OBJECTIVE To analyze the reliability of Xpert MTB/RIF (referred to as Xpert) in detecting rifampicin resistance under varying bacterial loads of Mycobacterium tuberculosis (MTB) so as to reference for clinical diagnosis and treatment.

METHODS A total of 3 050 mycobacterial culture-positive test results from Jan. 2016 to Dec. 2023 at the First Affiliated Hospital of Xiamen University were selected. After excluding 220 non-tuberculous mycobacteria cases, 2 830 culture-positive specimens underwent Xpert testing, with 2 412 positive and 418 negative cases. Based on MTB bacterial load, specimens were divided into four treatment groups-high, medium, low and very low-with 552, 916, 656 and 288 cases, respectively. Solid phenotypic drug susceptibility testing (DST) was performed on 1 801 MTB strains that were both culture-positive and Xpert-positive.

RESULTS The positive rates of rifampicin in the high, medium, low and very-low treatment groups were 10.33% (57/552), 11.14% (102/916), 8.54% (56/656) and 8.68% (25/288), respectively. Using the solid phenotypic drug susceptibility testing results as the reference standard, the specificity and negative predictive value of Xpert for detecting rifampicin resistance in all four bacterial load groups exceeded 96%, with no statistically significant differences observed. No significant difference in sensitivity was noted among the groups (P=0.053), though the high bacterial load group exhibited relatively lower sensitivity at 84.62% (44/52), while the very-low bacterial load group showed the lowest sensitivity at 76.00% (19/25), which was lower than the medium bacterial load group′s 94.05% (79/84). The positive predictive values (PPV) across all groups were generally low but without statistically significant differences (P=0.239), with the high and very-low bacterial load groups registering PPVs of 77.19% (44/57) and 76.00% (19/25), respectively. The concordance rates of Xpert and solid phenotypic drug susceptibility testing were consistently high among all four bacterial load groups, including 94.95% (395/416), 97.51% (666/683), 96.75% (477/493) and 94.26% (197/209), respectively, with no significant difference(P=0.052). The Kappa values from high to very-low bacterial load groups were 0.778, 0.889, 0.831 and 0.727, respectively. Among the strains with discordant results between Xpert and phenotypic drug susceptibility testing, the rifampicin result consistency rates between Xpert and gene chip were 81.82% (9/11), 100.00% (9/9), 100.00% (2/2) and 25.00% (1/4), respectively for the high to very-low bacterial load groups.

CONCLUSIONS When the bacterial load of MTB is very low, the Xpert assay demonstrates a higher probability of both false-positive and false-negative results for rifampicin resistance, followed by high bacterial loads. The most reliable detection of rifampicin resistance occurs with medium bacterial loads.