Joint Survival Analysis
Message/Author
 Miguel Arce Renteria posted on Friday, October 05, 2018 - 11:20 am
Dear Drs. Muthen,

I am trying to run a joint model with a survival analysis to evaluate the effect of a factor on survival. I am evaluating the association of literacy (literate/illiterate) on risk of incident dementia, and seeing whether death or dropout affects this relationship. I was able to get the model to run using a similar example on chapter 6 of the user guide (ver 8), but the estimate I am getting for the death/dropout factor is so huge it makes me believe it is incorrect. The syntax is as follows:

missing are all (-999);
usevariables are demtime demstat lit ndv2-ndv5;
useobservations are demtime ge 0.00000001;

survival = demtime;
timecensored = demstat (1 = NOT 0 = RIGHT);
ANALYSIS: BASEHAZARD = on;
MODEL:
F by ndv2-ndv5@1;
F@0;
demtime on lit F;
F on lit;

output: cinterval standardized;

The estimates in the output are:
F ON LIT -0.005 (SE = 0.003)
DEMTIME ON F 40.969 (SE = 13.855)
DEMTIME ON LIT 0.921 (SE = 0.274)

The estimate of 40 for demtime on F seems incredibly large. Especially because when I exponentiate it to get the hazard's ratio is 620312637569017000.00. What am I doing wrong?

I appreciate the help!

Best,
Miguel
 Tihomir Asparouhov posted on Friday, October 05, 2018 - 3:23 pm
I think you should replace

F by ndv2-ndv5@1;
F@0;

with

F by ndv2-ndv5;

otherwise F=0.
 Miguel Arce Renteria posted on Monday, October 08, 2018 - 5:32 am
Thanks for the suggestion, but when I do that I get the following error:
" THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO AN ILL-CONDITIONED
FISHER INFORMATION MATRIX. CHANGE YOUR MODEL AND/OR STARTING VALUES.

THE MODEL ESTIMATION DID NOT TERMINATE NORMALLY DUE TO A NON-POSITIVE
DEFINITE FISHER INFORMATION MATRIX. THIS MAY BE DUE TO THE STARTING VALUES
BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION
NUMBER IS 0.160D-11.

THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE
COMPUTED. THIS IS OFTEN DUE TO THE STARTING VALUES BUT MAY ALSO BE
AN INDICATION OF MODEL NONIDENTIFICATION. CHANGE YOUR MODEL AND/OR
STARTING VALUES. PROBLEM INVOLVING THE FOLLOWING PARAMETER:
Parameter 6, F BY NDV4"

That's why I had chosen to constraint the model @1 in order to make it converge.
 Tihomir Asparouhov posted on Monday, October 08, 2018 - 11:16 am
In principle the model is identified but maybe there is something specific to your data that is causing the issue. Try removing only the line
F@0;
 Miguel Arce Renteria posted on Monday, October 08, 2018 - 11:30 am
After removing F@0; I continue getting a similar estimate and SE.

It can't be that that estimate is correct, right?

I use a similar syntax to jointly model that death/dropout factor on latent growth curve model and the estimates are appropriate there.
 Miguel Arce Renteria posted on Monday, October 08, 2018 - 2:58 pm
Forgot to mention, the LGCM uses the t-score function given that the participants have individually varying time points. Not sure if that may play a role in why that factor works one way in the LGCM and another here.

Thanks!