Bioequivalence and Bioavailability Forum

Hi ElMaestro,

wow! THX for enlightening me. Halving all values given in this post would add a new level of absurdity to the story.

Hi ElMaestro,

What do you think about my interpretation here below described?

I do not think that EMA provides broader limits for ISR than those widely knew and shared.
I believe that what is written in the BIOANALYTICAL GUIDELINE ON METHOD VALIDATION of 21 July 2011, it's just bad written.

If we start from the DRAFT 19 November 2009, at paragraph 6. Incurred Samples Reanalysis, at line 485 is reported: "The difference between the two values ​​obtained should be within 20% of the mean (30% for ligand-binding assays) for at least 67% of the repeats."

Continuing in the document "Overview of comments received on 'Guideline on validation of bioanalytical methods' (EMEA/CHMP/EWP/192217/2009)", the concept widely reported in the WPs is mentioned and i would highlight a couple of points:

Line no. 467-477
Stakeholder no. 24
Comment and rationale;proposed changes: Chapter should be adapted according to EBF recommendations. as described in “Incurred Sample Reproducibility: Views and Recommendations by the European Bioanalysis Forum, P. Timmerman et al” (PDF attached)
Outcome: It is considered that the revised paragraph covers sufficiently the recommendations of the EBF.
Regarding the criteria, 20% difference from the mean would be in line with the White paper (Fast et all., AAPS Journal, DOI 10.128/s12248-009-9100-9)

Line no. 467-486
Stakeholder no. 39
Comment and rationale;proposed changes: … And finally, the requirement for the difference of two measurements to be within 20% is more strict than that for QC samples.
Outcome:There was an international consensus on this criteria (see White Paper).

Finally, in the document "Appendix IV of the Guideline on the Investigation on Bioequivalence (CPMP/EWP/QWP/1401/98 Rev.1): Presentation of Biopharmaceutical and Bioanalytical Data in Module 2.7.1" at the point "Table 4.3 Sample analysis of <Study ID> ", under the heading Incurrent sample reanalysis, it is repeated the same acceptance criterion "Percentage of samples where the difference between the two values ​​was less than 20% of the mean for chromatographic assays or less than 30% for ligand binding assays. "

In conclusion, I believe that the ISR acceptance criteria for EMA is that well known and internationally recognized:

Variability (%)= (Repeat – Original) × 100/Mean

and that what is reported in the current GL is written in a not clear way and misleading.

What is your opinion on this?

Thanks at all and best regards

Roberto

Dear Ohlbe!

❝ My understanding is what is described in the Crystal City white papers:

❝

❝ (C₁-C₂)/((C₁+C₂)/2).

OK, that’s what we have done in the past. Which CCIII WP are you referring to? I can’t find anything in the Conference Report. Do you mean

Rocci ML, Devanarayan V, Haughey DB, and P Jardieu
Confirmatory Reanalysis of Incurred Bioanalytical Samples
AAPS J 9(3) Article 40 (2007)
online

which states in footnotes

Difference = (Repeat–original)/average expressed as a percentage

BTW, nice examples for the number of ISRs (astonishingly small).

What makes me wonder is that I had to use the ratio of the first to the second measurements in order to reproduce their results.
# Rocci et al. Example 1
first   <- c(478,107,826,108,248,696,141,194,548,676,
             636,635,244,527,139,107,664,187,690,187)
second  <- c(406,107,718,109,250,674,135,179,564,598,
             676,624,240,579,117,99.3,583,176,610,190)
Pct.diff<- 100*(second-first)/((first+second)/2)
Pos.d   <- length(Pct.diff[Pct.diff>0])
Zero.d  <- length(Pct.diff[Pct.diff==0])
Neg.d   <- length(Pct.diff[Pct.diff<0])
diff.r  <- log10(first) - log10(second)  # Step 1
Ratio   <- 10^(diff.r)
mean    <- mean(diff.r)
sd      <- sd(diff.r)
n       <- length(diff.r)
geo.mean<- sqrt(first*second)            # Step 2
MR      <- 10^(mean)                     # Step 3
sig     <- c(-1, +1)
RL      <- 10^(mean+sig*2*sd/sqrt(n))
LA      <- 10^(mean+sig*sd)              # Step 4
Comp    <- length(Ratio[Ratio>=LA[1] & Ratio<=LA[2]])
Comp.pct<- 100*Comp/n
plot(geo.mean, Ratio, las=1, log="xy", pch=16)
abline(h=MR, lwd=2, col="blue")
abline(h=RL, lwd=2, col="darkgreen")
abline(h=LA, lwd=2, col="red")
cat(" MR:", round(MR,2), "\n", "RL:", round(RL,2), "\n",
  "LA:", round(LA,2), "\n", "Ratios with LA:", Comp.pct, "%\n")

---

Edit: A more comprehensive code over there.