Stopped because of too many likelihood evaluations

[Jordani]

Hi everybody ,

I have a longitudinal data on Body Mass Index BMI  before and after medical intervention . I am running a mixed model that accounts for time interval between date of measurement  and the date of  intervention. I also included an interaction term. 

the data looks like this :

ID       BMI      Time_interval  Time 

1         29               -10              Before 

1         30              10                 After 

1         30.5           20                 After

2         30.1           -8                  Before 

2         29.5            11                 After 

2         28.7             18                After 

The model I used :

ods graphics on;

Proc mixed data = have  plots= (Maxpoints=none);

class ID Time ;
model      BMI=  Time Time_interval  Time*Time_interval  /solution residual outpm=marg outp=cond vciry 
influence(iter=0 effect=ID est) ;
random Time Time_interval /subject=pat_mrn_id type=UN g gcorr;
run;
ods graphics off;

I keep receiving this message :

"Stopped because of too many likelihood evaluations"

any ideas ?

[SteveDenham]

The fact that your subjects are not measured at the same timepoints complicates things immensely.

It looks to me as if you are trying to fit BMI to time_interval, but that there is something unusual about the pre-intervention values.  Using an unstructured covariance matrix with so many time_interval values, and with not every subject at those times leads to definite convergence problems.

If all of your data looks like this for the first two IDs, I might consider recoding time_interval to something like:

data want;

set have;

if time_interval<=0 then era=0;

if 0<time_interval<=10 then era=1;

if 10<time_interval<20 then era=2;

...more like this if needed...

run;

Then I would try to fit the following model:

Proc mixed data = want  plots= (Maxpoints=none);
class ID  era ;
model      BMI=  era /solution residual outpm=marg outp=cond vciry 
influence(iter=0 effect=ID est) ;
random intercept/subject=ID;
repeated era /subject=ID type=ARH(1) g gcorr;
run;

Now if you really don't like aggregating time_interval like this, I would suggest fitting some sort of spline, using the EFFECT statement.

Steve Denham