Xpert MTB/RIF assay for the diagnosis of rifampicin resistance in different regions: a meta-analysis

Background To estimate the diagnostic accuracy of Xpert MTB/RIF for rifampicin resistance in different regions, a meta-analysis was carried out. Methods Several databases were searched for relevant studies up to March 3, 2019. A bivariate random-effects model was used to estimate the diagnostic accuracy. Results We identified 97 studies involving 26,037 samples for the diagnosis of rifampicin resistance. The pooled sensitivity, specificity and AUC of Xpert MTB/RIF for rifampicin resistance detection were 0.93 (95% CI 0.90–0.95), 0.98 (95% CI 0.96–0.98) and 0.99 (95% CI 0.97–0.99), respectively. For different regions, the pooled sensitivity were 0.94(95% CI 0.89–0.97) and 0.92 (95% CI 0.88–0.94), the pooled specificity were 0.98 (95% CI 0.94–1.00) and 0.98 (95% CI 0.96–0.99), and the AUC were 0.99 (95% CI 0.98–1.00) and 0.99 (95% CI 0.97–0.99) in high and middle/low income countries, respectively. The pooled sensitivity were 0.91 (95% CI 0.87–0.94) and 0.91 (95% CI 0.86–0.94), the pooled specificity were 0.98 (95% CI 0.96–0.99) and 0.98 (95% CI 0.96–0.99), and the AUC were 0.98 (95% CI 0.97–0.99) and 0.99 (95% CI 0.97–0.99) in high TB burden and middle/low prevalence countries, respectively. Conclusions The diagnostic accuracy of Xpert MTB/RIF for rifampicin resistance detection was excellent.


Background
Tuberculosis (TB) remains a major global health problem and ranks as the leading cause of death from an infectious disease worldwide. In 2017, TB infected about 10.0 million people and approximately 16% (1.6 million) of infected patients died from the disease, which was a higher global total for new TB cases and deaths than previous one. Of the 1.6 million died cases, 300,000 occurred among people infected with human immunodeficiency virus (HIV) [1].
Drug-resistant TB, including multidrug-resistant TB (MDR-TB, defined as resistance to at least isoniazid and rifampicin, the two most important first-line anti-TB drugs) and extensively drug-resistant TB (XDR-TB, defined as MDR-TB plus resistance to any fluoroquinolone, such as ofloxacin or moxifloxacin, and to at least one of three injectable second-line drugs, amikacin, capreomycin, or kanamycin) has become a serious threat to global health [2]. In 2017, approximately 460,000 people, which means 3.5% of new and 18% of previously treated TB cases, were estimated to have had MDR-TB globally. And 9.0% of them had developed to XDR-TB. Rifampicin resistance (RR) was the most common resistance drug, affected approximately 558,000 people [1].
When TB is detected and effectively treated, the disease is largely curable. However, accurate and rapid detection of TB can be difficult, as challenging sample collection from deep-seated tissues and the paucibacillary characteristics of the disease [3]. Worldwide, approximately 35% of all forms of TB and 75% of patients with MDR-TB remain undiagnosed [4]. Notablely, under 3% of people who diagnosed with TB are tested to have certain pattern of drug resistance [5]. Xpert MTB/RIF was an effective, rapid, new method to diagnose TB and RR-TB, which was recommended by WHO [1].
Traditionally, the best available reference standard for TB diagnosis is solid and/or liquid culture. However, in clinical practice, prolonged turnaround times and limited laboratory infrastructure in resource-limited settings undermine the utility of culture-based diagnosis [6]. Histology is widely used for the diagnosis of TB where the technical pathologists are available However, it is time-consuming, technically demanding, and lacks specificity [7]. In early 2011, the World Health Organization (WHO) endorsed the Xpert® MTB/RIF assay (Cepheid, Sunnyvale, USA) [8], a novel, rapid, automated, cartridge-based nucleic acid amplification test (NAAT), for the initial diagnosis in patients with suspected pulmonary MDR-TB or HIV-associated pulmonary TB [9,10]. It can simultaneously detect TB through detection of the DNA of Mycobacterium tuberculosis and simultaneously identify a majority of the mutations that confer rifampicin resistance (which is highly predictive of MDR-TB). A high accuracy for pulmonary TB detection (sensitivity 89%, specificity 99%) was obtained [11]. In late 2013, WHO expanded its recommendations to include the diagnosis of TB in children and some forms of extrapulmonary TB (EPTB) [1].
A series of meta-analyses were carried out to determine the diagnostic accuracy of Xpert MTB/RIF in different forms of TB [12][13][14], however, evaluation of its accuracy in rifampicin resistance is rare [11]. More importantly, no study estimated the diagnostic accuracy of Xpert MTB/RIF for rifampicin resistance in countries with different TB prevalence and income till now. To replenish this, in this review, we synthesized the available data, taking into account the accuracy of Xpert MTB/ RIF in diagnosing rifampicin resistance.

Literature search strategy
We searched the MEDLINE, Cochrane library, EMBASE, and Web of Knowledge for published works without language restrictions. The key searching words were used were: "Xpert MTB/RIF", "Xpert", "Gene Xpert", plus "rifampicin resistance". Our last search was accomplished on March 3, 2019.

Study selection and data extraction
The study selection and data extraction procedures were performed by two researchers (Kaican Zong and Hui Zhou) independently. Any differences in the process were solved by discussing with a third author (Shiying Li).

Data extraction
The following data were extracted from each included study: first author, year of publication, country, study settings, gender, the number of patients, the number and type of samples, diagnostic characteristics of Xpert MTB/RIF such as TP, TN, FP and FN. We sent e-mails to the authors for more details when data of individual studies were insufficient for a meta-analysis. In the case of inability to obtain data from the authors, the studies were excluded.

Statistical analysis
MIDAS modules in the STATA statistical software (version 12.0; STATA Corporation, College Station, TX, USA) was used to perform the meta-analyses. The summary receiver operating characteristic (SROC) model and the bivariate random-effects model were used in our study to evaluate the diagnostic accuracy of Xpert MTB/RIF for rifampicin resistance detection. For each study, we calculated the sensitivity and specificity of Xpert MTB/RIF to diagnose rifampicin resistance along with 95% confidence intervals. Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool was introduced to assess the quality of each included study. The Review Manager software (version 5.3, The Nordic Cochrane Centre, Copenhagen, Denmark) was used to present the result of QUADAS assessment.
We assessed the heterogeneity between included studies by using a bivariate boxplot, which can describe the degree of interdependence including the central location and identification of any outliers with an inner oval representing the median distribution of the data points and an outer oval representing the 95% confidence bound (by visually examining the position of each individual study, within the range of boxplot suggesting more heterogeneity).

Description of included studies
Finally, we included 97 studies in this meta-analysis   (Fig. 1), including 26,037 samples for the diagnosis of rifampicin resistance. All studies were in English except five (three in Chinese [46,64,111] and two in Turkish [31,79]). Twenty-six studies (26.8%) were conducted in high income countries (the World Bank income classification 2018) and 52 studies (53.6%) were in the 22 countries with a high burden of TB [1].
The median number of samples per study was 268 for rifampicin resistance detection. The samples of 56 included studies were pulmonary, such as sputum and BAL. Another 15 studies were extrapulmonary samples (e.g. body fluid, FNA, stool and blood), 16 studies included samples of both pulmonary and extrapulmonary (Tables 1 and 2).

Methodological quality of included studies
The overall methodological quality of the included studies was summarized in Fig. 2. Approximately half of the included studies collected data consecutively (n = 41; 42.2%) ( Table 1) and no study used a casecontrol design. All studies were carried out either in tertiary care centers or reference laboratories. In index tests part, 15 studies (15.5%) were considered as unclear risk of bias. In reference standard part, 11 studies (11.3%) were considered as unclear risk of bias because the results of the reference standard were interpreted with unclear blind of the results of the index tests. In flow and timing part, 14 studies (24.7%) were considered as unclear risk of bias because not all patients were included in the analysis.
The heterogeneity of the studies included in this study was tested by a bivariate boxplot (Fig. 3a) and a Deek's funnel plot (Fig. 3b). Most of the included studies were in the bivariate boxplot, and the slope of Deek's funnel was almost horizontal, which all meant a good heterogeneity.

Detection of rifampicin resistance in different prevalence and income regions
The accuracy of Xpert MTB/RIF for rifampicin resistance detection was estimated in 59 studies. The pooled sensitivity, specificity and AUC of Xpert MTB/ Fig. 4 The SROC plot of Xpert MTB/RIF sensitivity and specificity for rifampicin resistance detection. The points represent the sensitivity and specificity of one study; the summary point represents the summary sensitivity and specificity RIF for detecting rifampicin resistance were 0.93 (95% CI 0.90-0.95), 0.98 (95% CI 0.96-0.98) and 0.99 (95% CI 0.97-0.99), respectively (Fig. 4).

Discussion
Several meta-analyses have focused on the diagnostic accuracy of Xpert MTB/RIF for pulmonary [12] or extra-pulmonary TB [13,14] detection either on adults or children [12]. However, to our knowledge, this is the first meta-analysis for Xpert MTB/RIF diagnostic accuracy for rifampicin resistance detection in different prevalence and income regions. Our systematic review demonstrated that Xpert MTB/RIF is high sensitive diagnostic tool for rifampicin resistance detection. Firstly, the accuracy of Xpert MTB/RIF for rifampicin resistance detection was estimated in our meta-analysis. As shown in Fig. 4, the accuracy of Xpert MTB/RIF for rifampicin resistance detection was impressive. The pooled sensitivity, specificity and AUC were 0.93 (95% CI 0.90-0.95), 0.98 (95% CI 0.96-0.98) and 0.99 (95% CI 0.97-0.99), respectively. As estimated, about 75% of multi-drug resistant TB remains undiagnosed [4]. We strongly hope Xpert Fig. 5 The SROC plot of Xpert MTB/RIF sensitivity and specificity for rifampicin resistance detection. a High income countries, b Middle/low income countries, c High TB burden countries, d Middle/low TB prevalence countries. The points represent the sensitivity and specificity of one study; the summary point represents the summary sensitivity and specificity MTB/RIF, which provided a quick and accurate result, will contribute to early and accurate diagnosis of rifampicin resistance.
The overall sensitivity of Xpert MTB/RIF for rifampicin resistance detection were almost the same between high TB prevalence countries and middle/low ones (0.91, 95% CI 0.87-0.94 versus 0.91, 95% CI 0.86-0.94). And for different income levels, the sensitivities of high income ones was also similar with the ones of middle/low income (0.94, 95% CI 0.89-0.97 versus 0.92, 95% CI 0.88-0.94). We can see, taking the different levels of TB prevalence and country income into account, no significant differences were found between subgroups, either in sensitivities, specificities and AUCs.
TB remains one of the world's deadliest communicable diseases. However, it is intensively distributed in several high burden countries. In 2017, more than half of the new TB was developed in the South-East Asia and Western Pacific Regions. To be specific, one quarter were in the African Region. India and China alone accounted for 24 and 13% of the total cases, respectively [4]. Interestingly, the tendency of TB prevalence was consisted with the economic development at some degree. The income levels of the 22 high TB burden countries all were all middle or low, except one (Russian) [4]. Therefore, it is of significant meanings to estimate the diagnostic accuracy of Xpert MTB/RIF in countries with different levels of TB prevalence and income. Some researchers discovered that the Xpert MTB/RIF showed a higher sensitivity of TB detection in lower TB prevalence countries, which could significantly help the physicians to make clinical decisions [112]. However, our result, from another aspect, showed the diagnostic accuracy of Xpert MTB/ RIF for rifampicin resistance detection was not differed between countries with different TB prevalence and incomes.
Advantages of this review were the use of a standard protocol, a bivariate random-effects model used for meta-analysis, and independent reviewers. The data set involved comprehensive searching to identify studies as well as repeated correspondence with authors of study to obtain additional data on the studies.
While there were still some limitations in our analysis. We may have missed some studies despite the comprehensive search. Secondly, sample processing was highly variable across and within studies, as there was no recommendation available on how to process non-respiratory samples from the manufacturer or the WHO.

Conclusions
In conclusion, based on our meta-analysis, the diagnostic accuracy of Xpert MTB/RIF for rifampicin resistance detection was excellent. The overall sensitivity of Xpert MTB/RIF for rifampicin resistance detection in different TB prevalence and income countries were not significant different. We believe that the information obtained from this study will aid the decision making of physicians who take care of patients with possible resistant tuberculosis infection.