Introduction

As the most common form of extrapulmonary tuberculosis (EPTB), lymph node tuberculosis (LNTB) exhibits swollen lymph node, fever, night sweat, representing approximately 35% of EPTB cases in developing countries [1]. Without a quick and accurate diagnostic method, treating and controlling LNTB poses a significant challenge [2]. Currently, culture remains the gold standard for diagnosis. Nonetheless, an extended turnaround time of 6–8 weeks and relatively low sensitivity pose challenges [3]. Similarly, acid-fast bacillus (AFB) smears have low sensitivity. The presence of granuloma and caseous necrosis in tissue histopathology supported the diagnosis of LNTB. However, these features are not specific, as they can be found in other diseases [4].

Xpert MTB/RIF is capable of detecting Mycobacterium tuberculosis complex (MTBC) DNA within a mere 2 hours [5]. This assay has been demonstrated to exhibit high diagnostic accuracy, specifically in detecting LNTB [6]. The World Health Organization (WHO) has provisionally recommended using Xpert MTB/RIF for testing lymph node samples in individuals with suspected EPTB. This recommendation was supported by limited and highly heterogeneous data, and underscored the need for further research to evaluate the performance of Xpert in EPTB patients, particularly in the LNTB [7]. Although Xpert Ultra exhibits greater sensitivity, have more molecular targets, and is able to detect rifampicin resistance more effectively [8], it lacks sufficient original articles and data. By contrast, Xpert MTB/RIF is used to detect various sample types of tuberculosis (TB) in numerous studies and exhibits greater specificity than Xpert Ultra [9–11]. In addition, the diagnostic accuracy of Xpert MTB/RIF has been continuously reported in newly published articles [12–14], indicating its high clinical significance.

A consensus has not been reached on the optimal specimens for patients suspected of LNTB [15]. Different sample processing procedures may also affect the interpretation of the results [6]. While FNA is an easy-to-perform and quick diagnostic technique, it exhibits variable sensitivity, potentially explained by the limited quantity of bacilli in FNA samples [16]. Pathological diagnosis is highly sensitive when utilizing the most invasive biopsy specimens [15]. Furthermore, we implemented a composite reference standard (CRS). CRS consists of results including histopathological examination, smears, results of biochemical tests, clinical manifestations, radiological findings, culture and/or response to antituberculosis therapy, serving as a benchmark in addition to culture for TB classification [17]. CRS classifies the presence of TB as positive if at least one component test is positive.

TB in children is typically paucibacillary and young children cannot voluntarily produce sputum specimens, making the microbiological confirmation of TB more difficult in children [18]. WHO found children younger than 14 years old gained poorer access to diagnosis and treatment. 16% of death cases from TB were children while children cases shared only 11% of whole age groups [19]. Hence, this meta-analysis was implemented to assess the diagnostic accuracy of the Xpert MTB/RIF assay for detecting various specimen types (including FNA and tissue samples) in individuals in different age groups suspected of LNTB compared to both CRS and mycobacterial culture methods.

Materials and methods

Results

Study evaluation

In this meta-analysis, we identified 614 candidate articles from relevant databases and 2 additional articles from other sources. Among these, 97 articles were from PubMed, 67 from Embase, 16 from the Cochrane Library, 185 from the Web of Science, and 247 from Scopus. Of the 614 articles, 317 were identified as duplicates. After reviewing the titles and abstracts, 57 articles were subjected to the full-text screening process. After excluding 21 articles due to inability to extract data, 3 articles with insufficient data, 1 article with inappropriate sample type, 1 article not reporting a gold standard, and 3 articles where the fulll text could not be found, 27 articles were ultimately included for full-text review and meta-analysis. Each article evaluated specimens ranging from 11 to 379, with a median of 82. All studies were conducted in English. The screening process is illustrated in Fig 1.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Literature retrieval flow chart.

A total of 97, 67, 185, 16 and 247 articles were found from PUBMED, EMBASE, WEB OF SCIENCE, COCHRANE LIBRARY and SCOPUS respectively.

https://doi.org/10.1371/journal.pone.0321507.g001

If an article utilized two distinct standards within the same research, we classified it as comprising two separate studies. According to these guidelines, 33 distinct studies were included: 9 assessed Xpert MTB/RIF against CRS, and 24 evaluated Xpert MTB/RIF against culture (Table 1). Among these, 22 studies employed FNA samples, while 11 utilized biopsy tissue sample

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of included articles.

https://doi.org/10.1371/journal.pone.0321507.t001

Study quality

A quality assessment of diagnostic accuracy studies against CRS or culture as a reference standard was carried out using QUADAS-2 (Fig 2).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Methodological quality graphs (risk of bias and applicability concerns) as percentages across the included studies.

(a) Composite reference standard. (b) Culture reference standard.

https://doi.org/10.1371/journal.pone.0321507.g002

Diagnostic accuracy of the Xpert MTB/RIF Assay for identifying LNTB

In nine studies, a total of 914 FNAs or tissue samples were compared with those from patients with CRS. The combined sensitivity of the Xpert MTB/RIF assay was 85% (95% CI: 71–93%; I2: 82%), while the aggregated specificity of the Xpert MTB/RIF assay was 97% (95% CI: 87–99%; I2: 90%) (Fig 3(a)). Notably, significant heterogeneity was observed in both sensitivity and specificity. The estimated area under the curve (AUC) against CRS, the gold standard in the Xpert assay, was 0.97 (95% CI: 9598%) (S1 Fig). Compared with culture, the overall sensitivity of Xpert MTB/RIF was 85% (95% CI: 77–90%; I2:89%), and the specificity was 78% (95% CI: 62–89%; I2: 95%), with 2973 FNAs or tissue specimens from 24 studies (Fig 3(b)), indicating significant heterogeneity. The AUC of the SROC was 0.89 (95% CI: 86–91%) compared with that of culture, indicating excellent overall diagnostic validity (S2 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. (a) Forest plot of the sensitivity and specificity of Xpert for tuberculosis detection compared with a composite reference standard. (b) Forest plot of the sensitivity and specificity of Xpert for tuberculosis detection compared with the culture reference standard. The squares represent the sensitivity and specificity of a study, and the black line represents their confidence intervals. The diamonds represent the pooled sensitivity and specificity and their confidence intervals.

https://doi.org/10.1371/journal.pone.0321507.g003

The combined sensitivity of Xpert MTB/RIF for FNA specimens compared with CRS specimens was 93% (95% CI: 50–99%; I2: 90%), while its specificity was 88% (95% CI: 68–96%; I2: 91%) (Fig 4(a)). The AUC of the SROC was 0.95 (95% CI: 93–97%), indicating excellent diagnostic accuracy (S3 Fig). The sensitivity of tissue specimens in the Xpert evaluation versus CRS was 74% (95% CI: 58–85%; I2: 91%), while its specificity was 100% (95% CI: 95–100%; I2 = 0) (Fig 4(b)). Furthermore, the estimated AUC of tissue specimens compared with that of CRS specimens in the Xpert trial was 0.95 (95% CI: 93–97%) (S4 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Forest plot of the sensitivity and specificity of Xpert for detecting tuberculosis versus a composite reference standard.

(a) FNA samples. (b) Tissue samples. The squares represent the sensitivity and specificity of a study, and the black line represents their confidence intervals. The diamonds represent the pooled sensitivity and specificity and their confidence intervals.

https://doi.org/10.1371/journal.pone.0321507.g004

The aggregated sensitivity for Xpert MTB/RIF using FNA samples against culture was 87% (95% CI: 78–93%; I2: 90%), while the combined specificity was 77% (95% CI: 63–87%; I2: 95%) (Fig 5(a)). The estimated AUC of FNA samples against culture in the Xpert application was 0.90 (95% CI: 87–92%) (S5 Fig). The combined sensitivity and specificity of the Xpert MTB/RIF assay utilizing tissue samples compared to culture were 74% (95% CI: 61–85%; I2: 89%) and 77% (95% CI: 20–98%; I2 = 96%), respectively (Fig 5(b)). Notably, in contrast to culture, there was significant heterogeneity observed among the tissue samples. The estimated AUC of tissue samples against culture in the Xpert application was 0.78 (95% CI: 74–81%) (S6 Fig).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Forest plot of the sensitivity and specificity of the Xpert for tuberculosis detection versus culture.

(a) FNA samples. (b) Tissue samples. The squares represent the sensitivity and specificity of a study, and the black line represents their confidence intervals. The diamonds represent the pooled sensitivity and specificity and their confidence intervals.

https://doi.org/10.1371/journal.pone.0321507.g005

Meta-regression analysis revealed that compared with CRS, studies with FNA samples and tissue samples showed different but not statistically significant sensitivity (93% versus 74%, meta-regression P = 0.42) (S7 Fig). Comparative analysis revealed that studies utilizing FNA samples exhibited a lower specificity (88%) in contrast to those employing tissue samples (100%) compared to those employing CRS samples, with statistical significance noted (meta-regression P=0.04) (S8 Fig). While the sensitivity of studies utilizing FNA samples surpassed that of those employing tissue samples in comparison to culture (87% versus 74%), the disparity lacked statistical significance (meta-regression P = 0.29) (S9 Fig). Studies utilizing FNA samples and tissue samples showed the same specificity compared with culture(77% versus 77%), with no significant difference found (meta-regression P = 0.05) (S10 Fig).

Through subgroup and meta-regression analyses, we investigated the heterogeneity among studies by examining predefined subgroups related to decontamination methods (with or without NALC-NaOH), homogenization techniques (mechanical or alternative), and sample conditions (fresh or frozen). Meta-regression analysis revealed that none of these factors had an impact on the diagnostic performance of Xpert MTB/RIF compared to CRS and culture (meta-regression P > 0.01) (S11-13 Fig). Consequently, these variables were not attributed to the heterogeneity observed across the studies.

Discussion

Systematic reviews and meta-analyses published from 2014 to 2019 revealed the excellent diagnostic performance of Xpert MTB/RIF for LNTB [6,25–27]. However, no consensus has been reached on the diagnostic performance of the test to specimens obtained by different methods, including FNA and biopsy specimens, compared with different reference standards.

Similar studies have reported the diagnostic accuracy of Xpert MTB/RIF for different sample types and different reference criteria [26,28]. However, Kohli Mikashmi et al. lack adequate data on lymph node biopsy specimens examined by Xpert MTB/RIF and lack sufficient discussion, which decreases the reliability of the results [28]. Yu, Guocan, et al. were limited by various biases and remarkable heterogeneity [26]. Therefore, our research updated the articles and data included to assess the effectiveness of Xpert MTB/RIF in both biopsy and FNA specimens compared with different reference standards, offering new insights into specimens and reference standards selection for diagnosing LNTB.

Our research revealed that the overall sensitivity of Xpert MTB/RIF employing CRS as a reference standard in LNTB patients was 85%, while its specificity was 97%. Compared with the 85% sensitivity and 78% specificity of Xpert versus culture, employing CRS as a reference standard improved diagnostic performance of Xpert in our study. Similarly, a meta-analysis by Denkinger et al. reported a sensitivity of 81.2% and a specificity of 99.1% for detecting TB with Xpert MTB/RIF versus CRS, respectively. Moreover, it indicated 83.1% sensitivity and 93.6% specificity versus culture for lymph node tissues or FNA samples [6]. The specificity improved when the CRS was used as the gold standard in comparison with culture as a reference standard. CRS is a combination of component tests. Compared with a single imperfect reference standard, CRS combining multiple tests can be transparent and reproducible to reach the final diagnosis of TB [22]. However, subjectivity might be introduced in this process [22]. Samples were categorized as positive if any component test indicated TB. If CRS included more component tests, more TB specimens could be classified as positive and more non-TB samples could be diagnosed as TB wrongly, potentially leading to more false negatives and fewer false positives with Xpert MTB/RIF. With more tests added into CRS, the gain in sensitivity or specificity of CRS decreases and the clinical interpretability may diminish [22]. There is always a trade-off between sensitivity and specificity when combining various component tests [29]. And the diagnostic performance of Xpert could be biased when not all samples receive the CRS including same component tests [30]. To define a standardized and robust CRS in further investigation, complementary component tests that catch each other’s mistakes in detecting TB can be incorporated into CRS by assigning weights to reach the final diagnosis [22]. The reduced specificity observed in lymph node specimens compared to culture methods can be attributed to potential factors such as inadequate collection of specimens [28,31] and the utilization of decontaminated samples for culture inoculation. The latter involves a digestion and purification process using the N-acetyl-L-cysteine and sodium hydroxide method (NALC/NaOH), which may lead to the loss of viable mycobacteria and, consequently, false-negative culture results, particularly in paucibacillary specimens [32,33]. While our results aligned with this study, we observed significant heterogeneity among the studies. This may be explained by multiple reasons. First, the processing of lymph node specimens exhibited significant variability across and within studies, encompassing differences in sample purification with NALC-NaOH and machine homogenization [33]. This may have contributed to high heterogeneous data. Second, LNTB can infect lymph nodes in different parts of the body. Some included studies collected specimens solely in cervical lymph nodes, whereas other studies collected specimens in various lymph node sites such as cervical, axillary, and inguinal [34,35], potentially leading to high heterogeneity among the studies. Third, some studies exclusively included children or HIV-infected patients, possibly introducing some bias in our results [36].

Through meta-regression analysis, the sensitivity of Xpert MTB/RIF for FNA specimens was found to exceed that for tissue samples (93% versus 74% compared with CRS; 87% versus 77% compared with culture). However, this difference was statistically insignificant, regardless of the gold standard. This may be partly explained by the sample reagent buffer may not sufficiently homogenize the tissues, causing a potential decrease on sensitivity [37], while FNA specimens are more likely to be homogenized by contrast. Furthermore, tissue samples obtained through biopsy may contain a mixture of cellular and noncellular components, potentially diluting the concentration of mycobacteria. In contrast, more targeted FNA specimens reduce the dilution effect, making it easier to identify tuberculosis-specific markers [38]. The specificity of Xpert MTB/RIF for FNA specimens was lower than that for tissue specimens (88% versus 100% compared with CRS; 77% versus 77% compared with culture). Overall, the diagnostic accuracy of FNA exceeded that of tissue specimen methods in our study. As a first line diagnostic procedure, FNA is a simple, cost effective and highly accurate method in diagnosing TB. However, due to the limited number of bacilli in the FNA sample and the possibility to miss out extracting tissue in the lesion site, some samples may not be representative in the case of TB lesion, yielding false negatives on Xpert [16]. Repeat aspiration and using ultrasound-guided FNA may improve the quality of FNA and the detection rate of TB with Xpert [39,40].

Nonetheless, as a type of LNTB, mediastinal tuberculous lymphadenitis, endosonography is the recommended technique for diagnosing mediastinal lymphadenopathies. Mondoni, Michele, et al. reported a lower sensitivity (61%) of Xpert MTB/RIF in the diagnosis of mediastinal TB lymphadenitis obtained by endoscopic ultrasound-guided needle aspiration techniques than that reported for FNA samples in our study (93% compared with CRS, 87% compared with culture) and a higher specificity of endosonography compared with our study (89% versus 87% against CRS; 89% versus 77% against culture) [41]. Similarly, Chhajed, Prashant N et al. reported 77% sensitivity of Xpert in diagnosing mediastinal tuberculous lymphadenitis through endosonography [42]. Kohli, Mikashmi et al. reported 88.9% sensitivity and 86.2% specificity of Xpert in diagnosing LNTB [28]. The discrepancy in diagnostic performance may be due to the following reasons. First, mediastinal lymph nodes are more difficult to sample than peripheral lymph nodes and have a low bacillary load. Second, the decontamination procedure or the existence of a caseous necrotic lesion may lead to negative culture results and false-positive Xpert results, lowering the sensitivity of Xpert since such false-positive results could be true-positivity [42]. In addition, if the concentration of mycobacteria is below the detection threshold of Xpert MTB/RIF, the diagnosis could result in Xpert MTB/RIF-negative but TB culture-positive results, decreasing the sensitivity [16]. Third, the indications for mediastinal tuberculous lymphadenitis at the site of aspiration were made only by the pulmonology staff performing endosonography, which can be subjective, and quite a few individuals had enlarged intrathoracic lymph nodes due to a history of malignant diseases, which may have led to incorrect and unnecessary aspiration [43].

According to our study, a multistep approach for managing suspected LNTB is recommended: less invasive FNA, followed by comprehensive examinations such as Xpert MTB/RIF and pathological tests, to enhance diagnostic sensitivity. If the first step yields a negative result, a more invasive technique (such as biopsy) is recommended. Pathological examination, rather than Xpert MTB/RIF, may be employed for further validation, as the assay did not enhance sensitivity when applied to biopsy-obtained samples.

Due to the limited presence of mycobacteria, CRS may be more suitable for detecting LNTB. However, the composition of the CRS differed among the studies analyzed in this research. One potential cause of variability among studies could stem from this factor. Variability existed in the processing procedure of lymph node specimens among studies, encompassing sample conditions, decontamination, and homogenization. Nonetheless, meta-regression analysis did not reveal that the variability markedly impacted the outcomes or contributed to heterogeneity.

We conducted a subgroup analysis of the child and adult populations. The diagnostic Xpert of the child group with FNA samples showed 87% sensitivity and 90% specificity. A review from the WHO reported 86% sensitivity and 81% specificity of Xpert MTB/RIF compared with culture for detecting peripheral lymph node TB in children, with high heterogeneity indicated [44]. The diagnostic Xpert of the adult group with FNA samples showed 80% sensitivity and 75% specificity with culture as the gold standard and 85% sensitivity and 88% specificity with CRS. Similarly, Kohli, Mikashmi, et al. reported 88.9% sensitivity and 86.2% specificity of Xpert MTB/RIF in lymph node aspirates compared with culture as a reference standard, and 81.6% sensitivity and 96.4% specificity against CRS were reported for adults [28]. As LNTB has different epidemiological characteristics in children and adults, we hope that this subgroup analysis of different age ranges can provide a reference and attract more attention to this issue.

Our study suggested that, as a rapid diagnostic tool, Xpert MTB/RIF has excellent performance in detecting rifampicin resistance in LNTB patients, with a pooled sensitivity and specificity of 90% and 99%, respectively. Similarly, a study from Liu, Aimei et al. reported a pooled sensitivity and specificity of 83.8% and 100%, respectively, for detecting rifampicin resistance in suspected suspected drug-resistant TB [45]. Kohli, Mikashmi et al. reported 96.5% sensitivity and 99.1% specificity of Xpert MTB/RIF in detecting rifampicin resistance in extrapulmonary specimens [28]. However, our results are based on only 5 studies and should be interpreted with caution.

Our meta-analysis has limitations. First, we acknowledge the possibility of missing studies despite a thorough search, some of which did not differentiate specimen types. Second, some included studies focused solely on one sample type, whereas one study [33] collected multiple specimens, potentially introducing bias to our results. Besides, our finding is unable to offer more clues about the main strains of LNTB since Xpert MTB/RIF assay can’t distinguish causative agents of TB lymphadenitis between M. tuberculosis, M. bovis or M. orygis [32]. Furthermore, CRS varied among studies. Typically, studies have employed a CRS involving culture, microscopy, or cytology of FNA samples [46,47]. Meanwhile, some studies reported clinical symptoms or responses to anti-TB treatment, which might bias our final research data. In addition, some included studies exclusively included children, whereas others included all age groups or specifically enrolled adults, possibly introducing some bias in our results.

Furthermore, some included studies specifically focused on HIV-infected patients, and certain studies utilized the response to anti-tuberculosis therapy as a standard to screen the target population [48]. Patients in the study were typically referred to the research center from primary health care clinics and presented with more severe cases [48], potentially introducing selection bias. Notably, significant heterogeneity existed among the studies, requiring cautious interpretation of the pooled estimates.

Conclusion

The meta-analysis revealed that in comparison to CRS, Xpert MTB/RIF exhibited a sensitivity of 85% and specificity of 97%. Conversely, compared to culture, the sensitivity and specificity were 85% and 78%, respectively. FNA samples demonstrated a sensitivity of 93% and specificity of 88% against CRS and 87% and 77% against culture, respectively. The tissue samples displayed a sensitivity of 74% and specificity of 100% against CRS and 74% and 77% against culture, respectively. Overall, Xpert MTB/RIF shows significant diagnostic accuracy for LNTB, with FNA samples outperforming tissue samples acquired through biopsy.

Supporting information