Re:

Dear Susan I am a beginner in NONMEM here is my input from one of my online resources. I appreciate if you others comment on this. Go through individual profiles one by one, perform NCA analysis in WINONLIN for initial estimates (for further information go through Chapter 10 of ENE ETTE BOOK) 1.Try different starting estimates. Sometimes NONMEM is very sensitive to the starting estimates of the parameters. For variability parameters, NONMEM tends to perform better if the initial estimates are too high, rather than too low. 2. Reduce the number of THETAs in the model. If a model is over-specified, then NONMEM can thrash trying to find the correct value for THETA. Reduce a 3 compartment model to a 2 compartment model. Eliminate covariate effects. The fewer parameters NONMEM tries to estimate, the more likely it is to successfully estimate the value of the parameters. 3. Reduce the number of variance parameters. Variance parameters can be particularly troublesome. Try having only 1 ETA and 1 EPS in the model. 4. Fix the values of ETA at 0. This is the naïve pooled data (NPD) approach, and it works surprisingly well. NONMEM sometimes has real problems estimating the structural model and the variance model at the same time. When this is the case, you may be able to get NONMEM to run by fixing every ETA value at 0 (e.g., $OMEGA 0 FIX 0 FIX 0 FIX ) and so on. With the ETA estimates fixed at 0, estimate the values of THETA. Then, fix the THETA values at those estimated, and now estimate the OMEGA matrix (i.e. ETA) for the fixed values of THETA. This way NONMEM isn’t trying to estimate both THETA and OMEGA at the same time. For further detailed information you should go through NONMEM manual 5 and this chapter helps you well in NONMEM 5 (Chapter 9 - Additional NM-TRAN Records) All the best.