Bioequivalence and Bioavailability Forum

Dear ElMaestro!

❝ I do have a strong opinion about this topic in general but I will spare you this time* as I promised earlier not to write lengthy posts.

Hey, you made a promise nobody asked for - I love lengthy posts.

❝ ❝ An additional approach presented was to assay one sample on either side of the PK repeat.

❝

❝ Does this mean on either side of the anomalous value within the time series, or it is on either side in the sample tray?

The former; the reason for the repeat is triggered by a plausibility review of PK data after analytical data are released (SOP in place, ). The idea behind is not to reanalyse only the suspected value, but neighbouring concentrations as some kind of internal validation as well. Most SOPs trigger the reanalysis by looking at the time course, e.g. value <LLOQ embedded between "normal" values,… We have some statistics behind (based on the PK of the drug and the analytical variability) defining a threshold. For example if the ratio of the suspected value > than x-times or < 1/x-times the log/lin-interpolated value of the neighbours reanalysis is triggered. One example (without the actual statistics of the analytical method included):
C(t) = 100·exp(-0.5·t) + ε*
 0 100.08     limits
 1  59.37
 2  38.09
 3  21.11
 4  15.28
 6   7.13
 8   1.80   1.13-18.93
12   2.25

Let's assume the suspected value is at 8 hours. The [6-12] log/lin interpolated value at 8 hours is 4.85, first we apply limits of 4.85/3 and 4.85×3 [1.62 - 14.56] and then allow for variability of the method 1.62×0.7, 14.56×1.3, getting [1.13 - 18.93].

A value outside these limits at 8 hours will trigger reanalysis.

If the neighbouring values in the repeated analysis are within the expected variability (rules similar to ISR are applicable), we can be pretty sure that the method performs well. If the suspected value is confirmed, use the original (and cross your finger for the PK analysis), if not, something happened in the analytical (or clinical) performance we can't explain. That's the problem with allowing reanalysis only for analytical reasons: It assumes that all potential causes of errors are noticed and documented. In other words, sample mix-up in the clinical part (yes, it happens!) can never trigger a confirmatory reanalysis, because no documented errors occured in the analytical part.

If we see bioanalytics isolated from the fact that we have information about PK of the drug we step into the boots of analytical chemists working in the field of environmental analysis. They have a single sample and no clue how the analyte entered the matrix. But we should know better. If we have a value which is 10times higher than adjacent values it's simply physiologically not possible. Just my 2¢.

---

• µ = 0, σ² = 2: the example looks like a two-compartment model by chance (w=1/y² iterative reweighting; 1-comp AIC 4.65, 2-comp AIC -1.78). The red line is the elimination from NCA and the blue ones from modeling. So far about simulations.