Hi Pharsight Modellers, I use BQL flag to handle my below BQL data, but I found there were no residual plots after model completion. I read some previous threads and look into some of the example files in the threads, which all suggested the same. I wonder if there is a reason for it (it is indeed all zero?) or any other rationale behind it. Thanks, Best, Shu-Pei
For likelihood models you will not get residuals. One way to get the residuals is to refit the model fixing all your paramters to the value you got and reruning it witouht BLQ and with niteration =0. This way you will get the residuals for the esimated parameters you got for BQL. Observations that are BQL will not have a residual. Samer
HI Samer, Thanks for the response. Yes, it makes sense that BQL points will not have a residual. Just to be more clear, refit the model with the parameters without BQLs. which means discard all the BQLs, not leave as the LLOQ value as preparing the sheet for flag-BQL format. Thanks, Best, Shu-Pei
Yes just re-fit ( in reality we are not refitting since we are keeping)all parameters estimates from the run with BQL) we are just re evaluating the model without the BLQ observation so we can get the residuals for the non blq Cobs.
In the Phoenix 1.4 release, the Residuals table and output such as WRES and CWRES will be given for the observations that are not BQL. Regards, Linda Hughes
Hi everyone.
I got my residuals as explained above but I’m still confused about why residuals are not computed with the censor option. Residuals are computed by DV - PRED and or DV- IPRED and since we have DV vs PRED and DV vs IPRED in the plot output for a model with the censor option, why not computing residuals and display plots ? Am I missing something ?
Another question about this censor option, if I understood well, the likelihood for all the data is computed and for the BQL data, it is taken as the likelihood for these data to really be BQL. The likelihood of these data is maximized w.r.t model parameters but for all the data or only the one above BQL ? And what about the likelihood for the BQL data to really be BQL, does this information is used in the estimation step later ?
Finally, this method used in NLME for censoring the data corresponds to the M3 or M4 method detailed in the Beal 2001 paper “Ways to fit a Pk model with some data Below the quantificaiton limit”? I would say M4 because it’s more realistic but so far it’s only a personal opinion, I didn’t found any clue during my investigations.
Thanks,
Estelle
Estelle, which version of Phoenix are you using, is it fully patched? Becasue I think in 1.4 you should be seeing residuals with BLQ.
Is it possible you can attach your project or email to support if confidentiality is an issue?
Simon.