Date:         Thu, 8 May 2003 12:51:47 -0400
Reply-To:     "Yampolskaya, Svetlana" <yampol@fmhi.usf.edu>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "Yampolskaya, Svetlana" <yampol@fmhi.usf.edu>
Subject:      Re: problems with survival analysis
Content-Type: text/plain; charset="iso-8859-1"

Dear List,

I have some problems with survival analysis, specifically, life tables, that
I cannot solve.

Here are the variables I have:

Subject id (SUBJID), the date when a subject entered services  (RECDATE),
the last date of tracking subjects (LDATE), and 5 districts (DISTR).

Some of these subjects entered the services for the second time. Therefore,
I have cases with the same subjects' id but different time of entry.

Using life tables I am trying to calculate the percentage of subjects who
have the second episode of services by the end of 12th month.

Here is my syntax:

SORT CASES BY SUBJID RECDATE.

STRING  L_SUBJID (A11).
COMPUTE L_SUBJID = LAG(SUBJID).

DO IF (L_SUBJID EQ SUBJID).
 COMPUTE INTRTIME = CTIME.DAYS((RECDATE - LRECDATE)/30.44).
 COMPUTE CENSOR = 1.
END IF.

RECODE INTRTIME (0,SYSMIS=-999).
DO IF (INTRTIME EQ -999).
  COMPUTE INTRTIME = CTIME.DAYS((LDATE - RECDATE) / 30.44).
 COMPUTE CENSOR = 0.
END IF.

VAL LAB CENSOR 0 'CENSORED' 1 'NOT' .

SURVIVAL TABLE = INTRTIME BY DISTR(1,5)
  /INTERVAL=THRU 12 BY 1
  /STATUS=CENSOR(1)
  /PRINT=TABLE .


The problem is that the percentages I get are at least twice as higher as
they should be (compared to  % of uncensored cases obtained from the
crosstabs).

I do know that percentages obtained from survival tables should a little bit
higher but I don't think what I get is right.

Please help,


Lana