Bayesian estimation of diagnostic accuracy measures when there is no gold standard: a Helicobacter Pylori data application

Purpose: Diagnostic tests are used for defining the disease status of person in a heterogeneous population which consists of healthy and diseased people. In this study, it was aimed to obtain the accuracy measures of the diagnostic test with the Bayesian approach when there was no gold standard test, and to compare it with other methods. Materials and Methods: In the diagnosis of Helicobacter Pylori, different tests are used such as HpSA, Culture, CLO, Histology and PCR. In this study, models containing a priori information (prevalence, sensitivity, etc.) were created to obtain parameter estimations with the Bayesian approach. Bayesian methods were used to correct the imperfect gold standard bias that may occur in determining the accuracy of these diagnostic tests. For this purpose, the data of a medical thesis on H. Pylori were analyzed with the help of WinBUGS and R package programs. Results: When Bayesian inference is made, the results of the positive and negative predictive values of the diagnostic test were found to be more reliable, since the prevalence and a priori information about the disease were taken into account. Similar to the predictive values, reliable results were obtained for sensitivity and specificity values. The credible interval obtained by using Bayesian approach is narrowed when a prior information is used. Conclusion: In the absence of a gold standard, Bayesian approaches using a prior information such as prevalence and diagnostic test information could facilitate diagnosis for clinicians.