Re: Estimates for Random effects

Shankar, Because NONMEM uses a first-order Taylor series approximation when fitting data, Y=F*(1+EPS(1)) <CCV> is equivalent to Y=F*EXP(EPS(1)) . Therefore to fit a log-normal error model, use log-transformed concentrations as DV and then fit an additive error model, Y=F + EPS(1), where F represents now represents the log of concentration. To obtain initial estimates, use the procedure for the additive model with log of concentration. Luann Phillips Director, PK/PD Cognigen Corporation Shankar Lanke wrote: > Dear All, > > I am going through NONMEM manual 5 and I came across How to obtain initial estimates for omega and sigma > > In chapter 9 they mentioned the additive and ccv model but not about Log Normal model. > > For Additive y =f +e sigmay = sigma e= rt (r*t is the fraction of its typical value or mean* typical value) > > var(e) =se^2=(rt )^2 = (r*t) ^2 > > For CCV > > y =f +f e > sy =f *se=r*t > var(e)=se^2= (rt)^ 2 /f^2 > > but if we identify t with the value of f (whatever it may be), we have then Var(e) = r^2 i.e the fraction of its typical value > > How we have to estimate for Log normal model. > > Thank you very much in advance for your feed back. > > Regards, Shankar Lanke Ph.D. University at Buffalo > > Office # 716-645-4853 > Fax # 716-645-2886 > > Cell # 678-232-3567