Re: predictors in survival analysis

[Yampolskaya, Svetlana]

Dear List,

Does anybody know if the predictors in Cox regression should have
constant effect over the length of time that was studied (i.e.,
dependent variable)?

For example, I am trying to examine predictors for time to child's death
as a result of maltreatment.

Can I include predictors like "child's enrollment in daycare?" I do know
that the child was not enrolled in daycare from the moment he/she was
born.

Thank you,

Lana

[Hector Maletta]

         Of course, Lana, predictors in Cox regression may be (and
ordinarily are) independent of time. Sex is an obvious predictor that is in
fact constant over time. Besides, you may also have predictors that vary
over time ("time-dependent covariates").

         Hector

         -----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Yampolskaya, Svetlana
Sent: 28 August 2007 12:38
To: [hidden email]
Subject: Re: predictors in survival analysis

         Dear List,

         Does anybody know if the predictors in Cox regression should have
         constant effect over the length of time that was studied (i.e.,
         dependent variable)?

         For example, I am trying to examine predictors for time to child's
death
         as a result of maltreatment.

         Can I include predictors like "child's enrollment in daycare?" I do
know
         that the child was not enrolled in daycare from the moment he/she
was
         born.

         Thank you,

         Lana

[Hector Maletta]

         Lana,
         You pose a very interesting question, one that we have discussed
some time ago in this forum. It is the difference between censored cases and
cases lost to follow up. A censored case is a case surviving to the end of
the study without suffering the event (not being adopted, in your research).
A case lost to follow up is a case that abandoned the study prior to the
established period of observation, not having suffered the event until that
time. These cases MIGHT suffer the event after they abandon the study, but
we do not know because we did not follow them up.
         I do not know whether in your study the children in the third
category may have been adopted elsewhere, after they were discharged for
other reasons, but let us suppose they might. If they do not have the
possibility of being adopted after their discharge, they are truly censored
(the other causes of discharge are "competitive risks", like dying from a
car accident in a study on survival after cardiovascular surgery). If they
still can be adopted after their non-adoption discharge (e.g. adopted
through some other adoption service), their chances of adoption are not
exhausted with their discharge.
         In practice, SPSS does not distinguish between the two, and does
not provide a means to treat the third group differently. Some analysts,
including Marija Norusis in her SPSS book, consider that both are the same.
Others regard them as different. I tend to belong to this latter group. At
any rate, your third category is composed of children whose time of exposure
to the risk of being adopted under your study is shorter than other
children's, since they were discharged before the end of the study for some
reason, though the reason might not necessarily be incompatible with later
adoption.
         If one accepts the idea that they should be treated as censored,
you should include them in the second group. The only other alternative
would be to exclude them from the study. You should be careful to take care
of the fact that their time of exposure under your study is shorter.

         Hector

         -----Original Message-----
From: Yampolskaya, Svetlana [mailto:[hidden email]]
Sent: 28 August 2007 17:41
To: Hector Maletta
Subject: RE: predictors in survival analysis

         Hector,

         Thank you very, very much for answering my question. I have another
one
         for you. I hope you don't mind.

         I am doing a survival analysis trying to predict time to adoption
after
         a child was placed in out-of-home care and I have three conditions:

         One is clear and refers to cases when the child was adopted.

         DO IF (DSY GE 1 AND ADOP = 1).  (DSY means that the child has a
         discharged date and adop = 1 means that the reason for discharge
was
         adoption).

          COMPUTE adoption = CTIME.DAYS((Dischargedate - placedate )/30.44).
          COMPUTE status = 1.
         END IF.

         **The second condition is pretty clear too. This is the situation
when
         the child did not have a discharge date and therefore this case is
a
         censored observation.

         DO IF (DSY  = 0).
          COMPUTE adoption  = CTIME.DAYS((July07 -  placedate) / 30.44).
          COMPUTE status = 0.
         END IF.


         *** I also have the third case when the child does have a discharge
date
         but was the reason for discharge was not an adoption. What should I
do
         with it?
         Should I ignore that and repeat the second line or should I
calculate
         separate time for this case: Time from placement to discharge?

         DO IF (DSY  = 0).
          COMPUTE adoption  = CTIME.DAYS((July07 -  placedate) / 30.44).
          COMPUTE status = 0.
         END IF.

         Any thoughts or clarifications would be greatly appreciated,

         Lana


         -----Original Message-----
         From: Hector Maletta [mailto:[hidden email]]
         Sent: Tuesday, August 28, 2007 1:17 PM
         To: Yampolskaya, Svetlana; [hidden email]
         Subject: RE: predictors in survival analysis

                  Of course, Lana, predictors in Cox regression may be (and
         ordinarily are) independent of time. Sex is an obvious predictor
that is
         in fact constant over time. Besides, you may also have predictors
that
         vary over time ("time-dependent covariates").

                  Hector

                  -----Original Message-----
         From: SPSSX(r) Discussion [mailto:[hidden email]] On
Behalf Of
         Yampolskaya, Svetlana
         Sent: 28 August 2007 12:38
         To: [hidden email]
         Subject: Re: predictors in survival analysis

                  Dear List,

                  Does anybody know if the predictors in Cox regression
should
         have
                  constant effect over the length of time that was studied
(i.e.,
                  dependent variable)?

                  For example, I am trying to examine predictors for time to
         child's death
                  as a result of maltreatment.

                  Can I include predictors like "child's enrollment in
daycare?"
         I do know
                  that the child was not enrolled in daycare from the moment
         he/she was
                  born.

                  Thank you,

                  Lana