topic Dichotomising a continuous variable for strata in Cox regression/ Kaplan Meier survival analysis in Statistical Procedures

Dichotomising a continuous variable for strata in Cox regression/ Kaplan Meier survival analysis

ammarhm — Sat, 12 Sep 2020 05:18:59 GMT

Hi everyone,

i have a dataset containing about 7000 observations, I am interested in a certain events (event), the incidence of which seems to increase with age. The increment is relatively slow at younger ages but increases more steeply with older age.

What I am trying to do is to find an optimal age cut off as a strata predicting the event of interest in survival analysis. Basically I am looking for the age at which the difference in hazards ratios is largest, of if you want the age cut off that separates the Kaplan Meier strata curves most.

Is there a way of doing this?

Thank you and appreciate your help.

Re: Dichotomising a continuous variable for strata in Cox regression/ Kaplan Meier survival analysis

PGStats — Sat, 12 Sep 2020 20:25:24 GMT

The brute force method would be something like:

data testData;
set myData;
do cutoffAge = 25 to 55 by 5;
	strata = age > cutoffAge;
	output;
	end;
run;

proc sort data=testData; by cutoffAge strata; run;

proc lifetest data=testData ... ;
by cutOffAge;
strata strata;
...
ods output ... ; /* capture analysis statistics for each cutoff age*/
run;

You could then choose your prefered age cutoff. Note that any such searching method will likely create a bias in favor of your hypothesis.

Re: Dichotomising a continuous variable for strata in Cox regression/ Kaplan Meier survival analysis

ammarhm — Sat, 12 Sep 2020 22:30:59 GMT

Thanks @PGStats
Appreciate your help.
The issue is that the suggested solution will be based on subjective assessment of the KM graphs, I was hoping there would be an objective statistical procedure or way, maybe something similar to Youden index to solve this kind of problems?
Thanks once again, really appreciate your help.