Error modeling

From: Toufigh Gordi - tgordi@buffalo.edu Subject: [NMusers] Error modeling Date: 1/19/2004 9:57 AM Dear all, I have been working on a set of data gathered from malaria patients after treatment with a potent antimalarial compound. The model works fine, capturing the observed number of the parasites very well. Ever since collecting the data, I have been interested in the error terms associated with the observed number of parasites. We collected several blood samples from each patient for a period of time and a technician counted the parasites by looking at the stained samples under a light microscope. Although we let the same technician look at the samples from an individual patient, there is still a concern on how consistent she was. Moreover, we had three people count the parasites in the group and I let 2 of them count several samples separately. Not surprisingly, it turned out that they did not come up with identical number of parasites on each sample. However, there was no trend in their results. While one reported a slightly higher parasite count in one sample compared to her colleague, she would report somewhat lower count in the next sample. My question is if there is any rational way in taking into account the variability when different people evaluate an outcome in the same subject. Will this be covered by the error model ($ERROR) in the nonmem model file or is there any other way to handle such situation? Regards, Toufigh Gordi

From: "Bachman, William (MYD)" - bachmanw@iconus.com Subject: RE: [NMusers] Error modeling Date: 1/19/2004 10:26 AM In general, the measurement error in the observation is included in the residual error. If the error is systematic, e.g. consistant within a given "counter", you might be able to assign and identify a covariate like counter in the residual error model. If not, it will just contribute to the overall variance. _______________________________________________________