2018 Guidance on Stability and Accuracy and Precision, Second Blush
Ed O'Connor
Open for contract remote work 1) GxP QA auditor. 2) GxP method Development. 3) GxP Method, Process, Software, Instrument and Computer System Validation. 4) Data Integrity. 5) Statistics- Minitab (ROC, etc.)
First concern from the text: What do we compare results against?
2018 suggests several options, by omission. in the text, Autosampler and Bench-top comparisons for acceptance are not described. Extract stability is determined against freshly prepared calibrators (and Nominal?). Freeze thaw and Long term stability are compared against freshly prepared curves, and QCs.
There is no guidance here. It would seem by reading, we have options to compare against fresh QC, against curves (nominal?) or even against the original initial values (not stated directly).
The tables rectify this and describe comparison against nominal. There is no description of precision acceptance here. Is there a precision requirement for stability? Why not present this in the text as well? Why have the text so confusing? Align it with the tables.
A guidance document should not be ambiguous or nebulous. Consider the EMA 2012 Guidance, 4.1.9, third paragraph, last sentence:
“The QC samples are analysed against a calibration curve, obtained from freshly spiked calibration standards, and the obtained concentrations are compared to the nominal concentrations. The mean concentration at each level should be within ±15% (±20% for LBA) of the nominal concentration.”
and 7.1.1.11. second paragraph:
“The mean concentration at each level should be within 20% of the nominal concentration.”
Accuracy and precision runs are the other area. Supposedly analysts are to accept (pass?) validation assays for accuracy and precision only based on the performance of the curve, not the accuracy and precision QCs for which these runs are designed. Ostensibly, this is to reduce or eliminate the cherry picking of assays to give the best front. I have had personal experience with this where initially the Sponsor insisted on following this approach. After running six assays which met curve acceptance but which had three assays where at least two levels of QC “failed”, the sponsor was incredulous that we had continued to finish the six A&P(?) runs when the QC failed. We pointed out that we followed their direction using curve performance as the only parameter for acceptance. Silence.
In Table I, associated with A&P, the entry states for LBA that “A & P should be established with at least six independent A & P runs, five QC levels per run (LLOQ, L, M, H, ULOQ QC), and ≥ three replicates per QC level.” It is similar to the CC which has fewer runs and fewer levels of QC (I do not know why ULOQ is omitted since there is an upper limit to MS chromatography assays as well).
So, following this and using only curve acceptance, we can establish A&P at values that do not meet this guidance (15/20% A&P). Can someone please explain this to me? Or do we reject the runs (based on?) and restart, etc?
Two other points for discussion;
1) Why are CC assays exempt from Total Error, since there are both accuracy and precision measures there?
Why are LBA assays exempt from carryover? EMA at least addresses this for both why does the FDA ignore the possibly of robotic interfaces and the potential for carryover for LBA?