Hi Pavel, I guess I would opt for using logtransformed concentrations as dv in your model. That would deal with the negative model predictions.
Best, Huub ________________________________________ From: [email protected] [[email protected]] on behalf of Pavel Belo [[email protected]] Sent: Friday, January 29, 2016 10:47 PM Cc: '[email protected]' Subject: [NMusers] model of error, linearisation, additive error, and VPC Hello NONMEM Users, When I tried to print log-scaled VPC, there was an error message about negative values. It can be caused by an additive error and/or linearization of the error model when NONMEM transforms Y = F*DEXP(ERR(1)*SD1) into Y = F+ F*ERR(1)*SD1. The error is inflated at small concentrations and removing the additive term is not an option in my case. Is there an easy way to solve it? It can be something like Y = F**GAMMA*DEXP(ERR(1)*SD1) or Y = F+ F*ERR(1)*SD1 + F**GAMMA*ERR(2)*SD2, where GAMMA<1. Thanks, Pavel
