Semiparametric Estimation of ROC Curve with Multiple Imperfect Gold Standards: Application to the Diagnosis of Endometriosis

• Activity Number: 643
• Type: Topic Contributed
• Date/Time: Thursday, August 4, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Danping Liu* and Beom Seuk Hwang and Zhen Chen
• Companies: Eunice Kennedy Shriver National Institute of Child Health and Human Development and Eunice Kennedy Shriver National Institute of Child Health and Human Development and Eunice Kennedy Shriver National Institute of Child Health and Human Development
• Keywords: ROC curve ; imperfect gold standard ; semiparametric model ; endometriosis ; expectation-maximization

Abstract

In assessing diagnostic accuracy of a disease, the gold standard may not be observed due to high cost, invasiveness to the patient, or lack of biotechnology for accurate results. Physician Reliability Study investigates the reliability of the diagnosis of endometriosis by residents and regional experts. The gold standard of endometriosis is not well established given the largely unknown pathology. The diagnosis made by multiple regional experts can be used as imperfect surrogates for the gold standard. In this paper, we propose a semiparametric approach for estimating the accuracy of the diagnosis made by the residents. The multiple imperfect surrogates are modelled by a mixed effects model given the latent disease status. The test distributions for the diseased and non-diseased subjects are estimated by a constraint nonparametric likelihood. The receiver operating characteristic (ROC) curve and its area can then be constructed in a nonparametric form. We propose an expectation-maximization (EM) estimation procedure for dealing with the latent disease status, and derive its asymptotic properties. We also examined the sensitivity of the estimated ROC curve to the model assumptions.