Dear Martin, Chenyan, and Devin, I agree that in a large number of cases it would probably not matter much which of the two (pcVPC or NPDE) you use, however there are particular cases where I think NPDE may have advantages over pcVPC. This is mainly when you are evaluating clinical data with highly variable dosing (for instance due to rescue medication or dose adjustments based on individual needs) or in situations with large covariate effects.
This is because pcVPC averages simulated DV values for all individual profiles combined over a certain bin and compares the observed DV values of all individuals to this, but when the variability in the bin is large due to inter-individual variability in dosing regimen or the influence of covariates, this may not be very relevant (If there is only one major covariate effect, stratifying the pcVPC may be an option, but with a large number of covariates or a large range in continuous covariate values this may become impractical as well). These issues can be overcome by NPDE, as the observed DV values of an individual are compared to the simulated range of DV values for each separate individual, taking individual dosing and covariate effects into account in the simulations. Kind regards, Elke Krekels From: [email protected] [mailto:[email protected]] On Behalf Of Martin Bergstrand Sent: Monday, September 14, 2015 3:42 PM To: Devin Pastoor; ZhaoChenyan; [email protected] Cc: [email protected] Subject: RE: [NMusers] pcVPC or NPDE Dear Chenyan and Devin, Chenyan: Nice of you quote me ;) I still stand by the old quote. In my opinion the advantage with the pcVPC is that it is easy to diagnose if the model accurately predicts both the central trend as well as the variability of the data. The NPDEs on the other hand usually makes for a faster diagnostic that do not require any binning of the data for a rough interpretation. For you own interpretation I recommend you use both types of diagnostics (possibly with different x-axis variables, time, dose etc.) and for a publication you should use whatever you think will be easiest for the audience to interpret. Devin: The posterior predictive check (PPC) you suggest will only be appropriate if you have a model that also predicts the dose alteration decisions (TDM). With TDM dosing a PPC with the actual dosing history will result in an over prediction of the variability just like VPCs has been demonstrated to do. Bergstrand M, Hooker AC, Wallin JE, Karlsson MO. Prediction-corrected visual predictive checks for diagnosing nonlinear mixed-effects models. AAPS J. 2011 Jun;13(2):143-51. doi: 10.1208/s12248-011-9255-z. Best regards, Martin From: [email protected]<mailto:[email protected]> [mailto:[email protected]] On Behalf Of Devin Pastoor Sent: den 11 september 2015 16:46 To: ZhaoChenyan; [email protected]<mailto:[email protected]> Cc: [email protected]<mailto:[email protected]> Subject: Re: [NMusers] pcVPC or NPDE Dear Chenyan, Appropriateness is largely a matter of what the ultimate purpose of the model is, and neither metric will be 'better' in all cases. Extrapolating into a new population may require different evaluation diagnostics than using a model to optimize the dose the observed population. Given you only have trough samples, using a posterior predictive check on trough levels or equivalence criteria such as proposed in: 1. Jadhav, P. R. & Gobburu, J. V. S. A new equivalence based metric for predictive check to qualify mixed-effects models. AAPS J 7, E523–E531 (2005). would likely work well. Devin Pastoor Clinical Research Scientist, PhD student Center for Translational Medicine University of Maryland, School of Pharmacy On Fri, Sep 11, 2015 at 10:38 AM ZhaoChenyan <[email protected]<mailto:[email protected]>> wrote: Dear all: I'm now having a set of TDM data, only troughs （C0 ） available. I intend to evaluated the appropriateness of the constructed model. My question is whether to use pcVPC or NPDE as a diagnostic tool in such a case? Which one is better? Or to use them both, as suggested by Bergstrand et al.: "The best practice most likely lies in using a wide toolbox of diagnostics, rather than one single universal test to decide whether a model is fit for purpose or not." Thank you in advance. Yours, Chenyan Zhao Email: [email protected]<http://hotmail.com> Mobile: +86 13917430219
