Running robustness tests

[AnalMD]

I am trying to determine the more preferred approach to testing the robustness parameters during a method validation. I want to limit the scope to instrument-controlled parameters (e.g. flow rate, temperature, wavelength, mobile phase composition).

Do I run each particular adjustment as a separate sample set (standards and samples run at the same condition) or can I run the changes in a single sample set, using the procedure-stated instrument parameters for my standard injections and samples as my baseline, then adjust the component parameters using different instrument methods, comparing the sample results with my baseline samples?

On a related topic, when a DOE is used to examine robustness, are the standards run at the procedure-stated parameters and then the DOE parameters are changed for the samples, with the samples quantitated from the "nominal" conditions?

[JGK]

Youden, W.J. & Steiner, E.H.: Statistical Manual of the AOAC, Association of Official Analytical Chemists, 1975. this seems to be a constantly used reference when discussing robustness testing. I believe it also contains a template for an experimental design.

[KM-USA]

Granted that modern instrumentation will not even let a sequence run unless all the parameters are in the "ready" state, we vary the conditions and document that system suitability under each set of conditions is maintained.

In the past, we injected standards and samples at each modified set of conditions and evaluated the results.

Which -if either - is correct? Apparently yet another FDA/cGMP gray area......

[shaun78]

I've always grouped to provide a worst case.

For example, for RP, more organic and a higher column temperature will result in earlier elution as well as reduced resolution. Run them both at once to produce a worst case and cut your work in half.

I would test system suitability from standards (must pass) and also a sample from which ever degradation condition (acid, base, heat, light, oxidation) produced the greatest degradation.

I've done this for NDA's, ANDA's, consent decrees, and validations that the FDA (or other regulatory agencies) have looked at. Never once was I questioned or cautioned about the approach used.

[aceto_81]

We use an full factorial experimantal design to evaluate robustness.
As we want to check how the method behaves when parameters are changed, we inject standards and samples on every condition.
We don't expect the users to change the flow between refs and samples, but want to account for instrument variations, so refs and samples are ran on the same parameters.

Afterwards we perform ANOVA on our results to see if some parameters are of significant influence.

HTH

Ace