Hi Dean, I have just gone through the process of validating an automated filling and packing line for a class III medical device that also had to be compliant to EU GMP requirements quite recently. There are no explicit stipulations in any of the regulations but the expectation is that there is a sound scientific rationale to the approach and that the PQ takes into consideration output from your process risk assessment with regards critical parameters and potential failure rates the,OQ work and an understanding of how the process will be monitored during routine use.

For you example with the limited information above, a start point would be to assess what would be a statistically valid sample size, this could be achieved by using ISO 2859-1:1999 sampling tables. The number of samples can then be sampled from through a typical run and then the data could be used to generate process capability data to assess how well the filling process is within control

David

Thomas Zelikman Hi Dean

You want to show with statistical confidence that your process is stable in producing according to your acceptance criteria. If anyhow possible, your measure should be continuous - then you will have the minimal number of tests and the number of tests and batches will depend on how close you are to your acceptance criteria. If you have a binary variable, e.g. Passed / Not passed, you can use ISO 2859-2 for the PQ (not 2859-1) to determine your sample size.

Thomas

Dean Harper Thanks David and Thomas, this is really useful. We have been going round in circles with this for a while now.
Our machine outputs, I. E. Fill volume, torque , seal integrity where applicable will all be measured for every run. The fill volume is a 100% check during filling. The torque applied to caps is through a torque transducer which will be sent for annual calibration. Release torque will also be tested using a sample g plan for every run. The seal integrity will be checked to a sampling plan for every run using a pressure decay machine which is calibrated annually.
It sounds like we have food justification for not validating a full batch size for each format that is to go through the filling machine.

Thomas Zelikman Hi Dean
Seems that most of your ouputs are variable, except the seal integrity. Do I understand it correctly, that you plan to evaluate release torque in your PQ using a 3951 G plan? This will leave you with something like a LQ of 11%, which means you will have a risk of 10% of accepting a batch with 11% release torques outside specification limits.
Theses standards are meant for continuing series of batches in production. Not for a PQ validation with the purpose of showing that future batches will fall within acceptance criteria.
Instead you could calculate the prediction limits on your future release torques with confidence. This would validate that your future batches are within specification limits.

Dean Harper Hi Thomas,

excuse my poor typing on my last message, I wrote that on my phone.

To explain the machine properly.

We have 4 filling needles, so can fill 4 bottles simultaneously (of the same reagent). And we have 4 banks of needles so if required can fill 4 different reagents in a single pass on the machine. We fill into multi compartment reagent packs.

So our machine has 4 weigh cells built into it. These weigh cells will be calibrated every 6 months and checked every day prior to use using a set of calibrated weights. When the cart with the 4 reagent packs stops at the dispense station the 4 individual weigh cells rise and lift up the 4 individual inserts and 'tares' them. It then dispenses reagent 1 into compartment 1 of the 4 packs at the same time using the 4 needles in bank 1. The machine volumetrically dispenses to a lower target then tops up if required. Ones the correct weight is achieved (as measured by the weigh cells) the machine stops dispensing. In theory we should never have an overfill. When all 4 bottles are correctly filled the weigh cells re- tare and the same process is repeated for reagent 2 using the bank of needles 2. Once dispensing is complete the cart then moves to the capping or sealing station depending on the product.
If capping, the caps are applied to an application torque, using annually calibrated torque transducers. During a normal run (post validation) we will take a sample (typically beginning, middle and end) of these capped packs and run them through a torque tester to check release torque. The torque tester will be calibrated annually.
If sealed we will take a similar sample size and run through a pressure decay machine (again independently calibrated annually) and check for leakage.
Since each of the processing steps of the machine are 'independently' verified for every run (to a sampling plan) I am assuming the amount of validation we need to do during IQ/OQ/PQ should be fairly minimum. For example we don't have to run the machine for a shift worth of every format to test worst case. Of course we will also have data from the FAT and SAT to fall back on also.

I hope that maybe is a little clearer.

Thomas Zelikman Hi Dean
To my best knowledge, if you do not have 100% verification, you need validation. Exactly what might be considered "enough" depends on various parameters, among others the drug type, process and packaging materials.

As the leakage test is not necessarily destructive, it could be performed as 100% testing, but I understand that is not the current setup. Because it is an attribute, it takes a large sample size to prove a small defect rate with confidence. To give you a qualified advice on exactly what to do, would require deeper process and product knowledge and involve colleagues of mine with in-depth GMP knowledge to sort out the details around the exact requirements for the validation.

The release torque - although the test is destructive - is variable and can probably be validated with just a few tests. How many, depends mainly on the specification limits and the process capability.

Here is an extract from a GHTF document with an example that resembles torque testing to some degree:

Heat sealing processes, as described in this example, use equipment to seal plastic pouches which perform as sterility barriers for disposable medical devices. Seal integrity is crucial for maintenance of sterility. Testing of seal integrity is usually destructive testing, and the process therefor is a special process which requires validation.

You may already know the document. If not you can download it from here: http://www.imdrf.org/docs/ghtf/final/sg3/technical-docs/ghtf-sg3-n99-10-2004-qms-process-guidance-04010.pdf

Regarding sample sizes: It is not difficult to validate with statistical confidence, so this would be my suggestion instead of using ISO sampling plans. You do not even necessarily need a fixed sample size, as it often depends on the actual process output and the specification limits.

E.g. when you see a small process variation far from the specification limits, a few samples are enough to prove stability and vice versa. This can even be put into a simple lookup tables for easy lookup during production. Another advantage is, that you can create optimal sampling plans for the situation instead of using a tables from the 60's that may either not be adequate or ideal for the situation.

If you are interested, I can send you some diagrams on how to decide on release strategies depending on situation and ensuring statistical confidence - in case, just provide me your email.