Evaluation of the Accuracy of Tuberculous Meningitis Diagnostic Tests Using Bayesian Latent Class Analysis

Abstract

Background: Tuberculous meningitis is the most dangerous form of tuberculosis which has high mortality and disability rates. However, the diagnosis process was often delayed due to the lack of information about the sensitivity and specificity of diagnostic tests. The objective of this study is to estimate the TBM prevalence and the performance of 4 routine diagnostic tests including MGIT culture test, Xpert MTB/RIFF test, cerebrospinal fluids (CSF) protein levels, and CSF leukocytes count. Methods: We conducted a study using the Bayesian latent class analysis to estimate the disease prevalence, specificity, and sensitivity of the 4 diagnostic tests with our chosen prior distribution and also the expert elicitation. Furthermore, the area under the receiver operating characteristic ROC curve (AUC) of CSF protein levels and leukocytes count were also compared and estimated using different thresholds values. Results: A total of 1213 suspected TBM patients were included in this study. According to the Bayesian models, the estimated posterior distribution (posterior mean and 95% credible interval (CI) of the TBM prevalence was 34.81% (95% CrI: 28.80-41.30). The sensitivity of the MGIT culture test was 62.69% (95% Crl: 52.53 -74.00). Likewise, for the Xpert MTB/RIFF test, the sensitivity was 57.52% (95% Crl: 51.00-64.00), and specificity was 95.89% (95% CI: 92.03 -99.81). The AUC ROC for CSF leukocytes count was 76.03, and for the CSF protein level was 73.4. Conclusion: The culture test was the imperfect gold standard but a good and reliable test with high sensitivity and specificity. The Xpert test had the sensitivity similar to culture test and the specificity nearly perfect. Both of the CSF protein level and leukocyte counts had only moderate accuracy for diagnosing TBM. However, the CSF leukocytes count had higher precision when compared with CSF protein concentration levels.