IIV decrease in mixture model

Dear NMusers, I have a PD model describing viral infection process. There is variable Vo for initial viral load and it has IIV, omega(V0). When I fit the data with simple model, omega(V0) is 52% and covariance step is ok. But when I try mixture model, omega(V0) becomes very low, and covariance step is not finished. But if I look at post-hos values of Vo, there are two peaks, for 2 groups of patients. Do you have any idea why it is like this? I mean it seems strange that omega(V0) decreases so much. Thank you all in advance, Svetlana

Svetiana, Without seeing the code, these are a few questions/thoughts that came to my mind. Do the actual measures of the initial viral load appear to have 2 peaks? When you fit with the mixture model, how many subjects is NONMEM assigning to each group? Are you using 2 different theta & etas for the 2 different groups with a BLOCK SAME option? SAME assumes the size of the variability is the same for the 2 groups. However, with the 2 different etas, as NONMEM is testing which group the subject should belong to it can assess 2 different eta values for an individual. $OMEGA BLOCK(1) 0.25 ;--eta1-IIV in V0 grp 1 [exp] $OMEGA BLOCK(1) SAME ;--eta2-IIV in V0 grp 2 [exp] ;assign mixnum to a variable to output in table EST=MIXEST ;Assign V0 group IF(MIXNUM.EQ.1) THEN TV0=THETA(1) V0=TV0*EXP(ETA(1)) ELSE TV0=THETA(2) V0=TV0*EXP(ETA(2)) ENDIF You might also try initializing the viral load compartment to the estimated value rather than the measured value. IF(A_0FLG.EQ.1) THEN A_0(N) = V0 ;initialize the viral load cmt to the estimated value Luann Phillips Director of PK/PD Cognigen Corporation Vinogradova Svetlana wrote: > Dear NMusers, > > I have a PD model describing viral infection process. There is variable Vo for > initial viral load and it has IIV, omega(V0). When I fit the data with simple > model, omega(V0) is 52% and covariance step is ok. But when I try mixture > model, omega(V0) becomes very low, and covariance step is not finished. But if > I look at post-hos values of Vo, there are two peaks, for 2 groups of patients. > Do you have any idea why it is like this? I mean it seems strange that > omega(V0) decreases so much. > > Thank you all in advance, > Svetlana