Statistical Properties of the Signal-to-Noise Crossover Dose Based on the Hill Model as a Point of Departure for Health Risk Assessment

No-observed-adverse-effect-level (NOAEL) and benchmark dose (BMD) are widely accepted points of departure for human exposure guidelines. BMD has considerable merits to the NOAEL. However, BMD estimates could be model dependent if the estimation requires extrapolation beyond the range of experimental data. Signal-to-noise crossover dose (SNCD) is recently introduced to overcome this limitation, where the focus was point estimation of the SNCD using NTP data sets. In this research, we introduce three new approaches to estimate the SNCD. Further to illustrate the characteristics of the BMD to SNCD, a new analytical approach to estimate the BMD is introduced and comparison of these PoDs is carried out using the NTP data sets. The first part involves estimation and assessment of Hill model parameters and an introduction of an analytical approach to estimate the benchmark dose. Accuracy and precision of these estimates are investigated for all possible combination of four shapes of the Hill curves, two dose levels and four number of trials per dose and effects of these factors on these estimates are reported. The second part includes estimation and assessment of SNCD using three different methods. Results show that the analytical approach with bias correction is the best. To estimate the realized bias, multiple regression equation as a function of Hill model parameters, number of dose levels, number of trials per dose and maximum dose; and additional five new approaches to estimate the bias are introduced and compared. Our analysis show that the best approach to estimate the bias is the multiple regression approach. It is used in the application of the above illustrated theories. The developed theories are applied in the analysis of NTP datasets and the corresponding PoDs are estimated. Comparison of BMD to SNCD, BMDL to SNCDL and computation of extra risks at SNCD and SNCDL are carried out to illustrate the advantages of one PoD over the other. Results show that SNCD0.67 is comparable to BMD10 and SNCDL0.67 to BMDL10 and confirming the conclusion that there is a need to choose smaller p (< 0.67) of SNCD which could be comparable to BMD05 or BMD01