Q: Can F not be used as a parameter?
A: Yes, if you obtained data from both iv and oral routes (ideally for each species). Here we have iv data only. Also note that unlike Classic WNL F is not a reserved name.
Q: Can you explain more the linear model components, regressor etc.?
A: This has been demonstrated during the WebEx. In the Linear Mixed Effects Model object you need to specify which of your columns correspond to Classification or Regressor variables. Typical Classification variables are conditions like treatment, period or sequence. Regressor variables are covariates like age or body weight. Then you would need to specify your dependent variable. In our LinMix model we have selected the log of body weight (LN_BW) as Regressor variable and the final parameter estimates as dependent variables, we have used the Parameter column as sort key so that the program will distinguish between the different parameters. We have chosen to transform the dependent variable so that the regression is done on the logarithm of the final estimates. When we execute the LinMix model the program is doing a log-log regression that corresponds to solving the general allometric equation:
Q: Would the regression function in the plot have worked similarly as the Lin Mix?
A: Unfortunately no, as it does not have an option to compute the regression for a ln/ln plot.
Q: Typically we only have animal data and want to predict man. Can you describe how you would do that?
A: Good question! Fit the model to the data you have, and then you can estimate the parameters for man as secondary parameters. e.g., Cl_man = a*70^b
Q: Could you still assume allometric scaling works for humans if it would work across several other species?
A: Not necessarily. There could be differences in binding or metabolism (or other factors). Note that you never know for sure if scaling works until after you collect the data.
Q: If we have multiple subjects from each species, do you suggest to calculate the mean value first then run the model? Or do you suggest to run the model directly, then it will take long time to fit individually?
A: In general, I prefer using a population approach whenever possible and fit the means only as a last resort.
Q: If you use a population pk model (only one subject from one species), do you assume there are no differences on exposure accrossing species?
A: In the PK29 example, we had one profile per species, so we did a naive pooled fit. That is, we ignored any between subject variability (BSV). In effect, we are also assuming the BSV is comparable across species.
Q: What is it exactly what the "covariate" statement does?
A: For standard PK models the only columns the program is looking for are t and C. However, for this model we need the body weight (BW) as well. The covariate statement enables an additional column on the data set to be mapped in and used in the model.
Q: Can't we just declare "BW" as a parameter?
A: BW is fixed – it does not change for a given individual. Parameters are values that need to be estimated.
Q: What is the validity of the allometric scaling approach in estimating pediatric dose from adult data since there will be only 2 points on the graph? In this case, shall we estimate or fix the exponents on CL and Vss?
A: For this type of modeling a population approach is best, so there are as many points as individuals. Ideally one would want adult data from a relatively broad range.
Q: Can you describe when you would use the allometric scaling and Dedrick plots method versus Wajima method to predict concentration-time profile in human from preclinical PK data?
A: You may want to read this reference:
J Pharm Sci. 2011 Oct;100(10):4111-26. doi: 10.1002/jps.22551. Epub 2011 Apr 7.
PhRMA CPCDC initiative on predictive models of human pharmacokinetics, part 4: prediction of plasma concentration-time profiles in human from in vivo preclinical data by using the Wajima approach.
Here is the last line of the abstract: “The overall ability of the Css -MRT approach to predict the curve shape of the profile was in general poor and ranged between low to medium level of confidence for most of the predictions based on the selected criteria.”
Q: How about modeling responses (not PK parameters)? Can you scale those parameters as well?
A: If you have a PK/PD model, the portion can be scaled as in the example. (System) Turnover model parameters like Kin and Kout can also be scaled. These are mentioned in the ppt file. In general, however, please note that the theory of allometry is based on changes in metabolism related to body size. Thus, pharmacokinetic variables associated with metabolic processes (e.g. CL) are well modeled using allometric methods. However, non-metabolic processes do not always follow general allometric principles. Often pharmacodynamics responses do not follow allometric principles, which could lead to overdosing in humans relative to animals. For this reason, caution should be exercised when scaling pharmacodynamics responses with allometry.
Q: Would 2 preclinical species (one rodent and one non-rodent) be sufficient to perform the allometry scaling? And how is the performance in general?
A: One would prefer more species as if you only have two points they will always fall on a line. So such an approach is risky. But regardless, we never know if scaling will work until we actually collect the data.
