Statistical Methods and Software for Meta-analysis of Diagnostic Tests

点击翻译按钮获取中文摘要

Statistical Methods and Software for Meta-analysis of Diagnostic Tests Principal Investigator: Haitao Chu, M.D., Ph.D. Summary Comparative effectiveness research relies fundamentally on accurate assessment of clinical outcomes. The growing number of assessment instruments, as well as the rapid escalation in the cost has generated the increasing need for scientifically rigorous comparisons of the diagnostic tests in clinical practice. Meta-analysis of diagnostic tests presents many additional statistical challenges compared to traditional meta-analysis applications such as meta-analysis of controlled clinical trials. In particular, diagnostic accuracy cannot be adequately summarized by one measure; two measures are typically used, most often sensitivity and specificity, or alternatively positive and negative likelihood ratios, and either two are correlated. Furthermore, diagnostic accuracy parameters may depend on disease prevalence. In response to AHRQ PAR-10-168, the overall goal of this proposal is to develop cutting-edge multivariate statistical methods, and to integrate them into publicly available, easy-to-use software to enhance the consistency, applicability, and generalizability of the meta-analysis of comparative diagnostic test studies. In this proposal, we assume that a gold standard exists; the problem of imperfect gold standard bias in a meta-analysis of diagnostic tests is a topic for future research. Specifically, we will focus on developing statistical methods and related software for: (1) Meta- analysis of diagnostic tests accounting for disease prevalence when some studies use case-control design and some studies use cohort design, which is common in practice but methodological ramifications have never been addressed; (2) Correcting verification bias from meta-analysis of diagnostic tests due to biased sampling of whom is being tested by the gold standard, which can lead to biased estimation of accuracy parameters including sensitivities and specificities if the missing data and verification bias are not appropriately handled. We propose to perform empirical assessment of the strengths and weaknesses of these methods through real data applications and simulations. The proposed statistical methodology will be broadly applicable to the meta- analysis comparing diagnostic tests. It will improve public health by facilitating the diagnosis of various cancers, cardiovascular, infectious and other diseases. Completion of these two aims will directly benefit the comparative effectiveness research program at AHRQ by providing state-of-the art methods implemented in user-friendly software using WinBUGS and R statistical languages that will be made freely available to the public.