Interpretation of sas output with Proc logistic

[Kyra]

Hi,

i am trying to find association between post op intolerance (PO_Intolerance__Y_N_1)- dependent variable and pattern of sewing during surgery (LOCATION_OS1)- exposure. Both have two categories.

PO_Intolerance__Y_N_1: yes vs no

LOCATION_OS1: entire vs partial

SAS code i used is:

proc logistic data=red.sleeve7;
class PO_Intolerance__Y_N_1(ref="N") LOCATION_OS1(ref="partial");
model PO_Intolerance__Y_N_1= LOCATION_OS1/scale=none aggregate;
run;

OR i received is :

Odds Ratio Estimates
Effect	Point Estimate	95% Wald Confidence Limits
LOCATION_OS1 entire vs partial	0.182	0.107	0.310

please see attched for details of output.

I assumed that post op intolerance will be more likely in entire sewing group vs partial sewing group.

But the results are opposite.

Please let me know if it sample size played a role? numbers in each group played a role? or it is statistically sound to report the results?

Thanks.

[Reeza]

You should add the PARAM=REF option.

[Kyra]

Hi!

with the code below;

proc logistic data=red.sleeve7;
   class PO_Intolerance__Y_N_1(ref="N") LOCATION_OS1(ref="partial")/param=reference;
   model PO_Intolerance__Y_N_1= LOCATION_OS1/scale=none aggregate;
   run;

output remains the same. Please see attched for details.

SAS Output

Odds Ratio EstimatesEffect Point Estimate 95% WaldConfidence LimitsLOCATION_OS1 entire vs partial

0.182

0.107

0.310

[Reeza · Accepted solution]

Check your log. Specifically look at the notes to see what was modeled as the outcome, perhaps that is coded reverse to what you expect.

You can specify the outcome desired by using the effect option.

model varName (effect='no') = .... ;

[Kyra]

Hi!

Thank you for your help. I checked my dataset and there was mistake in coding. OR is now greater than 1 which i expected and it is insignificant which i believe is due to small numbers in Post op intolerance yes group.

 the output now is

SAS Output

Response ProfileOrderedValue PO_Intolerance__Y_N_1 TotalFrequency12

N	340
Y	13

SAS Output

Odds Ratio EstimatesEffect Point Estimate 95% WaldConfidence Limits LOCATION_OS1 entire vs partial

1.405

0.451

4.383

[Reeza]

You have an imbalance in your sizes Y=13 is significantly smaller than the No group. You may want to consider adding a prior or oversampling.

[Kyra]

I am sorry but i do not know what a prior or oversampling is. Do you have any material on that i can read. Thanks for your help.

[Reeza]

https://gking.harvard.edu/category/research-interests/methods/rare-events

https://statisticalhorizons.com/logistic-regression-for-rare-events

Search for : "logistic regression rare events" for more material. 

For SAS specific articles, search on lexjansen.com

---

@Kyra wrote:

I am sorry but i do not know what a prior or oversampling is. Do you have any material on that i can read. Thanks for your help.

---