Back to search

Georgia Southern University

Evaluating The Performance of Various Multiple Imputation Techniques for The Analytical Solution of Joint Confidence Region of AUC and Youden Index Assuming The Ignorable Missing Data

Abstract

dc:description.abstract

The area under the ROC curve (AUC) and the Youden index are standard measures of biomarkers' accuracy in medical diagnostics. Literature suggested that the AUC and the Youden index's joint confidence region provide a better estimation of biomarker's accuracy than the individual confidence interval for the given data. This dissertation introduced the close form solution (CFS) for the joint confidence region of the AUC and the Youden index estimates for a single continuous biomarker and the binary disease variable and compared it with the bootstrap method. The ties in the observed biomarker values were also analyzed by the close form solution (CFS) and bootstrap methods for various sample sizes. The second part of the dissertation introduced the missing data in the "gold standard" test. It explored the various range of the missing data (10% - 70%), assuming the ignorable missing mechanism (missing at random) for the disease variable. Further, it evaluated the performance of complete case analysis and other multiple imputation methods such as logistic regression (LR), predictive mean matching (PMM), and multivariate normal (MVN) to estimate the coverage probability and joint area for the AUC and the Youden index. Intensive simulation for these different settings suggested that for the smaller sample sizes and higher estimates of the AUC and the Youden index, logistic regression imputation is recommended. In contrast, for the larger sample sizes, lower disease rate, and smaller AUC and Youden index, either of the predictive mean matching or logistic regression imputation methods can be applied. Finally, the secondary data of the national health and nutritional examination survey (NHANS) recognizing diabetes (binary disease variable diabetes, continuous biomarker glycated hemoglobin and covariates age, gender, and race) has been used to evaluate the application of our methods in a real-world setting and using logistic regression multiple imputations with close form solution (CFS) method to impute and infer the joint results for missing binary disease variable.

Degree

thesis:*
Name thesis:degree_name
Doctor of Public Health in Biostatistics (Dr.P.H.)
Level thesis:degree_level
Dissertation (restricted to Georgia Southern)
Discipline thesis:degree_discipline
College of Public Health

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Rathi, Kapil
Contributors dc:contributor
  • Hani Samawi
  • JingJing Yin

Subjects

dc:subject × 13

Rights

Identifiers

dc:identifier.*
Repository record dc:identifier
https://digitalcommons.georgiasouthern.edu/etd/2252

Chain of custody

source
Harvested from
Georgia Southern University
Base URL
digitalcommons.georgiasouthern.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Rathi, Kapil. Evaluating The Performance of Various Multiple Imputation Techniques for The Analytical Solution of Joint Confidence Region of AUC and Youden Index Assuming The Ignorable Missing Data. Dissertation (restricted to Georgia Southern) thesis, https://digitalcommons.georgiasouthern.edu/etd/2252