Re: Simulation vs. actual data

From: "Liping Zhang" ZHANG_LIPING@lilly.com Subject: Re: [NMusers] Simulation vs. actual data Date: Tue, June 14, 2005 2:57 pm I totally support your thoughts. If you want, other than the mean, you can also get the confidence band (or like you said +/-SD) around the simulated curve. Depend on how precise and what the simulation will be used for, you can elaborate the simulation": you can simulate use the estimated typical structure parameter value, with inter-subj variation and random error, so a large number of subjects within one simulation will do. The simulation does not account the uncertainty in thetas and the band reflects inter-subject variability and the size of random error. You can also simulate different populations. For each simulation, first sample the parameters from their variance-covariance metrics, and then simulate using the sampled subjects. You then do somekind of summary statistic on the simulated results. It needs large number of simulations and large number of subjects within each simulation. It is a labor-intensive process but gives you a more realistic view of potential outcomes. Best regards, Liping Liping Zhang, PhD Global PK/PD Modeling and Trial Simulation ELi Lilly & Company Work: 317-277-8687 Fax: 317-433-6661 DC 0734 Email: zhang_liping@lilly.com