Estimation of ROC Curve with Multiple Types of Missing Gold Standard

Abstract

In evaluating the diagnostic accuracy of a test, the gold standard might be missing because of high cost or harmfulness to the patient. The estimation of the diagnostic accuracy could be biased if the missingness is not handled appropriately. In this chapter, we propose a likelihood-based approach to jointly estimate the selection model and disease model when the missing data mechanism is a mixture of ignorable and nonignorable missingness. The receiver operating characteristic (ROC) curve and its area are estimated empirically using imputation and reweighting techniques. The proposed method extends the results of Liu and Zhou (2010, Biometrics, 66, 1119–1128), as the latter assumes a single source of nonignorable missingness. We perform simulation studies to compare the performance of the proposed method and the existing approaches in the literature. This methodology is motivated from and applied to a study in Alzheimer’s disease (AD), where two reasons of missingness are modeled separately.

Acknowledgments

The authors would like to thank the referees for their insightful comments, which greatly improved the quality of this chapter. This work was supported in part by NIH/NIA grant U01AG016976. Dr. Danping Liu’s research is supported by the Intramural Research Program of the National Institute of Health (NIH), Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD). This chapter does not necessarily represent the findings and conclusions of VA HSR&D. Dr. Xiao-Hua Zhou is presently a core investigator and biostatistics unit director at HSR&D Center of Excellence, Department of Veterans Affairs Puget Sound Health Care System, Seattle, Washington.