Coding time varying covariate for Cox proportional hazards regression

Lefty — Fri, 20 Sep 2024 00:46:00 GMT

Hi, I am trying to create a time-varying covariate that indicates whether someone was taking drug 1 (drugcat=0), drug 2 (drugcat=1), or drug 1+ drug 2 (drugcat=2). I want to run a Cox proportional hazards regression (proc phreg) that incorporates the fact that the drugs people are taking changes over time. Here is a simplified version of the data I have:

ID	Drug1StartDate1	Drug1EndDate1	Drug1StartDate2	Drug1EndDate2	Drug2StartDate1	Drug2EndDate1	Drug2StartDate2	Drug2EndDate2	age	sex	censor
1	June 1 2011	July 1 2011							50	0	0
2	Jan 1 2012	Dec 31 2012			June 1 2012	Dec 31 2012			40	1	1
3	Jan 1 2010	June 15 2010	July 1 2010	Dec 1 2010	Feb 1 2010	June 30 2010	Sept 15 2010	Dec 15 2010	60	1	0

I want the start and stop times in days that they are in each drug category. It’s not so hard when someone falls into only one category (ID=1) or two categories (ID=2) but it gets really tricky when people move between the categories (ID=3 is on only drug 1 from Jan 1-Jan 31, then on both drugs from Feb 1 to June 15, then on only drug 2 from June 16-30, then back to drug 1 from July 1 to Sept 14, then on both drugs from Sept 15 to Dec 1, then only on Drug 2 from Dec 2 to Dec 15. Some people have up to 20 start/stop dates so I can't really muscle through each possible variation.
Here is what I'd like my data to look like:

ID	Start	Stop	Drugcat	age	sex	censor
1	0	30	0	50	0	0
2	0	152	0	40	1	0
2	152	365	2	40	1	1
3	0	31	0	60	1	0
3	31	167	2	60	1	0
3	167	182	1	60	1	0
3	182	257	0	60	1	0
3	257	334	2	60	1	0
3	334	365	1	60	1	0

Ultimately I’ll use this code to run my regression (where drugcat=0 means drug 1 alone, drugcat=1 means drug 2 alone and drugcat=3 means both drugs at the same time):

Proc phreg data=want;

Class drugcat;

Model (start, stop)*censor(0)=drugcat age sex/rl;

Run;

I’ve tried using arrays to get the data I want but it quickly falls apart when someone switches between the drug categories more than once.

Can anyone tell me a better way to get to my goal? Thanks in advance!

Re: Coding time varying covariate for Cox proportional hazards regression

quickbluefish — Sat, 28 Sep 2024 02:10:27 GMT

Hi - sounds like you need a "counting process" dataset as input for PHREG. Are you still looking for a way to create this? I have a macro for it - let me know.

topic Re: Coding time varying covariate for Cox proportional hazards regression in Statistical Procedures

Coding time varying covariate for Cox proportional hazards regression

Re: Coding time varying covariate for Cox proportional hazards regression