topic PROC LOGISITIC, assigning a reference in Statistical Procedures

PROC LOGISITIC, assigning a reference

deengyn — Wed, 27 Oct 2021 16:09:38 GMT

Outcome of interest is any complications within 30 days of surgery.

Predictors include patient and disease characteristics, almost all of which are binary.

/*Creating dummy variables*/
data ga_1;
set groin;
/*OUTCOME: any complications*/
if comp ^= 'yes' then comp_num = 1;
else if comp = 'yes' then comp_num = 2;
else if comp = . then comp_num = 3;
label comp_num = 'Any complications categorical';

/*PREDICTOR: age*/
age = round((sx_date - birthdate)/365, 2.);
label age = 'Age at surgery';
/*PREDICTOR: sex*/
if sex = 'M' then sex_num = 1;
else if sex = 'F' then sex_num = 2;
label sex_num = 'Sex';
/*PREDICTOR: smoker*/
if smoker = 'no' then smoker_num = 1;
else if smoker = 'yes' then smoker_num = 2;
else if smoker = . then smoker_num = 3;
label smoker_num = 'Smoker categorical';
/*PREDICTOR: cardiac disease*/
if card_dis = 'no' then card_num = 1;
else if card_dis ='yes' then card_num = 2;
else if card_dis = . then card_num = 3;
label card_num = 'Cardiac/vascular categorical';run; 


proc format;
value sex 1 = 'Male' 2 = 'Female';

value smoker 1 = 'N' 2 = 'P' 3 = 'Missing';
value card_dis 1 = 'N' 2 = 'P' 3 = 'Missing';

value comp 1 = 'no' 2 = 'yes' 3 = 'Missing';

run;

I would like to make a non-smoker (SMOKER = NO) as the reference, but it's not working out as I'd hope it to. I'd like to see the OR between non smoker and smoker.

proc logistic data = ga_2 descending;/
class sex_num (param = ref ref = first)
smoker_num (param = ref ref = first)
 card_num (param = ref ref = first)

model comp_num (event = '1') = age sex_num
smoker_num   card_num

/ clodds = wald orpvalue;
output out = out p = new;
format sex_num sex.
smoker_num smoker.  card_num card_dis.

run;

Re: PROC LOGISITIC, assigning a reference

StatDave — Wed, 27 Oct 2021 16:21:25 GMT

Remove REF=FIRST after SMOKER_NUM in your CLASS statement, or change it to explicitly specify a reference level, such as REF="N". Note that the formatted value is specified, not the internal, numeric value.

Re: PROC LOGISITIC, assigning a reference

Reeza — Wed, 27 Oct 2021 16:24:45 GMT

In SAS, missing is the lowest so is the first technically.

You'll need to set the reference to the No value, using the formatted value.

class sex_num ( ref = first)
smoker_num ( ref = 'N')
 card_num (ref = first) / param=ref;

FYI - your code has some syntax errors such as missing semicolons but I'm assuming this is just an example and not your actual code.

@deengyn wrote:

Outcome of interest is any complications within 30 days of surgery.

Predictors include patient and disease characteristics, almost all of which are binary.

/*Creating dummy variables*/
data ga_1;
set groin;
/*OUTCOME: any complications*/
if comp ^= 'yes' then comp_num = 1;
else if comp = 'yes' then comp_num = 2;
else if comp = . then comp_num = 3;
label comp_num = 'Any complications categorical';

/*PREDICTOR: age*/
age = round((sx_date - birthdate)/365, 2.);
label age = 'Age at surgery';
/*PREDICTOR: sex*/
if sex = 'M' then sex_num = 1;
else if sex = 'F' then sex_num = 2;
label sex_num = 'Sex';
/*PREDICTOR: smoker*/
if smoker = 'no' then smoker_num = 1;
else if smoker = 'yes' then smoker_num = 2;
else if smoker = . then smoker_num = 3;
label smoker_num = 'Smoker categorical';
/*PREDICTOR: cardiac disease*/
if card_dis = 'no' then card_num = 1;
else if card_dis ='yes' then card_num = 2;
else if card_dis = . then card_num = 3;
label card_num = 'Cardiac/vascular categorical';run; 


proc format;
value sex 1 = 'Male' 2 = 'Female';

value smoker 1 = 'N' 2 = 'P' 3 = 'Missing';
value card_dis 1 = 'N' 2 = 'P' 3 = 'Missing';

value comp 1 = 'no' 2 = 'yes' 3 = 'Missing';

run;

I would like to make a non-smoker (SMOKER = NO) as the reference, but it's not working out as I'd hope it to. I'd like to see the OR between non smoker and smoker.

proc logistic data = ga_2 descending;/
class sex_num (param = ref ref = first)
smoker_num (param = ref ref = first)
 card_num (param = ref ref = first)

model comp_num (event = '1') = age sex_num
smoker_num   card_num

/ clodds = wald orpvalue;
output out = out p = new;
format sex_num sex.
smoker_num smoker.  card_num card_dis.

run;