Bioequivalence and Bioavailability Forum

Dear Ohlbe!

❝ ... when some FDA guys said they wanted long-term stability of the analyte in the samples to be re-demonstrated if a bioanalytical method was transferred from one site to another

Do you know the waiter Manuel in the TV-series ‘Fawlty Towers’?
He would just heartily reply: ¿Qué?

❝ Actually the summary slides of the 2006 Crystal City meeting used the word "reproducibility" but my understanding is that FDA is concerned with both precision and accuracy. They apparently had some studies with huge (not just 20-30 %) differences between initial and repeated concentrations (e.g. page 2 of the December 2004 Letter to MDS Canada. The letter mostly refers to contaminations, but apparently they also had other concerns (there or elsewhere)).

OK, IUPAC defines reproducibility as:

The closeness of agreement between independent results obtained with the same method on identical test material but under different conditions (different operators, different apparatus, different laboratories and/or after different intervals of time). The measure of reproducibility is the standard deviation qualified with the term ‘reproducibility’ as reproducibility standard deviation. In some contexts reproducibility may be defined as the value below which the absolute difference between two single test results on identical material obtained under the above conditions, may be expected to lie with a specified probability. Note that a complete statement of reproducibility requires specification of the experimental conditions which differ.

So I’m still not getting FDA’s point.
All – serious – labs have an SOP for re-analysis (including acceptance criteria) in place. What is FDA suggesting?
In order to set any internal rules for ‘incurred samples re-analysis reproducibility’ we would have to use some of these samples.
But when? Of course after the first study - otherwise we would not have any samples…
The maximum time interval would be the validated long-term stability. Which sponsor would be willing to voluntarily give away his/her biosamples?
Even if we were able to establish some kind of acceptance criteria, IMHO it would lead to a lot of 'not reportable' results.

Let’s look at an artificial example:
1 (unknown) sample analysed in triplicates in three batches. In order to keep it simple, let’s apply the common rules for valid intermediate QCs (±15% accuracy, 15% precision, ). Of course we don’t know the ‘true’ concentration; I just set it to 15/19.5/30 (‘true’, ‘true’ +30%, ‘true’ +100%). Don’t worry about the within-batch CVs; the given numbers were simply the first valid ones jumping out of my random generator…
            +------+------+------+
            |  # 1 |  # 2 |  # 3 |
+-----------+------+------+------+ 
|conc. calc.| 15.5 | 17.1 | 30.2 |
|           | 17.2 | 21.9 | 25.6 |
|           | 12.9 | 17.8 | 28.4 |
+-----------+------+------+------+
|mean       | 15.2 | 18.9 | 28.1 |
|SD         |  2.17|  2.59|  2.32|
|CV         | 14%  | 14%  |  8%  |
+-----------+------+------+------+
Now let’s look at the intra-batch variability (reproducibility according to IUPAC); we can’t talk about accuracy!
# 1 <-> # 2: CV 15%
# 1 <-> # 3: CV 42%
What limit should we suggest?

Many people’s SOPs right now (if there is insufficient volume for three analyses) end up with a ‘not reportable’ value if the deviation of second to first analysis is >30%.
We re-analyse values not for fun, but because we assume something has gone wrong (bad chromatography, equipment failure, …).

Whatever procedure one is using, looking only at single data points (at least in any type of PK study) would throw away all a priori information on the time course of concentrations.
Consequent concentrations are not independent! IMHO any re-analysis ignoring this fact is just some kind of bureaucracy, but definitely not science.
I always suggest to include two adjacent values, or – if the suspect values is the first/last one in a profile - two succeeding/preceding ones.

Another example. The second value was suspected for problems with internal standard addition, but only enough sample for one re-analysis was available.
+----+-------+-------+
|time|  # 1  |  # 2  |
+----+-------+------ +
|  1 |  78.7 |  82.4 |
|  2 | 341.2 |  85.5 |
|  4 |  51.3 |  53.7 |
+----+-------+-------+
If we only repeat a single value we would end up with no result; if we analyse three consequent samples, we have strong evidence for replacing the first result with 85.5.
OK, such an SOP needs a little bit of statistical background (actually accuracy + precision from the analytical method, total biological variability and some kind of ‘detection threshold’ are incorporated); but I think it’s worth it!