topic Re: Sensitivity,Specificity and Youden index in SAS Procedures

Sensitivity,Specificity and Youden index

Jems — Wed, 22 Jul 2015 13:31:13 GMT

How to calculate the sensitivity,specificity and Youden index for 18 factors,and its combination (6 factors in each) with an outcome measure.

Re: Sensitivity,Specificity and Youden index

PhilC — Wed, 22 Jul 2015 20:57:41 GMT

PROC MEANS/SUMMARY can be used to calculate a confusion matrix.

Re: Sensitivity,Specificity and Youden index

PhilC — Wed, 22 Jul 2015 20:58:57 GMT

We can communicate a solution more effectively if you give us a sample of your data. Then we can speak in terms of your data's variables names.

Re: Sensitivity,Specificity and Youden index

Jems — Thu, 23 Jul 2015 01:58:23 GMT

Thank you for your kind reply.I have added sample of my data.Here 'malnut'(malnutrition) is the outcome measure and others like chew,weilos......cc are the independent variables.I want to get the sensitivity,specificity and youden index of each of these variables and their combination having 6 factors(A+B+C+D+E+F) like 'chew,weilos,mobil,stress,dement,bmi' with the outcome measure.

Re: Sensitivity,Specificity and Youden index

PhilC — Thu, 23 Jul 2015 13:43:40 GMT

So you want sensitivity and specificity with respect to what predictive model? Is 'MALNUT' the result of your predictive model, or is it the known state of the patient? If you have established a predictive model then its outcome would be compared with the clinical data, this is how I know this type of statistics. I can refer you to https://en.wikipedia.org/wiki/Confusion_matrix. I don't know anything in SAS_STATS that will do this, but perhaps this question should be posed in that forum: https://communities.sas.com/community/support-communities/sas_statistical_procedures .

Re: Sensitivity,Specificity and Youden index

Jems — Thu, 23 Jul 2015 15:24:38 GMT

Thank you for your consistent support. Ya 'Malnut' is the known state of the patient,which indicates malnutrition.Here we wish to find the sensitivity and specificity of each factor and 'Malnut' status 0=not malnourished & 1=malnourished.Thanks for your directions.

Re: Sensitivity,Specificity and Youden index

Reeza — Thu, 23 Jul 2015 17:01:13 GMT

Run a proc freq between each set of variables. That will create your cross tabs and you can get the number from there to calculate your specificity/sensitivity.

https://en.wikipedia.org/wiki/Sensitivity_and_specificity

Re: Sensitivity,Specificity and Youden index

PhilC — Thu, 23 Jul 2015 17:07:39 GMT

Actually PROC MEANS is what I've used and it involved another step, but PROC TABULATE can be used using one step.

My question about predictive models is still relevant: I believe you need to choose dichotomous values to represent your independent variables to do this correctly. If both MALNUT and another variable, _x_ are dichotomous then this code will produce a dataset with up to four rows, each corresponding to a cell of the confusion matrix. Sensitivity, specificity and Youden's index can be calculated from these values.

PROC TABULATE DATA=Given_DS
              OUT=Want_X_vs_Malnut ;
CLASS _x_     ;
CLASS Malnut ;
TABLE _x_,Malnut * N     ;
;
RUN;

Re: Sensitivity,Specificity and Youden index

PhilC — Thu, 23 Jul 2015 17:19:39 GMT

I'll add if you want to do this with more than one variable you can do this:

data work.given;
do i=1 to 300; drop i;
    A=ranuni(0)>0.5;
    W=ranuni(0)>0.5;
    x=ranuni(0)>0.5;
    Y=ranuni(0)>0.5;
    Z=ranuni(0)>0.5;
    output;
end;
stop;
run;
PROC TABULATE DATA=&syslast
              OUT=work.Want ;
CLASS A W--Z ;
TABLE W--Z,A * N     ;
;
RUN;

Re: Sensitivity,Specificity and Youden index

Jems — Fri, 24 Jul 2015 02:02:53 GMT

Thank you very much, Phil for your extremely valuable SAS codes.In my data set i have coded the independent and dependent variable(s) as dichotomous ones; 0=no & 1=yes.I have 190 rows of 18 independent dichotomous variables and 1 dependent dichotomous variable.

Re: Sensitivity,Specificity and Youden index

Reeza — Fri, 24 Jul 2015 05:00:02 GMT

If you have your answer please mark the question answered

Re: Sensitivity,Specificity and Youden index

PhilC — Fri, 24 Jul 2015 12:45:35 GMT

Well we technically did not answer the question pertaining to the combinations of factors... so... what can we do?

Re: Sensitivity,Specificity and Youden index

Jems — Sat, 25 Jul 2015 04:04:06 GMT

Thank you for the support offered !.I just wanted to get sensitivity and specificity.proc freq procedure produces that but its mixed with other values,is there any way to cull out only those 2 items from that table.

Re: Sensitivity,Specificity and Youden index

Jems — Sat, 25 Jul 2015 04:10:05 GMT

Definitely i will do so :smileycool:

Re: Sensitivity,Specificity and Youden index

PhilC — Mon, 27 Jul 2015 13:58:21 GMT

Read this SAS support note: 24170 - Estimating sensitivity, specificity, positive and negative predictive values, and other statistics

Run that code first, this code can inform you about obtaining sensitivity and specificity as values.

PROC TABULATE
       DATA=WORK.FATCOMP
       OUT =WORK.STABSummaryTablesFATCOMP(LABEL="Summary Tables for WORK.FATCOMP");
CLASS Test / ORDER=UNFORMATTED DESCENDING MISSING;
CLASS Response / ORDER=UNFORMATTED DESCENDING MISSING;
TABLE Test*ColPctN,Response;
FREQ Count;
RUN;

data numbers (Keep=sensitivity specificity);
do until (eof);
      set &syslast end=eof;
      if Test=1 and Response=1 then sensitivity=PctN_01;
      if Test=0 and Response=0 then specificity=PctN_01;
end;
putlog "NOTE: " sensitivity= specificity=;
output;
stop;
run;

Re: Sensitivity,Specificity and Youden index

Jems — Tue, 28 Jul 2015 09:52:44 GMT

Thank you for SAS codes,Is there any provision(code) to find the Youden Index(sensitivity+specificity-1).

Re: Sensitivity,Specificity and Youden index

PhilC — Tue, 28 Jul 2015 12:21:23 GMT

I have never used Youden's index personally. It should be easy to calculate inside the data step I provided above.

Re: Sensitivity,Specificity and Youden index

Jems — Tue, 28 Jul 2015 12:35:22 GMT

Yeh i will make use of that possibility.Thank you very much for the support offered.