Q: Why do you recommend not to set bounds for fixed effects in Phoenix NLME? And what if we do not want lower bound to be negative?
A: The fundamental optimization algorithm used in Phoenix NLME is the well-known unconstrained BFGS algorithm. Hence, if bounds are used, then Phoenix NLME has to transform them into unconstrained ones. This may occasionally bring some numerical difficulties. If you are sure that the parameter, say clearance Cl, is always postive then, instead of using
stparm(Cl = tvCl * exp(nlogCl))
we would suggest writing it as follows:
stparm(Cl = exp(tvlogCl + nlogCl))
which ensure that Cl is always positive no matter what the values of tvlogCl (fixed effect) and nlogCl are.
Q: Can you please shed some light on dealing with multiple covariates of different nature. Eg. Effect of disease state (ordinal) and gender (binary covariate) on CL.
A: Ordinal covariates are those categorical variables that are ordered in nature. For example, the Effect of disease can be stated as “Normal”, “Mild”, “Moderate” or “Severe”. For the purpose of Phoenix NLME, both ordinal and binary covariates are treated statistically in the same manner as “Categorical Covariate” under Covariate type.
Q: Which is the better software: NLME or NONMEM?
A: You would not expect anything else then us saying that NLME is the better software, however that won’t help you much. That is why I would like to point you to an objective review of Phoenix NLME that was authored by Nathan Teuscher in 2011 (he joined Certara in 2015, so he was unbiased at the time he wrote the review):
Q: Is there a way to compare statistically in Phoenix the difference between covariate groups for fixed effects?
A: I think you are referring to categorical covariates when you are referring to “Covariate groups”. If so, Phoenix NLME will not assess this difference between two groups and only evaluates the overall significance of the effect of a particular covariate. WinNonlin Linear Mixed Effects model can evaluate the differences between different covariate groups through pair-wise comparisons.
Q: How to add disease related covariates in NONMEM control stream?
A: As you have seen in the example project, you need to add 3 “if” statements with each statement describing each of the disease status including the “Normal” disease status. For example, the NONMEM code for disease related covariate with 3 levels (0, 1, 2) on Volume of Distribution (V2) is as follows in the example project presented in the webinar:
DVDISEASESTATE1 = THETA(6)
DVDISEASESTATE2 = THETA(7)
IF (DISEASESTATE .EQ. 0) THEN
V2 = TVV * EXP(ETA(2))
IF (DISEASESTATE .EQ. 1) THEN
V2 = TVV * EXP(DVDISEASESTATE1) * EXP(ETA(2))
IF (DISEASESTATE .EQ. 2) THEN
V2 = TVV * EXP(DVDISEASESTATE2) * EXP(ETA(2))