A retrospective study on Xpert MTB/RIF for detection of tuberculosis in a teaching hospital in China

This is a retrospective survey and analysis of data collected for clinical purposes. Inpatients simultaneously tested with Xpert MTB/RIF, AFB smear microscopy, and IGRA at PKUPH from November 2016 to October 2018 were included. PKUPH is a non-TB specialized, comprehensive teaching hospital, which serves patients in Beijing, as well as surrounding cities in northern China, and even throughout China. Equipped with over 2000 beds, the hospital admits more than 78,000 inpatients and handles nearly 2.6 million outpatient visits each year.

Smear microscopy was performed according to the Clinical and Laboratory Standards Institute (CLSI) M48-A guideline [13]. Specimens were stained for acid-fast microscopic examination using the Ziehl-Neelsen stain (BaSO Diagnostics Inc., Zhuhai, China). Smear-positive specimens were graded from 1+ to 4+ according to the American Thoracic Society scale [14].

The heparinized blood of patients was collected and used for the T-SPOT®.TB test (Oxford Immunotec, Oxford, UK) in accordance with the manufacturer’s instructions. When the negative and positive quality controls were under control, the results were considered positive if either panel A or panel B had ≥6 spots number.

Xpert MTB/RIF assay was performed on the GeneXpert Dx instrument system according to the manufacturer’s recommendations (Cepheid, Sunnyvale, CA, USA) (Detailed in supplementary material 1). The Xpert software was used to interpret the results, and semiquantitative results were provided based on the cycle threshold (CT) defined by the manufacturer as follows: high (CT ≤ 16), medium (16 < CT ≤ 22), low (22 < CT ≤ 28), and very low (CT > 28).

Active TB cases were considered comprehensively according to a composite reference standard (CRS), which combined the WHO guidelines [15] and WS 288--2017 (released on 2017-11-09 and implemented on 2018-05-01) by the National Health and Family Planning Commission of the People’s Republic of China (NHFPC, is now National Health Commission of the People’s Republic of China) [16]. Briefly, its contents are as follows: (1) microbiologically confirmed PTB: patients with positive MTB culture, or a pulmonary case with one or more positive initial sputum smears and chest imaging; (2) molecular biologically confirmed PTB: patients with positive MTB nucleic acid test and chest imaging; (3) histopathological examination confirmed PTB: patients with positive histopathological examination; (4) EPTB: patients with definite TB involving organs other than the lungs, with MTB isolated from a non-pulmonary source or histological or strong clinical evidence consistent with active EPTB, as well as improvement observed in anti-TB specific therapy. Thus, CRS was a combination of many pertinent aspects, which could diagnose the culture-negative TB and provide a higher accuracy. The cases were categorized as confirmed TB (PTB, EPTB, or tuberculous pleurisy, i.e., TP) or non-TB; the concurrent PTB and EPTB/TP cases were classified as PTB [15]. Then, different methods were evaluated and compared.

From November 2016 to October 2018, 1423 patients were tested using the Xpert in PKUPH. Patients who did not undergo AFB smear microscopy and/or IGRA (n = 602), patients whose Xpert specimen was inconsistent with AFB smear microscopy (n = 20) or had non-corresponding diagnostic results (n = 14) were excluded. In all, 787 patients were enrolled in this survey (Fig. 1). The mean age of the patients was 55.4 ± 18.5 years, and 55.1% of them were males (Table S1). Combined with presumptive symptoms, specimens, and clinical diagnosis, the patients were divided into three categories: presumptive PTB (533 cases), presumptive EPTB (172 cases) and presumptive TP (82 cases) (Table 2). When using CRS as the gold standard, 89 patients (11.3%) were confirmed as active TB: 52 PTB, 17 EPTB, and 20 TP (Fig. 1).

The positive ratios of AFB, Xpert, and IGRA were 3.4% (27/787), 8.3% (65/787), and 48.3% (380/787), respectively. Figure 2 shows the received operating characteristic (ROC) curves with area under the curve (AUC) for Xpert, AFB smear microscopy, and IGRA in comparison to the gold standard. The diagnostic performance of IGRA (AUC = 0.754, 95% confidence interval, i.e., 95% CI 0.722–0.783; sensitivity = 93.3, 95% CI 86.1–96.9%; specificity = 53.5, 95% CI 53.8–61.1%) was significantly higher than AFB (AUC = 0.614, 95% CI 0.579–0.648; sensitivity = 23.6, 95% CI 16.0–33.4%; specificity = 99.1, 95% CI 98.1–99.6%) (P < 0.001, Z test). Xpert showed the best diagnostic performance, with AUC value 0.846 (95% CI 0.819–0.871), sensitivity 69.7% (95% CI 59.5–78.2%), specificity 99.6% (95% CI 98.7–99.9%), and significantly higher than the other two methods (P < 0.001) (Fig. 2 and Table 1).

In presumptive PTB, EPTB, and TP cases, the positive ratios of Xpert were 9.0, 8.1, and 3.7%, respectively, which were higher than AFB (4.5, 1.7%, and 0), but lower than IGRA (49.2, 37.8, and 64.6%). Among the 89 confirmed TB cases, 19 cases (21.3%) tested positive using all the three methods, and the use of these three methods simultaneously could screen 96.6% (95% CI 90.5–98.9%) TB patients (Fig. 3a); 62 (69.7%, 46 PTB, 13 EPTB, and 9 TP), 21 (23.6%, 19 PTB, 2 EPTB, and 0 TP) and 83 (93.3%, 50 PTB, 14 EPTB, and 19 TP) cases tested positive using Xpert, AFB, and IGRA, respectively (Fig. 3b), while 3 (3.4%) confirmed TB cases (1 TP and 2 EPTB, i.e., 1 lumbar spine TB and 1 arthritis TB) tested negative using all the three methods. As shown in Table 1, the sensitivity of Xpert in detecting PTB, EPTB, and TP was 88.5% (95% CI 77.0–94.6%), 76.5% (95% CI 52.7–90.4%) and 15.0% (95% CI 5.2–36.0%), respectively, slightly lower than IGRA (96.2, 82.4, and 95.0%), but higher than AFB (36.5, 11.8, and 0%). IGRA had the highest sensitivity, but the specificity (55.9, 67.1, and 45.2%) was significantly lower than Xpert (99.6, 99.4, and 100%) and AFB (99.0, 99.4, and 100%) (P < 0.001); notably, the positive predictive values (PPV) (7.3, 95% CI 4.9–10.7%) is poor when only IGRA is positive.

As shown in Table 2, the highest sensitivity of Xpert in pulmonary specimens was in lung tissue (100%), followed by sputum (88.5%), BALF (85.7%), and FOB (81.2%); the specificity of all pulmonary specimens was higher than 99%. When detecting extrapulmonary specimens, the sensitivity and specificity of CSF, joint cavity fluid, and lymph node specimen were 100%; while the sensitivity for urine was relatively low (33, 95% CI 6.2–79.2%). The specificity of intestine tissue is 98.1%; other extrapulmonary specimens (except pleural fluid) showed a specificity of 100%. Notably, the sensitivity of pleural fluid was the lowest, only 15.0% (95% CI 5.2–36.0%), and significantly lower than pulmonary tissue, CSF, joint cavity fluid, extrapulmonary lymph node, sputum, BALF, and FOB (P < 0.05).

The correlation between Xpert semiquantitative category and AFB smear grade is presented as cross-tabulated data (Table 3). Of 65 Xpert positive cases, 12 were positive high and 13 were positive medium, and PPV were all 100%; two and one false positive cases were obtained in positive low and positive very low cases, respectively. The sensitivity of AFB decreased gradually, accompanied with the Xpert results from 75.0% in Xpert positive high cases to 3.7% in Xpert negative cases. Notably, the PPV of AFB suddenly dropped from 100% in Xpert positive cases to 14.3% in Xpert negative cases, i.e., six of the seven (85.7%) positive AFB cases that occurred in Xpert negative cases were false positive. Totally, the PPV of Xpert (95.4, 95% CI 87.3–98.4%) were significantly higher than AFB (77.8, 95% CI 59.2–89.3%) (P < 0.05). Besides, the smear grades increased as the Xpert CT values decreased among the 62 Xpert true positive cases (Figure S1), and there was a strong, negative correlation between smear grades and Xpert CT values, which was statistically significant (R = -0.632, P < 0.001).

Among the 27 AFB positive cases, 21 confirmed TB cases (23.6%, 21/89, 19 PTB, and 2EPTB) were detected, with a PPV of 77.8%; the pooled sensitivity, specificity, PPV, and negative predictive value (NPV) of Xpert were 95.2, 100, 100, and 85.7%, respectively, all higher than IGRA (90.5, 50.0, 86.4, and 60.0%, respectively). Obviously, 68 confirmed TB (76.4%, 68/89, 33 PTB, 15 EPTB, and 20 TP) were detected using Xpert and/or IGRA in AFB smear-negative patients (n = 760, Table S2); the sensitivity of Xpert (61.8, 95% CI 49.9–72.4%) was significantly lower than IGRA (94.1%, 95 CI 85.8–97.7%) (P < 0.001), while the specificity of the former (99.6, 95% CI 98.7–99.9%) was significantly higher than the latter (57.5, 95% CI 53.8–61.2%) (P < 0.001). In detail, the sensitivity of Xpert was 84.9, 73.3, and 15.0% for detecting presumptive PTB, EPTB, and TP cases, respectively, which were lower than IGRA (100, 80.0, and 95.0%); while the specificity of Xpert (99.6, 99.4, and 100%) was significantly higher than IGRA (56.1, 66.9, and 45.2%) (P < 0.001).