topic Re: Age adjusted Median estimation in Statistical Procedures

Age adjusted Median estimation

AllEpi — Fri, 31 Jan 2025 12:52:43 GMT

Hello

I am looking to distribution of biomakers. I am interested to estimate age and sex adjusted medians for these variables. Is there a way to do it? This paper has decscribed it but unfortunately the SAS code is not provided (https://pubmed.ncbi.nlm.nih.gov/19028825/).

Any support will be appreciated.

Best wishes
AllEpi

Re: Age adjusted Median estimation

StatDave — Fri, 31 Jan 2025 15:17:54 GMT

Look into using PROC QUANTREG.

Re: Age adjusted Median estimation

Ksharp — Sat, 01 Feb 2025 06:45:08 GMT

Look into @Rick_SAS blogs:
https://blogs.sas.com/content/iml/2018/08/13/quantile-regression-chess-ratings.html
https://blogs.sas.com/content/iml/2018/08/06/score-quantile-regression-sas.html
https://blogs.sas.com/content/iml/2013/04/17/quantile-regression-vs-binning.html

Re: Age adjusted Median estimation

AllEpi — Mon, 03 Feb 2025 12:14:36 GMT

Thank you for the helpful posts. My question was related to getting adjusted medians. I am aware of how to get quartile with confidence intervals from a post by @Rick_SAS (https://blogs.sas.com/content/iml/2017/02/22/difference-of-medians-sas.html). However, that gives me unadjusted medians as there is nothing on the right side of the model!.

There is also option to predict the quantiles. Using this approach, I proceed to calculate the adjusted (predicted medians, using q=0.5). For example in the sashelp.heart study if I am interested to know the median of the diastolic blood pressure across the quartiles of the systolic pressure, this is how I got the medians adjusted for age at start and sex.

*First I make quartiles of the systolic blood pressure;

proc rank data=sashelp.heart groups=4 out=heart;
var Systolic;
ranks sys;
run;

*Then I get age and sex adjusted predicted diastolic blood pressure;
proc quantreg data=heart ci=sparsity/iid algorithm=interior(tolerance=1.e-4);
class sex;
model diastolic =ageatstart sex / quantile=0.5;
output out=predictquant p=predquant;
run;

"Then I get adjusted medians (from predicted systolic blood pressure called predquant) across the quartiles of systolic blood pressure (sys), I keep original variable to get the difference in medians;

proc means data=predictquant median;

class sys;
var diastolic predquant ;
run;

I can see that the difference between median of the original variable and predicted (which is adjusted for age and sex). On a side note, to cross check my result when I added used height only for adjustment (assuming it will be less related to diastolic blood pressure), I get median of the original as well as predicted diastolic blood pressure as same.

Now the question is: Is this a correct approach to get adjusted medians (The question is not about whether this is a correct data anlaysis approach!)

Thanks in advance

Re: Age adjusted Median estimation

Ksharp — Tue, 04 Feb 2025 01:09:48 GMT

"adjusted medians " for what ? or how to get it ?

If you means it like Linear Regression Model , you can use LSMEANS or ESTIMATE to get it ,as you showed Rick blog:

https://blogs.sas.com/content/iml/2017/02/22/difference-of-medians-sas.html

proc quantreg data=heart ci=sparsity algorithm=interior(tolerance=1.e-4);
class sex;
model diastolic =ageatstart sex / quantile=0.5;
estimate 'Adjusted Medians of Female' sex 1 0 / CL;
estimate 'Adjusted Medians of Male' sex 0 1 / CL;
run;

Re: Age adjusted Median estimation

AllEpi — Tue, 04 Feb 2025 08:26:16 GMT

Thank you for the suggestion. To clarify my question, this is about how to get the adjusted medians

in SAS. Now I am interested in getting Medians of the diastolic blood pressure across the quantiles of

of the systolic blood pressure but adjusted for ageatstart and sex in the sashelp.heart dataset.

When I run your code I get "Non-est". I get the same message, if I change the variable. Any further toughts on this?

Re: Age adjusted Median estimation

StatDave — Tue, 04 Feb 2025 15:16:57 GMT

The analysis you showed with PROC RANK and PROC QUANTREG looks reasonable to me.

Re: Age adjusted Median estimation

Ksharp — Wed, 05 Feb 2025 01:24:04 GMT

Opps. I missed a INTERCEPT term to score a set of X variables when using ESTIMATE.

proc quantreg data=sashelp.heart ci=sparsity algorithm=interior(tolerance=1.e-4);
class sex;
model diastolic =systolic ageatstart sex / quantile=0.5;
estimate 'Adjusted Medians of diastolic when systolic=148,ageatstart=33,sex=Female' 
         intercept 1 systolic 148  ageatstart 33 sex 1 0 / CL;
run;

Here 148 is 75 percentile of systolic ,33 is 50 percentile of ageatstart.

Re: Age adjusted Median estimation

AllEpi — Wed, 05 Feb 2025 14:58:54 GMT

Thank you so much. It was really very helpful. I have two short questions i) if systolic is a categorized in 4 levels (quartile). How can I specify each level? ii) How is this approach different from the predicted approach (the one that I showed in the original post?). Kind regards

Re: Age adjusted Median estimation

Ksharp — Thu, 06 Feb 2025 01:09:55 GMT

"if systolic is a categorized in 4 levels (quartile). "
No. Here I take systolic as a continuous variable not category variable, if you want bin it as a category varible with four levels. you need put it in CLASS statment.
proc quantreg data=sashelp.heart ci=sparsity algorithm=interior(tolerance=1.e-4);
class sex systolic ;
model diastolic =systolic ageatstart sex / quantile=0.5;
estimate 'Adjusted Medians of diastolic when systolic=1,ageatstart=33,sex=Female'
intercept 1 systolic 1 0 0 0 ageatstart 33 sex 1 0 / CL;
run;

" How is this approach different from the predicted approach"
Honestly, I don't understand your code . It looks like you take systolic as a category variable , but I take it as a continuous variable.
Also I have no time to go through the paper you posted .