vpc under PsN

Dear all, I am new to vpc and have a basic technical question regarding the vpc under PsN to ask. Currently I am doing a vpc to validate a model, whose data was constructed in such a way that the dependent variable is log-transformed and labeled LNDV. The original model was built with data from both healthy subject as well as patients, while the vpc separated them. In the first step a plot was produced for the healthy subjects in which I found several outlier observations. This caused the 97.5% quantile of the observations to have an abnormal profile. So I tried to locate the variable, and run a separate vpc to check the influence of them. I looked up both observations in the vpctab file produced by PsN, and tried to look for the points in the model dataset. However to my surprise I did not find the observation in the dataset, whether the exact number (DV) or the ID. According to the vpc manual the ID is copied exactly from the dataset file, but the ID was also not found. Could you please help me out of this, as how to locate the points? Thanks a million for your kind help. I am attaching the command I am using in the vpc for your information: vpc final.nmctl -predcorr -lnDV=1 -dv=LNDV -bin_by_count=0 -bin_array=0.375,0.75,1.5,2.5,4,6,8.5,12,17.5,24.5,35,49,63,77,91 -stratify_on=STRT -clean=3 -seed=19860322 -samples=200 -run_on_sge -sge_queue=all.q -parafile=template.pnm -nodes=30 -sge_prepend="-pe mpich 30" Best Regards -- Xinting

Hi Xinting, if you used -predcorr when creating the VPC, your DVs in each bin will be rescaled using their own PRED and the median PRED in that bin, so you won't be able to find the exact value in your original dataset. As for the ID renumbering, I am not able to comment, but I would post this on the user list for PsN, where they may be able to help. Until you get an answer, I would simply suggest re-running the VPC without -predcorr and spot the outliers with that, if they are still visible. This way you could also try to infer whether the outlier values were introduced with the rescaling for the pred-correction, or they were there already in the original data. My experience is that pred-corrected VPCs may introduce unduly large rescaling when the bin is large and the concentration profile changes a lot within that bin. Think of the extreme case of a delay in absorption (or a very sharp transit absorption): samples drawn before or after the delay will have wildly different PREDs, and will thus be scaled very differently, cuz the scaling factor is not calculated for the median PRED at the time of the sample that is rescaled, but for the median PRED throughout the bin. I hope this helps you sort out your problem, and I would welcome other people's comment on this implementation of PC-VPC (or on my misunderstanding). Ciao, Paolo

Hi Xinting, The file vpctab is intended solely as input to Xpose when visualizing the vpc, it is not the table file described in the userguide. If you want to see the table files you should look at the *npctab.dta files in the m1 subdirectory. Best regards, Kajsa On 06/27/2014 05:17 PM, Paolo Denti wrote: Hi Xinting, if you used -predcorr when creating the VPC, your DVs in each bin will be rescaled using their own PRED and the median PRED in that bin, so you won't be able to find the exact value in your original dataset. As for the ID renumbering, I am not able to comment, but I would post this on the user list for PsN, where they may be able to help. Until you get an answer, I would simply suggest re-running the VPC without -predcorr and spot the outliers with that, if they are still visible. This way you could also try to infer whether the outlier values were introduced with the rescaling for the pred-correction, or they were there already in the original data. My experience is that pred-corrected VPCs may introduce unduly large rescaling when the bin is large and the concentration profile changes a lot within that bin. Think of the extreme case of a delay in absorption (or a very sharp transit absorption): samples drawn before or after the delay will have wildly different PREDs, and will thus be scaled very differently, cuz the scaling factor is not calculated for the median PRED at the time of the sample that is rescaled, but for the median PRED throughout the bin. I hope this helps you sort out your problem, and I would welcome other people's comment on this implementation of PC-VPC (or on my misunderstanding). Ciao, Paolo On 2014/06/27 15:47, Xinting Wang wrote: Dear all, I am new to vpc and have a basic technical question regarding the vpc under PsN to ask. Currently I am doing a vpc to validate a model, whose data was constructed in such a way that the dependent variable is log-transformed and labeled LNDV. The original model was built with data from both healthy subject as well as patients, while the vpc separated them. In the first step a plot was produced for the healthy subjects in which I found several outlier observations. This caused the 97.5% quantile of the observations to have an abnormal profile. So I tried to locate the variable, and run a separate vpc to check the influence of them. I looked up both observations in the vpctab file produced by PsN, and tried to look for the points in the model dataset. However to my surprise I did not find the observation in the dataset, whether the exact number (DV) or the ID. According to the vpc manual the ID is copied exactly from the dataset file, but the ID was also not found. Could you please help me out of this, as how to locate the points? Thanks a million for your kind help. I am attaching the command I am using in the vpc for your information: vpc final.nmctl -predcorr -lnDV=1 -dv=LNDV -bin_by_count=0 -bin_array=0.375,0.75,1.5,2.5,4,6,8.5,12,17.5,24.5,35,49,63,77,91 -stratify_on=STRT -clean=3 -seed=19860322 -samples=200 -run_on_sge -sge_queue=all.q -parafile=template.pnm -nodes=30 -sge_prepend="-pe mpich 30" Best Regards -- Xinting -- ------------------------------------------------ Paolo Denti, PhD Pharmacometrics Group Division of Clinical Pharmacology Department of Medicine University of Cape Town K45 Old Main Building Groote Schuur Hospital Observatory, Cape Town 7925 South Africa phone: +27 21 404 7719 fax: +27 21 448 1989 email: [email protected] ------------------------------------------------ -- ----------------------------------------------------------------- Kajsa Harling, PhD System Developer Department of Pharmaceutical Biosciences Uppsala University [email protected] +46-(0)18-471 4308 http://www.farmbio.uu.se/research/researchgroups/pharmacometrics/

Dear Kajsa and Paolo, Thanks very much for your explanation. I would try to delete the -predcorr option in the command line and run the vpc again. But another question is, if I spotted an unusual point in one of the VPC runs with (-predcorr), is it possible to detect this observation in the original data set, so that I might be able to see where the problem is? Thank you. Best Regards