Please note that Product Notification PHX032 has been issued for Phoenix 1.4 that affects NLME and Connect users only. It does not affect WinNonlin users. The Product Notification is also posted on the Certara Support website. This notification contains two defects:
PHOENIX MODEL OBJECT DEFECT (affects Phoenix NLME 1.3):
Quantiles are computed incorrectly for prediction-correction VPC (QC 15573):
In Phoenix 1.4, the quantiles of the original observation data are being determined before the original observation data is prediction-corrected. Therefore, the values are incorrectly reported as being identical before and after the correction. The incorrect output is in the Observations worksheet and in the VPC plot for the quantiles of the observed values (lines in red). There is no known workaround.
CDISC Data Preparer can produce incorrect Sample results with multiple Subject Identifiers (QC 15247/CRM 154491):
The CDISC Data Preparer merges data from the DM (Demographics) and the PC (Pharmacokinetic Concentrations) domains using the columns mapped to the Subject Identifier mapping context. The resulting “Sample” dataset contains the merged DM and PC data. In Phoenix 1.4, if multiple columns were mapped as identifiers, the code did not properly compare all mapped columns when executing the merge. The only workaround in Phoenix 1.4 is to map a single column to the Subject Identifier mapping context, if one identifier is sufficient.
For the VPC(QC 15573)issue I suggest that the user can do the following to mitigate the impact:
Use the model fit PRED (and observed DV data ) to correct the observation and compute the observed quantiles and then do a plot that correctly compare pred-corrected observed and simulated quantiles. Note that binning should be done in the same way it is done on the simulated data from VPC.
Perform a VPC which is not pred corrected but appropriately stratified by Dose, study and or any other relevant data design elements.
Use simulation based residuals PWRES to look at the model deficiency based on simulation ( this can be thought of as similar to using NPDE).
Output raw simulation table and observed data and feed it into R npde package to compute npde or R vpc package to do the vpc computations in R.
If you’re in Europe be aware that there is a one day (8am to 4pm) Roadshow 26 April, Basle and 28 April Cambridge UK, that will include demonstrations as well as the some presentations of interest; e.g.
2016 Software Roadmap[list]
[*]New features and functions for Phoenix
[*]Conceptualization and benefits of a biosimulation model platform
[*]Presenting and reporting PK, PD and Tox results to maximize impact
[/list] Leveraging Workflow Templates [list]
[*]Pre-clinical
[*]Clinical
[/list] Case Studies: Creating a Pharmacometric Strategy The Value Chain of Modeling
Automating Pivotal PK/PD Processes using Plug-ins [list]
[*]Interoperability with LIMS, allometric scaling algorithms, and CDISC submission systems
[/list] Introducing Certara University’s new e-learning platform and content
Princeton, NJ – May 5, 2016
Durham (Raleigh), NC – May 12, 2016
San Diego, CA – May 17, 2016
Lincolnshire (Chicago), IL – May 19, 2016
Boston, MA – June 9, 2016
Any News on a bug fix for the prediction corrected VPC?
In my hands (serveral models tried, covariates, dosages etc.) the prediction corrected predictions are also wrong (the variation always get bigger than before).
I beleive this pvVPC bug, (QC number is 15573), is fixed in PHX 7.0, which will be released soon (this quarter). it would be interesting if you could expand on your statemnet;
“the prediction corrected predictions are also wrong (the variation always get bigger than before).”
Do you have a project perhaps that illustrates what you’re seeing that I can pass to development.