st: modelling asynchronous interval-censored data with time-varyingcovariates

[Angus Campbell <a.campbell@unimelb.edu.au>]

Dear Stata-listers,

I inadvertantly left the subject off this post when I put it up 2 days ago, 
so am re-posting in case it was overlooked.

I am attempting survival analysis of asynchronously interval-censored 
data.  Some of my covariates start out equal to zero (before the subjects 
acquire particular attributes) but at later, different, times take on a 
value which then remains constant.  I've used -stsplit- to set this up.  I 
would like to analyse this time-varying characteristic in the model if 
possible.

I can find techniques that handle interval censored data, or time-varying 
covariates, but not both.  The asynchronous interval censoring seems to be 
the sticking point.  I have tried using -stpm- but can't make it work with 
the stsplit data if I also try to specify interval-censoring.

Here is an example of a 'fail' & 'non-fail' record:

. st
-> stset dayssurv, id(record) failure(fail)

                id:  record
     failure event:  fail != 0 & fail < .
obs. time interval:  (dayssurv[_n-1], dayssurv]
 exit on or before:  failure

. list record _t0 _t _d dayssurv fail left invwwt gr2t if record ==1 | 
record ==801, noobs

  +------------------------------------------------------------------+
  | record   _t0    _t   _d   dayssurv   fail   left   invwwt   gr2t |
  |------------------------------------------------------------------|
  |      1     0    60    0         60      .     60     .048      0 |
  |      1    60   100    0        100      .    100     .048      0 |
  |      1   100   150    0        150      .    150     .048      0 |
  |      1   150   180    0        180      .    180     .048   .051 |
  |      1   180   210    0        210      .    210     .048   .051 |
  |      1   210   288    0        288      0    288     .048   .051 |
  |------------------------------------------------------------------|
  |    801     0    60    0         60      .     60     .364      0 |
  |    801    60   100    0        100      .    100     .364      0 |
  |    801   100   143    1        143      1    100     .364      0 |
  +------------------------------------------------------------------+

* 'invwwt' isn't time-varying; gr2t is;  'left' == _t if no failure within 
the interval; left == _t0 if failure occurs within the interval

. xi: stpm invwwt gr1 gr2t gr3t i.sex, left(left) scale(hazard) df(2)
i.sex             _Isex_0-100         (naturally coded; _Isex_1 omitted)
cannot have both interval censoring and late entry
r(198);

Is anyone aware of techniques I can use in Stata to model this kind of 
situation, please?

Thanks in advance for any thoughts,

Angus Campbell
---------------------------------
Dr Angus Campbell
The Mackinnon Project
Faculty of Veterinary Science, University of Melbourne
250 Princes Highway
Werribee, Victoria, Australia 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/