Concerns about composite reference standards in diagnostic research

Composite reference standards are used to evaluate the accuracy of a new test in the absence of a perfect reference test. A composite reference standard defines a fixed, transparent rule to classify subjects into disease positive and disease negative groups based on existing imperfect tests. The acc...

Full description

Saved in:
Bibliographic Details
Published in:BMJ (Online) Vol. 360; p. j5779
Main Authors: Dendukuri, Nandini, Schiller, Ian, de Groot, Joris, Libman, Michael, Moons, Karel, Reitsma, Johannes, van Smeden, Maarten
Format: Journal Article
Language:English
Published: England BMJ Publishing Group LTD 18.01.2018
Subjects:
Accuracy
Bias
Chlamydia
Chlamydia trachomatis
Classification
Diagnostic tests
Disease
Medical diagnosis
Patients
Tuberculosis
ISSN:0959-8138, 1756-1833, 1756-1833
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Composite reference standards are used to evaluate the accuracy of a new test in the absence of a perfect reference test. A composite reference standard defines a fixed, transparent rule to classify subjects into disease positive and disease negative groups based on existing imperfect tests. The accuracy of the composite reference standard itself has received limited attention. We show that increasing the number of tests used to define a composite reference standard can worsen its accuracy, leading to underestimation or overestimation of the new test’s accuracy. Further, estimates based on composite reference standards vary with disease prevalence, indicating that they may not be comparable across studies. These problems can be attributed to the fact that composite reference standards make a simplistic classification and then ignore the uncertainty in this classification. Latent class models that adjust for the accuracy of the different imperfect tests and the dependence between them should be pursued to make better use of data
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
content type line 23
ISSN:0959-8138
1756-1833
1756-1833
DOI:10.1136/bmj.j5779