Visual inspection of the fit! [Study As­sess­ment]

posted by Helmut Homepage – Vienna, Austria, 2023-06-27 11:56 (327 d 01:18 ago) – Posting: # 23619
Views: 1,115

Hi NK,

❝ […] T½ is >1000 hrs […] After reaching Tmax at 2.00 hr the Concentration declined and reached BQL in 12 hours and after that few measurable concentrations was observed in 14 hr, 16 hr & 18 hr in elimination phase.

It doesn’t need rocket science to tell that such a t½ obtained by an automatic algorithm (e.g., \(\small{\textsf{max}\,\textsf{R}_\textsf{adj}^2\,}\)) is plain nonsense. Recall that the A&P of the bioanalytical method at the LLOQ is 20%. Furthermore, concentrations follow a lognormal distribution. Hence, it is not uncommon to face some values >LLOQ.

❝ How to handle this subject’s data? Outlier? Can the regulatory accept the based on Cmax and AUC0-t results?

When you explore ANDAs you will find many cases were the number of AUC0–∞ is less than the one of AUC0–t. Hence, it is acceptable in principle. In the recent ICH M13A draft guideline AUC0–∞ is not a primary parameter. It serves only to assess the validity of the study. Once the guideline will be final, the FDA will update theirs.

❝ Can we select the time point manually for Kel calculation instead of best fit method by the tool?

Of course, if specified in an SOP (see Divyen’ post above). Manual selection / exclusion of data is provided in any software I know. Certara states1

Using this methodology, Phoenix will almost always compute an estimate for Lambda Z. It is the user’s responsibility to evaluate the appropriateness of the estimated value.
(my emphasis)

Similarly in Lixoft’s PKanalix2 any many others.
For transparency I state the procedure also in the protocol (see this article). In 40+ years I never received a single (‼) deficiency letter in this respect.

Since you obviously didn’t have a suitable SOP in place, you can only present a sensitivity analysis with the AUC0–∞ of the one subject excluded. Supportive would also be an outlier test (Lund’s, Thompson’s τ, Cook’s distance, …) and/or graphical EDA (box plots, QQ-plots, histograms, scatter plots of studentized model residuals). See also this post about the method recommended by Health Canada. Though reanalysis based on PK grounds is not acceptable according to current bioanalytical guidances, I would do it anyway – not only the suspect values but the entire profile. Discuss it, and cross fingers.

I strongly recommend to update your SOP – mandating visual inspection of fits (see references 3, 5, 8, 20 in the article linked above) and allow to exclude data in the estimation of λz. Then you don’t have to exclude the nonsensical AUC0–∞ obtained by an automatic algorithm and keep the subject with a realistic AUC0–∞ in the analysis.
Maybe over the top, but I recommend also to perform the entire NCA in a blinded manner (see this thread). Avoids the impression of cherry-picking.

  1. Certara USA, Inc. Princeton, NJ. 7/9/20. Lambda Z or Slope Estimation settings. Online.
  2. LIXOFT, Antony, France. 2022. PKanalix Documentation. Check lambda_z. Online.

Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

Complete thread:

UA Flag
 Admin contact
23,029 posts in 4,834 threads, 1,644 registered users;
33 visitors (0 registered, 33 guests [including 7 identified bots]).
Forum time: 13:15 CEST (Europe/Vienna)

We must be careful not to confuse data with the abstractions
we use to analyze them.    William James

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz