This paper investigates estimating and testing treatment effects in randomized control trials where imperfect diagnostic device is used to assign subjects to treatment and control group(s). The paper focuses on pre-post design and proposes two new methods for estimating and testing treatment effects. Furthermore, methods for computing sample sizes for such design accounting for misclassification of the subjects are devised. The methods are compared with each other and with a traditional method that ignores the imperfection of the diagnostic device. In particular, the likelihood-based approach shows a significant advantage in terms of power, coverage probability and, consequently, in reduction of the required sample size. The application of the results are illustrated with data from an aging trial for dementia and data from electroencephalogram (EEG) recordings of alcoholic and non-alcoholic subjects.
|Number of pages||18|
|Journal||Statistics in Medicine|
|State||Published - Dec 20 2016|
Bibliographical noteFunding Information:
The authors would like to express their gratitude to the three anonymous reviewers for critically reading the manuscript and providing valuable suggestions that have led to a significant improvement from the original version. The authors are also grateful to the editor and associate editor for their efficient handling of the manuscript. This publication was made possible by a grant from the National Institute of General Medical Sciences (5 U54 GM104944) from the National Institutes of Health.
Copyright © 2016 John Wiley & Sons, Ltd.
- diagnostic accuracy
- expectation–maximization algorithm
- mixture of normals
- negative predictive value
- positive predictive value
- sample size
ASJC Scopus subject areas
- Statistics and Probability