OOS: killing me

[superkidz]

1. Let say in LC analysis, 2 sample preps, 2 inj each. The test is valid if the RSD of 2 inj is NMT 2% and the RSD of 2 prep is NMT 2%. What would then be the difference if I set the criteria between 2 preps to NMT 2% difference and not the 2% RSD? And will the 2 prep, 2 inj each enough or do I need to make it 3 prep, 3 inj each? Average of the preparations will then be my reportable value.

2. If my reportable value is OOS, will the 3 analysts with 3 preps each be ok for retest? Each prep has 2 inj each. Is it ok if each prep is treated as 1 retest having a total 9 retests? Is it ok if I set the intra analyst to NMT 2% RSD for the injection replicates? And what would be better for the intra analyst preparation? I set the RSD to NMT 2% or the NMT 2% difference among the 3 preps?

The inter analyst requirement passed if the % difference among the 3 analysts is NMT 2%.
The original is an outlier if it is outside of calculated average ±3σ for the retest results.

3. Considering the above situation, will my retest be invalidated if I did not meet the criteria I have set above? What if there is no laboratory error or no obvious error in our part. What will happen in my retest data?

4. What If I have mix of fail and pass result in my retests? What will to do then? How will I report my result?

5. Is this OOS procedure can be used if I have aberrant result in my routine analysis?

6. We are testing lab and not manufacturer so we cannot investigate on the manufacturing part. Any references given are very much appreciated. Other OOS insights other than HPLC analysis are also welcome. Thanks

[unmgvar]

To begin, in statistics it is not correct to use RSD for only 2 injections. the base of your SOP is wrong by definition

now in 1. from what i get, you do 4 injections. 2 samples injected in duplicates. so the duplicates, show you the accuracy of the sampler, do you need that? you are probably doing accuracy for that by doing duplicates of the standards. and in the end you want to show the difference between the averages of the 2 samples you prepared. what is the allowed absolute or relative difference you want to allow between the averages of the samples? doing more then 2 will let you do more mistakes. go by the assumption that if you are making mistake then it will occur in one of the samples and not all of them, otherwise why not make 10 preparations?

2. but what is OOS? the difference between the results or all the results? each case is different and requires a different approach. if the problem is the difference then you have an accuracy problem and the solution is different then to have a OOS of the results itself. and maybe you have both problems. your SOP needs to solve each issue.

3. if the problem is in the difference between the samples, then you most probably have a problem of accurately doing the samples. if the difference is good but the results are one OOS, and one in specs, then, what result is correct?
if you cannot find a lab error then yes cancel the batch. but before that you will have done further testing most probably, and then the real problem is if now you get good results and you cannot find out why in this case the results came out good.
the pressure here is now to find the "nice way" out. and it is most probable that you yes have a lab error, but you cannot point it out.
review with QA, statistics. you cannot do RSD for duplicates. 2 injections of the same vial are only looking at the performance of the sampler. more preparations to find out the mistakes, can also give more room for errors, especially for hard methods.

keep it simpler