Cox model for "Remission before time T" outcome

[F. Danjou]

Dear everybody,

I need help on a Cox model I ran.

I want to analyze predictors of long term treatment status (i.e.:
patients with still no remission after 3 years) so I used the outcome as
being "remission before 3 years". I have state of patients (under
treatment or not), time to event, and covariates. I choose to apply a
survival analysis to make use of lost at follow-up, but my outcome is
more a transversal one (state of patient at a time T), so I am scared
that it might be wrong to use a Cox model and should use a Logistic
Regression. I'd rather not use a logistic regression as I will loose
lost at follow-up patients. Is it statistically wrong to use a survival
analysis with such an outcome?

To be precise I need to say that I codified the outcome as "occurred"
when patients remission happened before 3 years, and "not occurred" when
remission occurred after 3 years or when patients left treatment before
3 years without remission (for any other reason).

Thank you in advance for your precious help.

Best regards,

Fabien.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Hector Maletta]

If you have patients followed at given dates during 3 years, you should used
Cox: it is precisely for those cases. Log Reg is completely out of the
question.
Your coding of "remission" as the event (coded with 1) is correct. I suppose
the date of remission would be the date in which remission was first
observed (this may be difficult with certain medical conditions: you observe
remission at a certain date, but remission may have occurred before without
you noticing). Cases abandoning the study (or dying) before 3 years are
censored, as well as those continuing without remission for 3 years when the
study ended.
Hector

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of F.
Danjou
Sent: 04 March 2009 09:33
To: [hidden email]
Subject: Cox model for "Remission before time T" outcome

Dear everybody,

I need help on a Cox model I ran.

I want to analyze predictors of long term treatment status (i.e.:
patients with still no remission after 3 years) so I used the outcome as
being "remission before 3 years". I have state of patients (under
treatment or not), time to event, and covariates. I choose to apply a
survival analysis to make use of lost at follow-up, but my outcome is
more a transversal one (state of patient at a time T), so I am scared
that it might be wrong to use a Cox model and should use a Logistic
Regression. I'd rather not use a logistic regression as I will loose
lost at follow-up patients. Is it statistically wrong to use a survival
analysis with such an outcome?

To be precise I need to say that I codified the outcome as "occurred"
when patients remission happened before 3 years, and "not occurred" when
remission occurred after 3 years or when patients left treatment before
3 years without remission (for any other reason).

Thank you in advance for your precious help.

Best regards,

Fabien.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Hector Maletta]

Fabien,
The end of the study is whatever date you choose to stop measuring the
outcome. In your case it was three years, for reasons of your own (you may
have data for a longer period, and by not using them you lose the
information about late remissions).
However, if you are only interested in remissions occurring before 3 years,
it is OK to mark as zero all cases that had not remitted after 3 years (or
those that disappeared from the study before 3 years, without having shown
remission at the time of leaving the study).
Hector

-----Original Message-----
From: F. Danjou [mailto:[hidden email]]
Sent: 04 March 2009 11:43
To: Hector Maletta
Subject: Re: Cox model for "Remission before time T" outcome

Thank you so much!

I think you answered to my question, however I want to be sure of one
thing: I codified patients with remission after 3 years (which is NOT
the end of the study) as 0, whereas patients with remission before 3
years are codified as 1, because I am interested in prediciting "time to
quick-term remission" (to indirectly predict long term treatment without
remission) and not "time to remission". Is it correct?

thank you again, best regards,

Fabien.

Hector Maletta wrote:
> If you have patients followed at given dates during 3 years, you should
used
> Cox: it is precisely for those cases. Log Reg is completely out of the
> question.
> Your coding of "remission" as the event (coded with 1) is correct. I
suppose
> the date of remission would be the date in which remission was first
> observed (this may be difficult with certain medical conditions: you
observe
> remission at a certain date, but remission may have occurred before
without
> you noticing). Cases abandoning the study (or dying) before 3 years are
> censored, as well as those continuing without remission for 3 years when
the
> study ended.
> Hector
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
F.

> Danjou
> Sent: 04 March 2009 09:33
> To: [hidden email]
> Subject: Cox model for "Remission before time T" outcome
>
> Dear everybody,
>
> I need help on a Cox model I ran.
>
> I want to analyze predictors of long term treatment status (i.e.:
> patients with still no remission after 3 years) so I used the outcome as
> being "remission before 3 years". I have state of patients (under
> treatment or not), time to event, and covariates. I choose to apply a
> survival analysis to make use of lost at follow-up, but my outcome is
> more a transversal one (state of patient at a time T), so I am scared
> that it might be wrong to use a Cox model and should use a Logistic
> Regression. I'd rather not use a logistic regression as I will loose
> lost at follow-up patients. Is it statistically wrong to use a survival
> analysis with such an outcome?
>
> To be precise I need to say that I codified the outcome as "occurred"
> when patients remission happened before 3 years, and "not occurred" when
> remission occurred after 3 years or when patients left treatment before
> 3 years without remission (for any other reason).
>
> Thank you in advance for your precious help.
>
> Best regards,
>
> Fabien.
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [hidden email] (not to SPSSX-L), with no body text except the
> command. To leave the list, send the command
> SIGNOFF SPSSX-L
> For a list of commands to manage subscriptions, send the command
> INFO REFCARD
>
>
>
>

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD