Mixture Modeling to Determine Population-Specific Cutoffs for Quantitative Diagnostic Tests without Gold Standards Público

Sullivan, Sarah Mary (2017)

Permanent URL: https://etd.library.emory.edu/concern/etds/5x21tg23q?locale=es
Published

Abstract

As Neglected Tropical Disease (NTD) programs succeed and transmission intensity declines, the ability to discriminate between positive and negative antibody tests becomes increasingly challenging. Previous techniques for defining diagnostic test cutoffs are no longer sufficient as prevalence rates decline towards zero. With varying degrees of non-specific background reactivity across populations and the absence of a gold standard, an objective yet flexible approach for cutoff determination is needed.

Mixture modeling allows for the probabilistic representation of subpopulations within an overall population. By fitting a mixture model to continuous data, members of the overall population can be assigned to groups (e.g., positive and negative), and the certainty of classification can be calculated based on the associated conditional probabilities. Certainty of classification can also be used to calculate absolute cutoffs and pre-specified indeterminate ranges (e.g. greater than 80% certainty of classification), resulting in positive, negative and indeterminate groups. The number of subpopulations and types of distribution (including Gaussian and skew-normal) may be specified or optimized by an algorithm using model selection criteria, such as the Bayesian Information Criterion.

However, current implementations of mixture modeling can be difficult to use for the public health and lab-based practitioners who determine diagnostic cutoffs. Therefore, we created an analytic tool which performs mixture modeling using the normal and skew-normal distributions for a variable number of subpopulations. The tool then calculates cutoffs, picks optimal models, and calculates indeterminate ranges with minimal user input.

We utilized this analytic tool to perform mixture modeling on standardized ELISA results from two post-treatment NTD settings. Antibody responses to a lymphatic filariasis recombinant antigen (Wb123) were analyzed via the cutoff tool and a two component, skew-normal model was found to be optimal. This yielded a cutoff of 0.115 and an indeterminate range of (0.100, 0.128), corresponding to 89.18% negative, 7.16% indeterminate and 3.66% positive results. The cutoff tool was also used to analyze responses to a recombinant onchocerciasis antigen (Ov-16), resulting in an optimal skew-normal, three component model, which yielded a cutoff of 0.609 (0.567, 0.646) with 96.31% negative, 1.5% indeterminate and 2.2% positive results.

These results demonstrate the utility of mixture modeling as a tool to provide population-specific diagnostic cutoffs with a corresponding indeterminate group that reflects our certainty regarding the cutoff. Such an approach may benefit other neglected and infectious disease programs driving towards elimination.

Table of Contents

Contents

Introduction 1

Methods 2

Mixture Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

Mixture Modeling Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

Script Development and Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

Results 7

Example A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Example B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Discussion 9

Figures 12

Figure 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Figure 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Figure 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Figure 4 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Figure 5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Figure 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

References 16

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
Department
Subfield / Discipline
Degree
Submission
Language
  • English
Research field
Palabra Clave
Committee Chair / Thesis Advisor
Committee Members
Partnering Agencies
Última modificación

Primary PDF

Supplemental Files