topic Re: Dichotomizing a continuous predictor variable in Statistical Procedures

Dichotomizing a continuous predictor variable

Gypsium — Wed, 24 Jan 2024 05:20:27 GMT

Hi all,

I am currently conducting logistic regression analysis with a continuous predictor variable. I'm interested in finding a proc solution that can automatically determine optimal cutpoints and dichotomize the continuous variable. I appreciate your help and thank you for your time!

Re: Dichotomizing a continuous predictor variable

Ksharp — Wed, 24 Jan 2024 05:58:48 GMT

Check Youden index.
@StatDave post many times about this question.
https://communities.sas.com/t5/Statistical-Procedures/Calculating-Optimal-Cutoff-Point-for-Repeated-Measurement-data/m-p/843880

https://support.sas.com/kb/30/333.html#errinvar

Re: Dichotomizing a continuous predictor variable

Ksharp — Wed, 24 Jan 2024 06:02:59 GMT

25018 - Plot ROC curve with cutpoint labeling and optimal cutpoint analysis (sas.com)

Re: Dichotomizing a continuous predictor variable

Gypsium — Wed, 24 Jan 2024 06:26:58 GMT

Thank you for sharing the link Ksharp! I will review the posts.

Re: Dichotomizing a continuous predictor variable

Ksharp — Wed, 24 Jan 2024 07:06:47 GMT

Here is the simple code to get CUTPOINT based on Youden index.

proc logistic data=sashelp.heart(obs=1000);
class bp_status;
model status=bp_status weight height / outroc=roc;
output out=out p=p;
run;

/*%rocplot(v,inpred=out, inroc=roc, p=p, id=_opty_, optcrit=youden)*/

data x;
set roc;
x=_SENSIT_;
y=1-_1MSPEC_;
youden=x+y-1;
run;

proc sql;
select _PROB_ as cutpint label='cutpoint',youden from x having youden=max(youden);
quit;