中文摘要：

目的探讨无金标准条件下诊断试验贝叶斯相关模型构建方法及应用条件。方法通过分析具有潜在真值的无金标准诊断试验评价模型,构建两个试验相关条件下的似然函数;利用共轭分布原理,构建灵敏度、特异度、患病率的先验分布;使用WinBUGS软件计算后验参数。通过234602名无偿献血员抗-HIV检测结果说明贝叶斯相关模型的应用。结果构建了无金标准时两次ELISA法检测抗-HIV的贝叶斯相关模型,发现两次ELISA的灵敏度相关系数为0.30,特异度相关系数为0.74;两次试验的联合灵敏度较单个试剂增高（P〈0.05）,特异度较单个试剂降低（P〈0.05）,但特异度降低的幅度明显小于灵敏度增高的幅度。结论应用贝叶斯相关模型可合理评价无金标准时联合试验的灵敏度和特异度。

英文摘要：

Objective To explore the methods of establish-ment of Bayesian conditional dependence model for evaluation of diagnostic tests in the absence of gold standard and to application of the model to practice. Methods By analysis of the model of diagnostic tests in the absence of golden standard with latent variables,the likelihood function was form. Using the principle of conjugate distribution,the prior distribution of sensi- tivity,specificity and prevalence were formed and all of posterior parameters were calculated by WinBUGS. The model was demonstrated by HIV-Ab tests from 234602 blood donors. Results Bayesian conditional depend- ence model were established for evaluation of the accuracy of two ELISA for antibody to HIV and the correlation of sensitivity was 0. 30 and of specificity 0. 74. The sensitivity of parallel test was higher than that of single test（ P 〈 0. 05） and the specificity of parallel test was lower than that of single test（ P 〈 0. 05% ） but the extent in which the specificity decreased was manifestly less than that of the sensitivity increased. Conclusion Bayesian conditional dependence model can evaluate the sensitivity and specificity of parallel test in the absence of golden standard reasonably.