无金标准诊断试验灵敏度和特异度的贝叶斯估计方法

Bayes estimation of sensitivity and specificity of diagnostic tests in the absence of a gold standard

目的 探讨在无金标准情况下诊断试验灵敏度和特异度的贝叶斯估计方法,并通过模拟实验验证此方法的有效性和适用性.方法 应用贝叶斯原理,结合一个或者两个对照诊断试验,采用对照诊断试验与待测诊断试验诊断结果相互独立或者存在相关模型;和两个对照诊断试验诊断结果存在相关,并且同时与待测诊断试验结果相互独立模型;以及一个对照诊断试验与待测诊断试验诊断结果存在相关,并且同时与另一个对照诊断试验诊断结果相互独立模型,对一种新的诊断方法的灵敏度、特异度进行估计.结果 模拟试验表明在先验信息相对准确的条件下,所给出的几种不同模型均能较准确的估计待测诊断方法的灵敏度和特异度,估测参数范围包括设定真值,并且估测参数均值基本与真实情况相同;此外,模拟实验结果表明,利用两个对照诊断试验模型估计参数的估计效果优于使用一个对照诊断试验模型时的参数估计效果.结论 本文介绍的贝叶斯方法能够有效地应用于无金标准诊断试验的评价,从而解决了无金标准诊断试验无法评价的难题.

Objective To provide a method to estimate sensitivity and specificity of a diagnostic test in the absence of a gold standard, and to study its properties through simulation experiments. Methods We considered estimation of the sensitivity and specificity of a diagnostic test in the following settings. There were two conditionally independent tests, one of which was a control group, two correlated tests, one of which was a control group, and three tests where two tests were correlated but jointly independent of the third test. For each scenario, we described a Bayesian model that incorporates parameters of interest. Results The simulation experiments verify that, given the beta prior distributions of every parameter, the models can evaluate them accurately. In addition, compared to the model with one control group, the estimating results of the model with two control groups is more close to the true values. Conclusion The Bayesian approach can be used to evaluate a new diagnostic test effectively in the absence of a gold standard.