Hierarchical Methods to Design Studies of Medical Test Accuracy

• Activity Number: 383
• Type: Contributed
• Date/Time: Wednesday, August 11, 2004 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Alicia Y. Toledano*+
• Companies: Brown University
• Keywords: sensitivity ; screening ; ROC curve ; hierarchical model ; diagnostic test

Abstract

Rapid advances in medical technology are leading to an increasing need for prospective, multicenter studies of medical test accuracy. Study planning is complicated by low prevalence of disease, variation in prevalence across centers, and often, variation in test accuracy across centers. This paper presents a framework for designing prospective, multicenter studies of medical test accuracy. We use hierarchical methods to address variation in prevalence and in test accuracy across institutions. We show that when there is low prevalence of disease, a design that specifies accruing a fixed number of positive subjects at each institution is more efficient than one that specifies accruing a fixed total number of subjects. We provide estimators of variation in accuracy that can be applied to pilot data to obtain values for input parameters, and normal approximations to the probability of achieving desired precision in estimating central tendencies in accuracy that can be used to reduce the number of parameter combinations evaluated with Monte Carlo methods. We illustrate use of the proposed method to design a national screening trial.