From: Paul Hutson <>
Subject: prediction testing
Date: Thu, 19 Apr 2001 12:00:54 -0500

Greetings, all:

I am trying to validate the use of limited (2 point) methods to predict the AUC of a drug. My study is designed to incorporate not only the two time points that for a similar treatment regimen yielded the best model, but also another 7 time points to permit the determination of the "true" AUC. I plan to use the Holford/Sheiner suggestion of using the prediction error (bias) and absolute error (RMSE, accuracy) rather than a correlation.

My question to the group is what test to use for accuracy. For the bias, I am testing the hypothesis that the mean bias is not different than zero. I am not sure what hypothesis to test for the accuracy (RMSE). I had planned to test whether the RMSE was equal to or less than the Standard Deviation of the "true" AUC in the patients. Does this seem reasonable? Are there better ways?

Paul Hutson, Pharm.D.
Associate Professor (CHS)
UW School of Pharmacy
425 N. Charter St
Madison, WI 53706-1515
Tel: (608) 263-2496
FAX: (608) 265-5421
Pager: (608) 265-7000, #7856