variance or modeled (non-
constant) variance, models for the mean response were tested for adequacy of fit using a likelihood ratio
test (BMDS Test 4, with yl p-value < 0.10 indicating inadequate fit). From among the models that
yielded an adequate fit, the model for POD determination was selected using the same procedure as for
the nested dichotomous models. For both the dichotomous and continuous model analyses, other factors
were also used to assess the model fit, such as scaled residuals, visual fit, and adequacy of fit in the low-
dose region and in the vicinity of the BMR.
With respect to the continuous model distribution-variance modeling options, responses were first
assumed to be normally distributed with constant variance across dose groups. If no model achieved
adequate fit to response means and response variances under those assumptions, models that assume
normal distribution with non-constant variance, variance modeled as a power function of the dose group
mean were considered (U.S. EPA (2012)). If no normal distribution model achieved adequate fit to
response means under the non-constant variance assumption (BMDS Test 3 p>0.05), models that
assume lognormal distribution with constant variance were considered and the same approach for
evaluating model fit for mean and variance used for the normal distribution data was applied. For each
body weight endpoint, the mean and standard deviation (SD) of litter means per dose group was
modeled, using the number of litters per group as the sample size.
For five endpoints, the constant variance model did not fit adequately when assuming normality, even
though some models fit the means adequately assuming constant variance, and either the non-constant
variance model did not fit adequately or did fit adequately but none of the models fit the means. For all
these endpoints, the constant variance model did not fit adequately when assuming lognormality.
Therefore, a sensitivity analysis was conducted to determine the influence of the variances on the results
for these endpoints by re-modeling the data assuming a different set of SDs. First, the data were
Page 123 of 244
-------
modeled by replacing the SD in all the groups by the minimum SD among the groups, assuming
constant variance and only fitting models that fit the means adequately in the observed SD case. This
procedure was repeated with the SD in all the groups replaced by the maximum SD among the groups.
For each case, a model was selected based on the procedure described above, provided all three cases
yielded usable results, and if the BMDLs among the three cases differed by at most threefold, the lowest
BMDL was selected as the POD. Table 5-2 provides the modeling results for the endpoints that
underwent sensitivity analysis, in addition to the NOAELs for these endpoints. For Day 1 F1A male
Wistar fetal weight (Section 5.6.2), the BMDLs differed by at most threefold, and the maximum SD case
yielded the lowest BMDL. Thus, the BMDL from this case was selected as the POD for this endpoint.
For Day 7 F2B female Sprague-Dawley pup weight (Section 5.5.3) and Day 21 F1A male Wistar pup
weight (Section 5.6.6), the minimum SD case did not yield a model that fit the means adequately. For
each of these endpoints, the lowest value from among the BMDL of the other two SD cases and the
NOAEL was selected as the POD.
Table 5-2 BMDspct and BMDLspct derivations from the variance (SD) sensitivity analysis of body
Section
Response a
St Dev
Case b
Selected
Model
Test 4
P-value
BMDspct
BMDLspct
5.5.3
Sprague-Dawley
Rat F2B Pup
Body Weight at
PND7 (Female)
Observed
Exp 3
0.116
1910
1230
Minimum
Exp 3
0.013
1910
1370
Maximum
Exp 3
0.323
1910
1080
NOAEL
—
—
—
2050
5.5.4
Sprague-Dawley
Rat F2B Pup
Body Weight at
PND7 (Male)
Observed
Exp 4
0.512
310
31.5
Minimum
Exp 4
0.310
310
142
Maximum
Exp 4
0.635
310
0
NOAEL
—
—
—
566
5.5.6
Sprague-Dawley
Rat F2B Pup
Body Weight at
PND21 (Male)
Observed
Exp 4
0.172
462
145
Minimum
Exp 4
0.033
462
238
Maximum
Exp 4
0.322
462
0
NOAEL
—
—
—
566
5.6.2
Wistar Rat F1A
Fetal Body
Weight at PND1
(Male)
Observed
Poly 3
0.373
2380
1800
Minimum
Poly 3
0.106
2610
2120
Maximum
Poly 3
0.652
2610
1760
NOAEL
—
—
—
1960
5.6.6
Wistar Rat F1A
Pup Body Weight
at PND21 (Male)
Observed
Poly 3
0.482
5960
3420
Minimum
Poly 3
0.085
5960
4640
Maximum
Poly 3
0.587
5960
2770
NOAEL
—
—
—
1960
a For all endpoints listed, results assuming constant variance are presented. Entries in parentheses are from
models that yielded unusable results, either because the model did not fit the means adequately (Test 4 p-
value <0.10; results for model selected in observed SD case presented) or the BMDL was zero.
b Case yielding the POD for each endpoint is shown in bold text and is highlighted in gray.
For Day 7 F2B male Sprague-Dawley pup weight and Day 21 F2B male Sprague-Dawley pup weight, the
maximum SD cases yielded BMDLs equal to 0. Thus, it was determined that there was too much
Page 124 of 244
-------
uncertainty in the BMD estimates across the three SD cases to rely on the modeling results for these
endpoints, and the NOAEL was selected as the POD for each. For each of these, the NOAEL was more
than three times higher than the BMDL based on the observed SD results. Furthermore, the mean weights at
the NOAEL for these endpoints were 9% and 4% lower than the mean weight at the control, respectively,
and thus their difference approximately corresponds to the BMR of 5% relative deviation from the control.
In other words, these NOAELs approximately yield a minimum biological response.
Page 125 of 244
-------
5.2 PBPK Analysis for NMP Producers Group (1999a, b)
The dose-response analyses in Section 5 use AUC (hr mg/L) internal doses predicted using the U.S.
EPA version of the NMP PBPK model, described in Appendix I of the risk evaluation for NMP. To
conduct this analysis a table was created that listed the mean maternal body weight (BW) for each dose
group/generation/pregnancy {e.g., for the 160 mg/kg-day dose group, F0 females, FIB litter, the mean
BW on GD 0 was 310.4 g) and the dose achieved for that group for GDs 0-7, 7-14, and 14-20. While the
mean maternal BW of each group was reported for each week of gestation, because the model already
predicts BW increase during pregnancy and the dose is specified as mg/kg-day (/".e., is multiplied by the
BW as predicted by the model, based on the measure GD 0 BW), these subsequent measured BW values
were not used. However, the fact that group-specific initial BW values and group- and time-specific
doses achieved were used, the model predictions are expected to reasonably incorporate the time-
dependence in BW and dose.
PBPK modeling was conducted for 7 days of dosing prior to the start of gestation, during which time the
maternal BW is treated as fixed at the GD 0 BW. Ingestion was assumed to occur at a constant rate for
12 hours per day {i.e., evenly over the rat's active period, during which time the ingestion rate is twice
the reported dose achieved, so the daily average dose matched what was reported). From testing with the
model, a simulation of 7 days was found to be sufficient to achieve "periodicity," meaning that that the
venous blood concentration was then predicted to repeat with the same pattern each day, given an
ongoing constant dosing schedule. The dose achieved during GD 0-7 was used for each simulation up to
GD 7. The dose was then set to the dose achieved for GD 7-14 and the simulation continued to GD 14.
Finally, the dose was set to the dose achieved for GD 14-20 and the simulation continued to that time
point. An example simulation is shown below. The result of a slight decrease in the dose achieved
during GD 7-14 versus GD 0-7, and then a larger drop during GD 14-20, can be seen. After each
simulation, the daily average venous blood AUC was calculated during pregnancy, simply as the AUC
from GD 0 to GD 20 divided by 20.
Sample PBPK simulation of venous blood concentration in a rat dam prior to and during
~350
| 300
.2 250
B
IB
1 200
v
u
8 150
-o
J 100
J2
(A
i 50
C
-6 -4 -2 0 2 4 6 8 10 12 14 16 18 20
Days relative to gestation day 0
For Wistar P0 male rats (testes weight endpoint), a time-weighted average achieved dose was calculated
from the reported achieved doses for weeks 0-17 and 17-28 of dosing. The highest dose was reduced
from a target of 500 mg/kg-day to 350 mg/kg-day after the first 17 weeks. Since animals grew
throughout the exposure period, simulations were first conducted to evaluate the effect of BW on
internal dose. These evaluation simulations were conducted with ingestion assumed to occur evenly over
Page 126 of 244
pregnancy.
-------
12 hours of each day, as described above, for 7 days to achieve periodicity. The 24-hour AUC on the last
day of exposure was used as the internal dose. For illustrative purposes, the table below provides
estimated doses for exposures of exactly 50, 160 and 450 mg/kg-d in animals with BW set to 200-600 g.
NMP AUC values (hr mg/L) as a function of dose and BW predicted by PBPK modeling for non-
pregnant rats
Dose
(mg/kg-d)
Body weight (g)
200
300
400
500
600
50
478
531
573
608
637
160
1,700
1,900
2,060
2,190
2,310
450
5,770
6,520
7,110
7,610
8,050
These results demonstrate that a lower internal dose is estimated for younger/smaller rats, which occurs
because, based on assumed BW°75 scaling, metabolism per BW is higher for smaller animals, but the
difference between 200 and 500 g animals is only around 20-25%. The testes weight increase may be a
developmental effect, determined primarily by exposure to younger animals, but data to define a
window of vulnerability do not exist. Therefore, the average dose achieved and BW during exposure
weeks 0-17 was used to estimate internal doses for this response. The corresponding BWs and doses
achieved were 400.6, 399.2, and 403.7 g and 48.7, 155.8, and 487.0 mg/kg-day for the low, medium,
and high doses, respectively.
Page 127 of 244
-------
5.3 Comparison of PODs for Critical Effects and for Effects Reported in
the NMP Producers Group Studies
Table 5-3 provides a summary of acute PODs for effects reported in the NMP Producers Group Studies
(1999a. b), including increased incidence of stillbirths (Sections 5.7 and 5.8). However, there is
uncertainty around whether stillbirths should be considered most relevant for acute or chronic exposures
and it is unknown whether this effect was the result of a single dose at a critical stage of development or
a result of repeated exposure to NMP. Therefore, BMD modeling of the stillbirth endpoint was
conducted using both Cmax and AUC as dose metrics. Table 5-4 provides a summary of chronic PODs
for critical effects reported in the NMP Producers Group Studies (1999a. b), including increased
absolute testes weight (Section 5.4), decreased pup body weight (Sections 5.5 and 5.6), and increased
incidence of stillbirths (Sections 5.7 and 5.8). Acute and chronic PODs derived for critical effects in the
NMP risk evaluation are shown for comparison.
EPA selected a POD derived from post-implantation loss in a developmental study Saillenfait et al.
(2003; 2002) as the basis for risk calculations for acute exposures to NMP. The selected POD (i.e., a
BMDL of 437 mg/L Cmax) is not the lowest POD among those EPA modeled for acute endpoints. As
demonstrated by Table 5-3, several studies were not amenable to BMD modeling, and for these studies
NOAELs were selected as PODs, several of which were lower than the selected POD for post-
implantation loss. For example, a NOAEL of 265 mg/L was selected as the POD for the fetal mortality
endpoint (Sitarek et al. (2012)). However, fetal mortality in the study by Sitarek et al. occurred in a
similar dose-range as post-implantation losses in the combined Saillenfait et al. oral and inhalation
studies (i.e., NOAELs for post-implantation losses and fetal mortality were 250 and 265 mg/L,
respectively, and LOAELs were 669 and 531 mg/L, respectively). Further discussion regarding EPA's
choice of acute POD is provided in Section 3.2.5.6 of the Final NMP Risk Evaluation.
EPA selected the POD derived from decreased male fertility (i.e., a BMDL of 183 hr mg/L AUC) in a
two-generation reproductive study (Exxon (1991a)) as the basis for risk calculations for chronic
exposures to NMP. The selected POD is not the lowest POD among those EPA modeled for chronic
endpoints. For example, BMD modeling of PND21 pup body weights in the NMP Producers Group
(1999a) study identified a POD of 100 hr mg/L. Although reduced pup body weight is considered a
sensitive endpoint, it is not the ideal basis for a chronic POD as there is uncertainty around actual
internal serum levels achieved in rat pups during lactation. Further discussion regarding EPA's choice of
chronic POD is provided in Section 3.2.5.6 of the Final NMP Risk Evaluation.
Page 128 of 244
-------
Table 5-3 Acute PODs: Comparison of PODs for critical effects and for effects reported in the
NMP Producers Group Studies (1999a, b)
Endpoint and reference
(exposure duration/route)
Dose
Metric or
NOAEL
Model
BMR
BMD
BMDL
POD
Internal
dose b
Equivalent
oral dose
mg/kg/day a
Post-implantation Loss
Saillenfait et al. (2002)
(GD 6-20, oral, post-
implantation loss)
Cmax (mg/L
blood)
Log-Probit
1%
RD
474
437
437
418
AUC (hr
mg/L blood)
Log-Probit
1%
RD
5010
4592
4592
419
Saillenfait et al. (2003; 2002)
Cmax (mg/L
blood)
Log-probit
1%
RD
470
437
437
418
(GD 6-20, oral and inhalation)
AUC (hr
mg/L blood)
Log-probit
1%
RD
4990
4590
4590
419
Resorptions
Saillenfait et al. (2002)
(GD 6-20, oral, post-
implantation loss)
NOAEL
Cmax, (mg/L
blood)
N/A
N/A
N/A
N/A
250 c
250
Becci et al. (1982)
(GD 6-15, dermal)
NOAEL
Cmax, (mg/L
blood)
N/A
N/A
N/A
N/A
662 d
612 (oral)
237 (dermal)
Fetal Mortality
Sitarek et al. (2012)
(GD1-PND1, oral)
NOAEL
Cmax,( mg/L
blood)
N/A
N/A
N/A
N/A
265 e
264
Stillbirths f
NMP Producers GrouD (1999a)
(Sprague-Dawley)
(Dietary exposure throughout
gestation, lactation, growth, pre-
mating)
NOAEL
Cmax (mg/L
blood
NA
NA
NA
NA
142
147
NOAEL
AUC (hr
mg/L blood)
NA
NA
NA
NA
2,120
216
NMP Producers GrouD (1999b)
(Wistar)
(Dietary exposure throughout
gestation, lactation, growth, pre-
mating)
Cmax (mg/L
blood)
NLogistic-
ICC
1%
ER
429
58
58
62
AUC (hr
mg/L blood)
NLogistic-
ICC
1%
ER
6440
855
855
96
Exxon(1991a)
(Sprague-Dawley)
(Dietary exposure throughout
gestation, lactation, growth, pre-
mating)
AUC (hr
mg/L blood)
NLogistic
- ILC
1%
ER
6744
1183
1183
129
Page 129 of 244
-------
Dose
Metric or
NOAEL
POD
Endpoint and reference
(exposure duration/route)
Model
BMR
BMD
BMDL
Internal
dose b
Equivalent
oral dose
mg/kg/day a
ER = extra risk; RD = relative deviation
The POD selected for calculating risk of acute NMP exposures is bolded and highlighted in gray.
a Assuming daily oral gavage and initial BW 0.259 kg (i.e., the same experimental conditions as the Saillenfait et al. (2002)
study) for the purposes of comparison across the studies.
b Internal doses refer to maternal blood concentrations (as opposed to fetal blood concentrations which are not predicted by
the PBPK model).
0 BMD models were considered unacceptable due to uncertainty caused by lack of model fit; the internal serum dose is
based on a NOAEL of 250 mg/kg-bw/day.
d Dose-response data were not considered amenable to BMD modeling. The internal serum dose is based on a NOAEL of
237 mg/kg bw/day dermal exposure. An oral dose of 612 mg/kg bw/day, given on GD 6-20, is predicted to yield the same
peak concentration (662 mg/L).
e BMD modeling failed to calculate an adequate BMD or BMDL value by either dose metric. The internal serum dose is
based on a NOAEL of 450 mg/kg bw/day.
f The relevance of stillbirth for acute exposure is unclear, as these effects were only observed following exposure
throughout gestation. In addition, the effect was reported in dietary studies in which exposure occurs throughout the day
rather than through a single bolus (which would result in a greater peak exposure). PODs for the stillbirth endpoint are
provided in terms of AUC and Cmax for reference. BMD modeling was attempted for stillbirth data reported in the NMP
Producers Group (1999a) study with Sprague-Dawley rats; however, no models adequately fit the dataset.
Table 5-4 Chronic PODs: Comparison of PODs for critical effects and for effects reported in the
NMP Producers Group Studies i
1999a. b)
Endpoint and reference
(exposure duration/route)
Selected
Model or
NOAEL
BMR
BMD
AUC
(hr
mg/L)
BMDL
AUC
(hr
mg/L)
POD
AUC (hr
mg/L
blood)a
Equivalent
oral dose b
mg/kg/day
Fetal Body Weight
Saillenfait et al. (2002)
Exponential 3
c,d
5%
RD
1400
981
981
109
(oral exposure GD 6-20)
Saillenfait et al. (2003)
Exponential 3 c
5%
RD
654
414
414
48
(inhalation exposure GD 6-20)
E. I. Dupont De Nemours & Co
Exponential 3 c
5%
RD
315
223
223
27
(1990)
(inhalation exposure preconception
and GD 1-20)
Becci et al. (1982)
NOAEL= 237
mg/kg/day e
NA
NA
NA
2052
210
(dermal exposure GD 6-15)
Reduced Male Fertility
Exxon (1991a)
Log-logistic
10%
ER
49211
3411"
262fl
1831"
183
28
(Dietary exposure throughout
gestation, lactation, growth, pre-
mating)
Reduced Female Fecundity
Exxon (1991a)
Log-logistic
10%
ER
862 11
420 G
401fl
202 c
202
31
(Dietary exposure throughout
gestation, lactation, growth, pre-
mating)
Page 130 of 244
-------
Endpoint and reference
(exposure duration/route)
Selected
Model or
NOAEL
BMR
BMD
AUC
(hr
mg/L)
BMDL
AUC
(hr
mg/L)
POD
AUC (hr
mg/L
blood)a
Equivalent
oral dose b
mg/kg/day
Alternate NMPProducers Group 1999 and Exxon 1991 Endpoints
Testes weights- absolute
NMP Producers Group (1999b)
(Wistar, dietary exposure
throughout gestation, lactation,
growth, pre-mating)
Exponential 4
5%
RD
1,610
601
601
69
PND 21 Pup body weights- females
NMP Producers Grouo (1999a)
(Sprague-Dawley, dietary exposure
throughout gestation, lactation,
growth, pre-mating)
Exponential 4
5%
RD
612
100
100
12
PND 21 Pup body weights- females
NMP Producers GrouD (1999b)
(Wistar, dietary exposure
throughout gestation, lactation,
growth, pre-mating)
Polynomial 3
5%
RD
6,940
3,350
3,350
321
Stillbirth 8
NMP Producers Group (1999a)
(Sprague-Dawley, dietary exposure
throughout gestation, lactation,
growth, pre-mating)
NOAEL= 160
mg/kg/day
NA
NA
NA
2,120
321
Stillbirth 8
NMP Producers Group (1999b)
(Wistar, dietary exposure
throughout gestation, lactation,
growth, pre-mating)
NLogistic- ICC
1%
ER
6,440
855
855
216
Stillbirth 8
Exxon (1991a)
(Dietary exposure throughout
gestation, lactation, growth, pre-
mating)
NLogistic -
ILC
1%
ER
6,744
1,183
1,183
96
Page 131 of 244
-------
Endpoint and reference
(exposure duration/route)
Selected
Model or
NOAEL
BMD
BMDL
BMR
AUC
AUC =
(hr
(hr
mg/L)
mg/L)
POD
AUC (hr
mg/L
blood)a
Equivalent
oral dose b
mg/kg/day
RD = relative deviation; ER= extra risk
The POD selected for calculating risk of chronic NMP exposures is bolded and highlighted in gray.
a Internal doses for fetal body weight reflect maternal blood concentrations during gestation and internal doses for fertility
reflect blood concentrations in pups post-weaning.
b Assuming daily oral gavage GDs 6-20 and initial BW 0.259 kg (i.e., the same experimental conditions as the Saillenfait
et al. (2002) study) for the purposes of comparison across the studies.
0 Since standard models gave adequate results for all endpoints, non-standard models were not considered. Since fits to the
means were obtained using normal distribution models, lognonnal models were not applied
dFor Saillenfait et al. (2002). the BMD and BMDL reported are from modeling the data with all the SDs equal to the
maximum SD across the groups.
e The data in Becci (1982) were not amenable to BMD modeling. The mean weight increased gradually from the control to
the middle dose group and then decreased significantly at the high dose group. This dose-response pattern is essentially
equivalent to one where only the highest dose has a response and thus the model estimates of the parameters and BMDs
would not be reliable. The internal serum dose is based on a NOAEL of 237 mg/kg bw/day dermal exposure.
f In the Exxon (1991a) study, each dam had two sets of mating periods. Each mating period was analyzed separately; dl
indicates results for the first mating period and d2 indicates results from the second mating period. PODs for male
fertility and female fecundity in this study are calculated based on exposure levels in 50g rats immediately post-
weaning.
g The relevance of stillbirth for acute vs. chronic exposure is unclear. These effects were observed following exposure
throughout gestation. In addition, the effect was reported in dietary studies in which exposure occurs throughout the day
rather than through a single bolus (which would result in a greater peak exposure). BMD modeling was attempted for
stillbirth data reported in the NMP Producers Group (1999a) study with Sprague-Dawley rats; however, no models
adequately fit the dataset.
Page 132 of 244
-------
5.4 Results for Benchmark Dose Modeling of Absolute Testes Weight in PO
Male Wistar Rats (NMP Producers Group (1999b))
Wistar Rat Absolute Testes Weight (PO Adult Males) Data used for BMD Modeling.
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
3.59
0.2
557.5
25
3.634
0.246
1995
25
3.769
0.41
7862
25
3.782
0.277
Table 5-5 Model Predictions for AUC (hr mg/L) versus Wistar Rat Absolute Testes Weight (PO
Adult Males) (NMP Producers Group (1999b)).
3MR = 5% Relative Deviation (RD)
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.0003904
45.81758667
8544.569
4816.473
33594.218
Constant variance
model did not fit
adequately, but non-
constant variance
model did fit. Only
exponential model 4
fit the means
adequately assuming
non-constant variance,
so it was selected.
Exponential 3
0.0003904
45.81758466
8544.868
4816.477
34459.048
Exponential 4
0.1205979
34.35160178
1606.791
601.1339
10227.64
Polynomial 3°
0.0004012
45.76301492
8455.645
4680.749
34005.204
Polynomial 2°
0.0004012
45.76301491
8455.675
4689.325
33533.859
Power
0.0004012
45.76301831
8455.992
4681.267
33905.848
Linear
0.0004012
45.76301472
8457.294
4682.196
33752.343
a Results assuming non-constant variance presented (BMDS Test 2 p < 0.01, Test 3 p = 0.41); selected model in bold.
3.95
3.9
3.85
3.8
3.75
-------
USER INPUT
Info
Model
frequentist Exponential degree 4 vl.l
Dataset Name
Absolute testes weight in FO male Wistar rats
Dose-Response Model
M[dose] = a * [c-(c-l) * exp(-b * dose)]
Variance Model
Var[i] = exp(log-alpha + log(mean[i]) * rho)
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Non-Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
1606.790852
BMDL
601.133903
BMDU
10227.64044
AIC
34.35160178
Test 4 P-value
0.120597877
d.f.
2
Model Parameters
# of Parameters
5
Variable
Estimate
a
3.570619756
b
0.001200959
c
1.058492801
rho
Bounded
log-alpha
-26.09921678
Page 134 of 244
-------
Goodness of
Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
GSD
Calc'd
GSD
Observed
SD
Scaled
Residual
0
25
3.570619756
3.59
3.59
0.20291751
0.2
0.2
0.477539964
557.5
25
3.672552383
3.634
3.634
0.26142016
0.246
0.246
-0.737364388
1995
25
3.760450768
3.769
3.769
0.32343443
0.41
0.41
0.132163292
7862
25
3.77945874
3.782
3.782
0.33844925
0.277
0.277
0.037542698
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-17.37746746
5
44.7549349
A2
-10.17281491
8
36.3456298
A3
-11.06050729
6
34.1210146
fitted
-13.17580089
4
34.3516018
R
-21.40147557
2
46.8029511
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
22.45732132
6
0.00100018
2
14.40930512
3
0.00239779
3
1.775384772
2
0.41160448
4
4.230587192
2
0.12059788
Page 135 of 244
-------
5.5 Results for BMD Modeling for Reduced Fetal and Pup Body Weight
for Sprague-Dawley Rats (NMP Producers Group (1999a))
5.5.1 Sprague-Dawley Rat F2B Fetal Body Weight at PND1 (Females)
;ht Data at PND1 (Females) used for BMD Modeling
Sprague-Dawley
iat F2B Fetal Body Weig
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
6.9
0.66
566.5
26
6.5
1.04
2053
23
6.5
0.76
5235
23
6.2
0.82
Table 5-6 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Fetal Body
Weight at PND1
(Females) Using Daily Average A1
JC as the
)ose Metric
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.351778
250.5541583
3269.079
1953.617
9632.114
Assuming constant
variance, exponential
model 4 had a BMDL
of zero, indicating that
there is excessive
uncertainty in the
BMD estimate. No
model was selected.
Exponential 3
0.351778
250.5541583
3269.139
1953.351
9632.2906
Exponential 4
0.148401
252.5532581
3267.961
0
9631.5953
Polynomial 3°
0.3461015
250.5866945
3339.144
2047.999
9628.0067
Polynomial 2°
0.3461015
250.5866945
3339.144
2048.117
9623.9732
Power
0.3461015
250.5866945
3339.143
2052.163
9563.3035
Linear
0.3461015
250.5866945
3339.144
2048.046
9623.8352
a Results assuming constant variance presented (BMDS Test 2 p = 0.07).
Estimated Probability
Response at BMD
O Data
— BMD
BMDL
2000 3000
Dose
Figure 5.5-1 Plot of Mean Response by Dose, with Fitted Curve for Linear Model with Constant
Variance for Sprague-Dawley Rat F2B Fetal Body Weight at PND1 (Females)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 136 of 244
-------
5.5.2 Sprague-Dawley Rat F2B Fetal Body Weight at PND1 (Males)
Sprague-Dawley Rat F2B Fetal Body Weight Data at PND1 (Males) used for BMD Modeling
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
7.3
0.66
566.5
26
6.9
1.04
2053
24
6.9
0.71
5235
23
6.6
0.93
Table 5-7 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Fetal Body
Modela
Goodness of Fit
BMD
(hr mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for Model
Selection
Test 4
P-value
AIC
Exponential 2
0.351778
250.5541583
3269.079
1953.617
9632.114
Assuming constant
variance, exponential
model 4 had a BMDL of
zero, indicating that there
is excessive uncertainty
in the BMD estimate. No
model was selected.
Exponential 3
0.351778
250.5541583
3269.139
1953.351
9632.2906
Exponential 4
0.148401
252.5532581
3267.961
0
9631.5953
Polynomial 3°
0.3461015
250.5866945
3339.144
2047.999
9628.0067
Polynomial 2°
0.3461015
250.5866945
3339.144
2048.117
9623.9732
Power
0.3461015
250.5866945
3339.143
2052.163
9563.3035
Linear
0.3461015
250.5866945
3339.144
2048.046
9623.8352
a Results assuming constant variance presented (Test 2 p = 0.07).
7.8
Dose
Figure 5.5-2 Plot of Mean Response by Dose, with Fitted Curve for Linear Model with Constant
Variance for Sprague-Dawley Rat F2B Fetal Body Weight at PND1 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 137 of 244
-------
5.5.3 Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Females)
Sprague-Dawley Rat F2B Pup Body Weight Data at PND7 (Females) used for BMD Modeling
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
16.2
1.72
566.5
26
14.6
2.6
2053
23
14.8
1.67
5235
21
13.6
3.32
Table 5-8 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.1163921
441.7245971
1905.072
1225.905
4118.7554
The constant variance model
did not fit adequately, and
none of the models fit the
means adequately assuming
non-constant variance. In the
context of a sensitivity
analysis, exponential model 3
was selected from among the
models that fit the means (Test
4 p-value > 0.10), assuming
constant variance.
Exponential 3
0.1163921
441.7245971
1905.072
1228.589
4118.7554
Exponential 4
0.0938171
442.2306413
192.4316
0
2976.4058
Polynomial 3°
0.1119924
441.8016643
2007.858
1339.323
4555.5464
Polynomial 2°
0.1119924
441.8016643
2007.858
1339.409
4367.2799
Power
0.1119924
441.8016643
2007.86
1339.381
4183.2379
Linear
0.1119924
441.8016643
2007.863
1339.437
4201.9577
a Results assuming constant variance presented (BMDS Test 2 p < 0.01).
18
17
16
o
c
o
Q.
15
14
13
12
(
)
1000
2000
3000
4000
Dose
5000
Estimated Probability
Response at BMD
O Data
BMD
BMDL
Figure 5.5-3 Plot of Mean Response by Dose, with Fitted Curve for Exponential 3 Model with
Constant Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Females)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 138 of 244
-------
USER INPUT
Info
Model
frequentist Exponential degree 3 vl.l
Dataset Name
Day 7 pup body weight in F2B female Sprague-Dawley rats
Dose-Response Model
M[dose] = a * exp(±l * (b * dose)Ad)
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
1905.072384
BMDL
1228.588778
BMDU
4118.75535
AIC
441.7245971
Test 4 P-value
0.116392057
d.f.
2
Model Parameters
# of Parameters
4
Variable
Estimate
a
15.56786252
b
2.69246E-05
d
Bounded
log-alpha
1.748697651
Page 139 of 244
-------
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
15.567863
16.2
16.2
2.3973137
1.72
1.72
1.318428778
566.5
26
15.332211
14.6
14.6
2.3973137
2.6
2.6
-1.557392921
2053
23
14.730682
14.8
14.8
2.3973137
1.67
1.67
0.138670184
5235
21
13.521197
13.6
13.6
2.3973137
3.32
3.32
0.150636477
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-215.7115076
5
441.423015
A2
-208.1511332
8
432.302266
A3
-215.7115076
5
441.423015
fitted
-217.8622986
3
441.724597
R
-222.4913993
2
448.982799
* Includes additive constant of -87.29916. This constant was not included in the
LL derivation prior to BMDS 3.0
Tests of Interest
Test
-2 *Log(Likelihood
Ratio)
Test
d.f.
p-value
1
28.68053237
6
<0.0001
2
15.12074884
3
0.00171631
3
15.12074884
3
0.00171631
4
4.301581969
2
0.11639206
Table 5-9 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body
Weight at PND7 (Females) Using Daily Average AUC as the Dose Metric.
All SDs set to minimum SD across the group.
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.0132848
377.5904324
1905.072
1367.171
3097.7833
Assuming constant
variance, no model fit the
means adequately (Test 4 p-
value < 0.10 for all models).
No model was selected.
Exponential 3
0.0132848
377.5904324
1905.072
1371.022
3097.7833
Polynomial 3°
0.0123167
377.7417503
2007.868
1477.946
3786.5275
Polynomial 2°
0.0123167
377.7417503
2007.858
1478.013
3534.3112
Power
0.0123167
377.7417503
2007.84
1480.133
3186.5526
Linear
0.0123167
377.7417503
2007.858
1477.883
3188.3811
aResults assuming constant variance presented (BMDS Test 2 p = 1.00).
Page 140 of 244
-------
17.3
Dose
Figure 5.5-4 Plot of Mean Response by Dose, with Fitted Curve for Exponential 3 Model with
Constant Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Females)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L; with constant variance and
all SDs set to the minimum SD across the group
Table 5-10 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body
Weight at PND7 (Females) Using Daily Average AUC as the Dose Metric.
All SDs set to maximum SD across the group.
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.3226975
501.7670583
1904.883
1079.55
7149.0406
Assuming constant
variance,
exponential model 3
was selected based
on lowest AIC.
Exponential 3
0.3226976
501.767058
1905.072
1082.045
7149.0315
Polynomial 3°
0.3161543
501.8080281
2007.858
1195.667
7239.7544
Polynomial 2°
0.3161543
501.8080281
2007.863
1195.548
7238.2775
Power
0.3161543
501.8080281
2007.859
1195.81
7234.9478
Linear
0.3161543
501.8080281
2007.858
1195.581
7242.2815
a Results assuming constant variance presented (BMDS Test 2 p = 1.00); selected model in bold.
Page 141 of 244
-------
18
Estimated Probability
Response at BMD
O Data
BMD
BMDL
1000
2000
3000
4000
5000
Dose
Figure 5.5-5 Plot of Mean Response by Dose, with Fitted Curve for Exponential 3 Model with
Constant Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Females)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L; all SDs set to the
maximum SD across the group
USER INPUT
Info
Model
frequentist Exponential degree 3 vl.l
Dataset Name
Day 7 pup body weight in F2B female
Sprague-Dawley rats-max SD
Dose-Response Model
M[dose] = a * exp(±l * (b * dose)Ad)
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
Page 142 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
1905.072384
BMDL
1082.044675
BMDU
7149.031453
AIC
501.767058
Test 4 P-value
0.322697556
d.f.
2
Model Parameters
# of Parameters
4
Variable
Estimate
a
15.56786338
b
2.69246E-05
d
Bounded
log-alpha
2.380723761
Goodness of
Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
15.56786338
16.2
16.2
3.28827095
3.32
3.32
0.961199104
566.5
26
15.33221177
14.6
14.6
3.28827095
3.32
3.32
-1.135418028
2053
23
14.73068203
14.8
14.8
3.28827095
3.32
3.32
0.101097904
5235
21
13.5211947
13.6
13.6
3.28827095
3.32
3.32
0.109824057
Likelihooc
s of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-246.7524892
5
503.504978
A2
-246.7521789
8
509.504358
A3
-246.7524892
5
503.504978
fitted
-247.883529
3
501.767058
R
-250.3999906
2
504.799981
* Includes additive constant of -87.29916. This constant was not incluc
BMDS 3.0.
ed in the LL derivation prior to
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
7.29562342
6
0.29437121
2
0.000620701
3
0.99999589
3
0.000620701
3
0.99999589
4
2.262079509
2
0.32269756
Page 143 of 244
-------
5.5.4 Sprague-Dawley Rat F2B Pup Body Weight at PND7 (Males)
Sprague-Dawley Rat F2B Pup Body Weight Data at PND7 (Males) used for BMD Modeling
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
17.2
1.82
566.5
26
15.7
2.73
2053
24
15
1.58
5235
21
14.4
3.39
Table 5-11 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body
Model
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.0973768
451.2052401
1709.332
1146.915
3260.3212
Neither variance model
fit adequately. A
sensitivity analysis
indicated that there was
too much uncertainty in
the BMD estimate to
use dose-response
modeling results. No
model was selected.
Exponential 3
0.0973768
451.2052401
1709.337
1149.303
3260.3231
Exponential 4
0.5119478
448.9769894
310.3716
31.48148
1268.6804
Polynomial 3°
0.0857581
451.4593557
1837.814
1274.571
3582.3314
Polynomial 2°
0.0857581
451.4593557
1837.809
1274.531
3474.3836
Power
0.0857581
451.4593557
1837.812
1274.512
3403.9466
Linear
0.0857581
451.4593557
1837.814
1274.511
3415.2812
a Results assuming constant variance presented (BMDS Test 2 p < 0.01).
18.8
17.8
16.8
15.8
14.8
13.8
12.8
Estimated Probability
Response at BMD
O Data
BMD
BMDL
1000
2000
3000
4000
5000
Dose
Figure 5.5-6 Plot of Mean Response by Dose, with Fitted Curve for Linear Model for Sprague-
Dawley Rat F2B Pup Body Weight at PND7 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 144 of 244
-------
5.5.5 Sprague-Dawley Rat F2B Pup Body Weight at PND21 (Females)
Sprague-Dawley Rat
AUC (hr mg/L)
0
566.5
2053
5235
N
25
26
23
20
2B Pup Body Weight Data at PND21 (Females) used for BMD Modeling
Mean
51.7
49.1
47.0
46.1
Std. Dev.
4.35
5.87
6.66
7.04
Table 5-12 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body
Model
Test 4
P-value
AIC
BMD
BMDL
BMDU
Basis for model
selection
Exponential 2
0.2091959
608.4509171
2567.306
1636.342
5782.4747
Exponential model 4
assuming constant
variance was selected
based on lowest BMDL
(BMDLs differed by
more than threefold).
This model also had a
much better visual fit
and lower residuals
than the other models.
Exponential 3
0.2091959
608.4509171
2567.306
1637.785
5782.4747
Exponential 4
0.8453858
607.3599775
611.6493
99.99824
2752.6331
Polynomial 3°
0.1957569
608.583712
2675.642
1749.609
5922.1867
Polynomial 2°
0.1957569
608.583712
2675.652
1750.346
5912.9217
Power
0.1957569
608.583712
2675.65
1753.461
5890.2517
Linear
0.1957569
608.583712
2675.642
1749.831
5916.3703
a Constant variance case presented (Test 2 p = 0.12); selected model in bold.
56
T5
2000
3000
4000
5000
Dose
^—Estimated Probability
Response at BMD
O Data
BMD
BMDL
Figure 5.5-7 Plot of Mean Response by Dose, with Fitted Curve for Exponential 4 Model with
Constant Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND21 (Females)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 145 of 244
-------
USER INPUT
Info
Model
frequentist Exponential degree 4 vl. 1
Dataset Name
Day 21 pup body weight in F2B female Sprague-Dawley rats
Dose-Response Model
M[dose] = a * [c-(c-l) * exp(-b * dose)]
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
[Custom]
Independent Variable
[Custom]
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
611.649313
BMDL
99.99824307
BMDU
2752.633086
AIC
607.3599775
Test 4 P-value
0.845385758
d.f.
1
Model Parameters
# of Parameters
4
Variable
Estimate
a
51.65433514
b
0.001045796
c
0.894186168
log-alpha
3.538294035
Page 146 of 244
-------
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled Residual
0
25
51.65433514
51.7
51.7
5.86584776
4.35
4.35
0.038924347
566.5
26
49.2110082
49.1
49.1
5.86584776
5.87
5.87
-0.096496361
2053
23
46.82716494
47
47
5.86584776
6.66
6.66
0.141307424
5235
20
46.21150063
46.1
46.1
5.86584776
7.04
7.04
-0.085008337
Likelihooc
s of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-299.6609745
5
609.321949
A2
-296.7494779
8
609.498956
A3
-299.6609745
5
609.321949
fitted
-299.6799888
4
607.359978
R
-305.5360764
2
615.072153
* Includes additive constant of -86.38022. This constant was not included in the LL derivation prior to
BMDS 3.0.
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
17.573197
6
0.00739219
2
5.822993199
3
0.12054684
3
5.822993199
3
0.12054684
4
0.038028582
1
0.84538576
Page 147 of 244
-------
5.5.6 Sprague-Dawley Rat F2B Pup Body Weight at PND21 (Males)
Sprague-Dawley Rat F2B Pup Body Weight Data at PND21 (Males) used for BMD Modeling
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
54
4.52
566.5
26
51.8
6.46
2053
24
47.2
9.82
5235
20
49.4
6.64
Table 5-13 Model Predictions for AUC (hr mg/L) versus Sprague-Dawley Rat F2B Pup Body
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.0223161
651.5750995
3026.784
1697.867
12736.279
Constant variance model did not
fit adequately. Non-constant
variance model fit adequately,
but no model fit means
adequately with this variance
model. A sensitivity analysis
indicated that there was too
much uncertainty in the BMD
estimate to use dose-response
modeling results No model was
selected.
Exponential 3
0.0223161
651.5750995
3026.784
1699.496
12736.279
Exponential 4
0.1715587
647.8394728
461.9646
145.2656
1933.1638
Polynomial 3°
0.0208586
651.7101835
3173.233
1828.495
13573.1
Polynomial 2°
0.0208586
651.7101835
3173.233
1828.947
13562.995
Power
0.0208586
651.7101835
3173.256
1833.186
13542.77
Linear
0.0208586
651.7101835
3173.253
1828.535
13546.869
a Results assuming constant variance presented (BMDS Test 2 p < 0.01).
58
Estimated Probability
Response at BMD
O Data
BMD
BMDL
2000
Dose
Figure 5.5-8 Plot of Mean Response by Dose, with Fitted Curve for Linear Model with Constant
Variance for Sprague-Dawley Rat F2B Pup Body Weight at PND21 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 148 of 244
-------
5.6 Results for BMD Modeling for Reduced Fetal and Pup Body Weight
for Wistar Rats (NMP Producers Group (1999b))
5.6.1 Wistar Rat F1A Fetal Body Weight at PND1 (Females)
Wistar Rat F1A Fetal Body Weig
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
6.2
0.46
538.0
25
6.0
0.55
1965
24
5.9
0.50
7793
13
5.1
0.85
it Data at PND1 (Females) used for BMD Modeling
Table 5-14 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Fetal Body Weight at
PND1 (Females) Using Daily Average AUC as the Dose Metric
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.7063237
151.0124763
2140.151
1640.755
3032.9951
Exponential model 3
assuming constant
variance was selected
based on lowest AIC.
Exponential 3
0.7063237
151.0124763
2140.151
1645.211
3032.9951
Exponential 4
0.7063237
151.0124763
2140.151
1640.755
3032.9951
Polynomial 3°
0.7042845
151.0182587
2288.456
1802.442
5406.6394
Polynomial 2°
0.7042845
151.0182587
2288.456
1802.331
4750.7133
Power
0.7042845
151.0182587
2288.418
1804.786
5426.4704
Linear
0.7042845
151.0182587
2288.412
1802.472
3160.3402
a Results assuming constant variance presented (Test 2 p = 0.05); selected model in bold.
6.5
6.3
6.1
5.9
-------
USER INPUT
Info
Model
frequentist Exponential degree 3 vl.l
Dataset Name
Day 1 fetal body weight in F1A female Wistar rats
Dose-Response Model
M[dose] = a * exp(±l * (b * dose)Ad)
Variance Model
Var[i] = alpha
Model Options
BMR Type
Std. Dev.
BMRF
0.5
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL OUTPUT
Benchmark Dose
BMD
2140.151258
BMDL
1645.210527
BMDU
3032.995056
AIC
151.0124763
Test 4 P-value
0.706323686
d.f.
2
Model Parameters
# of Parameters
4
Variable
Estimate
a
6.152774757
b
2.39671E-05
d
Bounded
log-alpha
-1.171066995
Page 150 of 244
-------
Goodness of
Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
6.152774757
6.2
6.2
0.55680873
0.46
0.46
0.4240706
538
25
6.073948193
6
6
0.55680873
0.55
0.55
-0.664036
1965
24
5.86972465
5.9
5.9
0.55680873
0.5
0.5
0.2663721
7793
13
5.10452406
5.1
5.1
0.55680873
0.85
0.85
-0.0292950
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-72.15855649
5
154.317113
A2
-68.28868638
8
152.577373
A3
-72.15855649
5
154.317113
fitted
-72.50623816
3
151.012476
R
-86.92819994
2
177.8564
* Includes additive constant of -79.94765. This constant was not included in the LL derivation prior to
BMDS 3.0.
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
37.27902712
6
<0.0001
2
7.739740213
3
0.05170815
3
7.739740213
3
0.05170815
4
0.695363336
2
0.70632369
Page 151 of 244
-------
5.6.2 Wistar Rat F1A Fetal Body Weight at PND1 (Males)
Wistar Rat F1A Fetal
Jody Weight Data at
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
6.6
0.41
538.0
25
6.3
0.67
1965
24
6.3
0.47
7793
16
5.5
0.95
Table 5-15 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Fetal Body Weight at
PND1 (Males) Using Daily Average AUC as the Dose Metric
Model a
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.3716029
174.2465268
2383.601
1804.475
3461.8727
Constant variance model
did not fit adequately.
Non-constant variance
model fit adequately, but
no model fit means
adequately with this
variance model. In the
context of a sensitivity
analysis, the polynomial
3° model was selected,
assuming constant
variance.
Exponential 3
0.3716029
174.2465268
2383.593
1807.673
3461.8886
Exponential 4
0.3716029
174.2465268
2383.593
1804.471
3461.8886
Polynomial
3°
0.3731475
174.2382308
2612.253
1963.694
5880.3753
Polynomial
2°
0.3726155
174.2410841
2526.986
1963.472
5182.3553
Power
0.3726155
174.2410842
2526.92
1966.249
7406.5677
Linear
0.3726154
174.2410849
2527.32
1963.667
3577.1007
a Results assuming constant variance presented (BMDS Test 2 p-value < 0.01, Test 3 p-value = 0.26).
7.3
2000
3000
4000
Dose
5000
6000
7000
Estimated Probability
Response at BMD
O Data
BMD
BMDL
Figure 5.6-2 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with
Constant Variance for Wistar Rat F1A Fetal Body Weight at PND1 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 152 of 244
-------
USER INPUT
Info
Model
frequentist Polynomial degree 3 vl.l
Dataset Name
Day 1 fetal body weight in F1A male Wistar rats
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
2612.253174
BMDL
1963.693651
BMDU
5880.375282
AIC
174.2382308
Test 4 P-value
0.373147505
d.f.
2
Model Parameters
# of Parameters
5
Variable
Estimate
8
6.502864556
betal
-0.00012395
beta2
Bounded
beta3
Bounded
alpha
0.379635268
Page 153 of 244
-------
Goodness of
Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
6.502864556
6.6
6.6
0.616145
0.41
0.41
0.78825087
538
25
6.436167867
6.3
6.3
0.616145
0.67
0.67
-1.10499767
1965
24
6.258726554
6.3
6.3
0.616145
0.47
0.47
0.32816562
7793
16
5.500922185
5.5
5.5
0.616145
0.95
0.95
-0.00598680
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-83.1333339
5
176.266668
A2
-74.41376771
8
164.827535
A3
-83.1333339
5
176.266668
fitted
-84.11911538
3
174.238231
R
-97.11291497
2
198.22583
* Includes additive constant of -82.70z
47. This constant was not incluc
ed in the LL derivation prior to
BMDS 3.0.
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
45.39829452
6
<0.0001
2
17.43913238
3
0.00057397
3
17.43913238
3
0.00057397
4
1.971562962
2
0.37314751
Table 5-16 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Fetal Body Weight at
PND1 (Males) Using Daily Average AUC as the Dose Metric.
All SDs set to Minimum SD Across the Group.
Model a
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.6511585
248.9425337
2383.593
1599.703
4482.8186
Assuming
constant
variance, the
polynomial 3°
model was
selected based
on lowest AIC.
Exponential 3
0.6511585
248.9425337
2383.593
1603.702
4482.8186
Exponential 4
0.6511585
248.9425337
2383.593
1599.703
4482.8186
Polynomial 3°
0.6523374
248.9389161
2612.313
1764.59
6500.3761
Polynomial 2°
0.6519317
248.9401603
2526.993
1764.605
5959.5439
Power
0.6519317
248.9401603
2526.98
1764.555
7541.4133
Linear
0.6519317
248.9401603
2526.986
1764.597
4571.7527
a Results assuming constant variance presented (BMDS Test 2 p-value = 1.00).
Page 154 of 244
-------
7
Dose
Figure 5.6-3 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with
Constant Variance for Wistar Rat F1A Fetal Body Weight at PND1 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L; all SDs set to the
minimum SD across the groups
USER INPUT
Info
Model
frequentist Polynomial degree 3 vl.l
Dataset Name
Day 1 fetal body weight in F1A male Wistar rats
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
Page 155 of 244
-------
MODEL OUTPUT
Benchmark Dose
BMD
2612.31263
BMDL
2120.291103
BMDU
5121.060892
AIC
101.324428
Test 4 P-value
0.105675358
d.f.
2
Model Parameters
# of Parameters
5
Variable
Estimate
8
6.502860455
betal
-0.000123946
beta2
Bounded
beta3
Bounded
alpha
0.168856677
Goodness of
Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
6.502860455
6.6
6.6
0.41092174
0.41
0.41
1.18197136
538
25
6.436165633
6.3
6.3
0.41092174
0.41
0.41
-1.65683170
1965
24
6.258728964
6.3
6.3
0.41092174
0.41
0.41
0.49203033
7793
16
5.500924526
5.5
5.5
0.41092174
0.41
0.41
-0.0089995
Likelihooc
s of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-45.41483043
5
100.829661
A2
-45.41306385
8
106.826128
A3
-45.41483043
5
100.829661
fitted
-47.66221398
3
101.324428
R
-72.91234561
2
149.824691
* Includes additive constant of -82.70447. This constant was not incluc
BMDS 3.0.
ed in the LL derivation prior to
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
54.99856352
6
<0.0001
2
0.003533171
3
0.9999442
3
0.003533171
3
0.9999442
4
4.494767098
2
0.10567536
Page 156 of 244
-------
Table 5-17 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Fetal Body Weight at
PND1 (Males) Using Daily Average AUC as the Dose Metric. All SDs set to Maximum SD Across
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for
model
selection
Test 4
P-value
AIC
Exponential 2
0.6511585
248.9425337
2383.593
1599.703
4482.8186
Assuming
constant
variance, the
polynomial 3°
model was
selected based
on lowest AIC.
Exponential 3
0.6511585
248.9425337
2383.593
1603.702
4482.8186
Exponential 4
0.6511585
248.9425337
2383.593
1599.703
4482.8186
Polynomial 3°
0.6523374
248.9389161
2612.313
1764.59
6500.3761
Polynomial 2°
0.6519317
248.9401603
2526.993
1764.605
5959.5439
Power
0.6519317
248.9401603
2526.98
1764.555
7541.4133
Linear
0.6519317
248.9401603
2526.986
1764.597
4571.7527
a Results assuming constant variance presented (BMDS Test 2 p-value = 1.00).
7.3
6000
Estimated Probability
Response at BMD
O Data
BMD
BMDL
Figure 5.6-4 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with
Constant Variance for Wistar Rat F1A Fetal Body Weight at PND1 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L; all SDs set to the
maximum SD across the groups
USER INPUT
Info
Model
frequentist Polynomial degree 3 vl.l
Dataset Name
Day 1 fetal body weight in F1A male Wistar rats
Dose-Response Model
M[dose] = g + b 1 *dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha
Page 157 of 244
-------
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL OUTPUT
Benchmark Dose
BMD
2612.31263
BMDL
1764.589754
BMDU
6500.376091
AIC
248.9389161
Test 4 P-value
0.652337387
d.f.
2
Model Parameters
# of Parameters
5
Variable
Estimate
8
6.502862348
betal
-0.000123946
beta2
Bounded
beta3
Bounded
alpha
0.870621557
Goot
of
ness
Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observe
d SD
Scaled
Residual
0
25
6.502862348
6.6
6.6
0.9330710
0.95
0.95
0.5205265
538
25
6.436167543
6.3
6.3
0.9330710
0.95
0.95
-0.7296740
1965
24
6.258730828
6.3
6.3
0.9330710
0.95
0.95
0.2166789
7793
16
5.500920556
5.5
5.5
0.9330710
0.95
0.95
-0.0039463
Page 158 of 244
-------
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-121.0422647
5
252.084529
A2
-121.0404981
8
258.080996
A3
-121.0422647
5
252.084529
fitted
-121.469458
3
248.938916
R
-127.6007931
2
259.201586
* Includes additive constant of -82.70447. This constant was not includec
BMDS 3.0.
in the LL derivation prior to
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
13.12059002
6
0.04116044
2
0.003533146
3
0.9999442
3
0.003533146
3
0.9999442
4
0.854386774
2
0.65233739
Page 159 of 244
-------
5.6.3 Wistar Rat F1A Pup Body Weight at PND7 (Females)
Wistar Rat F1A
AUC (hr mg/L)
0
538.0
1965
7793
up Body Weight Data at PND7 (Females) used for BMD Modeling
N
25
25
24
Mean
14.3
13.4
13.7
11.1
Std. Dev.
1.36
1.56
1.6
4.23
Table 5-18 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at
PND7 (Females) Using Daily Average AUC as the Dose Metric
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.0975118
313.3363909
1971.475
1124.407
5002.8131
Assuming non-
constant variance,
of the models that
fit the means
adequately (Test
4 p-value > 0.10),
the polynomial 3°
model was
selected based on
lowest AIC.
Exponential 3
0.0323356
315.2614419
2849.393
1132.245
6431.9847
Exponential 4
0.0989096
313.307925
2113.589
1127.362
4987.7772
Polynomial 3°
0.1362271
312.667691
3045.909
1277.225
10384.373
Polynomial 2°
0.1243995
312.8493423
2784.37
1255.667
11080.809
Power
0.0341136
315.1698264
2319.472
1220.992
7714.5917
Linear
0.1058252
313.1727608
2158.999
1220.729
11324.451
a Results assuming non-constant variance presented (BMDS Test 2 p < 0.01, Test 3 p = 0.85); selected model in
bold.
16
15
IB
12
11
10
9
8
7
Estimated Probability
Response at BMD
O Data
BMD
BMDL
1000
2000
3000
4000
Dose
5000
6000
7000
Figure 5.6-5 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with Non-
constant Variance for Wistar Rat F1A Pup Body Weight at PND7 (Females)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 160 of 244
-------
USER INPUT
Info
Model
frequentist Polynomial degree 3 vl.l
Dataset Name
Day 7 pup body weight in F1A female Wistar rats
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha * mean[i] A rho
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Non-Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
3045.909437
BMDL
1277.225416
BMDU
10384.37289
AIC
312.667691
Test 4 P-value
0.136227112
d.f.
2
Model Parameters
# of Parameters
6
Variable
Estimate
8
13.97477502
betal
-0.000203959
beta2
Bounded
beta3
Bounded
alpha
-8.483523005
Page 161 of 244
-------
Goodness of
Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
13.97477502
14.3
14.3
1.43641314
1.36
1.36
1.1320733
538
25
13.86461787
13.4
13.4
1.48544976
1.56
1.56
-1.563896
1965
24
13.55318758
13.7
13.7
1.63572057
1.6
1.6
0.4397029
7793
6
11.0874095
11.1
11.1
3.83388349
4.23
4.23
0.0080441
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-159.2746464
5
328.549293
A2
-150.1744151
8
316.34883
A3
-150.3404137
6
312.680827
fitted
-152.3338455
4
312.667691
R
-166.6684684
2
337.336937
* Includes additive constant of -73.51508. This constant was not incluc
BMDS 3.0.
ed in the LL derivation prior to
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
32.98810654
6
<0.0001
2
18.20046269
3
0.0003999
3
0.331997137
2
0.84704745
4
3.986863692
2
0.13622711
Page 162 of 244
-------
5.6.4 Wistar Rat F1A Pup Body Weight at PND7 (Males)
Wistar Rat F1A Pup Body Weight Data at PND7 (Males) used for BMD Modeling
AUC (hr mg/L)
N
Mean
Std. Dev.
0
25
15
1.2
538.0
25
13.7
2.03
1965
24
14.7
1.66
7793
7
12
4.24
Table 5-19 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at
Model
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 2
0.0232561
350.7071703
2346.337
1424.574
5949.4775
Constant variance model
did not fit adequately.
Only polynomial 3°
model fit the means
adequately assuming
constant variance, but its
residual at the low dose
group was high (1.9).
Non-constant variance
model fit adequately, but
no model fit means
adequately w/ this
variance model. No
model was selected.
Exponential 3
0.014954
351.1066788
7131.763
1668.443
7587.3348
Exponential 4
0.0232561
350.7071703
2346.337
1424.574
5949.4775
Polynomial 3°
0.1100963
347.2161199
5181.943
1739.039
6691.0253
Polynomial 2°
0.0959664
347.5301189
4240.252
1686.923
6287.3627
Power
0.014954
351.1066787
7443.559
7229.447
7773.2364
Linear
0.0250331
350.5599121
2433.142
1557.746
5800.1603
a Results assuming constant variance presented (BMDS Test 2 p < 0.01, Test 3 p = 0.66).
17
Estimated Probability
Response at BMD
O Data
BMD
BMDL
Figure 5.6-6 Plot of Mean Response by Dose, with Fitted Curve for Lines Model with Constant
Variance for Wistar Rat F1A Pup Body Weight at PND7 (Females)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 163 of 244
-------
5.6.5 Wistar Rat F1A Pup Body Weight at PND21 (Females)
Wistar Rat F1A
AUC (hr mg/L)
0
538.0
1965
7793
up Body Weight Data at PND21 (Females) used for BMD Modeling
N
25
25
24
Mean
47.9
46.6
47.6
44
Std. Dev.
3.09
4.24
4.05
3.71
Table 5-20 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.3004956
440.1433811
6106.267
3050.063
Infinity
Polynomial 3°
model assuming
constant variance
was selected
based on lowest
AIC.
Exponential 3
0.198527
441.3919138
7456.866
3304.092
33820.167
Exponential 4
0.3004956
440.1433808
6104.661
3050.074
Infinity
Polynomial 3°
0.6376365
437.4355767
6935.914
3353.844
34987.938
Polynomial 2°
0.6080195
437.5706289
6572.341
3304.633
41584.807
Power
0.1985298
441.3918931
7690.254
6557.246
7918.5146
Linear
0.304673
440.1157687
6078.114
3132.698
Infinity
a Results assuming constant variance presented (BMDS Test 2 p = 0.42); selected model in bold.
50
38
Estimated Probability
^^Response at BMD
O Data
BMD
BMDL
1000
2000
3000
4000
Dose
5000
6000
7000
Figure 5.6-7 Plot of Mean Response by Dose, with Fitted Curve for Polynomial 3 Model with
Constant Variance for Wistar Rat F1A Pup Body Weight at PND21 (Females)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 164 of 244
-------
USER INPUT
Info
Model
frequentist Polynomial degree 3 vl.l
Dataset Name
Day 21 pup body weight in F1A female Wistar rats
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
6935.913883
BMDL
3353.843846
BMDU
34987.93818
AIC
437.4355767
Test 4 P-value
0.637636501
d.f.
3
Model Parameters
# of Parameters
5
Variable
Estimate
8
47.38002134
betal
Bounded
beta2
Bounded
beta3
Bounded
alpha
14.13493282
Page 165 of 244
-------
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
47.38002134
47.9
47.9
3.75964531
3.09
3.09
0.6915262
538
25
47.37891573
46.6
46.6
3.75964531
4.24
4.24
-1.0358899
1965
24
47.3261519
47.6
47.6
3.75964531
4.05
4.05
0.35683585
7793
5
44.01979256
44
44
3.75964531
3.71
3.71
-0.0117717
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-215.8693682
5
441.738736
A2
-214.4497435
8
444.899487
A3
-215.8693682
5
441.738736
fitted
-216.7177884
2
437.435577
R
-218.5281122
2
441.056224
* Includes additive constant of -72.59614. This constant was not included in the LL derivation prior to
BMDS 3.0.
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
8.156737342
6
0.22684381
2
2.839249353
3
0.41707946
3
2.839249353
3
0.41707946
4
1.696840284
3
0.6376365
Page 166 of 244
-------
5.6.6 Wistar Rat F1A Pup Body Weight at PND21 (Males)
Wistar Rat F1A
AUC (hr mg/L)
0
538.0
1965
7793
up Body Weight Data at PND21 (Males) used for BMD Modeling
N
25
25
24
Mean
50.5
49.1
50.£
44.5
Std. Dev.
2.58
5.34
4.75
2.59
Table 5-21 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 2
0.0733481
467.037135
4047.756
2406.045
11736.316
Constant variance
model did not fit
adequately. Non-
constant variance
model fit adequately,
but no model fit means
adequately with this
variance model. In the
context of a sensitivity
analysis, the
polynomial 3° model
was selected, assuming
constant variance.
Exponential 3
0.1294555
466.1110791
7067.155
3517.961
7766.675
Exponential 4
0.0733481
467.037135
4047.756
2406.045
11736.316
Polynomial 3°
0.4819038
462.2756656
5960.325
3423.292
7685.371
Polynomial 2°
0.3956477
462.7860807
5257.935
3136.386
7838.2217
Power
0.129462
466.1110017
7560.324
5771.249
7787.9527
Linear
0.0782586
466.9075313
4053.597
2494.705
11188.116
a Results assuming constant variance presented (BMDS Test 2 p < 0.01, Test 3 p < 0.01).
54
52
50
(L
8 48
c
o
GL
in
£ 46
44
42
40
Estimated Probability
Response at BMD
O Data
BMD
BMDL
1000
2000
3000
4000
Dose
5000
6000
7000
Figure 5.6-8 Plot of Mean Response by Dose, with Fitted Curve for Polynomial Degree 3 Model
with Constant Variance for Wistar Rat F1A Pup Body Weight at PND21 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L
Page 167 of 244
-------
USER INPUT
Info
Model
frequentist Polynomial degree 3 vl.l
Dataset Name
Day 21 pup body weight in F1A male Wistar rats
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
MODEL RESULTS
Benchmark Dose
BMD
5960.324782
BMDL
3423.29248
BMDU
7685.370953
AIC
462.2756656
Test 4 P-value
0.48190384
d.f.
3
Model Parameters
# of Parameters
5
Variable
Estimate
8
50.15039221
betal
Bounded
beta2
Bounded
beta3
Bounded
alpha
18.00362096
Page 168 of 244
-------
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
50.15039221
50.5
50.5
4.2430674
2.58
2.58
0.4119753
538
25
50.14854812
49.1
49.1
4.2430674
5.34
5.34
-1.2356015
1965
24
50.06054123
50.8
50.8
4.2430674
4.75
4.75
0.85376757
7793
6
44.54573298
44.5
44.5
4.2430674
2.59
2.59
-0.0264013
Likelihoods of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-227.9060288
5
465.812058
A2
-220.1176472
8
456.235294
A3
-227.9060288
5
465.812058
fitted
-229.1378328
2
462.275666
R
-233.6652209
2
471.330442
* Includes additive constant of -73.51508. This constant was not included in the LL derivation prior to
BMDS 3.0.
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
27.09514741
6
0.00013898
2
15.5767632
3
0.00138457
3
15.5767632
3
0.00138457
4
2.463608041
3
0.48190384
Table 5-22 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at
PND21 (Males) Using Daily Average AUC as the Dose Metric.
Model
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model selection
Test 4
P-value
AIC
Exponential 3
0.012827
388.7658213
7412.757
4697.234
7599.4962
Assuming constant
variance, no model fit the
means adequately (Test 4
p-value < 0.10 for all
models). No model was
selected.
Polynomial 3°
0.0848587
385.1980681
5960.421
4635.693
6808.6243
Polynomial 2°
0.0469322
386.5288966
5257.727
4201.064
6474.939
Power
0.0128272
388.7657993
7579.644
4737.149
7726.4385
a Results assuming constant variance presented (BMDS Test 2 p = 1.00).
Page 169 of 244
-------
54
52
50
(L
48
44
42
40
()
()
1000
2000
3000
4000
Dose
5000
6000
7000
^^Estimated Probability
^^Response at BMD
O Data
BMD
BMDL
Figure 5.6-9 Plot of Mean Response by Dose, with Fitted Curve for Polynomial Degree 3 Model
with Constant Variance for Wistar Rat F1A Pup Body Weight at PND21 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L; all SDs set to the
minimum SD across the groups
Table 5-23 Model Predictions for AUC (hr mg/L) versus Wistar Rat F1A Pup Body Weight at
PND21 (Males) Using Daily Average AUC as the Dose Metric.
Modela
Goodness of fit
BMD
(hr
mg/L)
BMDL
(hr
mg/L)
BMDU
(hr mg/L)
Basis for model
selection
Test 4
P-value
AIC
Exponential 3
0.2223544
500.4519173
7071.418
2795.856
7908.6121
Assuming constant
variance, the polynomial
3° model was selected
based on lowest AIC.
Polynomial 3°
0.6602299
496.5591027
5960.421
2772.076
8727.6197
Polynomial 2°
0.587235
496.8919861
5258.084
2640.539
9314.8981
Power
0.2223618
500.4518697
7559.672
7419.501
7839.4866
a Results assuming constant variance presented (BMDS Test 2 p = 1.00).
Page 170 of 244
-------
54
52
50
48
. 46
44
42
40
38
Estimated Probability
Response at BMD
O Data
BMD
BMDL
1000
2000
3000
4000
Dose
5000
6000
7000
Figure 5.6-10 Plot of Mean Response by Dose, with Fitted Curve for Polynomial Degree 3 Model
with Constant Variance for Wistar Rat F1A Pup Body Weight at PND21 (Males)
BMR = 5% relative deviation; daily average AUC as dose shown in hr mg/L; all SDs set to the
maximum SD across the groups
USER INPUT
Info
Model
frequentist Polynomial degree 3 vl.l
Dataset Name
Day 21 pup body weight in F1A male Wistar
rats-max Sprague-Dawley
Dose-Response Model
M[dose] = g + bl*dose + b2*doseA2 + ...
Variance Model
Var[i] = alpha
Model Options
BMR Type
Rel. Dev.
BMRF
0.05
Tail Probability
-
Confidence Level
0.95
Distribution Type
Normal
Variance Type
Constant
Model Data
Dependent Variable
Dose
Independent Variable
Mean
Total # of Observations
4
Adverse Direction
Automatic
Page 171 of 244
-------
MODEL RESULTS
Benchmark Dose
BMD
5960.421398
BMDL
2772.076153
BMDU
8727.619723
AIC
496.5591027
Test 4 P-value
0.660229866
d.f.
3
Model Parameters
# of Parameters
4
Variable
Estimate
8
50.15035834
betal
Bounded
beta2
Bounded
beta3
Bounded
alpha
27.63579123
Goodness of Fit
Dose
Size
Estimated
Median
Calc'd
Median
Observed
Mean
Estimated
SD
Calc'd
SD
Observed
SD
Scaled
Residual
0
25
50.15035834
50.5
50.5
5.25697548
5.34
5.34
0.332550213
538
25
50.14851434
49.1
49.1
5.25697548
5.34
5.34
-0.99726006
1965
24
50.0605119
50.8
50.8
5.25697548
5.34
5.34
0.689129526
7793
6
44.54598299
44.5
44.5
5.25697548
5.34
5.34
-0.02142579
Likelihooc
s of Interest
Model
Log Likelihood*
# of Parameters
AIC
A1
-245.4814031
5
500.962806
A2
-245.454905
8
506.90981
A3
-245.4814031
5
500.962806
fitted
-246.2795513
2
496.559103
R
-249.2864426
2
502.572885
* Includes additive constant of -73.51508. This constant was not incluc
BMDS 3.0.
ed in the LL derivation prior to
Tests of Interest
Test
-2*Log(Likelihood Ratio)
Test d.f.
p-value
1
7.663075137
6
0.2638408
2
0.052996268
3
0.99680631
3
0.052996268
3
0.99680631
4
1.596296449
3
0.66022987
Page 172 of 244
-------
5.7 Results for BMD Modeling for Stillbirths, and PND4 and PND21 Pup
Deaths in Sprague-Dawley Rats (NMP Producers Group (1999a))
5.7.1 Sprague-Dawley Rat F1A stillborn/total delivered (NMP Producers Group (1999a))
Sprague-Dawley Rat F]
A stillborn/total delivered (NMP Producers Group (1999a))
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
Stillborn
Covariate
(mg, LD1 Dam BW)
0
0
13
0
245
0
0
8
0
273
0
0
11
0
278
0
0
13
0
280
0
0
16
0
281
0
0
11
0
283
0
0
12
1
284
0
0
17
0
289
0
0
11
0
294
0
0
13
0
303
0
0
13
1
308
0
0
15
0
309
0
0
15
0
311
0
0
16
0
311
0
0
12
0
313
0
0
13
0
315
0
0
16
0
315
0
0
16
0
317
0
0
15
0
319
0
0
16
0
319
0
0
15
0
320
0
0
16
0
323
0
0
17
4
323
0
0
14
0
324
0
0
14
0
366
589.1
40.87
14
0
272
589.1
40.87
11
0
276
589.1
40.87
14
0
281
589.1
40.87
14
0
285
589.1
40.87
16
0
288
589.1
40.87
15
0
288
589.1
40.87
14
0
291
589.1
40.87
14
0
294
589.1
40.87
16
0
295
589.1
40.87
14
0
296
589.1
40.87
1
1
298
589.1
40.87
10
1
300
589.1
40.87
11
0
302
589.1
40.87
15
0
302
589.1
40.87
14
0
306
589.1
40.87
17
2
313
589.1
40.87
18
0
314
Page 173 of 244
-------
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
Stillborn
Covariate
(mg, LD1 Dam BW)
589.1
40.87
6
0
316
589.1
40.87
13
0
317
589.1
40.87
16
0
317
589.1
40.87
11
0
318
589.1
40.87
7
0
324
589.1
40.87
18
0
326
589.1
40.87
14
1
328
589.1
40.87
14
0
328
589.1
40.87
13
0
333
2117
142.35
11
0
231
2117
142.35
16
1
253
2117
142.35
16
0
260
2117
142.35
14
0
280
2117
142.35
14
0
288
2117
142.35
15
0
292
2117
142.35
12
0
294
2117
142.35
13
0
295
2117
142.35
14
0
299
2117
142.35
16
0
301
2117
142.35
15
0
302
2117
142.35
14
1
304
2117
142.35
14
0
309
2117
142.35
10
0
312
2117
142.35
14
0
314
2117
142.35
17
0
314
2117
142.35
14
0
315
2117
142.35
13
0
316
2117
142.35
16
0
321
2117
142.35
16
0
323
2117
142.35
16
0
324
2117
142.35
10
0
329
2117
142.35
14
0
331
2117
142.35
15
0
344
8511
557.5
14
0
243
8511
557.5
12
0
243
8511
557.5
9
0
250
8511
557.5
6
0
255
8511
557.5
11
4
256
8511
557.5
15
1
261
8511
557.5
11
0
266
8511
557.5
15
0
269
8511
557.5
11
0
274
8511
557.5
17
1
276
8511
557.5
13
1
280
8511
557.5
12
0
282
8511
557.5
13
0
282
8511
557.5
15
0
283
8511
557.5
15
0
287
Page 174 of 244
-------
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
Stillborn
Covariate
(mg, LD1 Dam BW)
8511
557.5
15
0
287
8511
557.5
14
1
288
8511
557.5
14
0
292
8511
557.5
13
0
293
8511
557.5
11
1
294
8511
557.5
15
1
299
8511
557.5
9
0
300
8511
557.5
13
0
300
8511
557.5
18
2
301
8511
557.5
4
0
306
8511
557.5
15
8
306
8511
557.5
15
0
318
8511
557.5
15
0
329
8511
557.5
9
0
336
Table 5-24 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Sprague-
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model
Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
No model is chosen
because all model p-
values are below 0.1.
Nlogistic (b. seedb = 1597161083)
0.0537
276.885
7445.74
1555.12
NCTR (b. seed = 1597161085)
0.0483
274.881
7460.59
6217.16
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1597161079)
0
304.173
7349.34
2549.34
NCTR (b. seed = 1597161080)
0
302.116
7369.93
6141.61
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed = 1597161067)
0.051
272.956
7442.34
1546.16
NCTR (b. seed = 1597161072)
0.0523
270.956
7465.2
6221
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1597161075)
0
300.939
7438.23
2810.64
NCTR (b. seed = 1597161077)
0
298.939
7459.66
6216.39
aLitter-specific data were fit using BMDS NLogistic and NCTR nested dichotomous models. Adequate model fit (p-value
>0.1) was not achieved for either standard restricted (shown) and unrestricted (not shown) model forms.
b b. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
Page 175 of 244
-------
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
dose
11:51 08/11 2020
Figure 5.7-1 Plot of NLogistic (no LSC; ICC estimated) model for AUC (hr mg/L) versus Sprague-
Dawley Rat F1A stillborn/total delivered.
Table 5-25 Summary of BMDS nested modeling results for Cmax (mg/L) versus Sprague-Dawley
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model
Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
No model is chosen
because all model p-
values are below 0.1.
Nlogistic (b. seedb = 1597185415)
0.0533
276.885
488.742
102.875
NCTR (b. seed = 1597185417)
0.046
274.881
489.813
408.177
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1597185410)
0
304.173
482.277
170.785
NCTR (b. seed = 1597185412)
0
302.116
483.956
403.297
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed = 1597185401)
0.0537
272.956
488.56
102.333
NCTR (b. seed = 1597185404)
0.0507
270.956
490.11
408.425
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1597185407)
0
300.939
488.285
187.934
NCTR (b. seed = 1597185409)
0
298.939
489.746
408.122
a Litter-specific data were fit using BMDS NLogistic and NCTR nested dichotomous models. Adequate model fit (p-value
>0.1) was not achieved for either standard restricted (shown) and unrestricted (not shown) model forms.
bb. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
Page 176 of 244
-------
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
dose
18:36 08/11 2020
Figure 5.7-2 Plot of NLogistic (no LSC; ICC estimated) model for Cmax(mg/L) versus Sprague-
Dawley Rat F1A stillborn/total delivered.
Page 177 of 244
-------
5.7.2 Sprague-Dawley Rat F2B Pup death at PND4/total delivered (NMP Producers
Group (1999a))
Sprague-Dawley Rat F2B Pup Death at PND4/total Delivered (NMP Producers Group (1999a))
AUC
(hr mg/L)
Total Delivered
PND4 Pup Deaths
Covariate
(mg, LD1 Dam BW)
0
17
6
342
0
18
0
346
0
11
0
355
0
17
0
356
0
16
0
358
0
16
1
358
0
14
0
358
0
17
0
365
0
12
1
368
0
12
0
369
0
14
3
369
0
19
3
373
0
18
0
377
0
19
1
378
0
17
0
381
0
10
0
384
0
16
0
385
0
12
1
386
0
13
0
387
0
15
1
387
0
18
4
389
0
17
2
394
0
16
0
394
0
18
0
417
0
18
2
421
566.5
16
8
279
566.5
13
0
321
566.5
8
0
324
566.5
16
0
330
566.5
14
1
334
566.5
12
0
338
566.5
15
0
342
566.5
15
0
345
566.5
19
1
347
566.5
13
4
348
566.5
17
0
349
566.5
17
0
359
566.5
18
0
372
566.5
20
1
372
566.5
15
0
381
566.5
8
2
385
566.5
14
0
386
566.5
17
0
390
566.5
9
0
394
Page 178 of 244
-------
r mg/
566.5
566.5
566.5
566.5
566.5
566.5
566.5
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
Total Delivered
19
19
10
19
20
19
14
14
12
17
13
14
12
19
18
16
17
16
15
17
14
14
15
15
21
18
14
17
19
15
14
13
19
14
16
13
18
15
16
13
14
16
12
19
PND4 Pup Deaths
12
19
Covariate
(mg, LD1 Dam BW)
403
413
413
419
427
447
456
290
308
309
318
323
324
324
325
337
340
347
358
363
369
381
381
381
388
394
401
407
409
423
433
294
306
319
326
337
337
350
359
366
367
370
371
375
378
381
381
389
Page 179 of 244
-------
AUC
(hr mg/L)
Total Delivered
PND4 Pup Deaths
Covariate
(mg, LD1 Dam BW)
5235
10
0
389
5235
17
0
395
5235
16
4
395
5235
19
0
398
5235
15
1
423
5235
8
0
445
5235
15
0
456
Table 5-26 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Sprague-
Dawley Rat F2B Pup death at PND4 /total delivered (NMP Producers Group (1999a)); BMR = 1%
extra risk.
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
While some models met the
p-value fit criteria (p-value >
0.1), no model was deemed
to appropriate after visual
inspection of model plots,
which indicates considerable
model uncertainty and a
dose-response pattern
analogous to having a
positive response at only the
highest dose.
Nlogistic (b. seedb = 1597167183)
0.2783
624.069
21778.9
212.473
NCTR (b. seed = 1597167185)
0.469
612.588
4422.47
3685.39
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to be
zero
Nlogistic (b. seed = 1597167179)
0
751.826
4733.93
3044.98
NCTR (b. seed = 1597167181)
0
764.134
4501.04
3750.86
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed = 1597167169)
0.1837
620.686
21779.5
201.176
NCTR (b. seed = 1597167173)
0.3973
611.342
4450.73
3708.94
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1597167176)
0
787.278
4526.5
2061.5
NCTR (b. seed = 1597167177)
0
785.278
4533.37
3777.81
aLitter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models. No model
was chosen due the considerable model uncertainty indicated by visual inspection of model plots.
bb. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
Page 180 of 244
-------
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
Nested Logistic ¦
0.15 : ~r
17:38 07/17 2020
100000 150000
dose
Figure 5.7-3 Plot of NLogistic (no LSC; ICC estimated) model for AUC (hr mg/L) versus Sprague-
Dawley Rat F2B Pup Death at PND4/Total Delivered.
Page 181 of 244
-------
5.7.3 Sprague-Dawley Rat F2B Pup death at PND21/PND4 post-cull (NMP Producers
Group (1999a))
Sprague-Dawley Rat F2B Pup Death at PND2]
/PND4 Post-cull (NMP Producers Group (1999a))
AUC
(hr mg/L)
PND4 Live Post-cull
PND21 Pup Deaths
Covariate
(mg, LD1 Dam BW)
0
10
0
342
0
10
0
346
0
10
0
355
0
10
0
356
0
10
0
358
0
10
0
358
0
10
0
358
0
10
0
365
0
10
0
368
0
10
0
369
0
10
0
369
0
10
1
373
0
10
0
377
0
10
0
378
0
10
0
381
0
10
0
384
0
10
0
385
0
10
0
386
0
10
0
387
0
10
0
387
0
10
0
389
0
10
0
394
0
10
0
394
0
10
0
417
0
10
0
421
566.5
8
0
279
566.5
10
0
321
566.5
8
0
324
566.5
10
3
330
566.5
10
0
334
566.5
10
0
338
566.5
10
0
342
566.5
10
0
345
566.5
10
0
347
566.5
9
0
348
566.5
10
0
349
566.5
10
0
359
566.5
10
0
372
566.5
10
0
372
566.5
10
0
381
566.5
6
0
385
566.5
10
0
386
566.5
10
0
390
566.5
9
0
394
Page 182 of 244
-------
r mg/
566.5
566.5
566.5
566.5
566.5
566.5
566.5
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
2053
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
5235
PND4 Live Post-cull
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
PND21 Pup Deaths
0
10
Covariate
(mg, LD1 Dam BW)
403
413
413
419
427
447
456
290
308
309
318
323
324
324
325
337
340
347
358
363
369
381
381
381
388
394
401
407
409
423
433
294
306
319
337
337
350
359
366
367
371
375
378
381
381
389
389
395
Page 183 of 244
-------
AUC
(hr mg/L)
PND4 Live Post-cull
PND21 Pup Deaths
Covariate
(mg, LD1 Dam BW)
5235
10
0
395
5235
10
1
398
5235
10
1
423
5235
8
0
445
5235
10
0
456
Table 5-27 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Sprague-
Dawley Rat F2B Pup death at PND21/PND4 post-cull (NMP Producers Group (1999a)).
BMR =1% extra risk.
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
Nlogistic (b. seedb = 1597171302)
0.4993
136.056
2190.56
407.944
NCTR (b. seed = 1597171304)
0.4923
136.595
2063.58
1031.79
The NLogistic model that
estimated intra-litter
correlations but did not make
use of a litter-specific
covariate was selected based
on estimating the lowest
BMDL within a range of
BMDLs from acceptable
models (P-value > 0.1) that
varied more than 3-fold.
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to
be zero
Nlogistic (b. seed = 1597171298)
0.0157
184.305
3227.07
1468.34
NCTR (b. seed = 1597171299)
0.008
192.4
2157.95
1078.98
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed = 1597171290)
0.3293
135.305
1829.66
313.814
NCTR (b. seed = 1597171292)
0.3297
135.299
1816.24
908.119
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1597171294)
0
203.974
1697.58
555.973
NCTR (b. seed = 1597171296)
0
203.961
1674.73
837.367
aLitter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models. Selected
model is bolded.
bb. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
Page 184 of 244
-------
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
dose
14:41 08/11 2020
Figure 5.7-4 Plot of NLogistic model (LSC = LD1 dam weight; ICC estimated) for AUC (hr mg/L)
versus Sprague-Dawley Rat F2B Pup Death at PND21/PND4 Live Post-cull.
NLogistic Model. (Version: 2.20; Date: 04/27/2015)
Input Data File: C:/Users/jgift/BMDS2704/Data/SDF2b_Day21_pl563/Correct
Doses/BMRO l/nln_SDF2b_Day2 l_p 1563_Nln-BMRl -Restrict-IC. (d)
Tue Aug 11 14:41:30 2020
BMDS Model Run
The probability function is:
Prob. = alpha + thetal*Rij + [1 - alpha - thetal*Rij]/
[l+exp(-beta-theta2*Rij-rho*log(Dose))],
where Rij is the litter specific covariate.
Restrict Power rho >= 1.
Total number of observations = 97
Total number of records with missing values = 0
Total number of parameters in model = 9
Total number of specified parameters = 2
Maximum number of iterations = 500
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Page 185 of 244
-------
Number of Bootstrap Iterations per run: 1000
Bootstrap Seed: 1597171290
User specifies the following parameters:
thetal = 0
theta2 = 0
Default Initial Parameter Values
alpha = 0.0051889
beta = -23.4938
thetal = 0 Specified
theta2 = 0 Specified
rho = 2.51584
phil = 0
phi2 = 0.274833
phi3 = 0.205111
phi4 = 0.730024
Parameter Estimates
Variable
alpha
beta
rho
phil
phi2
phi3
phi4
Estimate
0.0051889
-23.4939
2.51584
0
0.274833
0.205111
0.730024
Std. Err.
0.00385081
0.509863
0.367021
Bounded
0.534557
NA
NA
Log-likelihood: -61.6524 AIC: 135.305
Litter Data
Lit.-Spec. Litter Scaled
Dose Cov. Est. Prob. Size Expected Observed Residual
0.0000
342.0000
0.005
10
0.052
0
-0.2284
0.0000
346.0000
0.005
10
0.052
0
-0.2284
0.0000
355.0000
0.005
10
0.052
0
-0.2284
0.0000
356.0000
0.005
10
0.052
0
-0.2284
0.0000
358.0000
0.005
10
0.052
0
-0.2284
0.0000
358.0000
0.005
10
0.052
0
-0.2284
0.0000
358.0000
0.005
10
0.052
0
-0.2284
0.0000
365.0000
0.005
10
0.052
0
-0.2284
0.0000
368.0000
0.005
10
0.052
0
-0.2284
0.0000
369.0000
0.005
10
0.052
0
-0.2284
0.0000
369.0000
0.005
10
0.052
0
-0.2284
0.0000
373.0000
0.005
10
0.052
1
4.1730
Page 186 of 244
-------
0.0000 377.0000 0.005
0.0000 378.0000 0.005
0.0000 381.0000 0.005
0.0000 384.0000 0.005
0.0000 385.0000 0.005
0.0000 386.0000 0.005
0.0000 387.0000 0.005
0.0000 387.0000 0.005
0.0000 389.0000 0.005
0.0000 394.0000 0.005
0.0000 394.0000 0.005
0.0000 417.0000 0.005
0.0000 421.0000 0.005
566.5000 279.0000 0.006
566.5000 321.0000 0.006
566.5000 324.0000 0.006
566.5000 330.0000 0.006
566.5000 334.0000 0.006
566.5000 338.0000 0.006
566.5000 342.0000 0.006
566.5000 345.0000 0.006
566.5000 347.0000 0.006
566.5000 348.0000 0.006
566.5000 349.0000 0.006
566.5000 359.0000 0.006
566.5000 372.0000 0.006
566.5000 372.0000 0.006
566.5000 381.0000 0.006
566.5000 385.0000 0.006
566.5000 386.0000 0.006
566.5000 390.0000 0.006
566.5000 394.0000 0.006
566.5000 403.0000 0.006
566.5000 413.0000 0.006
566.5000 413.0000 0.006
566.5000 419.0000 0.006
566.5000 427.0000 0.006
566.5000 447.0000 0.006
566.5000 456.0000 0.006
2053.0000 290.0000 0.018
2053.0000 308.0000 0.018
2053.0000 309.0000 0.018
2053.0000 318.0000 0.018
2053.0000 323.0000 0.018
2053.0000 324.0000 0.018
2053.0000 324.0000 0.018
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.052
0
-0.2284
0.046
0
-0.1254
0.057
0
-0.1286
0.046
0
-0.1254
0.057
3
6.6241
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.051
0
-0.1272
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.034
0
-0.1205
0.057
0
-0.1286
0.057
0
-0.1286
0.051
0
-0.1272
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.057
0
-0.1286
0.184
0
-0.2569
0.184
0
-0.2569
0.184
1
1.1366
0.184
0
-0.2569
0.184
0
-0.2569
0.184
0
-0.2569
0.184
0
-0.2569
10
10
10
10
10
10
10
10
10
10
10
10
10
8
10
8
10
10
10
10
10
10
9
10
10
10
10
10
6
10
10
9
10
10
10
10
10
10
10
10
10
10
10
10
10
10
Page 187 of 244
-------
2053.0000
325.0000
0.018
10
0.184
0
-0.2569
2053.0000
337.0000
0.018
10
0.184
0
-0.2569
2053.0000
340.0000
0.018
10
0.184
3
3.9234
2053.0000
347.0000
0.018
1
0.018
0
-0.1370
2053.0000
358.0000
0.018
10
0.184
0
-0.2569
2053.0000
363.0000
0.018
10
0.184
0
-0.2569
2053.0000
369.0000
0.018
10
0.184
0
-0.2569
2053.0000
381.0000
0.018
10
0.184
0
-0.2569
2053.0000
381.0000
0.018
10
0.184
0
-0.2569
2053.0000
381.0000
0.018
10
0.184
0
-0.2569
2053.0000
388.0000
0.018
10
0.184
0
-0.2569
2053.0000
394.0000
0.018
10
0.184
0
-0.2569
2053.0000
401.0000
0.018
10
0.184
0
-0.2569
2053.0000
407.0000
0.018
10
0.184
0
-0.2569
2053.0000
409.0000
0.018
2
0.037
0
-0.1766
2053.0000
423.0000
0.018
10
0.184
0
-0.2569
2053.0000
433.0000
0.018
10
0.184
0
-0.2569
5235.0000
294.0000
0.129
10
1.291
0
-0.4424
5235.0000
306.0000
0.129
2
0.258
2
2.7932
5235.0000
319.0000
0.129
10
1.291
10
2.9858
5235.0000
337.0000
0.129
10
1.291
0
-0.4424
5235.0000
337.0000
0.129
10
1.291
0
-0.4424
5235.0000
350.0000
0.129
10
1.291
0
-0.4424
5235.0000
359.0000
0.129
10
1.291
0
-0.4424
5235.0000
366.0000
0.129
10
1.291
0
-0.4424
5235.0000
367.0000
0.129
10
1.291
0
-0.4424
5235.0000
371.0000
0.129
10
1.291
0
-0.4424
5235.0000
375.0000
0.129
7
0.903
2
0.5330
5235.0000
378.0000
0.129
3
0.387
0
-0.4251
5235.0000
381.0000
0.129
10
1.291
0
-0.4424
5235.0000
381.0000
0.129
10
1.291
0
-0.4424
5235.0000
389.0000
0.129
10
1.291
0
-0.4424
5235.0000
389.0000
0.129
10
1.291
0
-0.4424
5235.0000
395.0000
0.129
10
1.291
0
-0.4424
5235.0000
395.0000
0.129
10
1.291
0
-0.4424
5235.0000
398.0000
0.129
10
1.291
1
-0.0996
5235.0000
423.0000
0.129
10
1.291
1
-0.0996
5235.0000
445.0000
0.129
8
1.032
0
-0.4405
5235.0000
456.0000
0.129
10
1.291
0
-0.4424
Scaled Residual(s) for Dose Group Nearest the BMD
Minimum scaled residual for dose group nearest the BMD = -0.2569
Minimum ABS(scaled residual) for dose group nearest the BMD = 0.2569
Average scaled residual for dose group nearest the BMD = -0.2569
Average ABS(scaled residual) for dose group nearest the BMD = 0.2569
Maximum scaled residual for dose group nearest the BMD = -0.2569
Page 188 of 244
-------
Maximum ABS(scaled residual) for dose group nearest the BMD = 0.2569
Number of litters used for scaled residual for dose group nearest the BMD = 1
Observed Chi-square = 101.3408
Bootstrapping Results
Number of Bootstrap Iterations per run: 1000
Bootstrap Chi-square Percentiles
Bootstrap
Run P-value 50th 90th 95th 99th
1 0.3340 78.0736 174.9388 219.4932 366.0397
2 0.3290 77.4467 186.1788 252.2181 403.7505
3 0.3250 76.9188 180.0700 253.0186 377.4700
Combined 0.3293 77.5709 182.1937 238.5844 383.6190
The results for three separate runs are shown. If the estimated p-values are sufficiently
stable (do not vary considerably from run to run), then then number of iterations is
considered adequate. The p-value that should be reported is the one that combines
the results of the three runs. If sufficient stability is not evident (and especially
if the p-values are close to the critical level for determining adequate fit, e.g., 0.05),
then the user should consider increasing the number of iterations per run.
To calculate the BMD and BMDL, the litter specific covariate is fixed
at the mean litter specific covariate of all the data: 370.257732
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 1829.66
BMDL = 313.814
Page 189 of 244
-------
5.8 Results for BMD Modeling for Stillbirths, and PND4 and PND21 Pup
Deaths in Wistar Rats (NMP Producers Group (1999b))
5.8.1 Wistar Rat F1A stillborn/total delivered (NMP Producers Group (1999b))
Wistar Rat F1A Stillborn/Total Delivered (NMP Producers Group (1999b))
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
Stillborn
Covariate
(mg, LD1 Dam BW)
0
0
12
0
294
0
0
15
0
295
0
0
14
1
299
0
0
15
0
300
0
0
14
0
303
0
0
14
0
304
0
0
9
0
308
0
0
11
0
308
0
0
13
0
314
0
0
9
0
314
0
0
16
0
315
0
0
16
0
315
0
0
16
0
321
0
0
10
0
322
0
0
13
1
322
0
0
12
0
326
0
0
7
0
327
0
0
11
0
327
0
0
9
0
328
0
0
11
1
329
0
0
15
1
332
0
0
19
0
336
0
0
15
2
339
0
0
14
0
343
0
0
17
1
344
538.0
37.49
8
0
264
538.0
37.49
17
3
281
538.0
37.49
16
0
287
538.0
37.49
13
3
290
538.0
37.49
17
0
294
538.0
37.49
12
0
296
538.0
37.49
14
0
302
538.0
37.49
13
0
303
538.0
37.49
14
0
304
538.0
37.49
15
0
306
538.0
37.49
15
0
307
538.0
37.49
17
0
307
538.0
37.49
12
1
308
538.0
37.49
13
0
308
538.0
37.49
5
1
308
538.0
37.49
17
0
314
538.0
37.49
16
2
315
Page 190 of 244
-------
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
Stillborn
Covariate
(mg, LD1 Dam BW)
538.0
37.49
10
1
315
538.0
37.49
13
0
316
538.0
37.49
15
0
316
538.0
37.49
5
0
330
538.0
37.49
13
0
334
538.0
37.49
13
0
336
538.0
37.49
13
0
339
538.0
37.49
14
0
339
1965
136.35
18
2
285
1965
136.35
14
0
288
1965
136.35
16
0
295
1965
136.35
15
0
295
1965
136.35
14
2
299
1965
136.35
8
0
301
1965
136.35
13
0
303
1965
136.35
16
0
303
1965
136.35
17
1
311
1965
136.35
5
0
311
1965
136.35
8
0
311
1965
136.35
13
0
313
1965
136.35
19
0
313
1965
136.35
15
0
318
1965
136.35
12
0
318
1965
136.35
8
0
323
1965
136.35
12
0
324
1965
136.35
14
0
326
1965
136.35
14
1
328
1965
136.35
13
0
329
1965
136.35
15
0
333
1965
136.35
12
1
341
1965
136.35
17
1
345
1965
136.35
18
0
354
7793
515.01
16
0
280
7793
515.01
13
1
283
7793
515.01
14
1
284
7793
515.01
10
0
286
7793
515.01
13
1
288
7793
515.01
16
0
288
7793
515.01
11
1
290
7793
515.01
15
1
292
7793
515.01
13
0
294
7793
515.01
12
4
295
7793
515.01
3
0
296
7793
515.01
16
1
301
7793
515.01
10
1
304
7793
515.01
13
0
305
7793
515.01
12
1
306
7793
515.01
17
0
308
Page 191 of 244
-------
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
Stillborn
Covariate
(mg, LD1 Dam BW)
7793
515.01
12
0
309
7793
515.01
11
2
318
7793
515.01
2
0
319
7793
515.01
14
1
320
7793
515.01
14
2
322
7793
515.01
14
2
333
7793
515.01
12
5
338
7793
515.01
13
0
338
Table 5-28 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat F1A
stillborn/total delivered (NMP Producers Group (
1999b)): B]
MR =1% extra risk.
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
Nlogistic (b. seedb =1597172141)
0.457
410.726
6297.7
1276.04
NCTR (b. seed =1597172143)
0.456
407.339
6320.06
5266.72
The NLogistic model that
estimated intra-litter
correlations and did not use a
litter-specific covariate was
selected based on estimating
the lowest BMDL within a
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to
be zero
Nlogistic (b. seed =1597172137)
0.095
410.058
6366.27
1944.49
NCTR (b. seed =1597172139)
0.0637
409.736
6345.17
5287.64
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed =1597172129)
0.4443
407.919
6440.69
855.343
range of BMDLs from
acceptable models (P-value
>0.1) that varied more than 3-
fold.
NCTR (b. seed =1597172131)
0.4547
405.919
6461.71
5384.76
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed =1597172134)
0.032
412.787
6477.21
960.487
NCTR (b. seed =1597172135)
0.0287
410.787
6497.12
5414.26
aLitter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models. Selected
model is bolded.
bb. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
Page 192 of 244
-------
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
dose
14:55 08/11 2020
Figure 5.8-1 Plot of NLogistic (LSC = LD1 dam weight; ICC estimated) model for AUC (hr mg/L)
versus Wistar Rat F1A Stillborn/Total Delivered.
NLogistic Model. (Version: 2.20; Date: 04/27/2015)
Input Data File: C:/Users/jgift/BMDS2704/Data/WFla_stillborn_p_558/Correct
Doses/BMR01/nln_WFla_stillborn_p_558_Nln-BMRl-Restrict-IC.(d)
Tue Aug 11 14:55:29 2020
BMDS Model Run
The probability function is:
Prob. = alpha + thetal*Rij + [1 - alpha - thetal*Rij]/
[l+exp(-beta-theta2*Rij-rho*log(Dose))],
where Rij is the litter specific covariate.
Restrict Power rho >= 1.
Total number of observations = 98
Total number of records with missing values = 0
Total number of parameters in model = 9
Total number of specified parameters = 2
Maximum number of iterations = 500
Relative Function Convergence has been set to: le-008
Page 193 of 244
-------
Parameter Convergence has been set to: le-008
Number of Bootstrap Iterations per run: 1000
Bootstrap Seed: 1597172129
User specifies the following parameters:
thetal = 0
theta2 = 0
Default Initial Parameter Values
alpha = 0.0250345
beta = -88.7152
thetal = 0 Specified
theta2 = 0 Specified
rho = 9.59136
phil = 0
phi2 = 0.0870947
phi3 = 0.011941
phi4 = 0.0521665
Parameter Estimates
Variable
alpha
beta
rho
phil
phi2
phi3
phi4
Estimate
0.0250345
-88.7151
9.59137
0
0.0870947
0.011941
0.0521665
Std. Err.
0.00558747
0.390723
0.0687291
Bounded
0.0426229
NA
NA
Log-likelihood: -197.959 AIC: 407.919
Litter Data
Lit.-Spec. Litter Scaled
Dose Cov. Est. Prob. Size Expected Observed Residual
0.0000
294.0000
0.025
12
0.300
0
-0.5551
0.0000
295.0000
0.025
15
0.376
0
-0.6206
0.0000
299.0000
0.025
14
0.350
1
1.1111
0.0000
300.0000
0.025
15
0.376
0
-0.6206
0.0000
303.0000
0.025
14
0.350
0
-0.5996
0.0000
304.0000
0.025
14
0.350
0
-0.5996
0.0000
308.0000
0.025
9
0.225
0
-0.4807
0.0000
308.0000
0.025
11
0.275
0
-0.5315
0.0000
314.0000
0.025
13
0.325
0
-0.5778
0.0000
314.0000
0.025
9
0.225
0
-0.4807
0.0000
315.0000
0.025
16
0.401
0
-0.6410
Page 194 of 244
-------
0.0000 315.0000 0.025
0.0000 321.0000 0.025
0.0000 322.0000 0.025
0.0000 322.0000 0.025
0.0000 326.0000 0.025
0.0000 327.0000 0.025
0.0000 327.0000 0.025
0.0000 328.0000 0.025
0.0000 329.0000 0.025
0.0000 332.0000 0.025
0.0000 336.0000 0.025
0.0000 339.0000 0.025
0.0000 343.0000 0.025
0.0000 344.0000 0.025
538.0000 264.0000 0.025
538.0000 281.0000 0.025
538.0000 287.0000 0.025
538.0000 290.0000 0.025
538.0000 294.0000 0.025
538.0000 296.0000 0.025
538.0000 302.0000 0.025
538.0000 303.0000 0.025
538.0000 304.0000 0.025
538.0000 306.0000 0.025
538.0000 307.0000 0.025
538.0000 307.0000 0.025
538.0000 308.0000 0.025
538.0000 308.0000 0.025
538.0000 308.0000 0.025
538.0000 314.0000 0.025
538.0000 315.0000 0.025
538.0000 315.0000 0.025
538.0000 316.0000 0.025
538.0000 316.0000 0.025
538.0000 330.0000 0.025
538.0000 334.0000 0.025
538.0000 336.0000 0.025
538.0000 339.0000 0.025
538.0000 339.0000 0.025
1965.0000 285.0000 0.025
1965.0000 288.0000 0.025
1965.0000 295.0000 0.025
1965.0000 295.0000 0.025
1965.0000 299.0000 0.025
1965.0000 301.0000 0.025
1965.0000 303.0000 0.025
0.401
0
-0.6410
0.401
0
-0.6410
0.250
0
-0.5067
0.325
1
1.1975
0.300
0
-0.5551
0.175
0
-0.4240
0.275
0
-0.5315
0.225
0
-0.4807
0.275
1
1.3985
0.376
1
1.0321
0.476
0
-0.6985
0.376
2
2.6848
0.350
0
-0.5996
0.426
1
0.8917
0.200
0
-0.3572
0.426
3
2.5833
0.401
0
-0.4221
0.325
3
3.3201
0.426
0
-0.4271
0.300
0
-0.3967
0.350
0
-0.4106
0.325
0
-0.4040
0.350
0
-0.4106
0.376
0
-0.4166
0.376
0
-0.4166
0.426
0
-0.4271
0.300
1
0.9238
0.325
0
-0.4040
0.125
1
2.1566
0.426
0
-0.4271
0.401
2
1.6853
0.250
1
1.1361
0.325
0
-0.4040
0.376
0
-0.4166
0.125
0
-0.3086
0.325
0
-0.4040
0.325
0
-0.4040
0.325
0
-0.4040
0.350
0
-0.4106
0.451
2
2.1312
0.350
0
-0.5578
0.401
0
-0.5903
0.376
0
-0.5745
0.350
2
2.6254
0.200
0
-0.4354
0.325
0
-0.5403
16
16
10
13
12
7
11
9
11
15
19
15
14
17
8
17
16
13
17
12
14
13
14
15
15
17
12
13
5
17
16
10
13
15
5
13
13
13
14
18
14
16
15
14
8
13
Page 195 of 244
-------
1965.0000
303.0000
0.025
16
0.401
0
-0.5903
1965.0000
311.0000
0.025
17
0.426
1
0.8171
1965.0000
311.0000
0.025
5
0.125
0
-0.3500
1965.0000
311.0000
0.025
8
0.200
0
-0.4354
1965.0000
313.0000
0.025
13
0.325
0
-0.5403
1965.0000
313.0000
0.025
19
0.476
0
-0.6337
1965.0000
318.0000
0.025
15
0.376
0
-0.5745
1965.0000
318.0000
0.025
12
0.300
0
-0.5219
1965.0000
323.0000
0.025
8
0.200
0
-0.4354
1965.0000
324.0000
0.025
12
0.300
0
-0.5219
1965.0000
326.0000
0.025
14
0.350
0
-0.5578
1965.0000
328.0000
0.025
14
0.350
1
1.0338
1965.0000
329.0000
0.025
13
0.325
0
-0.5403
1965.0000
333.0000
0.025
15
0.376
0
-0.5745
1965.0000
341.0000
0.025
12
0.300
1
1.2153
1965.0000
345.0000
0.025
17
0.426
1
0.8171
1965.0000
354.0000
0.025
18
0.451
0
-0.6198
7793.0000
280.0000
0.083
16
1.323
0
-0.8995
7793.0000
283.0000
0.083
13
1.075
1
-0.0591
7793.0000
284.0000
0.083
14
1.158
1
-0.1180
7793.0000
286.0000
0.083
10
0.827
0
-0.7832
7793.0000
288.0000
0.083
13
1.075
1
-0.0591
7793.0000
288.0000
0.083
16
1.323
0
-0.8995
7793.0000
290.0000
0.083
11
0.910
1
0.0803
7793.0000
292.0000
0.083
15
1.240
1
-0.1712
7793.0000
294.0000
0.083
13
1.075
0
-0.8489
7793.0000
295.0000
0.083
12
0.992
4
2.5131
7793.0000
296.0000
0.083
3
0.248
0
-0.4948
7793.0000
301.0000
0.083
16
1.323
1
-0.2196
7793.0000
304.0000
0.083
10
0.827
1
0.1640
7793.0000
305.0000
0.083
13
1.075
0
-0.8489
7793.0000
306.0000
0.083
12
0.992
1
0.0065
7793.0000
308.0000
0.083
17
1.406
0
-0.9139
7793.0000
309.0000
0.083
12
0.992
0
-0.8290
7793.0000
318.0000
0.083
11
0.910
2
0.9678
7793.0000
319.0000
0.083
2
0.165
0
-0.4139
7793.0000
320.0000
0.083
14
1.158
1
-0.1180
7793.0000
322.0000
0.083
14
1.158
2
0.6311
7793.0000
333.0000
0.083
14
1.158
2
0.6311
7793.0000
338.0000
0.083
12
0.992
5
3.3487
7793.0000
338.0000
0.083
13
1.075
0
-0.8489
Scaled Residual(s) for Dose Group Nearest the BMD
Minimum scaled residual for dose group nearest the BMD = -0.8290
Minimum ABS(scaled residual) for dose group nearest the BMD = 0.8290
Average scaled residual for dose group nearest the BMD = -0.8290
Page 196 of 244
-------
Average ABS(scaled residual) for dose group nearest the BMD = 0.8290
Maximum scaled residual for dose group nearest the BMD = -0.8290
Maximum ABS(scaled residual) for dose group nearest the BMD = 0.8290
Number of litters used for scaled residual for dose group nearest the BMD = 1
Observed Chi-square = 96.6123
Bootstrapping Results
Number of Bootstrap Iterations per run: 1000
Bootstrap Chi-square Percentiles
Bootstrap
Run P-value 50th 90th 95th 99th
1 0.4560 93.2688 135.9214 154.8401 208.1552
2 0.4400 93.4333 133.4073 152.3201 180.5382
3 0.4370 92.8919 134.4350 148.2379 177.4640
Combined 0.4443 93.1017 134.4672 152.1400 187.5065
The results for three separate runs are shown. If the estimated p-values are sufficiently
stable (do not vary considerably from run to run), then then number of iterations is
considered adequate. The p-value that should be reported is the one that combines
the results of the three runs. If sufficient stability is not evident (and especially
if the p-values are close to the critical level for determining adequate fit, e.g., 0.05),
then the user should consider increasing the number of iterations per run.
To calculate the BMD and BMDL, the litter specific covariate is fixed
at the mean litter specific covariate of all the data: 311.714286
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 6440.69
BMDL = 855.343
Page 197 of 244
-------
Table 5-29 Summary of BMDS nesting modeling results for Cmax (mg/L) versus Wistar Rat F1A
stillborn/total delivered (NMP Producers Group (
1999b)): I
IMR =1% extra risk.
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
The NLogistic model that estimated
intra-litter correlations and did not
use a litter-specific covariate was
selected based on estimating the
lowest BMDL within a range of
BMDLs from acceptable models
(P-value >0.1) that varied more
than 3-fold.
Nlogistic (b. seedb =1597185893)
0.4783
410.726
418.119
90.2154
NCTR (b. seed =1597185894)
0.4657
407.339
420.037
350.031
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to be zero
Nlogistic (b. seed =1597185890)
0.0833
410.058
422.009
134.725
NCTR (b. seed =1597185891)
0.0713
409.736
421.648
351.373
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed =1597185882)
0.453
407.919
429.396
57.6472
NCTR (b. seed =1597185885)
0.4447
405.919
429.188
357.657
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed =1597185887)
0.036
412.787
431.713
64.6766
NCTR (b. seed =1597185888)
0.0267
410.787
431.47
359.558
aLitter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models. Selected
model is bolded.
bb. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
dose
18:44 08/11 2020
Figure 5.8-2 Plot of NLogistic (LSC = LD1 dam weight; ICC estimated) model for C max (mg/L)
versus Wistar Rat F1A Stillborn/Total Delivered.
Page 198 of 244
-------
NLogistic Model. (Version: 2.20; Date: 04/27/2015)
Input Data File: C:/Users/jgift/BMDS2704/Data/WFla_stillborn_p_558/Correct
Doses/BMR01/Cmax/nln_WFla_stillborn_p_558_Nln-BMRl-Restrict-IC.(d)
Tue Aug 11 18:44:42 2020
BMDS Model Run
The probability function is:
Prob. = alpha + thetal *Rij + [1 - alpha - thetal*Rij]/
[l+exp(-beta-theta2*Rij-rho*log(Dose))],
where Rij is the litter specific covariate.
Restrict Power rho >= 1.
Total number of observations = 98
Total number of records with missing values = 0
Total number of parameters in model = 9
Total number of specified parameters = 2
Maximum number of iterations = 500
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Number of Bootstrap Iterations per run: 1000
Bootstrap Seed: 1597185882
User specifies the following parameters:
thetal = 0
theta2 = 0
Default Initial Parameter Values
alpha = 0.0250345
beta = -65.5512
thetal = 0 Specified
theta2 = 0 Specified
rho = 10.0548
phil = 0
phi2 = 0.0870952
phi3 = 0.0119409
phi4 = 0.0521674
Parameter Estimates
Variable Estimate Std. Err.
alpha 0.0250345 0.00558747
Page 199 of 244
-------
beta
rho
phil
phi2
phi3
phi4
-65.5512
10.0548
0
0.0870952
0.0119409
0.0521674
0.390724
0.0687296
Bounded
0.0426236
NA
NA
Log-likelihood: -197.959 AIC: 407.919
Litter Data
Lit.-Spec. Litter Scaled
Dose Cov. Est. Prob. Size Expected Observed Residual
0.0000
294.0000
0.025
12
0.300
0
-0.5551
0.0000
295.0000
0.025
15
0.376
0
-0.6206
0.0000
299.0000
0.025
14
0.350
1
1.1111
0.0000
300.0000
0.025
15
0.376
0
-0.6206
0.0000
303.0000
0.025
14
0.350
0
-0.5996
0.0000
304.0000
0.025
14
0.350
0
-0.5996
0.0000
308.0000
0.025
9
0.225
0
-0.4807
0.0000
308.0000
0.025
11
0.275
0
-0.5315
0.0000
314.0000
0.025
13
0.325
0
-0.5778
0.0000
314.0000
0.025
9
0.225
0
-0.4807
0.0000
315.0000
0.025
16
0.401
0
-0.6410
0.0000
315.0000
0.025
16
0.401
0
-0.6410
0.0000
321.0000
0.025
16
0.401
0
-0.6410
0.0000
322.0000
0.025
10
0.250
0
-0.5067
0.0000
322.0000
0.025
13
0.325
1
1.1975
0.0000
326.0000
0.025
12
0.300
0
-0.5551
0.0000
327.0000
0.025
7
0.175
0
-0.4240
0.0000
327.0000
0.025
11
0.275
0
-0.5315
0.0000
328.0000
0.025
9
0.225
0
-0.4807
0.0000
329.0000
0.025
11
0.275
1
1.3985
0.0000
332.0000
0.025
15
0.376
1
1.0321
0.0000
336.0000
0.025
19
0.476
0
-0.6985
0.0000
339.0000
0.025
15
0.376
2
2.6848
0.0000
343.0000
0.025
14
0.350
0
-0.5996
0.0000
344.0000
0.025
17
0.426
1
0.8917
37.4900
264.0000
0.025
8
0.200
0
-0.3572
37.4900
281.0000
0.025
17
0.426
3
2.5833
37.4900
287.0000
0.025
16
0.401
0
-0.4221
37.4900
290.0000
0.025
13
0.325
3
3.3201
37.4900
294.0000
0.025
17
0.426
0
-0.4271
37.4900
296.0000
0.025
12
0.300
0
-0.3967
37.4900
302.0000
0.025
14
0.350
0
-0.4106
37.4900
303.0000
0.025
13
0.325
0
-0.4040
Page 200 of 244
-------
37.4900 304.0000 0.025
37.4900 306.0000 0.025
37.4900 307.0000 0.025
37.4900 307.0000 0.025
37.4900 308.0000 0.025
37.4900 308.0000 0.025
37.4900 308.0000 0.025
37.4900 314.0000 0.025
37.4900 315.0000 0.025
37.4900 315.0000 0.025
37.4900 316.0000 0.025
37.4900 316.0000 0.025
37.4900 330.0000 0.025
37.4900 334.0000 0.025
37.4900 336.0000 0.025
37.4900 339.0000 0.025
37.4900 339.0000 0.025
136.3500 285.0000 0.025
136.3500 288.0000 0.025
136.3500 295.0000 0.025
136.3500 295.0000 0.025
136.3500 299.0000 0.025
136.3500 301.0000 0.025
136.3500 303.0000 0.025
136.3500 303.0000 0.025
136.3500 311.0000 0.025
136.3500 311.0000 0.025
136.3500 311.0000 0.025
136.3500 313.0000 0.025
136.3500 313.0000 0.025
136.3500 318.0000 0.025
136.3500 318.0000 0.025
136.3500 323.0000 0.025
136.3500 324.0000 0.025
136.3500 326.0000 0.025
136.3500 328.0000 0.025
136.3500 329.0000 0.025
136.3500 333.0000 0.025
136.3500 341.0000 0.025
136.3500 345.0000 0.025
136.3500 354.0000 0.025
515.0100 280.0000 0.083
515.0100 283.0000 0.083
515.0100 284.0000 0.083
515.0100 286.0000 0.083
515.0100 288.0000 0.083
14
0.350
0
-0.4106
15
0.376
0
-0.4166
15
0.376
0
-0.4166
17
0.426
0
-0.4271
12
0.300
1
0.9238
13
0.325
0
-0.4040
5
0.125
1
2.1566
17
0.426
0
-0.4271
16
0.401
2
1.6853
10
0.250
1
1.1361
13
0.325
0
-0.4040
15
0.376
0
-0.4166
5
0.125
0
-0.3086
13
0.325
0
-0.4040
13
0.325
0
-0.4040
13
0.325
0
-0.4040
14
0.350
0
-0.4106
18
0.451
2
2.1312
14
0.350
0
-0.5578
16
0.401
0
-0.5903
15
0.376
0
-0.5745
14
0.350
2
2.6254
8
0.200
0
-0.4354
13
0.325
0
-0.5403
16
0.401
0
-0.5903
17
0.426
1
0.8171
5
0.125
0
-0.3500
8
0.200
0
-0.4354
13
0.325
0
-0.5403
19
0.476
0
-0.6337
15
0.376
0
-0.5745
12
0.300
0
-0.5219
8
0.200
0
-0.4354
12
0.300
0
-0.5219
14
0.350
0
-0.5578
14
0.350
1
1.0338
13
0.325
0
-0.5403
15
0.376
0
-0.5745
12
0.300
1
1.2153
17
0.426
1
0.8171
18
0.451
0
-0.6198
16
1.323
0
-0.8995
13
1.075
1
-0.0591
14
1.158
1
-0.1180
10
0.827
0
-0.7832
13
1.075
1
-0.0591
Page 201 of 244
-------
515.0100
288.0000
0.083
16
1.323
0
-0.8995
515.0100
290.0000
0.083
11
0.910
1
0.0803
515.0100
292.0000
0.083
15
1.240
1
-0.1712
515.0100
294.0000
0.083
13
1.075
0
-0.8489
515.0100
295.0000
0.083
12
0.992
4
2.5131
515.0100
296.0000
0.083
3
0.248
0
-0.4948
515.0100
301.0000
0.083
16
1.323
1
-0.2196
515.0100
304.0000
0.083
10
0.827
1
0.1640
515.0100
305.0000
0.083
13
1.075
0
-0.8489
515.0100
306.0000
0.083
12
0.992
1
0.0065
515.0100
308.0000
0.083
17
1.406
0
-0.9139
515.0100
309.0000
0.083
12
0.992
0
-0.8290
515.0100
318.0000
0.083
11
0.910
2
0.9678
515.0100
319.0000
0.083
2
0.165
0
-0.4139
515.0100
320.0000
0.083
14
1.158
1
-0.1180
515.0100
322.0000
0.083
14
1.158
2
0.6311
515.0100
333.0000
0.083
14
1.158
2
0.6311
515.0100
338.0000
0.083
12
0.992
5
3.3486
515.0100
338.0000
0.083
13
1.075
0
-0.8489
Scaled Residual(s) for Dose Group Nearest the BMD
Minimum scaled residual for dose group nearest the BMD = -0.8290
Minimum ABS(scaled residual) for dose group nearest the BMD = 0.8290
Average scaled residual for dose group nearest the BMD = -0.8290
Average ABS(scaled residual) for dose group nearest the BMD = 0.8290
Maximum scaled residual for dose group nearest the BMD = -0.8290
Maximum ABS(scaled residual) for dose group nearest the BMD = 0.8290
Number of litters used for scaled residual for dose group nearest the BMD = 1
Observed Chi-square = 96.6120
Bootstrapping Results
Number of Bootstrap Iterations per run: 1000
Bootstrap Chi-square Percentiles
Bootstrap
Run P-value 50th 90th 95th 99th
1 0.4500 92.5235 133.3830 146.8070 182.9345
2 0.4400 92.4527 133.6012 148.2521 188.2813
3 0.4690 94.4295 138.5229 155.0223 183.1285
Combined 0.4530 93.0302 135.0792 149.8451 186.4510
The results for three separate runs are shown. If the estimated p-values are sufficiently
stable (do not vary considerably from run to run), then then number of iterations is
Page 202 of 244
-------
considered adequate. The p-value that should be reported is the one that combines
the results of the three runs. If sufficient stability is not evident (and especially
if the p-values are close to the critical level for determining adequate fit, e.g., 0.05),
then the user should consider increasing the number of iterations per run.
To calculate the BMD and BMDL, the litter specific covariate is fixed
at the mean litter specific covariate of all the data: 311.714286
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 429.396
BMDL = 57.6472
Page 203 of 244
-------
5.8.2 Wistar Rat F1A Pup death at PND4/total delivered (NMP Producers Group (1999b))
Wistar Rat F1A Pup Death/Total Delivered (>
MP Producers Group
1999b))
AUC
(hr mg/L)
Total Delivered
Stillborn
Covariate
(mg, LD1 Dam BW)
0
12
0
294
0
15
0
295
0
14
1
299
0
15
0
300
0
14
0
303
0
14
1
304
0
9
0
308
0
11
1
308
0
13
0
314
0
9
0
314
0
16
1
315
0
16
2
315
0
16
4
321
0
10
0
322
0
13
1
322
0
12
1
326
0
7
1
327
0
11
0
327
0
9
0
328
0
11
1
329
0
15
1
332
0
19
0
336
0
15
2
339
0
14
1
343
0
17
1
344
538
8
0
264
538
17
6
281
538
16
1
287
538
13
4
290
538
17
2
294
538
12
0
296
538
14
1
302
538
13
4
303
538
14
0
304
538
15
2
306
538
15
0
307
538
17
1
307
538
12
2
308
538
13
1
308
538
5
1
308
538
17
0
314
538
16
3
315
538
10
1
315
538
13
0
316
538
15
0
316
Page 204 of 244
-------
r mg/
538
538
538
538
538
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
1965
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
7793
Total Delivered
13
13
13
14
18
14
16
15
14
13
16
17
13
19
15
12
12
14
14
13
15
12
17
18
16
13
14
10
13
16
11
15
13
12
16
10
13
12
17
12
11
Stillborn
0
16
13
14
10
13
16
11
15
12
16
13
12
17
12
Covariate
(mg, LD1 Dam BW)
330
334
336
339
339
285
288
295
295
299
301
303
303
311
311
311
313
313
318
318
323
324
326
328
329
333
341
345
354
280
283
284
286
288
288
290
292
294
295
296
301
304
305
306
308
309
318
319
Page 205 of 244
-------
AUC
(hr mg/L)
Total Delivered
Stillborn
Covariate
(mg, LD1 Dam BW)
7793
14
14
320
7793
14
14
322
7793
14
3
333
7793
12
12
338
7793
13
2
338
Table 5-30 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat F1A
Pup death at PND4/total delivered (NMP Producers Group (1999b)); BMR = 1% extra risk.
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
The NCTR model that
estimated intra-litter
correlations and used LD1
dam weight as a litter-specific
covariate was selected based
on lowest AIC. BMDLs from
acceptable models (P-value
>0.1) did not vary more than
3-fold.
Nlogistic (b. seedb = 1597171727)
0.3343
641.926
5193.6
1707.85
NCTR (b. seed = 1597171729)
0.3203
640.119
5262.12
4385.1
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to
be zero
Nlogistic (b. seed = 1597171723)
0
751.242
5143.03
1888.53
NCTR (b. seed = 1597171725)
0
749.195
5179.67
4316.39
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed = 1597171713)
0.2783
642.357
5019.1
1731.28
NCTR (b. seed = 1597171715)
0.274
640.357
5250.64
4375.54
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1597171719)
0
788.458
4927.89
1820.82
NCTR (b. seed = 1597171721)
0
786.458
5168.83
4307.36
aLitter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models. Selected
model is bolded.
bb. seed: bootstrap seed. The bootstrap seed must be entered into BMDS 2.7.0.4 nested model to replicate results.
Page 206 of 244
-------
NCTR Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
1
0.8
0.6
0.4
0.2
0
0 1000 2000 3000 4000 5000 6000 7000 8000
dose
14:48 08/11 2020
Figure 5.8-3 Plot of NCTR model (LSC = LD1 dam weight; ICC estimated) for AUC (hr mg/L)
versus Wistar Rat F1A Pup Death at PND4/Total Delivered.
NCTR Model. (Version: 2.13; Date: 04/27/2015)
Input Data File: C:/Users/jgift/BMDS2704/Data/WFla_PND4_p_558/Correct
Doses/BMRO l/nct_WF 1 a_PND4_p_5 5 8_Nct-BMRl -Restrict-IC-LSC. (d)
Gnuplot Plotting File: C:/Users/jgift/BMDS2704/Data/WFla_PND4_p_558/Correct
Doses/BMRO l/nct_WF 1 a_PND4_p_5 5 8_Nct-BMRl -Restrict-IC-LSC .pit
Tue Aug 11 14:48:49 2020
BMDS Model Run
The probability function is:
Prob. = 1 - exp[-(alpha + thl*Rij) - (beta + th2*Rij)*DoseArho],
where Rij is the centralized litter specific covariate.
Restrict Power rho >= 1.
Total number of observations = 98
Total number of records with missing values = 0
Total number of parameters in model = 9
Total number of specified parameters = 0
Maximum number of iterations = 500
Relative Function Convergence has been set to: le-008
Page 207 of 244
-------
Parameter Convergence has been set to: le-008
Number of Bootstrap Iterations per run: 1000
Bootstrap Seed: 1597171729
Default Initial Parameter Values
alpha = 0.0708532
beta = 5.40955e-051
thetal = -0.00167557
theta2 = le-008
rho = 12.9748
phil = 0.00383523
phi2 = 0.0578419
phi3 = 0
phi4 = 0.732024
Parameter Estimates
Variable
Estimate
Std. Err.
alpha
0.0726995
0.00999493
beta
4.73061e-051
Bounded
thetal
-0.000459225
0.000552295
theta2
-9.93029e-053
NA
rho
12.9871
1.51715e-024
phil
0.00471897
0.0227674
phi2
0.0554074
0.0350542
phi3
0
Bounded
phi4
0.688877
0.600172
Log-likelihood: -313.059 AIC: 640.119
Litter Data
Lit.-Spec. Litter Scaled
Dose Cov. Est. Prob. Size Expected Observed Residual
0.0000
294.0000
0.078
12
0.932
0
-0.9800
0.0000
295.0000
0.077
15
1.158
0
-1.0852
0.0000
299.0000
0.076
14
1.057
1
-0.0564
0.0000
300.0000
0.075
15
1.127
0
-1.0689
0.0000
303.0000
0.074
14
1.034
0
-1.0255
0.0000
304.0000
0.073
14
1.028
1
-0.0276
0.0000
308.0000
0.072
9
0.645
0
-0.8185
0.0000
308.0000
0.072
11
0.789
1
0.2413
0.0000
314.0000
0.069
13
0.899
0
-0.9560
0.0000
314.0000
0.069
9
0.622
0
-0.8026
0.0000
315.0000
0.069
16
1.099
1
-0.0950
0.0000
315.0000
0.069
16
1.099
2
0.8601
Page 208 of 244
-------
0.0000 321.0000 0.066
0.0000 322.0000 0.066
0.0000 322.0000 0.066
0.0000 326.0000 0.064
0.0000 327.0000 0.064
0.0000 327.0000 0.064
0.0000 328.0000 0.063
0.0000 329.0000 0.063
0.0000 332.0000 0.061
0.0000 336.0000 0.060
0.0000 339.0000 0.058
0.0000 343.0000 0.057
0.0000 344.0000 0.056
538.0000 264.0000 0.090
538.0000 281.0000 0.083
538.0000 287.0000 0.081
538.0000 290.0000 0.079
538.0000 294.0000 0.078
538.0000 296.0000 0.077
538.0000 302.0000 0.074
538.0000 303.0000 0.074
538.0000 304.0000 0.073
538.0000 306.0000 0.073
538.0000 307.0000 0.072
538.0000 307.0000 0.072
538.0000 308.0000 0.072
538.0000 308.0000 0.072
538.0000 308.0000 0.072
538.0000 314.0000 0.069
538.0000 315.0000 0.069
538.0000 315.0000 0.069
538.0000 316.0000 0.068
538.0000 316.0000 0.068
538.0000 330.0000 0.062
538.0000 334.0000 0.061
538.0000 336.0000 0.060
538.0000 339.0000 0.058
538.0000 339.0000 0.058
1965.0000 285.0000 0.081
1965.0000 288.0000 0.080
1965.0000 295.0000 0.077
1965.0000 295.0000 0.077
1965.0000 299.0000 0.076
1965.0000 301.0000 0.075
1965.0000 303.0000 0.074
1965.0000 303.0000 0.074
1.058
4
2.8595
0.657
0
-0.8214
0.854
1
0.1586
0.768
1
0.2668
0.445
1
0.8479
0.699
0
-0.8444
0.568
0
-0.7645
0.690
1
0.3770
0.921
1
0.0820
1.134
0
-1.0544
0.876
2
1.1988
0.793
1
0.2319
0.956
1
0.0448
0.722
0
-0.7563
1.413
6
2.9334
1.290
1
-0.1967
1.032
4
2.3608
1.320
2
0.4486
0.922
0
-0.7876
1.040
1
-0.0308
0.960
4
2.4990
1.028
0
-0.8030
1.088
2
0.6809
1.082
0
-0.8104
1.226
1
-0.1544
0.860
2
1.0050
0.932
1
0.0565
0.359
1
1.0060
1.175
0
-0.8181
1.099
3
1.3880
0.687
1
0.3195
0.888
0
-0.7565
1.024
0
-0.7868
0.311
0
-0.5214
0.787
3
1.9942
0.776
0
-0.7040
0.759
2
1.1375
0.818
2
1.0276
1.466 2 0.4599
1.123 0 -1.1048
1.236 0 -1.1572
1.158 2 0.8139
1.057 3 1.9647
0.597 0 -0.8036
0.960 0 -1.0180
1.181 1 -0.1734
16
10
13
12
7
11
9
11
15
19
15
14
17
8
17
16
13
17
12
14
13
14
15
15
17
12
13
5
17
16
10
13
15
5
13
13
13
14
18
14
16
15
14
8
13
16
Page 209 of 244
-------
1965.0000
311.0000
0.070
17
1.197
2
0.7610
1965.0000
311.0000
0.070
5
0.352
1
1.1324
1965.0000
311.0000
0.070
8
0.563
0
-0.7785
1965.0000
313.0000
0.070
13
0.904
0
-0.9859
1965.0000
313.0000
0.070
19
1.322
1
-0.2902
1965.0000
318.0000
0.067
15
1.011
1
-0.0118
1965.0000
318.0000
0.067
12
0.809
1
0.2197
1965.0000
323.0000
0.065
8
0.522
0
-0.7475
1965.0000
324.0000
0.065
12
0.778
0
-0.9123
1965.0000
326.0000
0.064
14
0.896
1
0.1136
1965.0000
328.0000
0.063
14
0.884
1
0.1275
1965.0000
329.0000
0.063
13
0.815
0
-0.9326
1965.0000
333.0000
0.061
15
0.915
0
-0.9870
1965.0000
341.0000
0.058
12
0.690
1
0.3839
1965.0000
345.0000
0.056
17
0.949
1
0.0544
1965.0000
354.0000
0.052
18
0.934
0
-0.9925
7793.0000
280.0000
0.941
16
15.059
16
0.2971
7793.0000
283.0000
0.935
13
12.150
13
0.3132
7793.0000
284.0000
0.932
14
13.052
14
0.3195
7793.0000
286.0000
0.927
10
9.274
10
0.3298
7793.0000
288.0000
0.922
13
11.988
13
0.3442
7793.0000
288.0000
0.922
16
14.754
16
0.3453
7793.0000
290.0000
0.916
11
10.081
11
0.3565
7793.0000
292.0000
0.910
15
13.656
15
0.3724
7793.0000
294.0000
0.904
13
11.750
3
-2.7048
7793.0000
295.0000
0.900
12
10.805
12
0.3933
7793.0000
296.0000
0.897
3
2.691
0
-3.3130
7793.0000
301.0000
0.877
16
14.034
16
0.4447
7793.0000
304.0000
0.864
10
8.635
5
-1.2479
7793.0000
305.0000
0.859
13
11.162
13
0.4806
7793.0000
306.0000
0.854
12
10.243
12
0.4898
7793.0000
308.0000
0.843
17
14.330
17
0.5132
7793.0000
309.0000
0.837
12
10.048
12
0.5213
7793.0000
318.0000
0.777
11
8.547
5
-0.9149
7793.0000
319.0000
0.769
2
1.538
0
-1.9859
7793.0000
320.0000
0.761
14
10.652
14
0.6649
7793.0000
322.0000
0.743
14
10.408
14
0.6966
7793.0000
333.0000
0.623
14
8.718
3
-0.9993
7793.0000
338.0000
0.550
12
6.605
12
1.0689
7793.0000
338.0000
0.550
13
7.156
2
-0.9443
Scaled Residual(s) for Dose Group Nearest the BMD
Minimum scaled residual for dose group nearest the BMD = 0.5213
Minimum ABS(scaled residual) for dose group nearest the BMD = 0.5213
Average scaled residual for dose group nearest the BMD = 0.5213
Average ABS(scaled residual) for dose group nearest the BMD = 0.5213
Page 210 of 244
-------
Maximum scaled residual for dose group nearest the BMD = 0.5213
Maximum ABS(scaled residual) for dose group nearest the BMD = 0.5213
Number of litters used for scaled residual for dose group nearest the BMD = 1
Observed Chi-square = 105.5696
Bootstrapping Results
Number of Bootstrap Iterations per run: 1000
Bootstrap Chi-square Percentiles
Bootstrap
Run P-value 50th 90th 95th 99th
1 0.3160 94.4716 124.7141 132.5897 155.1097
2 0.3360 96.3317 125.7066 136.8936 155.4666
3 0.3090 95.3719 122.8344 132.9378 153.6561
Combined 0.3203 95.6022 124.4557 134.0383 155.2218
The results for three separate runs are shown. If the estimated p-values are sufficiently
stable (do not vary considerably from run to run), then then number of iterations is
considered adequate. The p-value that should be reported is the one that combines
the results of the three runs. If sufficient stability is not evident (and especially
if the p-values are close to the critical level for determining adequate fit, e.g., 0.05),
then the user should consider increasing the number of iterations per run.
To calculate the BMD and BMDL, the litter
specific covariate is fixed at the overall mean
of the litter specific covariates: 311.714286
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 5262.12
BMDL = 4385.1
Page 211 of 244
-------
5.8.3 Wistar Rat FIB stillborn/total delivered (NMP Producers Group (1999b))
Wistar Rat FIB Stillborn/Total Delivered (N1V
P Producers Group (1999b))
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
PND4 Pup Death
Covariate
(mg, LD1 Dam BW)
0
0
18
0
311
0
0
14
0
319
0
0
12
0
321
0
0
12
0
322
0
0
6
0
324
0
0
13
0
327
0
0
13
0
332
0
0
14
0
340
0
0
16
0
343
0
0
17
0
347
0
0
17
0
347
0
0
10
0
347
0
0
14
0
347
0
0
18
0
350
0
0
12
0
351
0
0
11
0
351
0
0
12
0
352
0
0
18
0
354
0
0
15
0
355
0
0
13
0
356
0
0
15
0
359
0
0
14
0
364
0
0
16
0
370
0
0
16
0
382
549.8
38.25
17
0
289
549.8
38.25
12
0
309
549.8
38.25
15
0
314
549.8
38.25
8
1
320
549.8
38.25
13
0
323
549.8
38.25
13
0
324
549.8
38.25
16
0
327
549.8
38.25
8
2
327
549.8
38.25
14
1
328
549.8
38.25
18
0
330
549.8
38.25
15
0
331
549.8
38.25
14
0
332
549.8
38.25
13
0
332
549.8
38.25
12
0
335
549.8
38.25
15
0
339
549.8
38.25
6
1
342
549.8
38.25
14
0
343
549.8
38.25
12
0
343
549.8
38.25
21
5
343
549.8
38.25
13
0
344
549.8
38.25
11
0
344
Page 212 of 244
-------
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
PND4 Pup Death
Covariate
(mg, LD1 Dam BW)
549.8
38.25
7
0
362
549.8
38.25
14
0
363
549.8
38.25
18
0
365
549.8
38.25
20
0
365
2006
135.21
13
0
317
2006
135.21
14
0
321
2006
135.21
15
0
323
2006
135.21
19
0
324
2006
135.21
19
0
324
2006
135.21
17
1
324
2006
135.21
10
0
325
2006
135.21
15
0
332
2006
135.21
18
0
334
2006
135.21
17
0
335
2006
135.21
9
0
341
2006
135.21
12
0
342
2006
135.21
14
0
344
2006
135.21
3
0
347
2006
135.21
4
0
347
2006
135.21
14
0
348
2006
135.21
12
0
349
2006
135.21
15
4
350
2006
135.21
13
0
352
2006
135.21
13
0
352
2006
135.21
3
0
354
2006
135.21
14
1
363
2006
135.21
13
0
382
2006
135.21
14
0
383
2006
135.21
17
0
385
6589
357.69
14
0
307
6589
357.69
14
0
315
6589
357.69
13
0
318
6589
357.69
15
0
321
6589
357.69
14
0
325
6589
357.69
8
0
325
6589
357.69
10
0
327
6589
357.69
12
1
329
6589
357.69
11
0
329
6589
357.69
10
0
340
6589
357.69
8
0
342
6589
357.69
10
0
345
6589
357.69
8
0
347
6589
357.69
9
0
350
6589
357.69
14
0
350
6589
357.69
15
0
352
6589
357.69
11
0
353
6589
357.69
9
0
353
6589
357.69
3
1
355
Page 213 of 244
-------
AUC
(hr mg/L)
Cmax
(mg/L)
Total Delivered
PND4 Pup Death
Covariate
(mg, LD1 Dam BW)
6589
357.69
8
0
359
6589
357.69
10
0
366
6589
357.69
13
0
366
6589
357.69
7
0
373
6589
357.69
14
1
379
6589
357.69
15
2
390
Table 5-31 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat FIB
stillborn/total delivered (NMP Producers Group (
999b)); BMR =1% extra risk
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
In all cases, models either
failed to compute BMD
values or reported p-values
that are below 0.1. Thus, no
model is chosen.
Nlogistic (b. seedb = 1595011547)
0.5787
196.62
CF
CF
NCTR (b. seed = 1595011553)
CF
195.195
CF
CF
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to be
zero
Nlogistic (b. seed = 1595011544)
0
217.921
67525.6
47680
NCTR (b. seed = 1595011546)
0
219.656
428161
1.10038
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed = 1595011532)
0.583
192.62
CF
CF
NCTR (b. seed = 1595011538)
CF
192.538
CF
CF
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1595011540)
0.0003
217.679
584759
45650.7
NCTR (b. seed = 1595011542)
0.0007
217.681
559444
1.10161
aLitter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models.
b. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
CF = Benchmark dose computation failed. Lower limit includes zero.
0.08
0.06
0.04
0.02
0
0 1000 2000 3000 4000 5000 6000
dose
Figure 5.8-4 Plot of NCTR model (LSC = LD1 dam weight; ICC estimated) for AUC (hr mg/L)
versus Wistar Rat FIB stillborn/total delivered.
Page 214 of 244
-------
Table 5-32 Summary of BMDS nested modeling results for Cmax (mg/L) versus Wistar Rat FIB
stillborn/total delivered (NMP Producers Group (
1999b)): B]
MR =1% extra risk.
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
In all cases, models either
failed to compute BMD values
or reported p-values that are
below 0.1. Thus, no model is
chosen.
Nlogistic (b. seedb = 1597186626)
0.571
196.635
CF
CF
NCTR (b. seed = 1597186632)
CF
195.195
CF
CF
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to
be zero
Nlogistic (b. seed = 1597186623)
0
217.921
492.387
244.904
NCTR (b. seed = 1597186624)
0.0003
219.512
307.326
0.272837
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed = 1597186610)
0.5783
192.635
CF
CF
NCTR (b. seed = 1597186619)
CF
192.538
CF
CF
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed = 1597186620)
0.0007
217.574
403.516
111.323
NCTR (b. seed = 1597186621)
0
217.577
405.879
0.316558
a Litter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models.
bb. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
CF = Benchmark dose computation failed. Lower limit includes zero.
0.08
0.06
0.04
0.02
0
0 50 100 150 200 250 300 350
dose
Figure 5.8-5 Plot of NCTR model (LSC = LD1 dam weight; ICC estimated) for Cmax (mg/L) versus
Wistar Rat FIB stillborn/total delivered.
Page 215 of 244
-------
5.8.4 Wistar Rat F2B Pup death at PND4/total delivered (NMP Producers Group (1999b))
Wistar Rat F2B Pup Death at PND4/Total Delivered (NMP Producers Group (1999b))
AUC
(hr mg/L)
Total Delivered
PND4 Pup Death
Covariate
(mg, LD1 Dam BW)
0
11
0
292
0
15
2
293
0
17
1
303
0
5
0
304
0
15
1
312
0
16
1
312
0
11
0
316
0
18
1
318
0
17
2
323
0
13
0
326
0
14
0
333
0
13
0
335
0
20
2
341
0
17
1
342
0
13
1
343
0
13
2
344
0
15
1
351
0
15
0
353
0
10
0
361
0
14
1
366
0
11
1
369
0
15
0
371
0
18
2
374
0
6
2
375
0
16
3
379
576.7
3
1
277
576.7
15
1
280
576.7
15
0
295
576.7
8
0
300
576.7
11
0
302
576.7
14
0
305
576.7
15
0
308
576.7
14
0
310
576.7
17
2
312
576.7
12
0
315
576.7
12
1
315
576.7
13
1
322
576.7
13
0
324
576.7
21
4
326
576.7
17
0
330
576.7
15
3
335
576.7
7
0
336
576.7
11
2
337
576.7
12
2
339
576.7
18
0
348
Page 216 of 244
-------
r mg/
5767
576.7
576.7
576.7
576.7
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
Total Delivered
16
18
12
15
13
13
14
12
14
14
15
18
16
19
11
16
14
14
13
12
16
12
17
11
18
15
13
13
13
17
16
13
12
12
10
14
18
12
16
14
13
11
PND4 Pup Death
0
10
Covariate
(mg, LD1 Dam BW)
351
352
357
370
380
282
298
298
304
308
311
315
316
316
317
318
320
322
323
323
324
325
327
331
335
336
345
347
363
392
268
294
300
301
302
309
309
314
319
320
328
335
337
340
342
345
347
349
Page 217 of 244
-------
AUC
(hr mg/L)
Total Delivered
PND4 Pup Death
Covariate
(mg, LD1 Dam BW)
5243
17
3
349
5243
15
5
350
5243
12
1
359
5243
10
0
361
5243
16
1
366
5243
19
2
385
Table 5-33 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat F2B
Pup death at PND4/total delivered (NMP Producers Group (1999b)); BMR = 1% extra risk.
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
While some models met the
p-value fit criteria (p-value >
0.1), no model was deemed to
appropriate after visual
inspection of model plots,
which indicates considerable
model uncertainty and a dose-
response pattern analogous to
having a positive response at
only the highest dose.
Nlogistic (b. seedb =1597174507)
0.701
656.055
4632.85
695.198
NCTR (b. seed =1597174509)
0.7017
653.707
4632.34
3860.28
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to
be zero
Nlogistic (b. seed =1597174503)
0
691.894
4619.65
2103.54
NCTR (b. seed =1597174505)
0
689.888
4625.05
3854.21
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed =1597174495)
0.7573
654.87
4624.92
726.435
NCTR (b. seed =1597174497)
0.7313
652.87
4631.62
3859.68
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed =1597174499)
0
692.473
4613.99
2138.4
NCTR (b. seed =1597174501)
0
690.473
4618.69
3848.91
aLitter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models. No model
was chosen due to considerable model uncertainty indicated by visual inspection of model plots.
bb. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
Page 218 of 244
-------
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
0.2
0.15
0.1
0.05
0
0 1000 2000 3000 4000 5000
dose
Figure 5.8-6 Plot of NCTR model (LSC = LD1 dam body weight; ICC estimated) for AUC (hr
mg/L) versus Wistar Rat F2B Pup Death at PND4/Total Delivered.
Page 219 of 244
-------
5.8.5 Wistar Rat F2B Pup death at PND21/PND4 post-cull (NMP Producers Group
(1999b))
Wistar Rat F2B Pup Death at PND21/PND4 Post-cull (NMP Producers Group (1999b))
AUC
(hr mg/L)
PND4 Live Post-cull
PND21 Pup Death
Covariate
(mg, LD1 Dam BW)
0
10
0
292
0
10
0
293
0
10
0
303
0
5
0
304
0
10
0
312
0
10
1
312
0
10
0
316
0
10
0
318
0
10
0
323
0
10
0
326
0
10
0
333
0
10
0
335
0
10
0
341
0
10
0
342
0
10
0
343
0
10
0
344
0
10
0
351
0
10
0
353
0
10
0
361
0
10
0
366
0
10
0
369
0
10
0
371
0
10
0
374
0
4
0
375
0
10
0
379
576.7
2
0
277
576.7
10
0
280
576.7
10
0
295
576.7
8
0
300
576.7
10
0
302
576.7
10
0
305
576.7
10
0
308
576.7
10
0
310
576.7
10
0
312
576.7
10
0
315
576.7
10
0
315
576.7
10
0
322
576.7
10
0
324
576.7
10
0
326
576.7
10
0
330
576.7
10
0
335
576.7
7
0
336
576.7
9
0
337
576.7
10
0
339
Page 220 of 244
-------
r mg/
5767
576.7
576.7
576.7
576.7
576.7
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
2024
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
5243
PND4 Live Post-cull
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
PND21 Pup Death
0
Covariate
(mg, LD1 Dam BW)
348
351
352
357
370
380
282
298
298
304
308
311
315
316
316
317
318
320
322
323
323
324
325
327
331
335
336
345
347
363
392
268
294
300
301
302
309
309
314
319
320
328
335
337
340
342
345
347
Page 221 of 244
-------
AUC
(hr mg/L)
PND4 Live Post-cull
PND21 Pup Death
Covariate
(mg, LD1 Dam BW)
5243
8
0
349
5243
10
0
349
5243
10
1
350
5243
10
0
359
5243
10
0
361
5243
10
0
366
5243
10
1
385
Table 5-34 Summary of BMDS nested modeling results for AUC (hr mg/L) versus Wistar Rat F2B
Pup death at PND21 /PND4 post-cull (NMP Producers Group (1999b)); BMR= 1% extra risk.
Modela
Goodness of fit
BMDoi
(hr mg/L)
BMDLoi
(hr mg/L)
Basis for Model Selection
P-value
AIC
Litter-specific covariate = LD1 dam weight; intra-litter correlations estimated
The NLogistic model that
estimated intra-litter
correlations but did not make
use of a litter-specific
covariate was selected based
on lowest AIC and BMDL.
BMDLs from acceptable
models (P-value >0.1) did not
vary more than 3-fold.
Nlogistic (b. seedb =1597184767)
0.4807
151.165
2068.11
649.506
NCTR (b. seed =1597184769)
0.4857
150.805
1633.38
816.692
Litter-specific covariate = LD1 dam weight; intra-litter correlations assumed to
be zero
Nlogistic (b. seed =1597184764)
0.0877
166.9
2193.49
843.599
NCTR (b. seed =1597184765)
0.0753
166.819
2140.17
1070.08
Litter-specific covariate not used; intra-litter correlations estimated
Nlogistic (b. seed =1597184753)
0.4777
147.545
2266.39
723.867
NCTR (b. seed =1597184758)
0.4793
147.546
2269.33
1134.67
Litter-specific covariate not used; intra-litter correlations assumed to be zero
Nlogistic (b. seed =1597184761)
0.08
162.964
2221.61
910.752
NCTR (b. seed =1597184762)
0.0857
162.965
2223.59
1111.79
aLitter-specific data were fit using standard (restricted) BMDS NLogistic and NCTR nested dichotomous models. Selected
model is bolded.
bb. seed: bootstrap seed. The bootstrap seed shown must be entered into BMDS 2.7.0.4 nested model to replicate results.
Page 222 of 244
-------
Nested Logistic Model, with BMR of 1% Extra Risk for the BMD and 0.95 Lower Confidence Limit for the BMDL
Figure 5.8-7 Plot of NLogistic model (no LSC; ICC estimated) for AUC (hr mg/L) versus Wistar
Rat F2B Pup Death at PND21/Live PND4 Post-cull.
NLogistic Model. (Version: 2.20; Date: 04/27/2015)
Input Data File: C:/Users/jgift/BMDS2704/Data/WF2b_PND21_p_942/Correct
Doses/BMR01/nln_WF2b_PND21_p_942_Nln-BMRl-Restrict-IC.(d)
Tue Aug 11 18:25:53 2020
BMDS Model Run
The probability function is:
Prob. = alpha + thetal*Rij + [1 - alpha - thetal*Rij]/
[ 1+exp(-beta-theta2 *Rij -rho* log(Dose))],
where Rij is the litter specific covariate.
Restrict Power rho >= 1.
Total number of observations = 99
Total number of records with missing values = 0
Total number of parameters in model = 9
Total number of specified parameters = 2
Maximum number of iterations = 500
Relative Function Convergence has been set to: le-008
Parameter Convergence has been set to: le-008
Number of Bootstrap Iterations per run: 1000
Bootstrap Seed: 1597184753
Page 223 of 244
-------
User specifies the following parameters:
thetal = 0
theta2 = 0
Default Initial Parameter Values
alpha = 0.00396861
beta = -20.6474
thetal = 0 Specified
theta2 = 0 Specified
rho = 2.0777
phil = 0
phi2 = 0
phi3 = 0.0926644
phi4 = 0.227904
Parameter Estimates
Variable
alpha
beta
rho
phil
phi2
phi3
phi4
Estimate
0.00396861
-20.6474
2.0777
0
0
0.0926644
0.227904
Std. Err.
0.00249283
0.42836
NA
Bounded
Bounded
NA
NA
Log-likelihood: -68.7726 AIC: 147.545
Litter Data
Lit.-Spec. Litter Scaled
Dose Cov. Est. Prob. Size Expected Observed Residual
0.0000
292.0000
0.004
10
0.040
0
-0.1996
0.0000
293.0000
0.004
10
0.040
0
-0.1996
0.0000
303.0000
0.004
10
0.040
0
-0.1996
0.0000
304.0000
0.004
5
0.020
0
-0.1411
0.0000
312.0000
0.004
10
0.040
0
-0.1996
0.0000
312.0000
0.004
10
0.040
1
4.8301
0.0000
316.0000
0.004
10
0.040
0
-0.1996
0.0000
318.0000
0.004
10
0.040
0
-0.1996
0.0000
323.0000
0.004
10
0.040
0
-0.1996
0.0000
326.0000
0.004
10
0.040
0
-0.1996
0.0000
333.0000
0.004
10
0.040
0
-0.1996
0.0000
335.0000
0.004
10
0.040
0
-0.1996
0.0000
341.0000
0.004
10
0.040
0
-0.1996
0.0000
342.0000
0.004
10
0.040
0
-0.1996
0.0000
343.0000
0.004
10
0.040
0
-0.1996
Page 224 of 244
-------
0.0000 344.0000 0.004
0.0000 351.0000 0.004
0.0000 353.0000 0.004
0.0000 361.0000 0.004
0.0000 366.0000 0.004
0.0000 369.0000 0.004
0.0000 371.0000 0.004
0.0000 374.0000 0.004
0.0000 375.0000 0.004
0.0000 379.0000 0.004
576.7000 277.0000 0.005
576.7000 280.0000 0.005
576.7000 295.0000 0.005
576.7000 300.0000 0.005
576.7000 302.0000 0.005
576.7000 305.0000 0.005
576.7000 308.0000 0.005
576.7000 310.0000 0.005
576.7000 312.0000 0.005
576.7000 315.0000 0.005
576.7000 315.0000 0.005
576.7000 322.0000 0.005
576.7000 324.0000 0.005
576.7000 326.0000 0.005
576.7000 330.0000 0.005
576.7000 335.0000 0.005
576.7000 336.0000 0.005
576.7000 337.0000 0.005
576.7000 339.0000 0.005
576.7000 348.0000 0.005
576.7000 351.0000 0.005
576.7000 352.0000 0.005
576.7000 357.0000 0.005
576.7000 370.0000 0.005
576.7000 380.0000 0.005
2024.0000 282.0000 0.012
2024.0000 298.0000 0.012
2024.0000 298.0000 0.012
2024.0000 304.0000 0.012
2024.0000 308.0000 0.012
2024.0000 311.0000 0.012
2024.0000 315.0000 0.012
2024.0000 316.0000 0.012
2024.0000 316.0000 0.012
2024.0000 317.0000 0.012
2024.0000 318.0000 0.012
0.040
0
-0.1996
0.040
0
-0.1996
0.040
0
-0.1996
0.040
0
-0.1996
0.040
0
-0.1996
0.040
0
-0.1996
0.040
0
-0.1996
0.040
0
-0.1996
0.016
0
-0.1262
0.040
0
-0.1996
0.009
0
-0.0957
0.046
0
-0.2139
0.046
0
-0.2139
0.036
0
-0.1913
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.046
0
-0.2139
0.032
0
-0.1790
0.041
0
-0.2029
0.046
0
-0.2139
0.046
0
-0.2139
0.041
0
-0.2029
0.046
0
-0.2139
0.046
1
4.4828
0.046
0
-0.2139
0.046
0
-0.2139
0.059 0 -0.2092
0.119 0 -0.2558
0.119 0 -0.2558
0.119 0 -0.2558
0.119 0 -0.2558
0.119 0 -0.2558
0.119 0 -0.2558
0.119 0 -0.2558
0.119 0 -0.2558
0.119 0 -0.2558
0.119 2 4.0583
Page 225 of 244
10
10
10
10
10
10
10
10
4
10
2
10
10
8
10
10
10
10
10
10
10
10
10
10
10
10
7
9
10
10
9
10
10
10
10
5
10
10
10
10
10
10
10
10
10
10
-------
2024.0000
320.0000
0.012
10
0.119
0
-0.2558
2024.0000
322.0000
0.012
10
0.119
0
-0.2558
2024.0000
323.0000
0.012
10
0.119
0
-0.2558
2024.0000
323.0000
0.012
10
0.119
1
1.9013
2024.0000
324.0000
0.012
10
0.119
0
-0.2558
2024.0000
325.0000
0.012
10
0.119
0
-0.2558
2024.0000
327.0000
0.012
10
0.119
0
-0.2558
2024.0000
331.0000
0.012
9
0.107
0
-0.2491
2024.0000
335.0000
0.012
10
0.119
0
-0.2558
2024.0000
336.0000
0.012
9
0.107
0
-0.2491
2024.0000
345.0000
0.012
10
0.119
0
-0.2558
2024.0000
347.0000
0.012
10
0.119
0
-0.2558
2024.0000
363.0000
0.012
10
0.119
0
-0.2558
2024.0000
392.0000
0.012
10
0.119
0
-0.2558
5243.0000
268.0000
0.058
10
0.583
0
-0.4505
5243.0000
294.0000
0.058
10
0.583
0
-0.4505
5243.0000
300.0000
0.058
10
0.583
0
-0.4505
5243.0000
301.0000
0.058
10
0.583
0
-0.4505
5243.0000
302.0000
0.058
8
0.466
0
-0.4369
5243.0000
309.0000
0.058
6
0.350
0
-0.4167
5243.0000
309.0000
0.058
9
0.525
3
2.0957
5243.0000
314.0000
0.058
10
0.583
1
0.3222
5243.0000
319.0000
0.058
7
0.408
0
-0.4279
5243.0000
320.0000
0.058
2
0.117
0
-0.3176
5243.0000
328.0000
0.058
10
0.583
0
-0.4505
5243.0000
335.0000
0.058
10
0.583
0
-0.4505
5243.0000
337.0000
0.058
10
0.583
6
4.1853
5243.0000
340.0000
0.058
10
0.583
0
-0.4505
5243.0000
342.0000
0.058
10
0.583
1
0.3222
5243.0000
345.0000
0.058
10
0.583
0
-0.4505
5243.0000
347.0000
0.058
10
0.583
0
-0.4505
5243.0000
349.0000
0.058
8
0.466
0
-0.4369
5243.0000
349.0000
0.058
10
0.583
0
-0.4505
5243.0000
350.0000
0.058
10
0.583
1
0.3222
5243.0000
359.0000
0.058
10
0.583
0
-0.4505
5243.0000
361.0000
0.058
10
0.583
0
-0.4505
5243.0000
366.0000
0.058
10
0.583
0
-0.4505
5243.0000
385.0000
0.058
10
0.583
1
0.3222
Scaled Residual(s) for Dose Group Nearest the BMD
Minimum scaled residual for dose group nearest the BMD = -0.2491
Minimum ABS(scaled residual) for dose group nearest the BMD = 0.2491
Average scaled residual for dose group nearest the BMD = -0.2491
Average ABS(scaled residual) for dose group nearest the BMD = 0.2491
Maximum scaled residual for dose group nearest the BMD = -0.2491
Maximum ABS(scaled residual) for dose group nearest the BMD = 0.2491
Page 226 of 244
-------
Number of litters used for scaled residual for dose group nearest the BMD = 1
Ob served Chi - square = 92.7301
Bootstrapping Results
Number of Bootstrap Iterations per run: 1000
Bootstrap Chi-square Percentiles
Bootstrap
Run P-value 50th 90th 95th 99th
1 0.4810 90.8819 170.6668 205.2285 260.8940
2 0.4710 90.1158 163.9175 188.5821 255.2733
3 0.4810 90.1495 168.9837 190.0508 267.9728
Combined 0.4777 90.3560 167.8649 194.2301 267.9728
The results for three separate runs are shown. If the estimated p-values are sufficiently
stable (do not vary considerably from run to run), then then number of iterations is
considered adequate. The p-value that should be reported is the one that combines
the results of the three runs. If sufficient stability is not evident (and especially
if the p-values are close to the critical level for determining adequate fit, e.g., 0.05),
then the user should consider increasing the number of iterations per run.
To calculate the BMD and BMDL, the litter specific covariate is fixed
at the mean litter specific covariate of all the data: 329.161616
Benchmark Dose Computation
Specified effect = 0.01
Risk Type = Extra risk
Confidence level = 0.95
BMD = 2266.39
BMDL = 723.867
Page 227 of 244
-------
6 References
Agresti. A. (1990). Categorical data analysis: Wiley.
http s: //b ooks. googl e. com/b ooks?i d=MCnv A A AAM A A J.
Barker. DJP. (2007). The origins of the developmental origins theory. J Intern Med 261: 412-417.
http://dx.doi.org/10.1111/i. 1365-2796.2007.01809.X.
Becci. PJ: Knickerbocker. MJ; Reagan. EL: Parent. RA; Burnette. LW. (1982). Teratogenicity study of
N-methylpyrrolidone after dermal application to Sprague-Dawley rats. Fundam Appl Toxicol 2:
73-76. http://dx.doi.org/10.1016/s0272-0590(82)80117-6.
Cochran. WG. (1977). Sampling Techniques (3 ed.). New York: John Wiley & Sons.
https://www.wilev.com/en-us/Sampling+Techniques%2C+3rd+Edition-p-9780471162407.
DuPont. (1990). Letter from E I DuPont de Nemours & Company to USEPA submitting comments
concerning the proposed test rule on n-methylpyrrolidone with attachment. (40-90107098). E I
Dupont De Nemours & Co.
E. I. Dupont De Nemours & Co. (1990). Initial submission: reproductive and developmental toxicity of
l-methyl-2-pyrrolidinone in the rat with cover letter dated 10/01/92. (OTS: OTS0555618; 8EHQ
Num: 8EHQ-1092-11957; DCN: 88-920010214; TSCATS RefID: 440618; CIS: NA).
Exxon. B. (1991a). Multigeneration Rat Reproduction Study with N-Methylpyrrolidone, Project Number
236535 [TSCA Submission], (OTS#: 0532510; New Doc ID: 40-91107125; Old Doc ID: 42114
Fl-2). Wayne, USA: GAF Corp.
Exxon. B. (1991b). Project No. 236535, 26 Nov 1991. ((sponsored by GAF Corp., Wayne, USA), (as
cited in OECD, 2007)). Wayne, USA: GAF Corp.
Fox. JR; Hogan. KA; Davis. A. (2016). Dose-response modeling with summary data from
developmental toxicity studies. Risk Anal 37: 905-917. http://dx.doi.org/10. Ill 1/risa. 12667.
Hothorn. LA. (2016). Statistics in Toxicology Using R. Boca Raton, Florida: CRC Press.
Kavlock. RJ; Allen. BC; Faustman. EM; Kimmel. CA. (1995). Dose-response assessments for
developmental toxicity .4. Benchmark doses for fetal weight changes. Toxicol Sci 26: 211-222.
http://dx.doi.org/10.1006/faat.1995.1092.
NMP Producers Group. (1999a). Two generation reproduction toxicity study with n-methylpyrrolidone
(NMP) in sprague dawley rats: Administration in the diet. (Project No. 97-4106). Millestone, NJ:
Huntingdon Life Science.
NMP Producers Group. (1999b). Two Generation Reproduction Toxicity Study with N-
Mythylpyrrolidone (NMP) in Wistar Rats - Administration in the Diet. (Project No.
70R0056/97008). Ludwigshafen, Germany: Department of Toxicology of BASF
Aktiengesellschaft.
Poet. TS; Kirman. CR; Bader. M; van Thriel. C; Gargas. ML; Hinderliter. PM. (2010). Quantitative risk
analysis for N-methyl pyrrolidone using physiologically based pharmacokinetic and benchmark
dose modeling. Toxicol Sci 113: 468-482. http://dx.doi.org/10.1093/toxsci/kfp264.
Reyes. L; Manalich. R. (2005). Long-term consequences of low birth weight [Review], Kidney Int
Suppl 68: S107-S111. http://dx.doi.Org/10.llll/i.1523-1755.2005.09718.x.
Saillenfait. AM; Gallissot. F; Langonne. I; Sabate. JP. (2002). Developmental toxicity of N-methyl-2-
pyrrolidone administered orally to rats. Food Chem Toxicol 40: 1705-1712.
http://dx.doi.org/10.1016/50278-6915(02)00115-1.
Saillenfait. AM; Gallissot. F; Morel. G. (2003). Developmental toxicity of N-methyl-2-pyrrolidone in
rats following inhalation exposure. Food Chem Toxicol 41: 583-588.
http://dx.doi.org/10.1016/50278-6915(02)00300-9.
Page 228 of 244
-------
Sasso. AF; Schlosser. PM; Kedderis. GL; Genter. MB; Snawder. JE; Li. Z; Rieth. S; Lipscomb. JC.
(2013). Application of an updated physiologically based pharmacokinetic model for chloroform
to evaluate CYP2E1-mediated renal toxicity in rats and mice. Toxicol Sci 131: 360-374.
http://dx.doi.org/10.1093/toxsci/kfs320.https://hero.epa.gov/hero/index.cfm?action=search.view
&reference id=6834307Shoukri. MM: Chaudhary. MA. (2018). Analysis of correlated data with
SAS and R. New York: R. Chapman and Hall.
http://dx.doi.Org/https://doi.org/10.1201/9781315277738.
Sitarek. K; Stetkiewicz. J: Wasowicz. W. (2012). Evaluation of reproductive disorders in female rats
exposed to N-methyl-2-pyrrolidone. Birth Defects Res B Dev Reprod Toxicol 95: 195-201.
http://dx.doi.org/10.1002/bdrb.21Q01.
Stiteler. WM; Knauf. LA: Hertzberg. RC: Schoeny. RS. (1993). A statistical test of compatibility of data
sets to a common dose-response model. Regul Toxicol Pharmacol 18: 392-402.
http://dx.doi.org/10.1006/rtph.1993.1065.
U.S. EPA. (2012). Benchmark dose technical guidance. (EPA/100/R-12/001). Washington, DC: U.S.
Environmental Protection Agency, Risk Assessment Forum.
https://www.epa.gov/risk/benchmark-dose-technical-guidance.
Page 229 of 244
-------
APPENDICES
Appendix A Analysis of Continuous Response Summary Data Subject
to Litter Effects
No individual fetal data were available for the studies analyzed here. For reference, when individual
fetal data are available, the preferable approach to determining the data to model is to apply a nested
analysis of variance to each dose group separately, with litter as main effect and offspring nested within
litters representing the individual replicates, and allowing for unequal litter sizes. In this case, to
determine the data to enter into BMDS, define the following:
n = number of litters in group
TTij = size of ith litter
JV = Yj?=i mi = total number of offspring in group
Fj = mean response in ith litter
To allow for an effect of the nesting of fetuses within litters on observed variance in the overall mean,
the following approach to BMDS analysis may be considered (applied separately for each group).
Sample size: JV, total number of offspring
Mean: Y = miYi, grand mean response of all offspring within the group
SD: yjMSA, the square root of the litter mean square (Cochran (1977)). where
n
MSA = Yjmi(?i-Y)2.
i=1
The last two quantities are the estimate of the mean among offspring and standard deviation of the mean
per offspring.
In cases where the individual fetal data are not available, other methods are necessary to approximate
the preferred analysis. Below are two methods applied here.
Method 1: Litter sizes and litter means are available. In this case, the litter sizes mi and litter means Yt
are available, so the quantities to enter into BMDS for the analysis of individual fetal data can be
calculated using these data as described above for the case where individual fetal data are available. This
approach was used as an alternate approach for some of the analyses presented in Section 3.3; however,
it was not utilized in the recommended modeling results.
Method 2: Means and SDs of litter means are available. When using any non-SD-based BMR, a
reasonable approximation of the preferred analysis can be made. In addition to the quantities defined
above, define the following:
— 1 —
Yl = ~Hf=i = mean of litter means
Si = ^-j-£f=1(Fj — Yl)2 = variance of litter means
The data to enter into BMDS for each group are as follows.
Sample size: n
Mean: YL
Page 23 Oof 244
-------
SD: SL = Si
Yl is generally similar to Y. SL is smaller than yjMSA by approximately a factor equal to the average litter
size (the difference is exactly equal to the individual litter size when all the litter sizes are equal).
However, the sample size n is also smaller than N by approximately the same factor, so these
differences cancel each other out. Therefore, in most cases the analysis of the means and SDs of litter
means provides a reasonable approximation of an analysis based on individual fetal data; however, high
inter-litter variability may result in poorer approximations.
Page 231 of 244
-------
Appendix B Tests for Differences and Trends in Saillenfait et al.
(2003; 2002) Post-Implantation Dose-Response Data
B.l Background and Objectives
The purpose of this appendix is to document statistical analyses of trend and trend difference for two
studies of the toxicity of NMP (Saillenfait et al. (2002) and (2003)). Saillenfait et al. (2002) is an oral
exposure study, while Saillenfait et al. (2003) is an inhalation exposure study. The data used in the
statistical analysis shown in this appendix is presented in TableApx B-l in Appendix B.2.
Two related, complementary analyses are reported in this Appendix. First is a deviance test for a
difference in dose-response relationships in the Saillenfait et al. (2002) and (2003) studies, restricted to
doses at which the dose-response curve was considered to be approximately linear, if not flat.
Restriction to a linear or flat dose range was based on graphical interpretation suggesting that the
nonlinear part of the combined curve was limited to higher doses, which were evaluated only in the oral
exposure study (Saillenfait et al. (2002)). Second is an analysis of trend based on a breakdown of the
Pearson chi-square, into chi-square statistics that represents a two-sided Cochran-Armitage test for
linear trend, and a chi-square test of nonlinearity (Agresti (1990)) as implemented in the software
EPITOOLS. According to the test of nonlinearity there were no significant deviations from linearity in
the dose range for which approximate linearity was assumed, in the deviance test.
The approach for modeling data from the Saillenfait et al. (2002) and (2003) studies was to combine the
two studies by fitting a single exposure-response curve, substituting a single internal dose metric that
can be estimated using a PBPK model for both oral and inhalation exposure routes, in place of the
external exposure concentrations. This analysis focused on dead fetuses expressed as a proportion of
implantations, "proportion dead fetuses" in Table Apx B-l below. The internal dose metric considered
in the analysis is Cmax (mg/1), as post-implantation loss is viewed as an acute response, and a statistical
test of the equivalence of the dose-response relationship in the lower dose range of the dose-response
curve was performed.
For purposes of testing for a statistical difference between the Saillenfait et al. (2002) and (2003)
studies, a very simple situation would be that the same set of doses has been evaluated in each study. A
conclusion on the role of study could then be made without a dose-response model. For a continuous
response, the analysis could be based on a two-way ANOVA, with dose and study as the two factors.
Absence of a main effect of dose, plus absence of a study-dose interaction, would together suggest that
response depends in no way on study, and might or might not depend on dose. One could then proceed
with some confidence to a dose-response model for the combined data. As an additional precaution, the
fit of such a model should still be examined separately for the two studies. An analogous approach may
in principle be developed for a dichotomous response (i.e., post-implantation losses). For the Saillenfait
et al. (2002) and (2003) studies, the controls were the only group that could be directly compared. The
comparison is necessarily based on fitted dose response models. The essential idea of the deviance test is
to evaluate whether a significantly better fit to the data is obtained by fitting the studies separately than
with the same dose-response curve. The null hypothesis is that all parameters of a dose-response model
are equal for the two studies. The idea of a parametric model-based evaluation of the compatibility of
dose-response datasets has been previously recognized (Stiteler et al. (1993)).
Based on a graphical evaluation, the dose-response relationship is practically flat up to a Cmax of 250
mg/1 for the Saillenfait et al. (2002) oral study, and is practically flat across the full range of doses
evaluated in the Saillenfait et al. (2003) inhalation study. If the combined data are modeled then (under
an assumption that dose-response parameters are equal in the two studies) the background level
Page 232 of 244
-------
parameter would be informed by data from both studies, primarily by data for Cmax < 250 mg/1.
Parameters defining the shape of the dose-response curves would, EPA expects, be informed primarily
by higher doses, which were evaluated only in the Saillenfait et al. (2002) oral study.
For the model-based comparison of this section, EPA approximated the dose-response curves for Cmax
up to 250 mg/1 using linear regressions. The approach would be substantially incorrect if there is
appreciable deviation from linearity in the dose range evaluated; however, substantial nonlinearity
appears only in the Saillenfait et al. (2002) oral study, at Cmax values > 250 mg/1. In practice any smooth,
nonlinear curve can be approximated to an arbitrary degree of precision by a straight line, in some range
of doses. The more nonlinear curve, the more narrow such a range of doses. A separate trend analysis
(Table Apx B-4) provides a test for nonlinearity based on a decomposition of chi-square and suggests
no statistical evidence of nonlinearity in the dose range of interest. While the comparison in this section
could in principle have been based on a nonlinear model that would apply to the entire dose range of
both studies, EPA did not think the essential results would be affected, because the estimated nonlinear
effects would be based on a higher dose range, evaluated in the Saillenfait et al. (2002) oral study.
As a general principle, it is suggested that such a statistical test is not necessarily to be treated as a
definitive rule by itself for deciding whether to combine the studies in dose-response modeling. If the
scientific arguments as a whole point to combining the datasets, non-significant results from the test may
be seen as having a "confirmatory" role, possibly suggesting that the scientific model is consistent with
the data as analyzed using a specific statistical criterion. Then, any apparent differences might be
considered consistent with sampling variability. However, data may be consistent with a variety of
interpretations, especially if few or highly variable. This viewpoint is similar to the concept of goodness
of fit testing and statistical model diagnostics.
The original design of this analysis was restricted to Cmax values < 250 mg/1 (based on interpretation of
the graphical analysis of all the data suggesting a linear response in that range). However, EPA repeated
the statistical tests with the dose of 531 mg/L (based on Cmax) from the Saillenfait et al. (2002) oral study
included. This extension did not change the conclusion that the dose-response relationships are similar
in the two studies.
Page 233 of 244
-------
B.2 Data
Data used in the statistical analyses are presented in Table Apx B-l. Saillenfait et al. (2002) is an oral
exposure study, while Saillenfait et al. (2003) is an inhalation exposure study. The dichotomous-
response data used for the statistical tests are shown in the columns labelled "RS-implants" and "RS-
dead." To account for potential litter effects in the developmental toxicity data, the data were adjusted
for clustering using the Rao-Scott (RS) approach (Shoukri and Chaudhary (2018); Fox et al. (2016)).
Estimated response proportions are shown in the column labelled "proportion dead fetuses." Note that
RS adjustment does not change the estimated response proportion at a given dose level. The aim of the
adjustment is to set the effective number on test to reflect the amount of information in the data, without
changing the estimated proportion that responded. Corresponding non-adjusted counts are shown in the
column labelled "Total Dead Fetuses" and "Total Implants." Note that these are also not necessarily
integer-valued. This is because the total number of dead fetuses is estimated as the product of a mean
number of fetuses, reported with limited precision, and a number of litters. The effect on calculations is
not expected to be severe.
After RS adjustment the pseudo-counts for number (number dead) and denominator (number of
implants) are not generally integer-valued. The software, which may be designed for dichotomous
responses, must process non-integer input correctly, using the same formula as used for integer-valued
inputs. The data used for this analysis are reported in the tables to 4-5 digits. Some intermediate
computations involve fewer digits. This precision is judged adequate for the type of result reported.
Page 234 of 244
-------
TableApx B-l Post-Implantation Losses/Implants from Oral (Saillenfait et al. (2002)) and Inhalation (Saillenfait et al. (2003))
Studies and Estimates of Internal C max
Reference
and
Endpoint
Cmax
(mg/
L)
Litters
w/
Implants
Mean
Implants
Total
Implants
Live
Litters
Mean
Live
Fetuses
Total
Live
Fetuses
Total
Dead
Fetuses
Proportion
Dead
Fetuses
RS-
Implants
RS-Dead
Saillenfait
0
21
13.3
279.3
21
12.7
266.7
12.6
0.0451
134.20
6.0541
et al.
120
22
13.6
299.2
21
13.1
275.1
24.1
0.0805
117.34
9.4516
(2002)
250
24
13.3
319.2
24
12.7
304.8
14.4
0.0451
153.37
6.9190
Post-
531
25
14
350
25
12.4
310
40
0.1143
121.42
13.877
implant-
ation loss
831
25
13.8
345
8
2.4
19.2
325.8
0.9443
57.044
53.870
Saillenfait
0
24
14.3
343.2
24
13.9
333.6
9.6
0.0280
194.94
5.4529
et al.
15
20
13.4
268
20
12.6
252
16
0.0597
116.73
6.9692
(2003)
30
20
14.1
282
19
14
266
16
0.0567
125.04
7.0946
Post-
implant-
ation loss
62
25
12.9
322.5
25
12
300
22.5
0.0698
133.01
9.2798
Page 235 of 244
-------
B.3 Statistical Approaches
The statistical test applied for the comparison of slopes of dose-response curves and intercepts is a
deviance test or likelihood ratio test. The test was applied to determine if response increase with dose is
sometimes termed a chi-squared test for trend. Both approaches are exemplified by a variety of tests
reported routinely by BMDS, and the general concepts are discussed in the Benchmark Dose Technical
Guidance manual, particularly in connection with the analysis of deviance table.
Given that the internal serum doses do not match in the Saillenfait et al. (2002) and (2003) studies, the
deviance test has to be based on a statistical modeling approach. Herein, the deviance test is based on
modeling with a form of linear regression, but with the response variable assumed to have a binomial
distribution (an ordinary, least-squares linear regression actually gives point estimates of slope and
intercept comparable to the estimates in Appendix B). The deviance test here was designed to be
sensitive to a difference in intercept or a difference in slope, when comparing the two Saillenfait et al.
studies (i.e., the null hypothesis is equivalence of intercepts and equivalence of slopes).
EPA used a generalized linear model (GLM) for the deviance test but with non-default software settings
as explained in Appendix B.4 (this is not exactly SAS Proc GLM, which implements the "general linear
model"). The literature on GLMs is very extensive and includes texts that are very application-oriented.
Hothorn (2016) provides considerable treatment specific to toxicological data analysis.
R code for the deviance test is provided in Appendix B.69. The test has been implemented here with base
R software that is well-established for the current analyses. Likewise, the chi-squared test for trend we
have applied using the EPITOOLS software is a common trend test, and the EPITOOLS software has
been available for many years, having been cited often for use in similar analyses.
9 R is available for download from the CRAN website at https://cran.r-project.org/ and EPITOOLS (Sergeant ESG (2018)) is
compilation of statistical tools developed "for the use of researchers and epidemiologists." The Cochran-Armitage trend test
available in BMDS 2.7 is not applied here because it requires integer data for incidences. The EPITOOLS software used here
appears to be completely distinct from an R package of the same name. The approach in EPITOOLS is based on a breakdown
of chi-square, apparently as described in Agresti (1990). The references given by the software are to be found at
https://epitools.ausvet.com.au/references.
Page 236 of 244
-------
B.4 Details of Deviance Test for a Difference in Dose-Response
Relationship
Because internal serum doses do not match in the two Saillenfait et al. studies, the test for a difference is
based on a statistical multiple regression model presented below.
Expected proportion = intercept + slope*Cmax + /study = 2*(D.intercept2 + D.slope2*Cmax)
where D.intercept2 is an intercept increment associated with the Saillenfait et al. (2003)
inhalation study;
D.slope2 is a slope increment associated with the Saillenfait et al. (2003) inhalation study; and
/study = 2 = 0 for Saillenfait et al. (2002). and /study = 2= 1 for Saillenfait et al. (2003) results.
This is a parametrization of a model that specifies two separate regressions, one for each study. In terms
of this parametrization, the null hypothesis may be stated as Ho: D.intercept2= 0 and D.slope2 = 0.
Rejection could result from a difference in intercept, a difference in slope, or a difference in both
intercept and slope.
The general approach of a deviance test is discussed in a toxicological-pharmacological journal, in an
article on evaluating "compatibility of two datasets to a common dose-response model" (Stiteler et al.
(1993)). However, the approach is well-known and more general literature sources are likely to provide
clearer descriptions of the degrees of freedom for an asymptotic chi-square test. As indicated in that
article, extension to nonlinear models (e.g., BMDS) is straightforward. Specialization to a linear
regression model here is convenient in avoiding a need for model selection. Also, issues related to
parameter constraints are avoided.
Here, the data was modeled in two ways (i.e., a single regression for both studies combined and separate
regressions for each study individually), and the results are compared based on deviance (log-likelihood
times negative). The R function anova() can be used to generate an analysis of deviance table if supplied
model objects corresponding to the two modeling options, provided parameters are named in each to
allow recognition of nesting. It is convenient to parametrize the separate-regressions model in terms of
intercept, slope, intercept difference, and slope difference. The anova() function will compute the test
statistic (difference of deviances) and degrees of freedom (i.e., two degrees of freedom here) for the test,
but does not compute ap-value. The conventional, asymptotic /> value may be computed by referring the
deviance difference to a chi-square distribution with two degrees of freedom.
Technical details
EPA used the R function glm() with binomial response family and linear link. The default link with a
binomial family is the logit link, resulting in a logistic regression. EPA assumed that the results are not
very sensitive to modeling the role of exposure in a range where the dose-response is practically flat, and
chose the link based on some concept of simplicity or familiarity.
The function glm() as such is not restricted to dichotomous responses, and EPA has found it most simple
in coding glm() to specify the binomial response family. It is reasonable to ask whether the
computations are correct, when an option designed for dichotomous data is used with non-integer
response data. EPA has found no specific statement on this to date; however, EPA does not believe that
this is an issue for the following reasons. First, the glm() function returns a warning when the responses
Page 23 7 of 244
-------
are non-integer, rather than a fatal error, suggesting that thought has been given to the possibility of non-
integer inputs. Second, EPA has been able to generate the same results in glm(), bypassing the binomial
response specification. In place of the binomial family we used the "quasi" (quasi likelihood approach),
with appropriate weights and a variance function based on the binomial distribution. Statistically, there
is no reason to restrict the quasi option to be restricted to a discrete response scale, and the warning
generated with the binomial response family does not appear. Finally, the glm() function is one of the
most used R modeling option for dichotomous responses, and EPA believes the issue of possibly non-
integer inputs has probably come up before in the long history of this function.
Page 238 of 244
-------
B.5 Results
TableApx B-2 presents an "analysis of deviance" including the p-value from the test for a difference
using the Rao-Scott transformed responses for the 0, 120 mg/L and 250 mg/L doses (based on C max )of
the Saillenfait et al. (2002) oral study and for all doses of Saillenfait et al. (2003) inhalation study shown
in TableApx B-l. Confidence intervals for parameters of the separate-regressions model of the data
from the Saillenfait et al. studies are shown in Table Apx B-3. Table Apx B-4 provides the results of
the chi-squared test for trend applied separately to each study for the same dose-response data, along
with an additional analysis of the Saillenfait et al. (2002) oral study with the 531 mg/1 dose included.
The p-value resulting from applying the deviance test to the 0, 120 mg/L and 250 mg/1 doses (based on
Cmax) of the Saillenfait et al. (2002) oral study and for all doses of Saillenfait et al. (2003) inhalation
study is 0.27, while the p-value of the additional analysis of the Saillenfait et al. (2002) oral study with
the 531 mg/1 dose included is 0.4. The Table Apx B-4 trend test results for the post-implantation loss
data from the Saillenfait et al. (2002) oral study without the 531 mg/L dose, Saillenfait et al. (2002) with
the 531 mg/L dose and the Saillenfait et al. (2003) inhalation study were 0.94, 0.053 and 0.11,
respectively. Thus, the results are not significant at cutoffs commonly used (i.e., at p-values of 0.05 and
0.01). These data are consistent with equal intercepts and equal slopes between the two studies, and no
significant increase in response with increasing dose for either study, in the dose ranges evaluated.
Table Apx B-2 Analysis of Deviance Results for Test for a Difference of Regressions
Model
Num. dose-response
model parameters
Deviance
AIC
(smaller is better)
Single binomial regression
for both studies.
2
5.3896
35.687
Separate regressions
4
2.7855
37.019
Absolute Difference
(test statistic and d.f. for test)
2
2.6041
P-value for test.
0.27
Table Apx B-3 Confidence Intervals for Parameters of the Separate-Regressions Models used in
the Deviance Test
Estimate
Standard
Error
Lower
bound
(95%)
Upper
bound
(95%)
P-value
Saillenfait et al. (2002) oral studv intercept
0.057
0.018
0.021
0.092
Saillenfait et al. (2003) inhalation studv
intercept
-1E-05
0.0001
-0.0002
0.0002
0.92
Intercept Difference (inhalation minus oral)
D.intercept2 in model.
-0.023
0.021
-0.065
0.019
0.29
Slope Difference (inhalation minus oral)
D.slope2 in model.
0.0007
0.0004
-0.0001
0.0015
0.099
Page 23 9 of 244
-------
Table Apx B-4 Trend Analysis Results
Data
Chi-square
statistic
Degrees
of
freedom
P-
value
Slope
Interpretation
Pearson's
Chi-square
Saillenfait et al.
(2002) (oral)a
2.0003
2
0.3678
Not-significant (at
the 5% level),
association between
score and outcome
not supported.
Saillenfait et al.
(2002) (oral) b
6.6774
3
0.0829
Saillenfait et al.
(2003)
(inhalation)
3.3979
3
0.3343
Chi-square
for slope
(linear trend)
Saillenfait et al.
(2002) (oral) a
0.0066
1
0.9353
0
Slope does not
differ from 0 (at the
strict 5%
significance level).
Some indication of
a trend if 531 mg/1
is included.
Saillenfait et al.
(2002) (oral) b
3.7515
1
0.0528
« 0.05
le-04
Saillenfait et al.
(2003)
(inhalation)
2.5611
1
0.1095
6e-04
Chi-square
for non-
linearity
Saillenfait et al.
(2002) (oral) a
2.9259
1
0.158
Trend does not
differ significantly
at the 1% level from
linearity
Saillenfait et al.
(2002) (oral) b
1.9937
2
0.2316
Saillenfait et al.
(2003)
(inhalation)
0.8368
2
0.6581
a Using Rao-Scott transformed responses for 0, 120 and 250 mg/1 doses.
b Using Rao-Scott transformed responses for 0, 120, 250 and 531 mg/1 doses.
Page 240 of 244
-------
B.6 Code (R)
m
## NMP data from Saillenfait 2002 (first) and 2003 (second).
## Obtained April 21 from Allen Davis, extracted from report table.
## Data were copied electronically.
## N = number of implantations
## r = number of fetal deaths
## Cmax = Cmax internal dose (mg/L) from PBPK
## ##
## Assume a linear regression of r/N on Cmax and test for a difference
## in regression lines (difference in slope or intercept)
## The approach has restricted applicability, as a rule to doses
## no larger than the lowest NOAEL. The regression is based on glm().
## The default use of the function with binomial response family would
## result in logistic regression. A linear link was chosen for
## perceived ease of explanation (the approach is linear regression
## for response proportions with binomial response).
## ##
m
## data has been included initially for all groups.
## data for doses LE 250 mg/L subsequently selected for analysis.
Cmax.cutoff <- 250
## Saillenfait(2002)
Cmax2002 <- c(0, 120, 250, 531, 831 )
r2002 <-c(6.0541, 9.4516, 6.919, 13.877, 53.87)
N2002 <- c(134.2, 117.34, 153.37, 121.42, 57.044)
stdy2002 <- rep("S112002", length(Cmax2002))
## Saillenfait(2003)
Cmax2003 <- c(0, 15,30,62)
r2003 <-c(5.4529, 6.9692, 7.0946,9.2798)
N2003 <- c(194.94, 116.73, 125.04, 133.01)
stdy2003 <- rep("S112003", length(Cmax2003))
## m
## combined data frame
##
d2002 <- data.frame( stdy=stdy2002, Cmax=Cmax2002, r=r2002, N=N2002 )
Page 241 of 244
-------
d2003 <- data.frame( stdy=stdy2003, Cmax=Cmax2003, r=r2003, N=N2003 )
dO <- rbind(d2002,d2003)
## m
## transformations and selection
## dummy for 2003 study (regressor for intercept diff)
d0$is2003 <- with(d0, as.numeric(stdy) == 2)
## analysis dataset is a selection based on dose
danly <- subset(d0, Cmax <= Cmax.cutoff)
View(danly)
## ##
## graph of response proportions (confidence intervals desirable)
##
dpi <- danly
dpl$pr <- with(danly, r / N)
with(dpl, plot(Cmax,pr, type = "n", ylab = "response proportion"))
with(subset(dpl, is2003), points(Cmax, pr, pch=19))
with(subset(dpl,!is2003), points(Cmax, pr, pch=17))
## m
## function to report params to specified number of digits, and a label
## for model.
printCoeffs <- function(model, mod.label, dig = 3 ) {
cat("\n", mod.label, signif(coef(model), dig), "\n")
invisible()
}
## ##
m
## Models with a single slope and intecept for both datasets.
## A preliminary ordinary linear regression of response proportion
## on dose is given for illustration, giving slope and intercept
Page 242 of 244
-------
## estimates comparable to those of the preferred, binomial response
## approach.
##
printCoeffs(
lm(r / N ~ Cmax, data=danly),
"ordinary linear regression (interc., slope)"
)
## model with same slope and intercept for both datasets.
model, lreg <- glm(
cbind(r, N - r) ~ Cmax,
data = danly,
family = binomial(link = "identity")
)
printCoeffs( model, lreg,
"binomial linear regression (interc., slope)" )
## same result without explicitly binomial family, using quasi family
## final model uses variance function (variance as function of mean)
# final lreg <- glm(
# pr ~ Cmax, data=danly, weights = N,
# family = quasi(link = "identity", variance = "mu(l-mu)"))
## ##
## model with slope and intercept estimated separately.
## The parametrization is in terms of a slope difference an d
## intecept difference for the 2nd dataset. This is convenient
## for computing an analysis of deviance table using anova(),
## requiring an recognizable nesting of the 2 models.
## anova([modell], model[2])
##
## regressor for estimating slope difference
danly$dslope03 <- with(danly, is2003*Cmax)
View(danly)
model.2reg <- glm(
cbind(r, N - r) ~ Cmax + is2003 + dslope03,
data = danly,
family = binomial(link = "identity")
Page 243 of 244
-------
)
printCoeffs( model.2reg, "separate binomial-response regressions")
##
## deviance test for difference in intercept or difference in slope
## analysis of deviance table (does not compute p-value)
print(anova(model. lreg, model.2reg))
## p-value
chi <- deviance( model.lreg ) - deviance( model.2reg )
degfr <- length(coef(model.2reg)) - length(coef(model.lreg))
cat("\n\nchi-square = signif(chi,4),
"\nd.f. =", degfr,
"\np = signif(pchisq(chi, degfr, lower.tail=FALSE ), 2)
)
##
## alternative - use binomial response as for logistic reg.
## see glm function in R manual
## default logit link
# m.glml <- glm(Ymat ~ Cmax, family = binomial, data=danly)
# coef(m.glml)
Page 244 of 244
------- |