PROC LOGISTIC "All Observations have the same response. No statistics are computed"

[deengyn]

I have looked at the previous posts, and have not been able to make sense of it. 

My data set is made of binary variables for a list of patient, disease, health characteristics (i.e. sex, diabetes, cardiac disease). Some are continuous variables (i.e. operation time in minutes) and others are categorical (i.e. tumour location).

The outcome was any complications post-op.

There is a total of 24 variables and 141 lines of data. 

Variable	Number of outputs available	Number of outputs missing
1.      Sex	141	0
2.      Thickness, categorical	141	0
3.      Histology, subtype	141	0
4.      Mitosis, binary	68	73
5.      Ulceration	78	63
6.      LVI	69	72
7.      Regression	63	78
8.      Smoker	116	25
9.      Diabetes	141	0
10.   Cardiac disease	141	0
11.   Hypothyroidism	141	0
12.   Staff/surgeon	141	0
13.   Presentation	141	0
14.   Indication	141	0
15.   Dissection type	139	2
16.   OP time	133	8
17.   Blood loss	91	50
18.   Number of drains left	141	0
19.   Duration of immobility	79	62
20.   Duration of post-op AB use	80	61
21.   Total hospital stay	80	61
22.   Total LN	140	1
23.   Total positive LN	141	0
24.   ER revisit	141	0

I've done correlation analyses; most had a RH0 of ~0.3-0.4. I couldn't perform VIF/tolerance tests, oddly. 

So to test the model, I did both univariate and multivariate analyses. An example of my code...

Where sex = 1 is male, where comp = 1 is any complication.

proc logisitic data = groin descending; 
class sex (ref = '0')/param = ref; 
model comp = sex/clodds = wald orpvalue; run;

And with each variable, the output states, "All Observations have the same response. No statistics are computed".

[maguiremq]

What does a PROC FREQ on `comp` tell you?

[deengyn]

Thank you and good catch, it seemed like the code didn't run through for comp. And it was all 0's. 

[ballardw · Accepted solution]

Your list of variables and those counts does not show COMP or anything that looks like the outcome "complications post-op".

You need to look closely at that variable and possibly in combination with your SEX variable to see if different levels of sex had any variation in Comp.

Such as

Proc freq data=groin;

   tables sex*comp/list;

run;

If there are no differences in COMP (when not missing) within levels of Sex then that is the reason.

[deengyn]

Ah, I see thank you!