| Author | Message | ||
     Sanette van der Merwe (Sanette) |
Hello all, I am working at a health food company manufacturing multivitamins, food, beverages, cosmetics and food supplements. I have validated about 35 different raw materials for identification by NIRS and like to start quantification on some of the products. I have a step-by-step protocol, which I received from the vendor to determine the lab error by using blind duplicates ( 11 samples covering the quantification range). My main concern is the time and resources needed to determine the laboratory error since I have numerous products using different methods. Is there another method of determining the laboratory error for our reference methods? We have validation data on some of the products, which we can use, but not for pharmacopial methods. I would also like to refer to a posting by Howard on 16 October 2001 under the topic 'Validation Methods', where he states that if the lab error is unknown, there are ways to estimate it from the calibration results. My chemometrics knowledge is limited and would appreciate some guidance on this topic. Kind regards Sanette | ||
| hlmark |
Sanette - the things I had in mind were more difficult and time-consuming than sending blind duplicates to the laboratory for reference analysis. They are useful when you're "desperate" but not recommended when available time and effort is the limiting factor Howard \o/ /_\ | ||
| Tony Davies (Td) |
Hello Sanette, I am concerned that you think that eleven analyses are going to be time consuming. In order to set up a single NIR application you should be thinking in terms of 100 samples; they will all require reference analysis, the extra to gain the vital information about the lab error is only going to cost another 10 percent of effort! You take one of your samples and send 10 samples of it to the laboratory and "hide" them (with different sample numbers) with the other 99 samples. The sample you choose should be typical and about average. You get one very well analysed sample (average of ten analyses) and the lab error (the std of the ten samples). It does take time and effort to set up NIR analyses but if you do it properly you get low cost and reliable analytical information. My advice is that you do not tell the analysts what you are doing. My experience is that it is more likely to upset than please them; even when you tell them how GOOD they are! Best wishes, Tony | ||
| Sanette van der Merwe (Sanette) |
Dear Tony and Howard, Thanks for your comments. They are much appreciated. Kind regards Sanette |