st: modeling "recovery" and "response" using xtmixed

[Adriaan Hoogendoorn <aw.hoogendoorn@gmail.com>]

Dear Statalist,

I use a linear mixed model on longitudinal data to study the
development of a "score on disease" over time. The measurements taken
at t0, t1, t2 and t3, my data are in "long format".
The mixed model deals appropriately with patients dropping out of the
study (under the MAR assumption). The parameter of interest is the
regression coefficient of the "treatment" indicator. I use Stata's
xtmixed and have no questions on this part.

My question is about modeling "response" and "recovery".

A patient is responding if the "score on disease" variable drops a
certain amount of points. I intend to model the (binary) indicator
"response" using Stata's xtmelogit. However, the indicator "response"
is missing at t0, and xtmelogit ignores the t0 observations. To me
this does not seem like an appropriate way of handling, since patients
who dropped out between t0 and t1 now disappear from the analysis
entirely. What to do?

A patient is recovered if the "score on disease" variable is below a
certain level. The indicator "recovery" is missing at t0, but could be
imputed based on the "score on disease" at t0. Since all patients sick
at t0, all recovery indicators at t0 obtain a value 0. Having only
zeroes at t0 gave (ML) convergence problems.

Do you have any suggestions on how to model response and recovery in
this context?

Kind regards,
Adriaan Hoogendoorn
GGZ inGeest, Amsterdam
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/