Dear Juergen, thanks for your comment. I was actually not aware such full non-parametric approach, apology for my ignorance. the approach is very intersting, I will try to understand it more.
with regards to non-parametric approach, I was thinking alone the line of estimation method for Eta only as offered in nonmem. so I went ahead tried $NONPARAMETRIC UNCONDITIONAL option, but the Eta for Ka still estimated to be very small, 5.50E-08 vs 0.13 estimated by using rich data only. ________________________________ From: Jurgen Bulitta <jbuli...@ordwayresearch.org> To: Ethan Wu <ethan.w...@yahoo.com> Cc: "nmusers@globomaxnm.com" <nmusers@globomaxnm.com>; Roger Jelliffe <jelli...@usc.edu>; "Neely, Michael" <mne...@usc.edu> Sent: Wednesday, June 17, 2009 2:42:31 PM Subject: RE: [NMusers] estimating Ka from dataset combining rich sample study and sparse sampling study Dear Ethan, Your first suggestion would be a pragmatic way of moving forward. I have no personal experience with the hybrid method. Your third suggestion, using a full non-parametric approach should work better and is mathematically more consistent. This approach should not suffer from shrinkage. I would expect this algorithm to behave as follows: 1) The subjects with rich data should be essentially completely unaffected by the subjects with sparse data. 2) The subjects with sparse data should have posterior (i.e. intra-individual) probability distributions of Ka which are similar to the inter-individual distribution of Ka for the population of subjects with rich data. Depending on how the distribution of individual Ka values of the subjects with rich data look, you may or may not get a multimodal intra-individual distribution of Ka for the patients with sparse data. This may become important for the covariate relationships which you are trying to develop subsequently.. Please let me know, if Roger’s group or I can be of help to set you up, if you want to use NPAG for solving this task. Best wishes Juergen From:owner-nmus...@globomaxnm.com [mailto:owner-nmus...@globomaxnm.com] On Behalf Of Ethan Wu Sent: Wednesday, June 17, 2009 11:21 AM To: nmusers@globomaxnm.com Subject: [NMusers] estimating Ka from dataset combining rich sample study and sparse sampling study Dear all, I am working on this pop PK analysis. the objective is, to explore some covariates on the exposure. the dataset has rich sampled study, with absorption phase well captured. and also sparse sampling study with only trough sample, and another sample around 1-2hr after dosing with rich sample study data, the ka and eta on Ka is well estimated using FOCE INT method and 1ct 1st order model. but when with pooled dataset, using the same model and method, eta on Ka is estimated to be almost 0, the fit to the data from rich sampled study became little worse on the peak. Is there way to keep a good estimation of Eta on Ka, which is to make sure the good capture of Cmax, at least for rich sampled subjects? with my limited knowledge, I was thinking: -- fixing Eta on ka with the estimate from rich sample study alone -- hybrid estimating methods -- nonparametric method Any comments will be highly appreciated.
