Graduation Year
2024
Document Type
Dissertation
Degree
Ph.D.
Degree Name
Doctor of Philosophy (Ph.D.)
Degree Granting Department
Public Health
Major Professor
Yougui Wu, Ph.D.
Committee Member
Getachew Dagne, Ph.D.
Committee Member
Yangxin Huang, Ph.D.
Committee Member
Feng Cheng, Ph.D.
Keywords
Diagnostic Test, Sensitivity, Specificity, Paired Design, Predictive Values, Incomplete Test Results, Missing Completely at Random
Abstract
Diagnostic tests are medical examinations essential for patient care, as they help detect the presence of diseases and rule out certain conditions. However, diagnostic tests can make errors, and misdiagnoses resulting from these errors can potentially lead to treatment delays or inappropriate treatments. Therefore, comprehensively evaluating the performance of the diagnostic tests and ensuring test accuracy is a fundamental priority in diagnostic studies. When a new diagnostic test is developed, it is customary to compare it with an existing test in a paired study. However, incomplete test results commonly occur in practice. The maximum likelihood-based method can be used to address the incomplete data issue for comparing two predictive values and achieve high efficiency, but its implementation requires an iterative algorithm, thereby increasing the burden in its practical application. Conversely, the simple method is easily implementable, but it sacrifices efficiency due to the application of an equal weighting scheme. Consequently, the research gap reveals a lack of easily implementable statistical methods that achieve high efficiency when comparing two predictive values with incomplete test results. Given that there is a lack of dedicated research on the optimal weighting selection in this research context, this dissertation addresses the research gap by developing a new optimal weight method, under the assumption that the missing test results are missing completely at random. Our findings consistently demonstrated that the proposed estimator not only achieved greater efficiency compared to the simple estimator using relative efficiency metric but also required a much lower level of implementation compared to the maximum likelihood-based approach. Additionally, the relative efficiency became particularly pronounced when the correlation coefficient between two diagnostic tests was strongly positive and/or when there was a relatively higher proportion of incomplete paired test results in a given dataset. Lastly, the superior empirical power demonstrated by the proposed optimal weight test statistics compared to the simple test statistic in simulations under all explored conditions underscores its value as a useful tool for researchers in diagnostic test studies.
Scholar Commons Citation
Wang, Meng, "Performance Comparison of Two Binary Diagnostic Tests with Incomplete Data" (2024). USF Tampa Graduate Theses and Dissertations.
https://digitalcommons.usf.edu/etd/11207
