Sample Size for Validation of Pharma product in Batches with only 1 measurement result

[raghu_1968]

Myself belongs to API (Active Pharmaceutical Industry) industry. Mostly ours is the R&D plant. We run only 3 batches as a process validation (after product development) and wait for the product approval from the agency. It may take long time for the approval.

During the recent agency inspection, inspector suggested to have some of the analysis (like RSD) to prove or ensure the consistency. We worked RSD with the past historical data and it comes around 30 % for few of the parameters. As mostly we take 3 batches it is very difficult to finetune. The constraints are very minimum batches and one batch will have only one result with respect to the parameter (e.g. one batch will have one Assay, Water content, Purity or impurity result, so totally 3 batches will have 3 results with respect to the parameter)

Any approach or tools for the above situation will help.

 

[Bev D]

So are you making an internal use assay for a pharma product? Or are you just making components for an assay? I assume you ar not selling this product but are makign it for internal use only? Please clarify.

Also was your ‘inspector’ also one of the agency reviewers for your product? If not it is unusual for an inspector to recommend an analysis fo r product validation (this lies with the stats group and the reviewers typically. The reviewers also typically defer to the stats guys. At least with the USDA and I assume the FDA is more rigorous not less.). Also not sure how it helps if you ‘do the math’ as opposed to the stats or reviewer groups. It’s a very simple straightforward calculation and most stats groups will redo your math anyway as a matter of fact standard practice.

Of course a 30% RSD (aka CV: SD/Mean *100%) would be concerning to anyone. Although RSD/CV is a very rough pseudo statistic no different than any percentage. It’s just a mathematical thing not a statistical thing.

It is helpful too if you could describe what data they want you to calculate a RSD/CV on? Is it for assay results (using this product as an assay for something else)? That is legitimate but a RSD/CV for water content for the product, etc. is not valid. Beyond that I would guess that they are asking for a CV/RSD from the 3 individual readings of the 3 batches…this is not really correct at all if there is a substantial difference between the ‘means’ (individual location measurement) of each batch. CV/RSD calculations are intended to be from a single source of the SD and the mean: in other words a homogenous distribution. Which you may not have and depending on what your product does it may or may not matter. So more clarity and some data will help us help you better…

For those who are not familiar: RSD is the relative standard deviation and CV is the coefficient of variance. They are the same thing.

 

[raghu_1968]

So are you making an internal use assay for a pharma product? Or are you just making components for an assay? I assume you ar not selling this product but are makign it for internal use only? Please clarify.

Also was your ‘inspector’ also one of the agency reviewers for your product? If not it is unusual for an inspector to recommend an analysis fo r product validation (this lies with the stats group and the reviewers typically. The reviewers also typically defer to the stats guys. At least with the USDA and I assume the FDA is more rigorous not less.). Also not sure how it helps if you ‘do the math’ as opposed to the stats or reviewer groups. It’s a very simple straightforward calculation and most stats groups will redo your math anyway as a matter of fact standard practice.

Of course a 30% RSD (aka CV: SD/Mean *100%) would be concerning to anyone. Although RSD/CV is a very rough pseudo statistic no different than any percentage. It’s just a mathematical thing not a statistical thing.

It is helpful too if you could describe what data they want you to calculate a RSD/CV on? Is it for assay results (using this product as an assay for something else)? That is legitimate but a RSD/CV for water content for the product, etc. is not valid. Beyond that I would guess that they are asking for a CV/RSD from the 3 individual readings of the 3 batches…this is not really correct at all if there is a substantial difference between the ‘means’ (individual location measurement) of each batch. CV/RSD calculations are intended to be from a single source of the SD and the mean: in other words a homogenous distribution. Which you may not have and depending on what your product does it may or may not matter. So more clarity and some data will help us help you better…

For those who are not familiar: RSD is the relative standard deviation and CV is the coefficient of variance. They are the same thing.

Yes. It is an internal consumption. The API will be formulated by our groups only. Inspector is not an agency reviewer.

 

[Bev D]

For those who don’t know API = Active Pharmaceutical Ingredient.

When materials are in batch form (homogenous within the batch), especially when small and in one ‘container’ there really isn’t any useful sampling beyond n=1 per batch. If the batch to batch variation is small (homogenous between batches) then you could ‘pool’ the data among the 3 batches to calculate a RSD/CV and it would be statistically ‘valid’. If batch to batch variation is large (non-homogenous batches) then any RSD/CV would be statistically invalid as the average (or in your case the single result per batch) would be an average of disparate values.

The assay value should consistent between batches OR you need to have a means of adjusting for this difference when producing the final product. Do you do this?

You could take multiple samples form each batch - at least for the Assay result - to understand the real within and between batch variability. Maybe n=3 to start…

What is your inspector concerned about? What do they think RSD/CV would do for you? This is the critical question, not anything about sample size at this point…beyond this I strongly suggest consulting with your reviewer and potentially your stats group. This will give you much better insight into what your agency group is looking for in validation and maybe why it takes so long…