Minimal number of runs for validation

[Fabiano]

I was discussing with other analysts, what would be a minimal number for runs in a method validation and if we could do this in overnight runs in a single autosample tray of 60 vials.
The intention is perform a total validation from a modified existent method (USP), sample is neat API.
New method would change mobile-phase and column type (ID, length and particle size)


1-Calibration curve 5 points-duplicate each---------10
2-Limit of detection and quantification dilutions---------5
3-Upper limit of detection (also for Range)125,150%- duplicate---------4
4-Recovery (accuracy) in 2 concentrations 50, 100% of theory---------5
5-Precision for sample %RSD 3 conc/3 replicates each---------9
6-Ruggedness- other analyst performing calibration
(5 points, recovery 2 spikes, reproducibility 6 replicates)---------14
7-Robustness- %B variation of + and - 5% , flow rate
variation + and – 20%---------4

Number of runs---------51


Of course, some runs are made to extrapolate a possible detection limit before do the dilutions.
I really don’t think that step 6 is necessary, since we have results of standard method that we can do a T test for assure equivalence but people insisted.
Would you add/remove something?

Fabiano

[ntruong]

Hi Fabiano,
I would follow the ICH guideline Q7A for method validation. In term of injection, I would do one injection of each solution except duplicate injections for LOD and LOQ.

[Alfred88]

Dear all:
I doubt that anyone can finish a 'full validation' overnight. Do you have a column switch? What is the cycle time for one injection?

We normally break each validation into several 'sequences', so that if one sequence fails (system suit failure may happen), we can repeat that sequence.
We have not attempted to minimize the number of injections. So, a validation would last for more than a week.
'Fabiano', you missed the check of injection precision (10 repeated injections, according to Guidance for validation of chromatographic methods). And you missed the system suit check of system (required before running anything).

Alfred.

[Fabiano]

Alfred,

I am in pharma so I was following the Q2B, for precision the 9 injections (3 conc/3 replicates) is fine.

Yes, I think that I can do a total Validation in overnight runs. I know people that don't modify methods because think that validation would took to long to complete, I believe that if my system is working fine - low RSD, low Residuals in linear regression and you are experienced in lab - you can get data to support it with "few" runs.

Why Do I need a column switch?
Total cycle will be about 5 minutes, with autosample. (short column with 3um particle)
System suitability was evaluated during method development: Rs, Tailing, LOD, LOQ, flow rate, wavelenght maximum, etc.

[DR]

We interpret ruggedness requirements to mean "different analysts on different instrument, different days" (we try to have 3 people run the same pooled sample on 3 instruments on 2-3 different days - each makes his/her own standards, phases, uses different column lot etc.). No overnighters for us.

[rc_12321]

WHAT ABOUT FOLLOWING PARAMETERS?

1) FILTER VALIDATION
2) TEMP .variation and wave length variation in robustness
3) DEGRADATION STUDIES Like acid ,base,peroxide and water etc
4) STABILITY STUDIES.......

There fore practically it is not possible to complete validation overnight.

[Fabiano]

RC:

My method is for bulk API, with available impurity. we don´t perform degradation studies fo that.

Filter validation? This is for a process not for an analysis. I just want to prove that my quantification is equivalent and VALID comparing to a USP monograph.

I think that most of people disagree that I can perform this during overnight but the number of runs is not so far from what I was planing.
With DR sugestion, it jumps for 79 runs.

Fabiano

[ym3142]

I know what I am running now:
1) system suit runs before any sample set;
2) accuracy: at 80%, 100% 120% with three preparation at each levels;
3) linearity at 60%, 80%, 100%, 120%, 140%;
4) precision and intermediate precision: two analysts, six sample preparation(this may be skip if only API);
5) three points, 48 hr stability
6) Specificity for both product and placebo at five conditions(on API only if test API);
7) robustness at temperture, flow, organic, buffer, pH
1) to 5) for related compounds as well as LOD LOQ.

I am alos interested in eliminating some tests. I am not sure which I can eliminate yet. I am feeling it is so much that we are wasting the time and energy.

[key4]

in my opinion, there are 2 things that you (Fabiano)skip and should alwas be there (as points 5 and 6 from ym3142 )
- stability for samples and std solutions
- and Specificity for both product and placebo (ICH)
as you dont have placebo you should stress your sample and show that there is no potential degradation peak that interfere your assay.

rc_12321: I believe that wavelength variations in robustness is too much, if you work under gmp system your instruments have a periodic review and usually preventive manteinance, most detectors check wavelength accuracy at every start.

K4

[Fabiano]

ym3142, I don't think that you have an excessive number of experiments, except for the 48 hour stability you should take 2-3 days in this routine (depending of run time), am I wrong?

K4, yes... I didn't take much attention for stability, except for ascorbic acid and C18:3 never worked with a very instable compound. I will add to my list.

Fabiano

[ym3142]

Though not every one agrees that my list is the minimum test for method validation there are many who agree.

some how I am not sure:
1) if I demonstrate the method is accurate from 140% to 60%, why do I need show the linearity for the range?
2) why do I need prepare three sample at the same concetration level to prove the accuracy? how about just one? say, I tested one sample at the target of 100% and I found it was 99.8%. Did this show that the method is accurate?
3) now we do six sample preparations for the precision test, (if) which demonstrate both the sample preparation and the instrument are precise, then why we need three accuracy sample preparations?
vice versa, if three accuracy sample preparations are showing both sample preparations and instrument are precise, why we do six precision samples?
so much for today. Thanks

[ntruong]

ym3142 wrote:

2) accuracy: at 80%, 100% 120% with three preparation at each levels

4) precision and intermediate precision: two analysts, six sample preparation(this may be skip if only API);

For 100% level, prepare six sample instead of three and use this data as repeatability for precision. The second anlyst will prepare six on his/her own.
Method validation applies for both API and finished drug. Why skip as in 4)?
5) three points, 48 hr stability

If your standard and assay are at 100% level, why bother with three points? one point at 100% level is sufficient.

6) Specificity for both product and placebo at five conditions(on API only if test API); [/quote] darn! I really hate this quote thingy. It may take me years to learn how to quote it right.
For specificity, one condition (100%) is good enough, provided all peaks are well resolved from each other. No twin peaks please
For related compounds, I think you should run it at a very low levels, says 0.1% to 1%. You could use the slope of this relared compound divide by the slope of your main compound to get the RRF, which will save you tons of time for not preparing these again for routine analysis.

if I demonstrate the method is accurate from 140% to 60%, why do I need show the linearity for the range?
2) why do I need prepare three sample at the same concetration level to prove the accuracy? how about just one? say, I tested one sample at the target of 100% and I found it was 99.8%. Did this show that the method is accurate?
3) now we do six sample preparations for the precision test, (if) which demonstrate both the sample preparation and the instrument are precise, then why we need three accuracy sample preparations?
vice versa, if three accuracy sample preparations are showing both sample preparations and instrument are precise, why we do six precision samples?

Accuracy does not tell you about whether the curve is linear, quadratic or...
FDA believe in the magic number 3. One is lucky, two is still lucky , three is proven to be real.
Got to go because my lady is hollering my name...

[Fabiano]

ntruong wrote:

FDA believe in the magic number 3. One is lucky, two is still lucky , three is proven to be real.

In other version:
One is accident, two is coincidence and three is validation.


Just wondering if outside Pharma industry, everybody agree that first list is enough....
I mean in food analysis, drugs in fluids, enviromental...etc

[Alfred88]

ntruong wrote: "I would follow the ICH guideline Q7A for method validation. In term of injection, I would do one injection of each solution."

Actually, ICH Q7A is for GMP, especialy about process/manufacturing, not about test method validation! One should follow ICH Q2A/B (now merged into one document), and USP <1225>. Also, it is good practice to do duplicate injections.

Fabiano wrote, "One is accident, two is coincidence and three is validation."

Actually, three instances (observations) cannot establish as proof of anything. I take one highschool induction example:
1. 1 is less than 100
2. 2 is less than 100
3. 3 is less than 100.
Induction: In all three cases examined, all numbers are less than 100. Therefore, all numbers are less than 100!!!
Question: What about 101?

Dear ym3142: Validation guideline is open for interpretation. You can use a wider range (e.g. 50-150% of target conc), or narrower. For linearity, you must have minimum 5 points, so you can use 50-75-100-125-150, or 50-80-100-120-150, or 50-65-100-130-150, or whatever.

Alfred

[bartjoosen]

ym3142 wrote:
For related compounds, I think you should run it at a very low levels, says 0.1% to 1%. You could use the slope of this relared compound divide by the slope of your main compound to get the RRF, which will save you tons of time for not preparing these again for routine analysis.

For related compounds, do you also perform linearity at 5 levels, 3 measurements (15 points at least), before dividing the slopes?
Or how do you measure RRF?

Bart