topic how to analyze logistic regression with low sample size in SAS Procedures

how to analyze logistic regression with low sample size

sms1891 — Fri, 26 Apr 2019 16:58:11 GMT

Hi all,

I have a data set with 423 pts but have only 10 pts with required outcome (Visit = 1 for Hospital visit and Visit = 0 for no hospital visit). I am trying to evaluate independent predictors for Hospital visit (Visit = 1). When I ran the logistic regression, some of the variables of interest have 95% CI values >999.999. How to handle such situations? And whether to include such variables in the logistic regression model or not?

Thank you in advance!

Re: how to analyze logistic regression with low sample size

PaigeMiller — Fri, 26 Apr 2019 17:04:27 GMT

When you have a response variable which is a rare event, often the solution is to oversample the data, and then run PROC LOGISTIC, and then transform the results back to the non-oversampled space. Here's a tutorial: http://support.sas.com/kb/22/601.html

Re: how to analyze logistic regression with low sample size

Ksharp — Sun, 28 Apr 2019 14:49:58 GMT

Try Exact Logistic Model or Penalty Logistic Model.

There are some examples in documentation.