Survival analysis with competing risks

[Andreas Völp]

Does anybody know how to perform survival analysis with competing risks (e.
g., death in remission when relapse is the event of interest) in SPSS?

[Karovaldas]

I don't have good ideas, but am curious about your experimental
question--what are you trying to test?

If you are modelling time to relapse, deaths are irrelevant and you
should simply censor them. For those who die, time-to-relapse is
unknown and survival analysis is designed for precisely this scenario.

Otherwise, if death and relapse are of equal relevance, you can code
both as terminal events, analyze time-to-event and add a predictor
variable that distinguishes between them or indicator variables for
each. This way you can get separate hazards for both and an estimate
of relative risk. I am not familiar with any weighing methods for
survival analysis, but I imagine they should be readily available.

Please post your findings since this is an interesting question.
Thanks.

Post by Andreas Völp
Does anybody know how to perform survival analysis with competing risks (e.
g., death in remission when relapse is the event of interest) in SPSS?
--
Andreas Völp
Psy Consult Scientific Services
Frankfurt, Germany

[Andreas Völp]

Post by Karovaldas
If you are modelling time to relapse, deaths are irrelevant and you
should simply censor them.

Not exactly. Let us assume that the disease under investigation is non-fatal
and that the reason for death is completely independent of treatment
efficacy or safety. A patient in remission who dies, e. g., during a traffic
accident cannot relapse, but he/she could have relapsed during the period of
observation, had he/she not died in the accident. In this case death in
remission is a competing risk because it modifies the risk of relapse.

For a more detailed discussion, see, e. g.,
http://www.ebmt.org/1WhatisEBMT/Op_Manual/OPMAN_StatGuidelines_oct2003.pdf#search=%22site%3Awww.ebmt.org%20competing%20risks%22
(particularly chapter 7). Methods for competing risk analysis (and their
rationale) are also presented in Fine & Gray, JASA 1999, and in Gooley,
Leisenring, Crowley, Storer, Stat. Med. 1999.

Hope this helps!

Andreas.

[Brendan Halpin]

Post by Andreas Völp

Post by Karovaldas
If you are modelling time to relapse, deaths are irrelevant and you
should simply censor them.

Not exactly. Let us assume that the disease under investigation is non-fatal
and that the reason for death is completely independent of treatment
efficacy or safety. A patient in remission who dies, e. g., during a traffic
accident cannot relapse, but he/she could have relapsed during the period of
observation, had he/she not died in the accident. In this case death in
remission is a competing risk because it modifies the risk of relapse.

No, the prior poster is correct. If the death is irrelevant to the
process being studied, it must be treated as censoring. Censoring
allows the observations up to the death to be taken into account.
An end of observation because you walked under a bus is
statistically the same as an end of observation because you
survived to the end of the study.

Competing risks analysis is required where there are more than one
substantively interesting ways of exiting the state. Using an
example that comes to my mind (not being medical), we may wish to
model exits from unemployment spells taking account of exits to
employment and exits to non-employment differently. The simplest
way of doing this is to run two analyses, treating exits to
employment as the outcome in the first, and all else as censoring,
and exist to non-employment as the outcome in the second. This
unfortunately makes the assumption that the two exits are
independent, which may well be unlikely.

Brendan

[d***@gmx.de]

Hi,

Competing risks with spss is not straightforward. It might be better to
use stata or R. I struggled myself with it. Here are some helpful
references:

This article shows how to do simple competing risks models using cox
regressions, you have to parts manually but ist is fairly
straightforward:
Lunn and McNeil (1995) Applying Cox regression to competing risk,
Biometrics 1995 51,524-532*

However, this method is said to be not the most sophisticated ones. I
just saw a few days ago an tutorial on competing rsiks with examples in
R but they mentioned that most can be doen in spss but "may require
some extra effort regarding data prep and programming":

Tutorial in biostatistics: competing risks and multi stage models
Statistics in medicine in press (online availaible)


Another overview paper is this one (with SAS syntax):

STATISTICS IN MEDICINE
Statist. Med. 2001; 20:661{684
Competing risks analysis of patients with osteosarcoma:
a comparison of four different approaches
Bee-Choo Tai1;; y, David Machin1; 2, Ian White3 and Val Gebski4

Hope this helps a bit
Dan iel

Post by Andreas Völp
Does anybody know how to perform survival analysis with competing risks (e.
g., death in remission when relapse is the event of interest) in SPSS?
--
Andreas Völp
Psy Consult Scientific Services
Frankfurt, Germany

[Andreas Völp]

... thanks very much! I was just thinking about getting NCSS which offers a
competing risks survival analysis ...

Andreas.

[n***@gmail.com]

Hi Andreas, is it possible to perform competitive risk analysis (using Fine and Gray model) with NCSS? Can you kindly explain me how can I do it?
Thank you

[Rich Ulrich]

Post by n***@gmail.com
Hi Andreas, is it possible to perform competitive risk analysis (using Fine and Gray model) with NCSS? Can you kindly explain me how can I do it?
Thank you

* * *
* Reply: That was a follow-on to the post from 2006, which my
* provider (Giganews) still has available -
* * *
... thanks very much! I was just thinking about getting NCSS which
offers a
competing risks survival analysis ...

Andreas.
* * *
* Reply: That, in turn, was a reply to this Q&A -
* * *

Competing risks with spss is not straightforward. It might be better
to use stata or R. I struggled myself with it. Here are some helpful
references:

This article shows how to do simple competing risks models using cox
regressions, you have to parts manually but ist is fairly
straightforward:
Lunn and McNeil (1995) Applying Cox regression to competing risk,
Biometrics 1995 51,524-532*

However, this method is said to be not the most sophisticated ones. I
just saw a few days ago an tutorial on competing rsiks with examples
in R but they mentioned that most can be doen in spss but "may require
some extra effort regarding data prep and programming":

Tutorial in biostatistics: competing risks and multi stage models
Statistics in medicine in press (online availaible)


Another overview paper is this one (with SAS syntax):

STATISTICS IN MEDICINE
Statist. Med. 2001; 20:661{684
Competing risks analysis of patients with osteosarcoma:
a comparison of four different approaches
Bee-Choo Tai1; ? ; y, David Machin1; 2, Ian White3 and Val Gebski4

Hope this helps a bit
Dan iel

Post by n***@gmail.com
Does anybody know how to perform survival analysis with competing risks (e.
g., death in remission when relapse is the event of interest) in SPSS?
--
Andreas Völp
Psy Consult Scientific Services
Frankfurt, Germany