Single testing for tablets

[AdrianF]

We routinely take two separate samples for each assay and related substances then report the mean result.

We are now considering moving to single testing.

Does anyone have experience in changing from double to single testing - how did you justify this change?

Currently to be accepted the two results have to be within 2% of the mean - is that a common requirement?

[zokitano]

Adrian,

I suppose that if you check the system suitability (if %RSD is less than 2% - or other acceptable value in the method, if the number of theoretical plates for the analyte is at least the minimum value given in the method, if the symmetry factor-tailing factor is less than the highest acceptable value in the method etc) then you will be sure that the result from the sample won't vary outside the accepted limits (which ensure you that the obtained results will be accurate), so you can make conclusions at the base of only one single sample injection.
I'm not discarding the way of injecting multiple aliquotes of the same sample. The more injections are made, the less errors will arise, and the results will be more accurate and precise.

Best regards

[tom jupille]

Go back and do "retrospective" control charting. What percentage of your runs did *not* meet the +/- 2% standard?

[Dan]

This has been a difficult issue in the pharmaceutical companies where I have worked. The preferece is split about 50-50 as to doing only one preparation or doing two preparations.

My preference is for having only one preparation. It is much more efficient and can be less problematic.

I will second Tom's suggestion. If you check the history of the results, you may find that you can get by with just one preparation. QA groups will (usually) go along with this if the data supports it.

One problem however with your reporting. You have two separate sample preparations. Therefore, you have two separate results to report. You cannot report the mean, the FDA and EMEA will not like that.

A big reason for this is the out-of-specification(OOS) issue. What happens if one sample results is within specification, the other result is OOS and the mean is within specification? If you report the mean only, then you are hiding the OOS.

Where I have been, we usually put the +/- 2% limit for the comparison between the two individual results not for the comparison of an individual result to the mean.

One last point for duplicate preparations and differing results. If you have two sample preparations and the results do not agree (to whatever limits you impose) then which is the "true" result? A former supervisor of mine had a great analogy: If you have two clocks and each shows a different time, then what time is it? His suggestion was that you either have three sample preparations or just stay with one sample preparation to avoid confusion and to be more efficient.

Regards,
Dan

[AdrianF]

Thanks for your answers so far, some comments:

1. An RSD of 2% is not the same as expecting most results to be +/-2% of the mean. In order for 95% of tests to be within 2% of mean you need an RSD of around 1.5%. I don't have enough statistical knowledge to prove this mathematically - I used the random number normal distribution number generator in EXCEL to produce 2 sets of a 1000 results with a mean of 100 and an SD of 1.5. By taking one set away from the other and then sorting them I found about 95% of the results were less than 4% different.

2. I have been studying historical data to justify this change. How good do you think the duplicates should be to justify change.

3. Is +/- 2% acceptance of duplicate results normal in the pharmaceutical industry - if so what is the rationale.

4. Dan- I should have said that we report the mean together with the individual results. If one is outside specification but the results agree we will do further tests.

5. I cannot agree with the clock analogy. If analysis is carried out correctly duplicate tests, will always give a truer answer than a single test. In order to get the best answer one would require testing to be done several times with different analysts using separate standards. However we are not looking for ultimate truth but just some assurance that we don't release a product which is out of specification.

[AdrianF]

Has anyone else got a view on this topic? If implemented it delivers a major cost saving. I have estimated that in the lab I work in we could save around £100,000, ($200,000) per year. Based on testing 400 batches a year. This is just a guestimate but significant savings are possible

[Bryan Evans]

I think one of the benefits for having 2 sample preps is for OOS results. Our QA group felt more comfortable signing off on OOS investigations when there were 2 results to support or reject the result.

One OOS and one "in-spec" result can be justification for re-dilution / re-injection (ex: OOS for assay, extra peak above reporting LOQ, ect.).
It can be helpful when trying to find the root cause of the error (sample dilution error, sample not fully dissolved, dirty glassware causing extra peak, ect.)

I'd recommend discussing it with your QA group, look over your past OOS results - and determine if you would be able to resolve those issues with 1 sample prep instead of 2 in a regulated environment.

[AdrianF]

I have estimated that in the lab I work in we could save around £100,000, ($200,000) per year. Based on testing 400 batches a year. This is just a guestimate but significant savings are possible

this should have been 400 batches a month!

[Chris]

Completely agree with Tom that it is best to plot the indivdual results and do some stats on the data. Consider using I-MR charts or X-bar- r charts.

Basically if you are going to make those sort of cost savings you should consider paying someone to design a study to demonstrate that the propsed change is acceptable. Look at how much variation is from the manufacturing process and how much is from the analytical process and what level of tolerance of you product specifcation these take up. e.g. if your upper control and lower control limits (e.g. 2sd) are close to your spec limits if you were to switch to single analysis then you would have to argue that your method was not fit for purpose.

Consider are there any times that taking a single result would have impacted on batch release. If you are going to do the stats yourself there are some great packages out there (we use minitab) and with a little training these can give you some great information your process controls.

For products with a specifcation of +-5% of mean we would expect to obatin an RSD of 1% or less if we to analyse the same product six times. If your method is capable, you have a decent OOS procedure (which it sounds like you do) and you use control charts to help identify out of trend results and constantly monitor your analytical and manufacturing processes then you should not have too much trouble.

[Dan]

Adrian,

Thanks for clarifying about your reporting of results (individual and mean). I shouldn't have assumed that you did it wrong (see my last post), I just made that conlcusion from the way you wrote it. Sorry for jumping to that conclusion.

For your cost savings estimate, did you include the cost of solvent disposal as well as the purchase cost?

That second preparation not only has solvent used by itself but also by the HPLC (or GC, etc.) for the run/analysis. Disposal costs for solvents are becoming increasingly a burden on us in the industry and some companies are looking to the analysts for ways to reduce solvent usage/disposal.

Regards,
Dan