Objective  To analyze the accuracy of Xpert MTB/RIF in the detection of rifampicin (RIF) resistance of Mycobacterium tuberculosis with different sputum bacterial loads (1+, 2+, 3+, 4+) and provide reference for the evaluation of clinical drug susceptibility test results.

  Methods  Positive results of liquid BACTEC MGIT960 culture of 1936 sputum samples of suspected tuberculosis (TB) patients were retrospectively collected from the reference laboratory of Tianjin Tuberculosis Control Center from January 2020 to December 2021, and 16 samples were excluded from the analysis. The results of 1158 Xpert MTB/RIF positive samples were included in the analysis, and 1142 M. tuberculosis strains tested positive for Xpert MTB/RIF and positive for liquid BACTEC MGIT960 culture were tested in vitro for drug phenotype sensitivity (DST) by routine culture method. Software SPSS 19.0 was used for statistical analysis, and paired χ² test was used. P<0.05 was considered statistically significant. Kappa test was used for the consistency of the two test methods.

  Results  In 1936 liquid BACTEC MGIT960 positive culture samples, the positive screening rate of TB detection by Xpert MTB/RIF was 59.81%, and the bacterial load of the test ranged from 1+ to 4+. The detection rates of RIF resistance were 8.88%, 14.91%, 11.37% and 15.15%, respectively. Phenotypic DST was used as a reference standard for the detection of RIF resistance. When the bacterial loads ranged from 1+ to 4+, the coincidence rates of DST and Xpert MTB/RIF for RIF resistance detection were 95.39% (290/304), 95.96% (380/396), 96.79% (332/343) and 96.97% (96/99), respectively, there were no significant differences (trend χ²=1.069, P=0.785); the sensitivity of Xpert MTB/RIF to RIF resistance showed an increasing trend, which were 80.95% (17/21), 97.96% (40/41), 93.75% (30/32), 100.00%(12/12), respectively, with no significant differences (trend χ²=7.280, P=0.063); the specificity was 96.47% (273/283), 95.77% (340/355), 97.11% (302/311), 96.55% (84/87), respectively, and there were no significant differences (χ²=0.860, P=0.835); the positive predictive values were 62.96% (17/27), 72.73% (40/55), 76.92% (30/39), 80.00% (12/15), there were no significant differences (χ²=2.047, P=0.563); The negative predictive values were 98.56% (273/304), 99.71% (340/341), 99.34% (302/304),100.00% (84/84), the differences were significant (χ²=64.661, P<0.001). The kappa values of Xpert MTB/RIF with bacterial loads from 1+ to 4+ were 0.68, 0.81, 0.83 and 0.87, respectively. In the strains with different RIF detection results of phenotype DST and Xpert MTB/RIF, if the bacterial load was 1+, the consistency rate of RIF detection results with the melting curve was 7.14% (1/14), if the bacterial load range was 2+–4+, the consistency rates of RIF detection results with the melting curve were all 100%.

  Conclusion  The sensitivity and specificity of Xpert MTB/RIF in the detection of RIF resistance under different bacterial loads were high, but the overall positive predictive value was low. When the bacterial load in sputum samples was ≤1+, the results of RIF resistance detection were generally consistent with the DST detection results, while when the bacterial loads ranged from 2+ to 4+, the results showed higher consistency with the DST detection results.