sas >> Cox model assumption check

by Larry, Raymond » Tue, 20 Dec 2005 22:49:53 GMT

I need to check the assumption of Cox model for the treatment group,
which is the main variable of interest, and there are also quite a lot
covariates in the data set. The follow-up time (survival time) is days
from randomization to either diagnosis of disease or censoring.

Is there a way that I can perform this assumption check in SAS? I
thought I can do a plot of residual of survivial time against time?
Will this work?

I appreciate any input from you folks!!!

sas >> Cox model assumption check

by davidlcassell » Wed, 21 Dec 2005 07:16:58 GMT

Before I get to PROC PHREG, I usually check the proportional hazards

If the proportional hazards assumptionl is appropriate, then the estimates
of the
log(-log(estimated survival distribution)) plotted against the log(TIME)
should be parallel lines.
If the Weibull model is a good fit, then these lines should be pretty
straight too.
If you have SAS 9.1 then you can plot this out using ODS Statistical

Express yourself instantly with MSN Messenger! Download today - it's FREE!

sas >> Cox model assumption check

by jweedon » Wed, 21 Dec 2005 07:56:16 GMT

On 20 Dec 2005 06:49:53 -0800, "Larry, Raymond" < XXXX@XXXXX.COM >

Which assumption are you referring to? If you mean proportional
hazards, the usual methods are to introduce terms involving
interactions of time or log(time) & the covariates. You may want to
categorize the time variable.

If these terms are important you'll want to retain them.


Similar Threads

1. PROC GLIMMIX--How to Check Linearity Assumption

2. Does SAS ETL Studio have the capability to check in/check out models

3. Statistics question - zero inflated models, skewed distributions, assumptions

An e-mail off list reminded me of some things I had forgotten to

It also reminded me that I had violated one of my own rules and not
given context

The data involves drug injectors.  There are 2 DVs: 'number of people
you loaned needles to after using them" and 'number of people you got
needles from after using them'.  The IVs include sociodemographics, and
frequency of use of various drugs by various methods.  The two DVs have
very similar distributions; it seems likely that methods which work with
one will work with the other

When data are so right-skew, one thinks of a log transform.  I did try
this (adding one before taking the log).  However, in this case, I am
not sanguine about using the results.  First, the DV in raw form has a
natural meaning.  The log does not.  Second, there is no easy way to
transform the results back to the raw scale.  Third, and central, it is,
to some extent, the outliers that are important.  People who share
needles extensively contribute greatly to spread of HIV and other
viruses.  This effect MAY be even greater than a linear one (see
extensive works on how risk networks operate) but, in any case, I'd like
to include those people.  I realize that the numbers they gave may be
guesses, but they probably have some relation to reality (i.e., someone
who says he shared needles with 200 people may have actually done so
with 198, or 202, or even 155, but probably NOT 500 or 50).

Thanks again


Peter L. Flom, PhD
Assistant Director, Statistics and Data Analysis Core
Center for Drug Use and HIV Research
National Development and Research Institutes
71 W. 23rd St
New York, NY 10010
(212) 845-4485 (voice)
(917) 438-0894 (fax)

4. Statistics question - zero inflated models, skewed distributions, assumptions etc

5. AW: Question with time-varying cox model output

6. Question with time-varying cox model output

7. cox model and multicollinearity


I am doing a cox model with time varying covarites, but several of my
variables are highly correlated. Is there a good way to deal with
multicollinearity in Cox models?

Another question is: In cox models with time varying covariates, it is
neccesary and possible to do fixed effects for time?

Thank you very much in advance.


8. subset variables in COX proportional Model