The use of diagnostic tests and biomarkers is an essential part of medical care, and plays an important role in guiding therapy decisions in the era of precision medicine. In this dissertation, we address two major aspects of test evaluation, the assessment of predictive accuracy (Chapters 1 and 2) and the assessment of the impact of tests on patient outcomes (Chapter 3). In the practice of evidence based medicine, the ability to synthesize evidence from primary studies of biomarkers is useful in optimizing health policy decision making. However methodological developments in the area of synthesizing predictive values have been limited. In Chapter 1, we put forth a new meta-analysis model to synthesize and compare predictive values of biomarkers. In Chapter 2, we undertake a critical evaluation of the widespread use of hazard ratio as a summary measure of the prognostic performances of biomarkers. From the results of this study, we obtain a better understanding of the implications of using hazard ratio to summarize, and compare prognostic performances of biomarkers. This study also identifies essential information that should accompany the reporting of hazard ratio to allow proper evaluation of the prognostic performances of a biomarker. A key challenge in evaluating the impact of diagnostic tests on patient outcomes is that the pathway from test to outcomes typically involves subsequent disease management and treatment interventions. Modeling approaches, such as decision analysis and micro-simulation, are commonly used to study the impact of tests. Randomized studies (also known as diagnostic randomized controlled trials, DRCT) have also been utilized, but to a lesser extent than modeling. In addition to the large sample size typically required, DRCT studies are also prone to selection bias arising from noncompliance by study participants to assigned tests and interventions. Recent work has laid out the formal framework for evaluating DRCT designs. However the impact of noncompliance has not been addressed. In Chapter 3, we adapt and apply modern methods in causal inference to estimate the causal outcomes of diagnostic tests in the presence of noncompliance. The performance of such causal estimates are evaluated via simulation of different scenarios.
Yue, Mun Sang,
"Evaluation of Predictive Accuracy of Tests and Impact of Tests on Patient Outcomes"
(2017).
Biostatistics Theses and Dissertations.
Brown Digital Repository. Brown University Library.
https://doi.org/10.7301/Z0V40SPP
