Table G.18.
Models | Number of parameters | Minus Log‐likelihood | p‐value | AIC | BMD10 | BMDL10 | Comment |
---|---|---|---|---|---|---|---|
(mg/kg bw per day) | |||||||
Full model | 5 | 5.04 | na | 20.08 | na | na | |
Null (reduced) model | 1 | 20.19 | na | 42.18 | na | na | |
Probit | 2 | 5.44 | 0.99 | 14.89 | 0.22 | 0.11 | Selected |
Logistic | 2 | 5.51 | 0.97 | 15.03 | 0.23 | 0.12 | Selected |
LogProbit | 2 | 5.41 | 1.00 | 14.83 | 0.22 | 0.11 | Selected |
LogLogistic | 2 | 5.45 | 0.99 | 14.90 | 0.22 | 0.11 | Selected |
Multistage | 2 | Faileda | na | ||||
Multistage Cancer | 1 | 5.89 | 0.91 | 13.78 | 0.14 | 0.054 | AIC minimum |
Quantal‐Linear | 1 | 8.08 | 0.25 | 18.15 | 0.045 | 0.027 | Not selected |
Weibull | 2 | 5.41 | 1.00 | 14.62 | 0.21 | 0.081 | Selected |
Gamma | Failed |
The models selected to identify the lowest BMDL10 following EFSA guidance (EFSA Scientific Committee, 2017) are identified in the column ‘Comments’ and include the model with the lowest AIC.
AIC: area under the curve; BMD10: benchmark dose response of 10%; BMDL10: 95% lower confidence limit for the benchmark dose response (BMR) of 10%; bw: body weight; na: not available, including cases where the BMD/Ls were not calculated, the fit was incomplete, or serious or conflicting comments were noted (e.g. parameters reaching boundary)
The multistage model was considered as failed since the output, see Suppl. Information Mink gave for model fit a statistically impossible p‐value of 2 which was taken as an indication for problems of model fitting using BMDS (see also the graphic in Supplementary Information WUDON3Ac (Annex A).