Diagnostic characteristics of Xpert MTB/RIF assay for the diagnosis of tuberculous meningitis and rifampicin resistance in Southern Brazil

ABSTRACT Background: The timely diagnosis of tuberculous meningitis (TBM) is challenging. Molecular diagnostic tools are necessary for TBM, particularly in low- and middle-income countries. Objectives: We aimed to calculate the diagnostics characteristics of Xpert MTB/RIF for the detection of Mycobacterium tuberculosis in the cerebrospinal fluid (CSF) and the frequency of rifampicin (RIF)-resistance in the CSF samples. Methods: A total of 313 consecutive CSF samples were studied and categorized into TBM definite, probable, possible, or not TBM cases based on the clinical, laboratory, and imaging data. Results: For the definite TBM cases (n=7), the sensitivity, specificity, efficiency, and positive likelihood ratio were 100, 97, 97, and 38%, respectively. However, for the TBM definite associated with the probable cases (n=24), the sensitivity decreased to 46%. All CSF samples that were Xpert MTB/RIF-positive were RIF susceptible. Conclusion: Xpert MTB/RIF showed high discriminating value among the microbiology-proven TBM cases, although the values for the probable and possible TBM cases were reduced. Xpert MTB/RIF contributes significantly to the diagnosis of TBM, mainly when coupled with the conventional microbiological tests and clinical algorithms.

Central nervous system (CNS) tuberculosis (Tb) is the most severe form of Tb, with a frequency rate of 5% as compared to 84% for pulmonary Tb and 11% for extrapulmonary Tb in the other sites 1 . Tb has re-emerged as a public health problem in the late 1980s because of the HIV epidemic and drug-resistant Mycobacterium tuberculosis (MTb) 2 .
The diagnosis of tuberculous meningitis (TBM) can neither be confirmed nor excluded based on the clinical features 3 . The early symptoms of TBM are non-specific, and diagnostic confirmation is difficult chiefly because TBM is a paucibacillary form of Tb, so the conventional microbiological tests do not help arrive at the diagnosis 4 . The identification of MTb in cerebrospinal fluid (CSF) by Ziehl-Neelsen smear microscopy is central to the diagnosis; however, its sensitivity is low, ranging from 10 to 60% 5 . In the same way, MTb culture is the reference method despite the slow-growing nature of MTb. The time to obtaining a positive result ranges from two to eight weeks, making the culture method ineffective for rapid clinical decisions 4 . Facing these limitations, several nucleic acid amplification (NAA) techniques have emerged and are in regular use to overcome the inadequacies of the conventional methods of laboratory diagnosis, although the exact role of NAA tests remains uncertain 6 . However, no commercial polymerase chain reaction (PCR) assay for MTb has been certified for non-respiratory or paucibacillary samples, including CSF. Xpert MTB/RIF is a robust assay, with the main advantages of a shortened turnaround time and automation of the procedure that reduces hands-on time, as well as the risk of cross-contamination. The assay is robust enough to be performed outside conventional laboratories at the district and sub-district levels of the health system, but requires an uninterrupted power supply. It provides accurate results and can allow rapid initiation of multidrug-resistant tuberculosis/ rifampicin-resistant tuberculosis (MDR-TB/ RR-TB) treatment for cases with pending conventional culture and antimicrobial susceptibility test (AST) results 7 .
The aims of this study were to calculate the diagnostic characteristics of Xpert MTB/RIF in CSF and the frequency of MTb rifampicin (RIF) resistance in the CSF samples. The results of the present study are expected to contribute to the findings of previous studies through further investigation of the diagnostic accuracy of Xpert MTB/RIF on a substantial number of CSF samples obtained from patients from an endemic region with suspected TBM categorized according to the international consensus case definitions 7 , favoring comparisons with other similar studies. The frequency of MTb RIF resistance in the CSF samples is unknown in Brazil.  Table 1.

Tuberculous meningitis case definition and classification
The TBM cases were categorized into definite, probable, possible, and not TBM according to the diagnostic criteria of the international consensus case definitions, which encompass clinical characteristics, CSF changes, neuroimaging criteria, and evidence of tuberculosis in the other sites 7 . These data were assessed by reviewing medical records. The group classified as "not TBM" included CSF samples which were positive for infectious or non-infectious diseases or a score of <6. CSF standard microbiological tests were used to test samples for bacteria, fungus, or viruses 5 . In 60 (55%) CSF samples, other etiologies apart from MTb were identified.

GeneXpert MTB/RIF
The MTb and RIF susceptibility were detected in uncentrifuged CSF samples by GeneXpert MTB/RIF (Cepheid, Sunnyvale, CA, USA) according to the manufacturer's recommendations. However, the GeneXpert MTB/RIF manufacturer has not claimed to use Xpert for extrapulmonary samples, including CSF. The CSF samples were processed as 300−1000 μL aliquots.

Cerebrospinal fluid microbiological testing for M. tuberculosis
The CSF sediment, after concentration by centrifugation, was used to prepare smears for direct examination of acid-fast bacilli (AFB) using Ziehl-Gabbett cold staining 8 . The CSF culture for M. tuberculosis was inoculated onto Löwenstein-Jensen solid medium culture plates, incubated at 37°C, and observed biweekly for two months. The isolates of M. tuberculosis were identified, and antimicrobial susceptibility testing was performed using the BD MGIT 320 system (Becton Dickinson, New Jersey, EUA) according to the manufacturer' s recommendations.

Statistical analyses
The diagnostic characteristics of the Xpert MTB/RIF (index test) in CSF for TBM were calculated using positive CSF acid-fast bacilli culture-proven (definite TBM) and/or probable TBM 7 as the reference method. The following diagnostic characteristics were calculated: sensitivity, specificity, accuracy (efficiency or test score), positive predictive value (PPV), negative predictive value (NPV), Youden' s index (J), error rate=( false positive+false negative)/true positive, false positive rate, presumptive positive, detection rate, error rate, and combined error 9 . The clinical utility index positive and negative (CUI+, CUI−) was classified as follows: utility excellent, ≥0.81; good, ≥0.64; fair, ≥0.49; poor, ≤0.49; and very poor, ≤0.36 10,11 . The positive and negative likelihood ratio (LR+, LR-) 12,13 and diagnostic odds ratio (DOR=LR+/LR-) were also calculated. An LR+ of around ≥10.0 indicated that a positive test almost confirmed the disease, a value of around 6.0 indicated that the disease was present, and a value of around 1.0 indicated that the test was unable to show whether the disease was present or not. An LR+ around ≤0.1 indicated that the disease was practically absent 12,13 . The receiver operating characteristic (ROC) curve was constructed using SPSS version 22 (IBM Corp., Armonk, NY, USA) 12,14 . The post-test probability for a positive and a negative test was calculated using the Fagan nomogram 15,16 .
The concordance between the CSF culture and Xpert MTB/RIF was determined by the kappa statistics (K): complete agreement, K=1, and lack of agreement (i.e., purely random coincidences of rates), K=0. The categorical variables were compared among the groups using the chi-square test, and the continuous variables were compared using the Mann-Whitney or Kruskal-Wallis test for non-parametric data, as appropriate. The results were considered significant at the 5% alpha level.

RESULTS
The Standards for Reporting of Diagnostic Accuracy Studies diagram of the flow of participants through the study is shown in Figure 1. The epidemiological, CSF cytological, and biochemical characteristics of all the groups studied are shown in Table 1. The categorized TBM groups were comparable in terms of age and sex.
The positivity rates of Xpert MTB/RIF in CSF in the different groups categorized according to TBM diagnosis are shown in Figure 2. The positivity rates of the four groups were   significantly different (p<0.001). Xpert MTB/RIF was positive in two cases in the not TBM cases, one of those was RIF susceptible with a diagnosis of CNS lymphoma (mantle cell lymphoma). In this study, it was considered a false positive result; however, this could be due to the association of lymphoma with TBM. The diagnostic characteristics of Xpert MTB/RIF in the different clinically categorized TBM groups are shown in Table 2 and Figure 3. Among the microbiology-proven cases of TBM (definite TBM; n=7), the sensitivity of Xpert MTB/ RIF in CSF was very high (100%), signifying that a positive result is often seen in patients with the disease having CSF-positive MTb culture or smear. The specificity of the test was high (97.39%), i.e., a negative result was often seen in patients without the condition. The PPV of the test was low, implying false positives were common in patients who screen positive, thereby limiting the value of the test for case definition. The NPV was very high, i.e., false negatives were rare in patients who screen negative, thereby suggesting a potentially useful screening test. Hence, the CUI+ of the test for case-findings (confirmation) was poor, and the CUI of the test for screening (ruling out) was excellent. The overall value  of this single test for combined screening and case finding showed a good score of 97.4 (out of 100). Considering definite associated with probable TBM cases (n=24), the sensitivity of Xpert MTB/RIF in CSF was low (45.83%), meaning a positive result was not seen in those with a disease. The specificity of the test was high (98.62%). The PPV of the test was modest, implying false positives are a little common in those who screen positive. The NPV was rather very high. Hence, the CUI of the test for case-finding (confirmation) was very poor, and the CUI of the test for screening (ruling out) was excellent. The overall value of this single test for combined screening and case finding showed a fair score of 94.6 (out of 100). Although considering the definite, probable and possible TBM cases, the sensitivity of Xpert MTB/RIF in CSF was extremely low, and the specificity was high. The overall value of this single test for combined screening and case finding was very poor with a score of 38.7 (out of 100). In the Fagan nomogram analysis of the definite TBM scenario, the posttest probability of a positive test result was 47%, while for definite or probable TBM a positive posterior probability of 73% could be considered to diagnose Tb, and the post-test probability was 4% for a negative test result ( Figure 4).

Rifampicin resistance
Eleven cases with Xpert MTB/RIF positive were RIF susceptible, of these five were in concordance with the M. tuberculosis AST, which is the gold standard 17 . In four cases, the RIF susceptibility by the Xpert MTB/RIF was not reported.

DISCUSSION
The exact role of NAA tests for TBM diagnosis in the CSF sample remains uncertain, although several meta-analyses have been published 6,18,19 . The different case definitions and the different reference standard tests used in studies make comparisons of research findings difficult 6 .
Our data showed, among the microbiology definite TBM, the sensitivity and specificity of the Xpert MTB/RIF in CSF were very high. The overall value of Xpert MTB/RIF for combined screening and case finding was good. Perhaps considering the association of cases with definite or probable TBM, the diagnostic capacity of Xpert MTB/RIF reduced considerably. It accurately identified only 46% of all clinically defined cases, although identified correctly 99% of those who did not have TBM. The area under the ROC curve (AUC) reflects the diagnostic value in distinguishing between patients with the disease and those without 14 . In the present study, MTb Xpert showed a high discriminating value among the microbiology-proven TBM cases, thereby reducing the value among the probable and possible TBM cases. These diagnostic characteristics for definitive and probable TBM were similar to those described previously 6,20,21 .
The low prevalence (5.4%) of the positive Xpert MTB/RIF results identified in this study is similar to those reported previously 22,23 . These findings reveal the low pretest probability of disease using Xpert MTB/RIF in routine practice and highlight the need to implement algorithms to optimize the test requested 23 . This is particularly important in scenarios with limited resources in which shortages are a constant concern. The possible reason for the limited sensitivity of Xpert MTB/RIF could be the low CSF volume (mostly the laboratory received <1 mL of CSF specimens) referred to the IDMBU to perform the assay, although the impact of CSF volume was not evaluated in this study. The other reason could be the presence of inhibitory factors previously documented in the CSF samples, such as the presence of red blood cells (RBC) or proteins 24,25,26 , although the literature on this subject is scarce. Traumatic CSF punctures are frequent in clinical practice, and the Xpert MTB/RIF was not validated for blood samples. However, a previous study with real-time PCR for the detection of M. tuberculosis (MTb qPCR) in the CSF showed that the number of RBC in the CSF did not inhibit the MTb qPCR 27 .
In 2013, the World Health Organization (WHO) recommended that Xpert should be used in place of conventional microscopy and culture as the first diagnostic test in CSF samples in patients with suspected TBM 19,28 irrespective of the fact that its sensitivity varied widely in the included studies 18 . Although the Xpert manufacturer has not recommended the use Xpert for samples other than sputum, several studies have been conducted to evaluate the utility of Xpert in the diagnosis of extrapulmonary TB, including TBM 19 . Therefore, although Xpert could be useful in patients with TBM, a negative result does not exclude the possibility 28 .
All CSF samples with Xpert MTB/RIF-positive were RIF susceptible; in five of these cases, the drug susceptibility was confirmed by CSF culture and DST. The bacteriological evidence served as the gold standard for the detection of MDR-TB/RR-TB 17 . To the best of our knowledge, the susceptibility of MTb to RIF in CSF samples has not been reported earlier in Brazil, however the number of samples studied was small for a definite conclusion, warranting further studies. Xpert detects the MTb rpoB gene and allows to detect mutations that may confer resistance to rifampicin 29,30 . MDR-TB/ RR-TB is emerging as a major problem due to poor management of drug-sensitive, as well as drug-resistant Tb 17 . MDR-TB/RR-TB has been an area of growing concern to human health worldwide and posing a threat to the control of Tb. The Global Tb Report 2016 estimated that of the 3.9% newly diagnosed Tb cases, 21% of previously treated Tb cases had MDR-TB 17 . In India, the prevalence of MDR M. tuberculosis was 5.6% in TBM patients 31 . In Brazil, the M. tuberculosis primary resistance to RIF (in patients that never received tuberculosis treatment, i.e. infected with MTb previously resistant to treatment) in sputum samples was 0.6% and the acquired resistance (in patients previously with MTb initially sensible, who turned resistant after the exposition to treatment) was 2.5%, combined resistance was 1.0% 32,33 .
Other methodologies have now been developed for MTb diagnosis, such as Xpert MTB/RIF Ultra (Cepheid, Inc., Sunnyvale, CA, USA), which have appealed as new diagnostic tools, although they have not been yet validated for CSF and tend to show several limitations when used in CSF samples 34,35,36,37 . The use of GeneXpert MTB/RIF Ultra for TBM diagnosis was also endorsed by the WHO in the 2017, due to its high sensitivity, especially in patients with HIV co-infection 38 . Hence, further studies are necessary to establish the real diagnostic accuracy of these tests in CSF. In addition, a comparison of the results among different methods would be useful. Several low-and middle-income countries, which are precisely the countries with a higher Tb prevalence, do not have access to these robust technology methods, highlighting the need to validate the efficiency of more accessible NAA tests in evaluating CSF.
The strength of this study is that the authors included all CSF samples evaluated by Xpert MTB/RIF within the study period in a sequential manner. For the first time, MTb RIF susceptibility to TBM CSF samples was studied in an area endemic for MTb in Brazil, especially the Southern parts. The main limitations are its retrospective design and the small number of CSF samples that were MTb culture-positive. Considering the association of definite and probable TBM, the number of cases in this study was similar to that reported in other studies 34,35 . The CSF samples of definite TBM are rare due to the difficulty to detect MTb in the CSF. In our study, the rate of CSF culture positivity for MTb was 29%; considering the groups with definite and probable TBM, this result was in accordance with that of the previous studies 27,39,40 . TBM accounts for 5% of all Tb cases; among which, 1.25% account for culture-proven TBM cases. This explains the low number of definite cases in the present study and the difficulty in conducting studies on the diagnosis and treatment of TBM 1 .
The analytical sensitivity of molecular biology methods is higher than that of the traditional culture methods (reference method). This is an intrinsic characteristic of molecular biology methods, and it is difficult to analyze the diagnostics characteristics of these new technologies. Because of these, it is suggested to add clinical criteria as a reference method with the use of molecular biology methods.
Commercial NAA tests generally show high specificities and can, therefore, be used to establish a definitive diagnosis in patients with symptoms or signs suggestive of TBM. However, more data are urgently needed to establish the robustness of these tests in field conditions; the specificities of both culture and PCR methods might be compromised in areas endemic for tuberculosis because of an increased risk of sample cross-contamination. The specificity of TBM diagnosis can be improved by molecular diagnostics tests 7 . Our data support the Xpert MTB/RIF assay for TBM diagnosis using CSF samples. It has the potential for use in clinical settings. In particular, in patients with positive TBM clinical scores, Xpert MTB/RIF should be used as an adjunct to conventional microbiological tests and clinical algorithms.