http://www.ncbi.nlm.nih.gov/pubmed/9584027

The Drescher and Schill paper addresses some especially complex arithmetic for computing AR from case-control. I'll presume that it is correct arithmetic. The paper doesn't address whether such a computation is sensible from any but a computational standpoint. Have a look at the attached article from seven years later. I would urge you, also, to find literature about AR in the 2000's.

From: Statistics Advisory Group (CDC-SAG)

Subject: Computation of Attributable Risks and their Confidence Interval via logistic regression

I am looking for a SAS program to compute Attributable Risks (AR) and their Confidence Interval (CI) via logistic regression in a case-control study. An article "Attributable risk Estimation from case-control data via Logistic Regression" published in Biometrics 47, 1247-1256, 1991 described the computation of AR and their confidence interval in GLIM. I am not familiar with GLIM and would greatly appreciate hearing from you if you have worked or know somebody who have worked in this area using SAS. If you have any references on methods to compute AR and CI via logistic regression using SAS, please let me know.

Thanks!