|
Activity Number:
|
132
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Monday, August 7, 2006 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #305681 |
|
Title:
|
Confidence Intervals on Intermediate Precision Measures in Analytical Method Validations and Transfers
|
|
Author(s):
|
Richard Burdick*+ and Shea Watrin
|
|
Companies:
|
Amgen Inc. and Amgen Inc.
|
|
Address:
|
4000 Nelson Road, Longmont, CO, 80503,
|
|
Keywords:
|
variance components ; random models ; mixed models
|
|
Abstract:
|
An important characteristic used to evaluate analytical method validations and transfers is intermediate precision (IP). IP is a measure of variation when analysts perform the method on different days or with different equipment. It is often appropriate to consider analyst as a random effect in such studies. However, it is usually not possible to sample more than two or three analysts, and resulting confidence intervals on IP are too wide to be useful. In this study, we consider three methods to model the analyst effect: random effect, fixed effect, and pooled with repeatability. We develop methods for constructing confidence intervals on IP under each condition using generalized inference and closed-form approximations. Recommendations are provided for selecting an appropriate method based on computer simulations.
|
