95% CI from NONMEM

Hello all, I need help calculating 95% confidence interval from NONMEM. Is mean±2SD accurate? or some paper mentioned use simulation 1000 times to get 95% CI, I just do not know how to do it. In addition, I have 4 different treatment groups, should I model them separately and get predicted concentration and PK parameters, then compared the statistically difference among these parameters, Or should I group them together and set a covariance, see if the covariance model has much lower OFV than base model. Thank you,

Hello Jocelyn, a simple way to calculate the 95 % confidence intervals [a,b] from 1000 simulations is to use the formula a=mean - 1.96*SD b=mean+1.96*SD (where 1.96 is the 97.5 % quantile of the standard normal distribution) The assumption you are using in this case is that the estimates are normally distributed. The mathematical theory about maximum likelihood estimates tells us that this is asymptotically true. But since NONMEM uses some approximations to find the maximum likelihood estimates this theory might not be completely right. Especially for random effects there is some criticism about using this simple symmetric confidence intervals centered at the mean (see for example the excellent book of Peter Bonate "Pharmacokinetic-Pharmacodynamic Modeling and Simulation"). A quite simple but more accurate and possible asymmetric interval can be obtained by taking the 2,5 % lower and 97,5 % upper quantile of your 1000 estimates(I think this is what is meant in the paper you are talking about). Your second question might better be answered by somebody else. Best regards, Sebastian Ueckert