how to deal with small sample size in one of the independent variabl in logistic regression?

[bhr-q]

Hello all,

As there is only one patient in one of the categories below, is that possible to run the logistic and get the odds ratio? Or should I leave this objective?

Not: there is no possibility to combine the groups.

Antibiotic    freq

a                  47

b                 20

c                 56

d                  3

e                 57

f                    1 

proc logistic data=tmp;
  class  antibiotic (ref='c')/ param=ref ;
  model Bf_u =  antibiotic;
run;

when I run my code, I will get the below results:

Odds Ratio Estimates
Effect	Point Estimate	95% Wald Confidence Limits
a vs c	1.023	0.461	2.269
b vs c	0.530	0.169	1.665
d vs c	1.029	0.483	2.193
e vs c	3.182	0.272	37.204
f vs c	>999.999	<0.001	>999.999

Thank you in advance!

[Ksharp]

Yes . you could "leave this objective".
Or try
1) penalty MLE by firth option

model Bf_u = antibiotic/ firth ;
2)exact logistic regression

model Bf_u = antibiotic;
exact antibiotic;

[bhr-q]

thanks so much for your help, I did this way, and received e a too big confidence interval for the last category which I think it is not appropriate,

something like this:

	OR	Confidence Limits
f	4.653	0.050	435.226

[Ksharp]

Yeah. That is because you have a very big std err for OR due to one obs for F .
I think you could "leave this objective out from model" if you do not want so wide CL .