Verification bias is a common problem in the diagnostic test efficacy studies. Verification bias refers to a distortion in the estimate of the diagnostic efficacy resulting from selectively verifying disease status of a subset of the initially tested patients. Verification bias can lead to grossly either underestimation of overestimation of the diagnostic test efficacy. Although verification bias could seriously distort the estimate of the diagnostic efficacy, available methods for retrospectively correcting for verification bias are not totally satisfactory. In this project, we will develop a comprehensive treatment of verification bias in the diagnostic efficacy studies. The first two aims are to develop verification bias correction procedures for: 1. estimating the efficacy of a diagnostic test with available covariates; 2. comparing the relative efficacious of two competing tests without covariates; 3. comparing the relative efficacious of two competing tests when some covariates are available. The last aim is to apply the methods developed to four data sets representing different types of studies in health services research and different degrees of available information on missing disease status mechanism. Thus, we will demonstrate that our methods are applicable to different types of studies which have different available information on missing data mechanism. This project will provide statistical methods for retrospectively correcting verification bias in the diagnostic test efficacy studies. Hence, these methods enable health-care providers to precisely assess and compare the efficacious of competing tests. Only then can health-care providers make informed choices of the most-effective diagnostic tests. This may contribute to the reduction of growing health care costs in the long run.
