RE: Simulation vs. actual data

From: "Perez Ruixo, Juan Jose [PRDBE]" JPEREZRU@PRDBE.jnj.com Subject: RE: [NMusers] Simulation vs. actual data Date: Tue, July 12, 2005 2:52 pm Dears, Interesting discussion. For the tolerance interval, I think it would be more accurate to sample (THETAs, OMEGAs and SIGMAs) from the non-parametric bootstrap replicates. For each replicate, NSUBPROBLEMS=100 should be implemented. Then, the 90% tolerance interval can be calculated from the empirical distribution of all individual predicted responses (with residual error) at each time. That's how I implemented it previously for a simulation exercise (see reference below). The tolerance interval constructed as described above should contain 90% of the observed responses at each time. In my experience, the prediction interval as Nick described could be too narrow to include the 90% of the observed response at each time, specially if uncertainty is high (relative to PPV and RUV). Reference: Jolling K, Perez-Ruixo JJ, Hemeryck A, Vermeulen V, Greway T. Mixed-effects Modelling of the Interspecies Pharmacokinetic Scaling of Pegylated Human Erythropoietin. Eur J Pharm Sci 2005; 24: 465-475. Regards, Juanjo.