[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: data set up for multiple failure survival model with time varying predictors

From   "Homish, Gregory" <>
To   <>
Subject   st: data set up for multiple failure survival model with time varying predictors
Date   Tue, 2 Oct 2007 09:55:28 -0400

I am trying to set up the data for an event history/survival/hazard analysis.  I want to implement a multiple failure model with time varying predictors.  The data are 6 assessments spaced one month apart.  For the time varying predictors, information is gathered for the past 30 days.  For the outcome event, it is captured on a daily level.  I know that I can collapse the daily level information into 6 one month time periods (e.g., did event happen in the past 30 days: yes/no), but I wanted to take advantage of all of the information available (for example, I would like to be able to capture events occurring at day 5 and at day 20, but in the earlier setup this would count as one event occurring in the first 30 day period).  I know that to incorporate time varying covariates, I would set up the data with 6 records per person (one per time), but I am confused how to indicate the occurrence of the event (when the event is a multiple failures).  Without time varying predictors,!
  I could set up my data as one row, with one variable representing time to each failure (t1 to t6) similar to example 3.2 in the Stata FAQ: analysis of multiple failure-time survival data).  However, the FAQ does not provide examples of multiple failures with time varying covariates.  Page 139 of the ST manual provides a nice example using -stgen nf=nfailures()- to create a new outcome variable which will handle multiple failures.  However, this example does not have time-varying covariates (there are X1 and X2 predictors, but it does not appear that they are meant to represent variable X at time 1 and variable X at time 2).  To sum up, I want to know how to set up my dataset to handle multiple failures with time varying predictors.  Also, I am not sure how to code my outcome.  In the case of recording any failure in the first 30 days as an event occurring in time 1, I would simply have a dichotomous failure variable at each of the 6 time points; however, I would prefer to !
 have the outcome modeled continuously.   Any help would be greatly app

reciated.  Thanks, in advance, for considering this question.

Greg Homish


*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index