Writing the OQ’s and PQ’s is okay but it is challenging to understand what sample plan needs to be?? It’s a Class II medical device being packaged on a new rotary blister package machine. A typical order will be a 1000 blister packages which will take1 hour to pack/seal in operations. I’ll be performing peeling testing and dye penetration for the OQ and PQ.
Main Question: How many samples per OQ or PQ?? Is there a model/tree to help figure the sample size out?
The steps involved in developing a sampling plan are:
a. identify the parameters to be measured, whether they are attributes or variables, the range of possible values, and the required resolution
b. design a sampling scheme that details how and when samples will be taken
c. select sample sizes
d. design data storage formats
e. assign roles and responsibilities
I’ve never felt completely comfortable with the answers I gotten when asking this question. When working with a more senior validation person than I, but there were a few strategies we came up with.
1- For IQ/OQ Use AQL (Assurance Quality Level) to determine the run size, the sample size, and the acceptable fail rate.
2 - For PQ, if you want to do Endurance runs, then I would do at least the same size as the actual commercial run. Do you know what your sample size will be during commerical runs? For PQ, you should increase the number of samples.
For this run (1000 packages). I would do a full run (1000 packages) for OQ and take samples according to the AQL tables, or another statistical model (which you should probably be able to find online. For the peel test will you collect data of a number where the peel sticks or will it just be pass/fail? This determines your sample size on the AQL tables.
This is always a tough question and people often mistakenly just default to an AQL. Technically the AQL isn’t correct. The AQL is the acceptable quality limit. The purpose of the AQL is to determine that the lot in question meets the acceptable quality limit or not. The question you need to answer is what is the confidence and reliability of your process to achieve the desired AQL for ALL lots? For the OQ the sample number is not as critical, as we are just showing the system works (I would just follow the same rationale as the PQ, but with lower values since the final test will really be the PQ). For the PQ it depends on whether or not there is another validation step or not. If there is one product a PQ may be the final validation step, alternatively if there are multiple products each product may have a process validation step specific to that product (I’ve seen it done both ways prefer the latter). Ideally you use a risk analysis to determine the confidence and reliability you want to achieve. Simply you could go with “high”, “moderate” and “low” risk levels. If it’s high then maybe you want 99% confidence with 95% reliability (99/95) for moderate 95/95 and for low 90/95). Then the simplest answer is to use a confidence/reliability table to determine the number of samples needed. That being said, it’s better to base the number off of the capability of the process and how close it is to the specification limit. For example, if you’re final ultimate test is peel force, and the system produces product that is 100+/-10 and your spec is 50… then you don’t need a lot of samples to prove you can achieve any desired AQL. But if your spec is 90 you’re going to need to test a lot to ensure you have confidence in achieving the desired AQL. All that said… a lot of places get away with just choosing an AQL and testing to it.
BTW: great article on process validation and choosing the number of runs and references to risk-based sample size methods. The Tolerance interval method is what I was referring to above when I was referring to the capability of the process and how close it is to the specification limit. See the related articles, just above the methods.