Gestational Diabetes validation data study: an example of testing differences in sensitivity and specificity of two diagnostic methods using non-random sampling in a paired study design Public

Abstract

Abstract

Gestational Diabetes validation data study: an example of testing differences in sensitivity and specificity of two diagnostic methods using non-random sampling in a paired study design

By Ganzhong Tian

Background: Comparing the accuracy of two diagnostic methods is a common problem in public health. This is especially the case when using a third ‘gold standard' to determine patients' true disease status is either too expensive or time-consuming. For this kind of problem, it is highly desirable to have an efficient way of sampling and comparing the diagnostic properties of the two diagnostic methods.

Methods: In this study, we used a CDC study of Gestational Diabetes data as an example and considered an efficient design for validation sampling, which gives us more useful information regarding the diagnostic properties of two diagnostic methods for Gestational Diabetes. Also, to match with this sampling design, we proposed a new Wald test based on a 12-level multinomial distribution to compare the difference of the two diagnostic methods, in terms of some commonly evaluated diagnostic properties (e.g., sensitivity and specificity).

Computer simulation based on SAS/STAT and SAS/IML was used to implement the sampling process and assess the results of the hypothesis test, under different sampling designs and disease prevalence. We compared the results of the multinomial-based test against a more conventional McNemar's test, assumptions for which might be partially violated under our study setting and proposed sampling design.

Results: The results show that validation sampling only from the discordant pairs (those with disagreeing diagnostic results from the two diagnostic methods) will greatly boost the statistical power of testing the difference in sensitivity and specificity of the two diagnostic methods. Also, the Wald test we proposed performs well under different parameter settings and different sampling designs. In addition, our new test is superior to conventional McNemar's test in terms of statistical power and type-I error under the null hypothesis.

Table of Contents

Contents
Introduction .................................................................................................................................... 1
Methods ........................................................................................................................................... 4
1. Data generation method .................................................................................................... 4
2. Reparameterization and sampling schemes .................................................................. 10
3. Hypothesis testing for difference in sensitivity and specificity .................................... 15
Results ........................................................................................................................................... 27
Discussion ..................................................................................................................................... 40
References ..................................................................................................................................... 43

About this Master's Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Rollins School of Public Health
Department	Biostatistics and Bioinformatics
Subfield / Discipline	Biostatistics - MPH & MSPH
Degree	MSPH
Submission	Master's Thesis
Language	English
Research Field	Biology, Biostatistics Statistics Health Sciences, Epidemiology
Mot-clé	Gestational Diabetes comparing diagnostic methods paired study
Committee Chair / Thesis Advisor	Lyles, Robert, Emory University
Committee Members	Satten, Glen Alan, Emory University
Partnering Agencies	CDC

Dernière modification

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Gestational Diabetes validation data study: an example of testing differences in sensitivity and specificity of two diagnostic methods using non-random sampling in a paired study design ()	2018-08-28 15:39:39 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions