Details of Operational Qualification (OQ) test design for plastic extrusion processes, effect of material property noise, and GHTF/SG3/N99-10

[Stoic]

I've been trying to identify the industry standard approach to OQ of plastic extrusion processes, and am instead finding a wide range of opinions and practices among medical device OEMs and extrusion suppliers. It's a complicated process with a high number of inputs and a significant noise factor in the resin material property (melt flow index), so some confusion is to be expected. Some individuals recognize the importance of extrusion process validation, while others are strongly disdainful of its engineering or business value or even feasibility. Others grudgingly acknowledge the reality of interpretation and enforcement: extrusion is similar enough to injection molding that most auditors regard validation as mandatory under 21 CFR 820 / MDSAP. However, the implementation can differ significantly from an OQ of a simple process with just one set of time, temperature, and pressure variables.

Practitioners I've encountered generally understand that they have to identify the most sensitive inputs as CTQ and focus the OQ on those. Some use final (as close to the die as possible) melt temperature and haul-off (puller) speed, others may also use screw speed. Some use final melt pressure instead of temperature to compensate for variation in resin MFI. Some of these answers may be better than others, but perhaps the best answer may be different for various combinations of equipment and material.

The compromises begin to creep in when interpreting OQ limits relative to process control limits. Even simpler processes are not immune to misinterpretation. I've encountered scenarios for a variety of process types in which:

• OQ is regarded as unnecessary because the process has just one "set point" for each CTQ control variable
• Calibration tolerances of the gauges are not taken into account when determining OQ limits relative to process control limits

In the most extreme cases of extrusion PV, OQ limits are regarded as just starting windows within which the operators will hunt for the combination of input variables that will produce conforming product during a given run. The common defense is that MFI variation (and possibly other noise sources) make it impossible to set OQ limits that guarantee conforming output. This seems to me to conflict with the very definition of OQ as presented in GHTF/SG3/N99-10. While the exercise appears on the surface to check the box for the OQ requirement, it actually represents an implicit statement that the process cannot feasibly be validated and must be actively "tuned" around the noise during every production run, or non-conformances will result.

A less extreme but still debatable approach involves OQ test runs that only vary one CTQ control parameter at a time. While it's a significant improvement over the previous school of thought, it's still far less rigorous than running at the worst-case high/low input combinations. (Imagine proposing this method for a sterile pouch sealing process.) As before, the defense is typically that noise factors make it impossible to run worst-case combinations and get conforming output. I.e., the process can't really be validated at the level of control the equipment provides, the level of precision & accuracy required in the output, and the level of noise that must be compensated.

What have you observed in practice? What are the most capable and effective organizations doing? Are they treating extrusion OQ the same as other processes and running worst-case combinations of the CTQ controls? Are they making compromises relative to validation of other processes? If so, how do they rationalize them?

 

[Tidge]

I've never done a OQ for an extrusion process, but your thinking seems correct to me. One thing that you report does jump out at me (with alarums similar to your own) is:

In the most extreme cases of extrusion PV, OQ limits are regarded as just starting windows within which the operators will hunt for the combination of input variables that will produce conforming product during a given run. The common defense is that MFI variation (and possibly other noise sources) make it impossible to set OQ limits that guarantee conforming output. This seems to me to conflict with the very definition of OQ as presented in GHTF/SG3/N99-10. While the exercise appears on the surface to check the box for the OQ requirement, it actually represents an implicit statement that the process cannot feasibly be validated and must be actively "tuned" around the noise during every production run, or non-conformances will result.

Any process with operational limits that can't guarantee a conforming output is not one that I would say is "validated", except possibly with some extreme caveats. If my back was against the wall, and I had to do something like a variable challenge and yet I knew there was non-conforming output... I think the best option is to do enough OQ studies to be able to quantify the fraction of non-conforming product so that meaningful downstream controls could be implemented.

 

[Stoic]

Thanks for your response, Tidge. Agreed that it's better to perform OQ and accept less than 100% conforming process output with sufficient post-process inspection rather than just declare OQ impossible, but only if all of the identified CTQs "can be fully verified". I don't see how this works if you can't pass a destructive OQ test. But often extrusions have only dimensional CTQs, and this is the hardest concept for engineers to accept about the regulations: even when all the (identified) CTQs can be fully verified, validation is still required for thermoplastic processes. (Whether this should be the rule is debatable, but the enforcement pattern since the QSR preamble was published is clear, and the language for the U.S. note about injection molding in the MDSAP documents makes it even clearer.)

In the case of only dimensional CTQs, let's assume that OQ performed at worst-case combinations of critical process control variables +/- calibration tolerance shows a capability that's less than desirable but not catastrophic. Should it not be an acceptable course of action to impose a higher sampling criteria on each lot, possibly up to 100%? You've met the requirement to challenge the process's sensitivity to inputs, your best efforts have shown that there is reason to be concerned, and you adjust your inspection requirements to compensate. As long as your company's quality system incorporates this option, the auditor should be OK with it.