PopPK Dataset Problem

Phoenix
1

Good morning,

I have been performing non-compartmental analyses (NCA) with WinNonlin for approximately two years. More recently, I have started using the NLME module to conduct population pharmacokinetic analyses. I am currently enrolled in the Certara University course, and I have no difficulties completing the exercises with the course datasets. However, I have encountered a problem when working with my own dataset.

The dataset consists of an Excel file containing plasma concentrations of a drug from 60 patients. The dosing regimen was as follows: the drug was administered orally three times daily at 8:00 a.m., 4:00 p.m. (8 hours after the first dose), and 10:00 p.m. (14 hours after the first dose of the day).

Drug concentrations were determined using a sparse sampling design. Samples were collected pre-dose (baseline), one additional sample between 1 and 5 hours post-dose (at 1, 2, 3, 4, or 5 hours), and another additional sample between 6 and 8 hours post-dose (at 6, 7, or 8 hours). PK sampling times were randomly assigned for each patient.

After the initial sampling, treatment was continued for two weeks and the same sampling scheme was repeated. Subsequently, a four-week washout period was implemented. Treatment was then reinitiated and samples were again collected at the same time points. Finally, treatment was maintained for another two weeks, followed by a final round of sampling at the same time points.

The issue I am facing is that, when I attempt to model these data, none of the structural models provide an adequate fit to the observed concentrations. The results show extremely large errors, high shrinkage for all parameters, and very poor CWRES diagnostic plots.

My question is whether this could be due to the limited number of samples per individual (i.e., the sparse design), or whether I may be making an error in the way I am structuring or importing the dataset.

When preparing the dataset for modeling, I assumed that I could not use the ADDL approach because the dosing intervals are not constant (doses are administered 8 hours and then 14 hours after the first daily dose). Therefore, I manually entered all dosing times from treatment initiation until just before the washout period; then I implemented a reset and re-entered the dosing records in the same way after treatment reinitiation.

Could you please advise whether I might be making a mistake in the dataset structure or model setup?

Thank you very much in advance.

2

Arturo - this should be feasible, the dataset sounds no more sparce than e.g the pheno set that you’ve probably looked at in the examples guide, however the struggle to fit a good model may reflect true ‘noise’ or variances in the data.

I would suggest you share your project file either here or by sending to support@certara.com and we can take a look at what you have done. incidentally you should be able to use ADDL in your case, i’ll have to think more closely if reset is appropriate in this case.
Simon.

3

Thank you very much for your prompt response.

I was unable to attach the project files directly, so I have uploaded the WinNonlin project file corresponding to the population modeling analysis to a shared Drive folder:

WNL Google Drive Project Files Folder

The dataset was originally constructed under the assumption that the drug was administered every 8 hours, as I was initially informed that the treatment was given three times daily. Accordingly, I implemented ADDL with an 8-hour dosing interval. However, after observing inconsistencies in the concentration data and seeking clarification, I was informed that the doses were actually administered at 8:00 a.m., 4:00 p.m. (8 hours after the first dose), and 10:00 p.m. (14 hours after the first dose of the day).

Despite this, there were no concentration measurements collected after the 14-hour post-dose time point. Therefore, I initially assumed that this discrepancy in dosing intervals would not substantially affect the model. Nevertheless, as mentioned previously, no structural model provides an adequate fit to the observed concentrations.

This leads to my main question: could the issue be related to the way the dataset was constructed, specifically that it does not accurately reflect the true dosing times (even though the total number of administered doses is correct)? If so, would it be possible to modify the dataset in a way that correctly accounts for the unequal dosing intervals without having to manually create a row for each individual administration?

Finally, if the “reset” function cannot be appropriately applied: given that the treatment scheme in periods 1 and 2 is identical to that in periods 3 and 4 (after the washout), would it be methodologically acceptable to treat periods 3 and 4 as a second cohort of participants receiving the same regimen?

Thank you very much in advance for your guidance.