Prev: Apply your understanding of Accounts and Statistics
Next: Dataset Comparison Using Proc Compare
From: Brittany Lapin on 8 Apr 2010 16:39
Hi all, I have a real stumper. Any advice would be greatly
appreciated!

I'm doing a multivariable analysis using proportional hazards
regression and have a co-variate I'm not sure how to include in the
model.
Anti-Arrhythmic (AA) Medication use after discharge (DC): It is a
variable that does not apply before discharge at all. Then, after
discharge, it is missing for all patients who died prior to
discharge. I want to somehow include the patients who died prior to
discharge while, at the same time, adjusting for impact of the post-DC
AA use on survival.

What are your thoughts on the pros and cons of the following
scenarios:
1. Use time-invariant coding for post-discharge AA medication status
as a three-category variable—Yes/No/DCDeath—and examine the effect of
the contrast between Yes and No

2. Code AA as time-dependent covariate with all pre-DC intervals
classified as “No” for post-DC AA use, and use “Missing,” “Yes,” and
“No” for the post-discharge period.

Also of note, we don’t have reliable data on AA use at admission or
between admission and discharge. How does that impact the coding for
the post-discharge AA use?
From: Shawn Haskell on 9 Apr 2010 10:20
On Apr 8, 4:39 pm, Brittany Lapin <brittany.la...(a)gmail.com> wrote:
> Hi all, I have a real stumper. Any advice would be greatly
> appreciated!
>
> I'm doing a multivariable analysis using proportional hazards
> regression and have a co-variate I'm not sure how to include in the
> model.
> Anti-Arrhythmic (AA) Medication use after discharge (DC): It is a
> variable that does not apply before discharge at all. Then, after
> discharge, it is missing for all patients who died prior to
> discharge.  I want to somehow include the patients who died prior to
> discharge while, at the same time, adjusting for impact of the post-DC
> AA use on survival.
>
> What are your thoughts on the pros and cons of the following
> scenarios:
> 1. Use time-invariant coding for post-discharge AA medication status
> as a three-category variable—Yes/No/DCDeath—and examine the effect of
> the contrast between Yes and No
>
> 2. Code AA as time-dependent covariate with all pre-DC intervals
> classified as “No” for post-DC AA use, and use “Missing,” “Yes,” and
> “No” for the post-discharge period.
>
> Also of note, we don’t have reliable data on AA use at admission or
> between admission and discharge.  How does that impact the coding for
> the post-discharge AA use?

to be clear, coding for time-dependency usually means that the
proportional hazards assumption is substantially violated and needs to
be rectified - there are at least 3 ways to do this - the PH
assumptiuon is often tested by examining Schoenfeld residuals. You
could model pre- and post-release from the hosptial separately, but
there probably are modifications to the Cox PH model that allow
examination of different time periods in a single modeling process - I
haven't done this, but maybe others will have better advice. SH
 | 
Pages: 1
Prev: Apply your understanding of Accounts and Statistics
Next: Dataset Comparison Using Proc Compare