0
$\begingroup$

I have a set of data where I would like to do logistic regression modeling the odds of a binary outcome variable (Therapy), with Stage as an ordinal explanatory variable (0,1,2,3,4). I am using SAS. The output I have been given is this:

Odds Ratio Estimates and Wald Confidence Intervals
Odds Ratio  Estimate    95% Confidence Limits
Stage 1 vs 0    0.873   0.547   1.394
Stage 2 vs 0    2.434   0.895   6.620
Stage 3 vs 0    0.915   0.431   1.941
Stage 4 vs 0    0.356   0.132   0.961
Stage 2 vs 1    2.788   0.980   7.935
Stage 3 vs 1    1.048   0.465   2.360
Stage 4 vs 1    0.408   0.144   1.156
Stage 3 vs 2    0.376   0.113   1.249
Stage 4 vs 2    0.146   0.038   0.567
Stage 4 vs 3    0.389   0.117   1.288

If I am reporting Stage as an ordinal variable, then is it correct that I create a table like this:

Stage 1 vs 0    0.873   0.547   1.394
Stage 2 vs 1    2.788   0.98    7.935
Stage 3 vs 2    0.376   0.113   1.249
Stage 4 vs 3    0.389   0.117   1.288

Or, should I report it like this:

Stage 1 vs 0    0.873   0.547   1.394
Stage 2 vs 0    2.434   0.895   6.62
Stage 3 vs 0    0.915   0.431   1.941
Stage 4 vs 0    0.356   0.132   0.961

The variable in question, Stage, is a measure of severity of structural variation in an eye that is diagnosed based on set criteria. The differences between 1,2,3, and 4 are different, and 0 is the absence of this structural variation.

For reference, the following is my SAS code.

PROC LOGISTIC data=new;
class EyeID Therapy (ref ="0") Stage (param = ordinal) Gender (ref="M") Ethnicity (ref="C")/ param = ref;
model Therapy = Stage Gender age A1c Ethnicity;
oddsratio Stage;
run; 
$\endgroup$
1
$\begingroup$

You could use either table. The choice depends on the question you want to use the odds ratios to answer.

Adjacent Category References

Stage 1 vs 0    0.873   0.547   1.394
Stage 2 vs 1    2.788   0.98    7.935
Stage 3 vs 2    0.376   0.113   1.249
Stage 4 vs 3    0.389   0.117   1.288

What are the odds of therapy in Stage I severity compared to no structural variation (zero)? What are the odds of therapy in Stage II severity versus Stage I severity?...so on and so forth. These would answer the questions such as - are the differences in odds with each subsequent category similar, or are Stage 4 and Stage 3 both more likely to get therapy and therefore the difference (OR) is not substantial?

Single Referent Category

Stage 1 vs 0    0.873   0.547   1.394
Stage 2 vs 0    2.434   0.895   6.62
Stage 3 vs 0    0.915   0.431   1.941
Stage 4 vs 0    0.356   0.132   0.961

With single reference category we often choose the lowest or highest extreme. You might see a threshold effect here and choose to dichotomize or make a recommendation that "all patients with Stage X and above get therapy" etc.

Interesting with your results is that they show a similar pattern, where something is going on with Stage 2 being much more likely to receive therapy than Stage 0 or Stage 1 (adjacent analysis) per the point estimate. Statistically, however, your confidence intervals are large and all cross the null value of 1.0 so you may be lacking power here.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.