Least Squares ROC Method for Tests with the Absence of the Gold Standard

• Activity Number: 246
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 5:50 PM
• Sponsor: Health Policy Statistics Section
• Author(s): Liansheng Tang* and Minh Huynh and Leighton Chan and John Collins and Ao Yuan and Xuan Che
• Companies: NIH Clinical Center/George Mason University and IMPAQ International, LLC and NIH Clinical Center and NIH Clinical Center and NIH and NIH Clinical Center
• Keywords: ROC ; Gold Standard

Abstract

The topics on diagnostic accuracy without the gold standard can be classified into several areas, including 1) binary test results with a perfect gold standard, 2) ordinal or continuous test results with a perfect gold standard, 3) binary test results without a gold standard, and 4) ordinal or continuous test results without a gold standard. Extensive literature is available on parametric, semiparametric and nonparametric methods to evaluate the accuracy of diagnostic tests with perfect gold standards. Sensitivities and specificities are commonly used for a binary test. These parameters can be estimated using the proportions when a perfect gold standard is available for every individual in the sample. The receiver operating characteristic (ROC) curve plotting pairs of sensitivities and specificities is a common statistical tool to evaluate the accuracy of ordinal or continuous tests. The ROC curve estimated from data without the gold standard is biased. To correct for the bias, a linear regression method is proposed to estimate the ROC curve from pairs of consistent sensitivity and specificity estimates. The proposed method first applies Hui and Walter's method to estimate a pair o