topic Re: which variable should be censored - cox proportional in SAS Programming

which variable should be censored - cox proportional

LarissaW — Wed, 06 Sep 2023 13:06:22 GMT

Hi, I'm cleaning a cohort dataset and want to use Cox proportional to analyze. However, I'm confused about assigning the censor data in the coding.

The study lasted 3000 days. The exposure is high salty food and the outcome is the death from stomach cancer. Among the 100,000 participants, 12 patients eventually died from stomach cancer, 38 patients died from other diseases, and the remaining were alive at the end of the study. I'd like to use the cox proportional hazard model to investigate the association between high salty food and the mortality of stomach cancer.

Based on my understanding, we should assign "38 patients died from other diseases" censored (0) , and all the others uncensored (1). But how do we differ the result of dying by stomach cancer and still alive? Will the SAS automatically separate them?

this is the code I use:

proc phreg data=stomach;
class salty(ref="0") bmi(ref="0") gender(ref="0") / param=ref;
model lenfol*censor(0) = salty bmi gender year los;
strata=age;
run;

Could you please check if I correctly understand the model and use the correct codes? Thank you!

Re: which variable should be censored - cox proportional

Reeza — Wed, 06 Sep 2023 15:16:56 GMT

I think you want a competing risks model?

https://support.sas.com/resources/papers/proceedings18/2159-2018.pdf

Re: which variable should be censored - cox proportional

LarissaW — Wed, 06 Sep 2023 17:08:34 GMT

Hey, thank you for the information. However, I don't think the competing risk model applied to my study. There is no competing relationship between patients alive and dead from stomach cancer. The question I have is I am confused with the define of the censored and uncensored groups here. I confirm that patients who died from other causes should be censored but I have no idea if patients survive until the end of the study belong to censored group or uncensored group. The event I am interested in here is the death from stomach cancer.

Re: which variable should be censored - cox proportional

Reeza — Wed, 06 Sep 2023 17:21:57 GMT

People alive until the end are also censored then, competing risks doesn't necessarily mean competing against each other, but dying for another cause is a competing risk. It is common to ignore this though, competing risks is an analytics level above.