Low diagnostic accuracy of Xpert MTB/RIF assay for extrapulmonary tuberculosis: A multicenter surveillance

Diagnostic accuracy of Xpert MTB/RIF assay for pulmonary tuberculosis (PTB) and extrapulmonary TB (EPTB) has not been investigated in Iran. This study was aimed to assess the diagnostic accuracy of Xpert MTB/RIF assay for both PTB and EPTB. A total of 2111 clinical samples (1218 pulmonary and 838 extra-pulmonary) were collected from 16 medical centers during the study period and were analyzed for detection of PTB and EPTB by both Xpert MTB/RIF assay and standard conventional methods (culture and direct smear microscopy). The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of Xpert MTB/RIF assay for PTB were found to be 95.5%, 96.7%, 83.8%, and 99.1% respectively. For EPTB, the sensitivity, specificity, PPV and NPV of Xpert MTB/RIF assay counted for 76.5%, 95.9%, 62%, and 97.9% respectively. Xpert MTB/RIF assay found to be highly sensitive, specific and comparable to standard conventional methods for the diagnosis of PTB. However, the sensitivity and specificity of Xpert MTB/RIF for EPTB specimens were highly variable; thus, Xpert MTB/RIF cannot be recommended to replace standard conventional tests for diagnosis of EPTB.

Specimens were collected from both outpatients and inpatients with a history of cough, unintentional weight loss, and fever for more than two weeks. Specimens were either from new cases or from patients with treatment failure or relapse.
The Ethics Committee of Shahid Beheshti University of Medical Sciences, Tehran, Iran approved the study and all research was performed in accordance with the relevant guidelines. All the patients and/or their legal guardians have signed informed consent.
Microscopy examination and culture. Specimens (2.5 to 10 mL) were decontaminated using 4% sodium hydroxide (NaOH) (Petroff method) 27 . Sediments of each treated sample were used to prepare a smear for Ziehl-Neelsen and Auramine fluorochrome staining and were cultured in Löwenstein-Jensen medium 27,28 . Only one culture isolated per study subject was considered for further analysis. Each isolate was examined regarding morphology, pigmentation, and date of growth. Bacterial isolates identified as M. tuberculosis using standard biochemical tests, including production of niacin, nitrate reduction and catalase 27 . Xpert MtB/Rif. One mL unconcentrated specimens used (without centrifuge) for Xpert MTB/RIF assay.
Specimens, firstly, homogenized with a 5% acid-alcohol buffer, a 2:1 ratio as the Cepheid Company (USA) prepared in sample reagent (SR) and then added to the Xpert MTB/RIF cartridge to do the assay in Genexpert instrument. Xpert MTB/RIF, also, uses five overlapping probes (A, B, C, D, and E) to detect five mutations in the rpoB gene that accord rifampicin resistance 2 . Data analysis. The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) with 95% confidence intervals were calculated for the Xpert MTB/RIF, using culture and direct smear microscopy as the gold standard. The summary receiver operating characteristic (SROC) curve was constructed based on a bivariate regression approach to show the level of accuracy of the Xpert MTB/RIF. The multiple regression analysis was used to shows any relationship between semi-quantitative Xpert MTB/RIF results and direct smear and culture results. All statistical analyses were performed with MedCalc (version 14.8.1, Medcalc Software). Table 1 (Table 1). "Invalid, " "error, " and "no result" of Xpert MTB/RIF were excluded from the study. The proportion of Xpert MTB/RIF invalid and error cases in extrapulmonary specimens was significantly   www.nature.com/scientificreports www.nature.com/scientificreports/ more than pulmonary specimens with Chi-squared 5.62 (p < 0.02). Comparison of false positives in pulmonary and extrapulmonary specimens have not shown meaningful difference whilst false-negative results illustrated a significant difference with Chi-squared 5.93 for two groups (p < 0.02). The percentage result for each subgroup specimen of pulmonary and extrapulmonary are showed in Table 1. Furthermore, 22 rifampicin resistance (7.7% of positive cases) and 13 rifampicin indeterminate (4.5% of positive cases) were also indicated by Xpert MTB/RIF. Sensitivity, specificity, PPV, and NPV. The overall sensitivity, specificity, PPV and NPV of Xpert MTB/ RIF assay for PTB were found to be 95.5%, 96.7%, 83.8%, and 99.1% respectively. For EPTB, the sensitivity, specificity, PPV, and NPV of Xpert MTB/RIF assay were found to be 76.5%, 95.9%, 62%, and 97.9%, respectively.

Microbiological findings. As shown in
Among extrapulmonary specimens, osteoarticular and pericardial specimens have lower sensitivity rather than average EPTB sensitivity (p < 0.0001). A comparison of sensitivity between PTB and EPTB showed a significant difference with 95% and 76% respectively (p < 0.0001). With the same respect, NPV for both groups was very high, while comparison of PPV between PTB and EPTB revealed a significant difference (p < 0.0001) ( Table 1). The forest plot (Fig. 1) shows the sensitivity, specificity, PPV, and NPV for each group of specimens. The inter-rater agreement test for PTB and EPTB showed Xpert MTB/RIF assay has substantial agreement (kappa 0.74) and moderate agreement (kappa 0.4) with standard conventional methods, respectively. The accuracy index for PTB and EPTB was 96% and 94%, respectively.

Scientific RepoRtS |
(2019) 9:18515 | https://doi.org/10.1038/s41598-019-55112-y www.nature.com/scientificreports www.nature.com/scientificreports/ Regression analysis. Based on our regression analysis, Xpert MTB/RIF results were used as regressand (dependent variables) and culture/direct smear results were used as regressors (independent variables). According to the Xpert MTB/RIF assay, results are semi-quantitative and are categorized as "Very low, " "Low, " "Medium, " and "High" based on the levels of detection. Likewise, the results of conventional methods are also semi-quantitative, which could be categorized as 1+, 2+, and 3+ in direct smear assay, and 1+, 2+, 3+ and 4+ in culture assay. Thus, we were able to compute the relationships among these variables. Accordingly, as shown in Fig. 3, there were correlations between culture/direct smear results and Xpert MTB/RIF results. Therefore, we were able to assess the relationship between the results of Xpert MTB/RIF and culture/direct smear results.

Discussion
This is the first assessment of Xpert MTB/RIF performance conducted on a large sample size in Iran. In the current study, sensitivity for pulmonary specimens was higher than in extrapulmonary samples (95.5% vs. 76.5%).
Although, previous studies showed sufficient accuracy of Xpert MTB/RIF assay for diagnosis of PTB, an acceptable accuracy for EPTB has not been well established 22,23,[35][36][37] . In the current study, Xpert has high specificity but limited sensitivity for the detection of TB in extrapulmonary specimens. Although positive results may be useful in rapidly identifying the disease, negative results provide less certainty for ruling out EPTB. The sensitivity of Xpert MTB/RIF for TB detection in extrapulmonary samples varied widely across different sample types explain more (From 40% to 100%) ( Table 1). As reported by other studies, it is possible that the decontamination step has determined the lowering of bacillary load and consequently the reduction of test sensitivity 8,38 . In agreement with these findings, our study indicated that specimens, which need smashing and homogenization (i.e. osteoarticular and biopsy specimens), had lower sensitivity.
Xpert MTB/RIF assay is based on real-time bacon PCR and subsequently is sensitive to PCR inhibitors; therefore, it needs special protocols and/or treatment buffer for extrapulmonary specimens, which includes variety of specimens 23 . The current Xpert MTB/RIF buffer has been developed for sputum. This may affect the outcome with false-negative results and decrease the sensitivity for specimens other than sputum.
We found that Xpert MTB/RIF results for pericardial and biopsy specimens are less sensitive with low PPV, which was likely due to the presence of blood and inhibitors for PCR reaction. Therefore, any treatment to remove blood before the reaction would increase the sensitivity of these specimens. Low PPV for osteoarticular specimens was also an important issue that should be addressed with further investigation.
Based on regression analysis, we were able to assess the relationship between the results of Xpert MTB/RIF and culture/direct smear results for PTB and EPTB specimens.
Limitation of the study. The current study is limited by a low sample size for blood and CSF specimens.

conclusions
Xpert MTB/RIF could be used as the first-line diagnostic tool for PTB cases. However, the overall low sensitivity of Xpert MTB/RIF for extrapulmonary specimens precludes the use of this test to rule out EPTB with certainty.