Bayesian methods for diagnostic test studies

本提案描述了贝叶斯统计方法的开发，该方法将用于1）评价，2）医学诊断测试的解释，以及3）涉及它们的研究设计。这里描述的项目建立在我以前关于不完美测试的贝叶斯方法的工作之上。在评价的主题下，方法将被描述为：i）用于调整条件依赖的分层潜在类模型的估计，和ii）用于调整条件依赖的比较竞争方法，和iii）在存在验证偏差和不完美的参考测试的情况下评价新的诊断测试。在解释的主题下，将描述i）基于多个不完美检验的复合结果分析临床试验结果的方法，以及ii）基于多个不完美检验构建质量控制图的方法。在研究设计主题下，将描述i）评估三种不完美检验或使用它们测量疾病患病率的研究的样本量计算方法，以及ii）基于决策理论标准的诊断试验研究的样本量计算方法。虽然这里描述的方法是由诊断测试中的问题所激发的，但它们可以应用于测量中存在误差的任何领域。

This proposal describes the development of Bayesian statistical methods to be used for 1) the evaluation, and 2) the interpretation of medical diagnostic tests, and 3) the design of studies involving them. The projects described here build upon my previous work on Bayesian methods for imperfect tests. Under the theme on evaluation, methods will be described for i) estimation of a hierarchical latent class model for adjusting for conditional dependence, and ii) comparing competing methods for adjusting for conditional dependence, and iii) evaluation of a new diagnostic test in the presence of both verification bias and an imperfect reference test. Under the theme on interpretation, methods will be described for i) analyzing clinical trial results based on a composite result from multiple imperfect tests, and ii) constructing quality control charts based on multiple imperfect tests. Under the theme on study design, methods will be described for i) sample size calculation for studies evaluating three imperfect tests or using them to measure disease prevalence, and ii) sample size calculation for diagnostic test studies based on decision theoretic criteria. Though the methods described here are motivated by problems in diagnostic testing they can be applied to any area where there are errors in measurement.