Background: Existing diagnostic approaches for paucibacillary tuberculosis (TB) are limited by the low sensitivity of testing methods and difficulty in obtaining suitable samples.
Methods: An ultrasensitive TB diagnostic strategy was established, integrating efficient and specific TB targeted next-generation sequencing and machine learning models, and validated in clinical cohorts to test plasma cfDNA, cerebrospinal fluid (CSF) DNA collected from tuberculous meningitis (TBM) and pediatric pulmonary TB (PPTB) patients.
Results: In the detection of 227 samples, application of the specific thresholds of CSF DNA (AUC = 0.974) and plasma cfDNA (AUC = 0.908) yielded sensitivity of 97.01 % and the specificity of 95.65 % in CSF samples and sensitivity of 82.61 % and specificity of 86.36 % in plasma samples, respectively. In the analysis of 44 paired samples from TBM patients, our strategy had a high concordance of 90.91 % (40/44) in plasma cfDNA and CSF DNA with both sensitivity of 95.45 % (42/44). In the PPTB patient, the sensitivity of the TB diagnostic strategy yielded higher sensitivity on plasma specimen than Xpert assay on gastric lavage (28.57 % VS. 15.38 %).
Conclusions: Our TB diagnostic strategy provides greater detection sensitivity for paucibacillary TB, while plasma cfDNA as an easily collected specimen, could be an appropriate sample type for PTB and TBM diagnosis.
Keywords: Diagnosis; Machine learning; Paucibacillary; Tuberculosis; cfDNA; target NGS.
Copyright © 2023 Elsevier B.V. All rights reserved.