cox Model time varying covariates and unequal repeated measurement per observation

Reply
Occasional Contributor
Posts: 5

cox Model time varying covariates and unequal repeated measurement per observation

[ Edited ]

Hi,

I'm working on a  survival analysis with repeated measurements in covariates (e.g., blood pressure, BMI). One way to deal with these covariates with repeated measurements is to treat them as time-dependent variables. An example is given here: http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_phreg_sect05...

But there are some major difference between that example and our data:

1) in that example, each subject has the same amount of repeated measurements taken at the same time, while in my data (which is medical record instead of a specially designed follow-up study) the amount and time of  measurements vary from subject to subject (e.g., blood pressure or BMI is not necessary taken on the same day for each subject in our cohort).

2) there is only one covariate with repeated measurements in the example, while we have multiple covariate with repeated measurements.

my data looks like this for the covariates:

1/1 to 3/31: BMI=unknown, blood pressure=unknown
4/1 to 4/30: BMI=25, blood pressure=unknown
5/1 to 12/31: BMI=25, blood pressure=120/80
this is for one observation. another observations couls have more or less rows with different start and end dates.

Has anyone done similar analysis and would like to cast some light.

Any help is appreciated.

Thank you

Frequent Contributor
Posts: 102

Re: cox Model time varying covariates and unequal repeated measurement per observation

There is no categorical difference between the example you provide and your scenario. Yours is just more complicated, that's all. You will need to write a compute statement where every day in your data on which covariate values change for at least one patient is treated as a 'repeated variable'. You will also need to write as many compute statements as you have time-dependent covariates. Not different, just more.

You have unknown values. As far as I know, they cannot be accommodated in the time-dependent analysis. You will have to impute or use some other missing data handling strategy.
Occasional Contributor
Posts: 5

Re: cox Model time varying covariates and unequal repeated measurement per observation

[ Edited ]

are you suggesting to brake down the intervals by day and for the entire follopwup study(3 years study)?  I think that is a good idea. Do you have  a  sample code for that? and yes  I have missing instead of unknown in my data. it was just to be clear.

Ask a Question
Discussion stats
  • 2 replies
  • 174 views
  • 0 likes
  • 2 in conversation