| Author | Message | ||
     Dave Ryan (Ryan) |
Need for Comparative data? I manage a network of nir instruments and have always incorporated an ongoing plan to gather and review comparative data using this to update the models as necessary. Recently I have gotten involved with developing PLS models for FTIR/ATR instrumentation. I am interested in opinions on the need for ongoing comparative data for the FTIR. In some cases it can be shown that FTIR calibrations are less matrix dependent than its NIR counterparts, but in some cases the reference methods are emperical or matrix dependent which I believe makes the calibration derived from this reference data equally matrix dependent. In other cases the FTIR calibration can use "pure" standards and develop a less matrix dependent calibration. I believe that our customers, government regulators, etc.. will not accept by faith that our FTIR calibrations are completely matrix dependent and never require some level of ongoing compararative data back to the reference or official method. I do acknowledge that spectroscopic methods that are official should not require comparative data back to a secondary reference method. Let me preface these comments in the fact that all our PLS methods are thoroughly validated initially and all run control samples routinely to define stability of the instrument. These control samples however do not validate the applicability of the model for the samples being run today from the process. Thanks ahead for any opinions. Dave Ryan | ||
| hlmark |
Dave - I tend to suspect that anything other than transmission through a clear liquid will probably require some sort of routine check procedure. The physics of ATR, for example, tells us that the phenomenon is completely dependent on the refractive index of the both the sample and the crystal; refractive indices are well-known to be temperature dependent, and the degree of penetration of the light into the sample will thus be a function of temperature from this cause alone. This would be equivalent to a pathlength change (a la "repack" variation?). If we thought about it enough we could probably come up with some more mechanisms that might cause trouble. As far as the customers and regulators, you'd have to deal with those on a case-by-case basis, I guess. Some might be satisfied and some not. I don't know how you'd find out: just asking is sure to elicit an answer of "of course you need to check". My gut feel is that a customer is more likely to be satisfied if the products turn out good, and less worried than regulators about how you proved it; regulators also want to have all their i's crossed, etc. Howard \o/ /_\ | ||
| Tony Davies (Td) |
David, Harald Martens still complains that the majority of professional statisticians regard PLS as an experimental technique! I think that I would always want to have some level of ongoing comparative analysis when using PLS to allow my own peaceful sleep � never mind the regulators! [My view is that we are still educating regulators; we need to tell them what should be required rather than ask them. As Howard said, we can predict what their answer will be.] I am assuming that you did NEED a PLS calibration. Best wishes, Tony PS How is the running? | ||
| hlmark |
Tony - considering some of the things that Harald has said about statisticians, I think he really has little cause for complaint about what THEY say! :-) But I still agree that Dave should institute some sort of check procedure. I think I'd start out with an equivalent to standard NIR practice, then reduce the rate of checking, if experience shows that's feasible. Here in the US we ARE educating the regulators. Now that they've seen the light (no pun intended) they realize that they need education, and have instituted formal policies to get themselves on the learning curve. Howard \o/ /_\ | ||
| David W. Hopkins (Dhopkins) |
Dave, I think we have to spend more time to educate ourselves that PLS may not be so much a black box. We need to show how the B-vector can be interpreted, and show how it indicates specificity for the analyte. I don't think we have to revert to the days of MLR with a few chosen wavelengths to demonstrate specificity. Please excuse me for responding so late, but apparently somehow my ISP starting blocking my messages from the discussion group. To all my friends, please change my address to [email protected]. I will keep the old address for "a while." While you are at it, the phone system is changing my Area Code, so my new phone is +1 269 964 9920. |