Dr. Bernd-Wolfgang Igl (Bayer AG)
Dose-response analysis is a central part in statistical evaluation of toxicological bioassays. Two approaches are used: simultaneous testing of order-restricted multiple contrasts and regression-based modeling. The first one considers the DOSE qualitatively, i.e. as randomized factor whereas the second assumes DOSE as quantitative covariate (in bioassays commonly for grouped dose levels). Both approaches are demonstrated by means of real data examples where robustness, e.g. against downturn effects is discussed. Moreover, a new approach is explained, where DOSE is jointly considered both quali- and quantitatively.
The recent p-value controversy is discussed from the perspective of regulatory toxicology where first confidence intervals for specific selected effect sizes are recommended. Secondly, the inclusion of individual data points within or without a prediction interval is proposed as an alternative to common-used null-hypothesis significance tests. The prediction intervals are defined for any single future value of a group with sample size n_i using the controls of multiple historical bioassays. The within- and between assay variance is considered by a mixed effect model.
Finally, the question will be discussed why the proof of safety („be safe in negative results“) is not widely used in routine up to now.
The “third main set” of statistics is: software must be available. And therefore all methods are demonstrated using //R-//CRAN packages.
Dr. Bernd-Wolfgang Igl (Bayer AG)
Dose-response analysis is a central part in statistical evaluation of toxicological bioassays. Two approaches are used: simultaneous testing of order-restricted multiple contrasts and regression-based modeling. The first one considers the DOSE qualitatively, i.e. as randomized factor whereas the second assumes DOSE as quantitative covariate (in bioassays commonly for grouped dose levels). Both approaches are demonstrated by means of real data examples where robustness, e.g. against downturn effects is discussed. Moreover, a new approach is explained, where DOSE is jointly considered both quali- and quantitatively.
The recent p-value controversy is discussed from the perspective of regulatory toxicology where first confidence intervals for specific selected effect sizes are recommended. Secondly, the inclusion of individual data points within or without a prediction interval is proposed as an alternative to common-used null-hypothesis significance tests. The prediction intervals are defined for any single future value of a group with sample size n_i using the controls of multiple historical bioassays. The within- and between assay variance is considered by a mixed effect model.
Finally, the question will be discussed why the proof of safety („be safe in negative results“) is not widely used in routine up to now.
The “third main set” of statistics is: software must be available. And therefore all methods are demonstrated using //R-//CRAN packages.
Dr. Bernd-Wolfgang Igl (Bayer AG)
Dose-response analysis is a central part in statistical evaluation of toxicological bioassays. Two approaches are used: simultaneous testing of order-restricted multiple contrasts and regression-based modeling. The first one considers the DOSE qualitatively, i.e. as randomized factor whereas the second assumes DOSE as quantitative covariate (in bioassays commonly for grouped dose levels). Both approaches are demonstrated by means of real data examples where robustness, e.g. against downturn effects is discussed. Moreover, a new approach is explained, where DOSE is jointly considered both quali- and quantitatively.
The recent p-value controversy is discussed from the perspective of regulatory toxicology where first confidence intervals for specific selected effect sizes are recommended. Secondly, the inclusion of individual data points within or without a prediction interval is proposed as an alternative to common-used null-hypothesis significance tests. The prediction intervals are defined for any single future value of a group with sample size n_i using the controls of multiple historical bioassays. The within- and between assay variance is considered by a mixed effect model.
Finally, the question will be discussed why the proof of safety („be safe in negative results“) is not widely used in routine up to now.
The “third main set” of statistics is: software must be available. And therefore all methods are demonstrated using //R-//CRAN packages.
Dr. Bernd-Wolfgang Igl (Bayer AG)
Dose-response analysis is a central part in statistical evaluation of toxicological bioassays. Two approaches are used: simultaneous testing of order-restricted multiple contrasts and regression-based modeling. The first one considers the DOSE qualitatively, i.e. as randomized factor whereas the second assumes DOSE as quantitative covariate (in bioassays commonly for grouped dose levels). Both approaches are demonstrated by means of real data examples where robustness, e.g. against downturn effects is discussed. Moreover, a new approach is explained, where DOSE is jointly considered both quali- and quantitatively.
The recent p-value controversy is discussed from the perspective of regulatory toxicology where first confidence intervals for specific selected effect sizes are recommended. Secondly, the inclusion of individual data points within or without a prediction interval is proposed as an alternative to common-used null-hypothesis significance tests. The prediction intervals are defined for any single future value of a group with sample size n_i using the controls of multiple historical bioassays. The within- and between assay variance is considered by a mixed effect model.
Finally, the question will be discussed why the proof of safety („be safe in negative results“) is not widely used in routine up to now.
The “third main set” of statistics is: software must be available. And therefore all methods are demonstrated using //R-//CRAN packages.
Dr. Bernd-Wolfgang Igl (Bayer AG)
Dose-response analysis is a central part in statistical evaluation of toxicological bioassays. Two approaches are used: simultaneous testing of order-restricted multiple contrasts and regression-based modeling. The first one considers the DOSE qualitatively, i.e. as randomized factor whereas the second assumes DOSE as quantitative covariate (in bioassays commonly for grouped dose levels). Both approaches are demonstrated by means of real data examples where robustness, e.g. against downturn effects is discussed. Moreover, a new approach is explained, where DOSE is jointly considered both quali- and quantitatively.
The recent p-value controversy is discussed from the perspective of regulatory toxicology where first confidence intervals for specific selected effect sizes are recommended. Secondly, the inclusion of individual data points within or without a prediction interval is proposed as an alternative to common-used null-hypothesis significance tests. The prediction intervals are defined for any single future value of a group with sample size n_i using the controls of multiple historical bioassays. The within- and between assay variance is considered by a mixed effect model.
Finally, the question will be discussed why the proof of safety („be safe in negative results“) is not widely used in routine up to now.
The “third main set” of statistics is: software must be available. And therefore all methods are demonstrated using //R-//CRAN packages.
Dr. Bernd-Wolfgang Igl (Bayer AG)
Dose-response analysis is a central part in statistical evaluation of toxicological bioassays. Two approaches are used: simultaneous testing of order-restricted multiple contrasts and regression-based modeling. The first one considers the DOSE qualitatively, i.e. as randomized factor whereas the second assumes DOSE as quantitative covariate (in bioassays commonly for grouped dose levels). Both approaches are demonstrated by means of real data examples where robustness, e.g. against downturn effects is discussed. Moreover, a new approach is explained, where DOSE is jointly considered both quali- and quantitatively.
The recent p-value controversy is discussed from the perspective of regulatory toxicology where first confidence intervals for specific selected effect sizes are recommended. Secondly, the inclusion of individual data points within or without a prediction interval is proposed as an alternative to common-used null-hypothesis significance tests. The prediction intervals are defined for any single future value of a group with sample size n_i using the controls of multiple historical bioassays. The within- and between assay variance is considered by a mixed effect model.
Finally, the question will be discussed why the proof of safety („be safe in negative results“) is not widely used in routine up to now.
The “third main set” of statistics is: software must be available. And therefore all methods are demonstrated using //R-//CRAN packages.
