Negative values in 95%CI on ETAs

From:"MANOJ KHURANA" manoj2570@yahoo.com Subject: [NMusers] Negative values in 95%CI on ETAs Date: Tue, November 16, 2004 2:03 pm Hello All, I am a novice nonmem user and out of my first crawls on this floor I am here with a querry. I am trying to model an oral absorption data aftter escalating doses using ADVAN4 TRANS1 with KA fixed due to rapid absorption (Data apppears to follow 2-comp and obviously parametrization using TRANS3 didnt work well even after fixing KA). With the current model I have good estimates of Thetas (K, K23, K32 and V2) but 95%CIs on the ETAs have negative values (I have all EXP ETAs on all THETAs) with very high %RSE. Can anybody suggest whats missing and if I am wrong somewhere. Thanks MK

From: "Bhattaram, Atul" BhattaramA@cder.fda.gov Subject: RE: [NMusers] Negative values in 95%CI on ETAs Date: Tue, November 16, 2004 2:43 pm Hello Manoj A couple of quick comments: 1. It is better to parameterize models in terms of CL, V, Q, VSS or V(peripheral) instead of the parameterization you have chosen. 2. You could try different values of Ka and check how it influences the results. 3. What is the estimation procedure you are using? Venkatesh Atul Bhattaram Pharmacometrics DPE-1, OCPB CDER, FDA.

From: "MANOJ KHURANA" manoj2570@yahoo.com Subject: RE: [NMusers] Negative values in 95%CI on ETAs Date: Tue, November 16, 2004 3:27 pm Hi All, Thanks for your input. I initially tried parametrization in terms of CL and V realizing they are meaningful parameters but I dont know for what reason it didnt work even after fixing KA to different values and I had underpredicted concentrations all over the data. I am using method=0 (FO). By the way the data I have is rich so please advice if FO or FOCE would be appropriate. Thanks MK

From: "Sam Liao" sliao@pharmaxresearch.com Subject: RE: [NMusers] Negative values in 95%CI on ETAs Date: Tue, November 16, 2004 4:24 pm Hi Manoj: I think the problem may be the initial estimates you used. It is more critical when you have a two-compartment model. Also, I would use FOCE for rich dataset. Sam Liao PharMax Research

From: "MADABUSHI,RAJANIKANTH" Subject: RE: [NMusers] Negative values in 95%CI on ETAs Date: Tue, November 16, 2004 6:06 pm Hi Manoj, Since you say the data is rich and you are fixing Ka, you should be using FOCE. Moreover, FOCE is a better method of estimation than FO ("quick and dirty"). raj

From: "Nick Holford" n.holford@auckland.ac.nz Subject: RE: [NMusers] Negative values in 95%CI on ETAs Date: Thu, November 25, 2004 4:20 am Manoj, The inclusion of a negative value in the 95% CI for an estimate of OMEGA arises from the assumptions that 1) NONMEM is correctly computing the standard error 2) the uncertainty distribution for the parameter is symmetrical. Both of these assumptions can be sidestepped by constructing a bootstrap distribution of the estimates (see http://wfn.sourceforge.net for some details). This is a good idea for any parameter estimated by NONMEM. IMHO any 95% CI constructed using the asymptotic SE from NONMEM is suspect. Nick -- Nick Holford, Dept Pharmacology & Clinical Pharmacology University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand email:n.holford@auckland.ac.nz tel:+64(9)373-7599x86730 fax:373-7556 http://www.health.auckland.ac.nz/pharmacology/staff/nholford/ _______________________________________________________