
Statistical Evaluation of Diagnostic Performance
Topics in ROC Analysis
Preview
Book Description
Statistical evaluation of diagnostic performance in general and Receiver Operating Characteristic (ROC) analysis in particular are important for assessing the performance of medical tests and statistical classifiers, as well as for evaluating predictive models or algorithms. This book presents innovative approaches in ROC analysis, which are relevant to a wide variety of applications, including medical imaging, cancer research, epidemiology, and bioinformatics.
Statistical Evaluation of Diagnostic Performance: Topics in ROC Analysis covers areas including monotone-transformation techniques in parametric ROC analysis, ROC methods for combined and pooled biomarkers, Bayesian hierarchical transformation models, sequential designs and inferences in the ROC setting, predictive modeling, multireader ROC analysis, and free-response ROC (FROC) methodology.
The book is suitable for graduate-level students and researchers in statistics, biostatistics, epidemiology, public health, biomedical engineering, radiology, medical imaging, biomedical informatics, and other closely related fields. Additionally, clinical researchers and practicing statisticians in academia, industry, and government could benefit from the presentation of such important and yet frequently overlooked topics.
Table of Contents
Introduction
Background and Introduction
Background Information
Gold Standard, Decision Threshold, Sensitivity, and Specificity
Kappa Statistics
Receiver Operating Characteristic Curve
Area and Partial Area under ROC Curve
Confidence Intervals, Regions, and Bands
Point of Intersection and Youden Index
Comparison of Two or More ROC Curves
Approaches to ROC Analysis
References
Methods for Univariate and Multivariate Data
Diagnostic Rating Scales
Introduction
Interpreter-Free Diagnostic Systems.
Human Interpreter as Integral Part of Diagnostic System
Remarks and Further Reading.
References
Monotone Transformation Models
Introduction
General Assumptions
Empirical Methods
Nonparametric Kernel Smoothing
Parametric Models and Monotone Transformations to Binormal Distributions
Confidence Intervals
Concordance Measures in Presence of Monotone Transformations
Intraclass Correlation Coefficient
Remarks and Further Reading
References
Combination and Pooling of Biomarkers
Introduction
Combining Biomarkers to Improve Diagnostic Accuracy
ROC Curve Analysis with Pooled Samples
Remarks and Further Reading
References
Bayesian ROC Methods
Introduction
Methods for Sensitivity, Specificity, and Prevalence
Clustered Data Structures and Hierarchical Methods
Assumptions and Models for ROC Analysis
Normality Transformation
Elicitation of Prior Information
Estimation of ROC Parameters and Characteristics
Remarks and Further Reading
References
Advanced Approaches and Applications
Sequential Designs of ROC Experiments
Introduction
Group Sequential Tests Using Large Sample Theory
Sequential Evaluation of Single ROC Curve
Sequential Comparison of Two ROC Curves
Sequential Evaluation of Binary Outcomes
Sample Size Estimation
Remarks and Further Reading
References
Multireader ROC Analysis
Introduction
Overall ROC Curve and Its AUC
Statistical Analysis of Cross-Correlated Multireader Data
Remarks and Further Reading
References
Appendix 7.A: Closed Form Formulation of DBM Approach for Comparing Two Modalities Using Empirical AUC
Appendix 7.B: Variance Estimators of Empirical AUCs
Free-Response ROC Analysis
Introduction
FROC Approach
Other Approaches of Detection–Localization Performance Assessment Remarks and Further Reading References
Machine Learning and Predictive Modeling
Introduction
Predictive Modeling
Cross-Validation
Bootstrap Resampling Methods
Overfitting and False Discovery Rate
Remarks and Further Reading
References
Discussions and Extensions
Summary and Challenges
Summary and Discussion
Future Directions in ROC Analysis
Future Directions in Reliability Analysis
Final Remarks
Appendix: Notation List
Index
Reviews
"… a useful addition to the ROC literature, which will prove valuable for both those involved in medical diagnosis and those whose primary interest is ROC analysis itself."
—David J. Hand, International Statistical Review (2013), 81, 2"This new book by Zou et al significantly contributes to the existing publications by providing short descriptions on basic issues and in-depth presentations on a few advanced, research-related issues. … the interested researcher can get inspired reading this book and discover new, unexplored research paths. Another pro of the book, useful for the interested researcher, is the extensive reference list at the end of each chapter. Overall, the book by Zou et al is a valuable starting point for those conducting basic research on ROC analysis and for applied researchers who are intrigued by the use of neat methodologies in applications."
—ISCB News, June 2012