Evaluation of the Xpert® MTB/RIF assay for the diagnosis of pulmonary tuberculosis in
a high HIV prevalence setting
Grant Theron1*, Jonny Peter1*, Richard van Zyl-Smit1, Hridesh Mishra2, Elizabeth Streicher3,
Samuel Murray1, Rodney Dawson1, Andrew Whitelaw4, Michael Hoelscher5, Surendra
Sharma2, Madhukar Pai6, Robin Warren3, Keertan Dheda1, 7, 8 †
1Lung Infection and Immunity Unit, Division of Pulmonology & UCT Lung Institute,
Department of Medicine, University of Cape Town, South Africa.
2Department of Medicine, All India Institute of Medical Sciences, India.
3DST/NRF Centre of Excellence for Biomedical TB Research/MRC Centre for Molecular and
Cellular Biology, Stellenbosch University, South Africa.
4Division of Medical Microbiology, University of Cape Town, Cape Town, South Africa.
5Department for Infectious Diseases and Tropical Medicine, Klinikum of the University of
Munich, Munich, Germany.
6Department of Epidemiology & Biostatistics, McGill University, Montreal, Canada.
7Institute of Infectious Diseases and Molecular Medicine, University of Cape Town, South
8Department of Infection, University College London Medical School, United Kingdom.
Requests for reprints should be addressed to Keertan Dheda, H47 Old Main Building, Groote
Schuur Hospital, Observatory, 7925, South Africa or firstname.lastname@example.org
*The first two authors contributed equally.
†Corresponding author: email@example.com, +27214047651 (fax), +27214046509
Page 1 of 37
AJRCCM Articles in Press. Published on April 14, 2011 as doi:10.1164/rccm.201101-0056OC
Copyright (C) 2011 by the American Thoracic Society.
Author contributions: GT, JP, RW and KD designed the study. GT, ES, HM and SM
generated the data. GT, JP and KD wrote the first draft. GT, KD, JP, RvZS, RD, AW, MH,
SS, MP and RW analysed the data and revised the manuscript.
Sources of support: Foundation for Innovative New Diagnostics, SA MRC (KD), SA DST
SARChI (KD), EU-FP7 and EDCTP (KD, GT, JP, RVZS, MP), and the Canadian Institute for
Health Research (MP, KD).
Running title: Xpert® MTB/RIF for TB diagnosis in a HIV prevalence setting
Descriptor number: 11.01
Word count: 4012
Page 2 of 37
At a Glance Commentary:
Scientific knowledge on the subject: Xpert® MTB/RIF is an accurate rapid diagnostic tool
for tuberculosis (TB) and rifampicin resistance. Although it has been recently endorsed by the
World Health Organization, there are limited data about the impact of HIV co-infection and
sample-related factors on test performance. The significance of Xpert® MTB/RIF-positive
culture-negative samples remains unclear.
What this study adds to the field: HIV co-infection, but likely not sputum volume and
processing methods, may impact on assay performance. Almost all Xpert® MTB/RIF-positive
culture-negative samples are likely true positives and diagnostically this approximately
doubles the number of detected TB cases over and above that of smear microscopy.
Page 3 of 37
Rationale/objective: Xpert® MTB/RIF is a novel automated molecular diagnostic recently
endorsed by the World Health Organization. However, performance-related data from high
HIV prevalence settings are limited. The impact of sample-related factors on performance and
the significance of Xpert® MTB/RIF-positive culture-negative discordance remain unclear.
Methods: Xpert® MTB/RIF was evaluated using single archived spot-sputum samples from
496 South African patients with suspected TB. Mycobacterium tuberculosis culture positivity
and phenotypic resistance to rifampicin served as reference standards.
Results: Overall, Xpert® MTB/RIF detected 95% (95% CI: 88-98%; 89/94) of smear-positive
culture-positive cases and the specificity was 94% (91-96%; 320/339). The sensitivity in
smear-negative cases was 55% (35-73%; 12/22) when the analysis was restricted to 1ml of
unprocessed sputum and culture time-to-positivity of ≤28 days. Compared to smear-
microscopy (n=94), Xpert® MTB/RIF detected an additional 17 cases (n=111) representing an
18% (11-27%; 111/94) relative increase in the rapid TB case detection rate. Moreover,
compared to smear-microscopy, the inclusion of Xpert® MTB/RIF-positive culture-negative
TB cases (ruled-in by an alternative diagnostic method) resulted in the detection of a further
16 cases (n=127), thus significantly increasing the rapid TB case detection rate to 35% (95%
CI 26-45%; 111/94 vs. 127/94; p<0.01), the overall specificity to 99.1% (97-100%; 320/323;
p<0.001), and sensitivity in smear-negative TB to 60% (p=0.12). Performance strongly
correlated with smear status and culture time-to-positivity. In HIV-infected compared to
uninfected patients Xpert® MTB/RIF showed a trend to reduced sensitivity (p=0.09) and
significantly reduced negative-predictive-value (p=0.01). The negative-predictive-value for
rifampicin resistance was 99.4%.
Conclusion: Xpert® MTB/RIF outperformed smear-microscopy, established a diagnosis in a
significant proportion of patients with smear-negative TB, detected many highly likely TB
Page 4 of 37
cases missed by culture, and accurately ruled-out rifampicin-resistant TB. Sample-specific
factors had limited impact on performance. Performance in HIV-infected patients, especially
those with advanced immunosuppression, warrants further study.
Key words: smear-negative TB, tuberculosis, human, diagnostics, HIV, PCR
Word count: 290
Page 5 of 37
Tuberculosis (TB) is a major global health priority and kills ~1.7 million people annually (1).
The incidence of multi-drug resistant (MDR) TB is increasing with almost 0.5 million
estimated new cases in 2008 (2). Although smear microscopy is widely used for the rapid
diagnosis of TB, it does not detect drug resistance and sensitivity in HIV co-infected
individuals varies between 20-50% (3). Results of mycobacterial culture often only become
available after 2 to 8 weeks (4). This creates a diagnostic delay that hampers disease control,
enhances transmission, and increases healthcare costs (5).
Xpert® MTB/RIF (Cepheid, Sunnyvale, USA) is an automated user-friendly real-time
polymerase chain reaction (PCR) assay designed for the rapid and simultaneous detection of
Mycobacterium tuberculosis and rifampicin resistance (6-8). The assay amplifies a M.
tuberculosis complex-specific region of the rpoB gene, which is probed with molecular
beacons to detect the presence of rifampicin resistance-determining mutations (9). In
December 2010, the World Health Organization (WHO) endorsed the scale-up of Xpert®
MTB/RIF and recommended its use as the initial test in HIV-TB co-infected patients and
patients with suspected multidrug resistant TB (10, 11). The performance of the test with the
1st generation software using both NALC-NaOH decontaminated and unprocessed sputum
from 1730 patients with suspected TB was recently assessed as part of a large multicentre
study (7). Using a single assay on a single unprocessed sputum sample M. tuberculosis
complex-specific DNA was detected in 98% of smear-positive cases and in 72% of smear-
negative cases using culture positivity as a reference standard. However, only 67 (39%) of all
smear-negative cases were from a high HIV prevalence setting. Moreover, there is no
information regarding Xpert® MTB/RIF performance stratified by CD4 count. Thus, data
Page 6 of 37
about performance in HIV-infected patients, particularly those with smear-negative TB, are
There are several other gaps in our knowledge: (i) There are limited data about the effects of
an alteration in sample volume (using the recommended 1 ml vs. less than 1ml) or processing
methods (raw vs. liquefied sputum) on assay performance. Additionally the relationship
between bacterial load [measured using smear grade and culture time-to-positivity (TTP)] and
assay performance is unclear. These factors have important implications for data
interpretation in sputum scarce patients, the integration of the assay into existing laboratory
work flows, and the design of future clinical trials. (ii) What additional yield Xpert®
MTB/RIF can offer, if any, over culture is unknown. Thus, the significance of Xpert®
MTB/RIF-positive, culture-negative samples remains unclear. (iii) The impact, given resource
constraints, of combining smear microscopy and Xpert® MTB/RIF requires clarification. (iv)
Finally we evaluated, hitherto untested, the specificity of a recently released 2nd generation
software algorithm for the simultaneous detection of M. tuberculosis and rifampicin
Our goal was to further validate test performance using a single cartridge in the context of
high HIV prevalence and to assess the impact of the above factors on assay performance.
These issues gain importance as countries prepare to roll-out and scale-up the Xpert®
MTB/RIF assay (10).
Page 7 of 37
Study sites and population
Sputa were collected from 496 consecutively recruited ambulant patients with suspected TB
(≥ 18 years of age) between February 2007 and April 2010 at 2 primary care clinics in Cape
Town, South Africa. Informed consent was obtained from all participants and the study was
approved by the University of Cape Town, Faculty of Health Sciences Research Ethics
Committee. Detailed patient and laboratory-specific information was recorded on a
standardised case record form and captured using double data entry. An HIV test was
performed after appropriate counselling. All chest radiographs were independently scored by
two trained readers using the CRRS scoring system (12, 13). Chest radiographs were scored
as compatible or unlikely to be compatible with active TB. Discrepant results were
adjudicated by a third senior reader.
Tuberculosis case definitions
Each patient was allocated to 1 of 3 diagnostic categories:
1. Definite TB: A clinical presentation compatible with tuberculosis with at least one
spot sputum sample culture-positive for M. tuberculosis.
2. Probable TB: A clinical-radiological picture highly suggestive of TB and/or anti-TB
treatment was initiated by an attending clinician based on clinical suspicion but the patient did
not meet the criteria for definite TB (no culture-based evidence of M. tuberculosis).
3. Non-TB: No evidence of TB based on smear microscopy and culture, no anti-TB
treatment initiated with response to alternative treatment where appropriate, and when
available, no radiological evidence to support the diagnosis of TB.
Page 8 of 37
At the first visit two paired spot sputa were concurrently collected from each patient. One
arbitrarily selected sample was decontaminated in NALC-NaOH, submitted for routine
concentrated fluorescence smear-microscopy and cultured for M. tuberculosis using the
BACTEC MGIT 960 system (BD Diagnostics, USA) (14). The second sputum sample was
stored (liquefied immediately or as raw sputum) at -20°C for later analysis using the Xpert®
MTB/RIF assay. Patients who returned for post-enrolment follow up provided additional
sputum samples at each visit. Smear grading according to the WHO/IUALTD method was
performed (15). Culture positive isolates underwent phenotypic drug susceptibility testing for
rifampicin and isoniazid using the MGIT 960 SIRE kit (16). Cultures with a TTP of > 28
days and only 1 out of 4 positive follow-up cultures (when available) were considered to be
possible cross contaminants (also analysed separately) (7). Unless otherwise stated, all Xpert®
MTB/RIF and culture results were generated from paired samples taken at the same visit, and
patients accordingly classified.
Sample processing and storage for later analysis
The second sputa from the first 101 patients were liquefied using a 2:1 volume of 0.1%
dithiothreitol (17) prior to storage at -20°C. The remaining 395 samples were unprocessed and
stored at -20°C upon collection.
Sample preparation and Xpert® MTB/RIF procedure
Sputum sample preparation was performed as described previously (6, 18, 19) by a trained
operator blinded to clinical information. Briefly, the sample reagent (Cepheid, USA) was
mixed at a 2:1 ratio with 1ml of sputum (either liquefied or unprocessed) and homogenized. If
the sample volume was <1ml, sterile phosphate buffer (Merck, USA) was added to bring the
Page 9 of 37
final volume to 1ml. 2ml of homogenized mixture was transferred into a Xpert® MTB/RIF
assay cartridge and inserted into the GeneXpert® instrument (6).
Resolution of discordant results
For all Xpert® MTB/RIF and culture discordant results, the cartridge-generated amplicon was
extracted, amplified and sequenced as previously described (6). In addition, a GenoType®
MTBDRplus test (Hain Lifescience, Nehren, Germany) and/or a PCR [using primers and
conditions described previously (6)], followed by sequencing of the reaction products, was
performed on the stored sputum sediment and/or stored sample (18). Post-enrolment sputum
cultures were also analysed and chest radiographs scored for likelihood of TB. Thus,
performance was evaluated based on culture alone (Tables 1 to 3) or a combination of culture
and these additional diagnostic investigations (Table 4). For the resolution of discordance in
rifampicin resistance (phenotypic MGIT culture vs. Xpert® MTB/RIF) a GenoType®
MTBDRplus test was performed on the culture isolate. When appropriate, the rpoB gene from
the sample sediment and/or the culture isolate was amplified and sequenced (6).
Test performance assessment and statistical analysis
For the analysis of assay sensitivity, culture positivity and phenotypic susceptibility to
rifampicin using simultaneously obtained paired samples (Xpert® MTB/RIF versus culture)
were the reference standards. Specificity calculations were based on paired culture-negative
samples from both culture-negative groups (probable and non-TB). Comparative specificity
using the non-TB group only was also obtained. Test performance assessment and chi-
squared analyses were performed using OpenEpi (version 2.3.1) (20). Graphpad Prism
(version 5.0; GraphPad Software, San Diego, USA, www.graphpad.com) was used for the
analysis of linear regression.
Page 10 of 37
After the exclusion of samples from 16 patients (see Figure 1 for details), 480 patients with
suspected TB were eligible for inclusion into the analysis. Figure 1 depicts how samples were
processed and archived, and patients categorised by diagnostic subgroup. Patient demographic
and clinical characteristics are shown in Table 1.
141 of 480 (29%) patients had definite TB. Of these, 94/141 (67%) patients had smear-
positive culture-positive TB whereas 47/141 (33%) had smear negative, culture positive TB.
182 (38%) patients were classified as probable TB whereas 157 (33%) patients were
classified as non-TB. Of the 141 patients with positive cultures from their first sputum, 2
(1%) were found by phenotypic drug susceptibility testing to have multidrug resistant isolates
(resistance to rifampicin and isoniazid), and 21 (15%) were found to have isoniazid mono-
Overall Xpert® MTB/RIF performance
The performance of Xpert® MTB/RIF versus liquid culture performed on a simultaneously
obtained paired spot sputum sample supplied at enrolment is shown in Table 2A. The overall
sensitivity of the assay was 78.7% (95% CI: 71.3-84.7; 111/141). In smear-positive culture-
positive cases the sensitivity was 94.7% (88.2-97.7; 89/94), whereas in smear-negative
culture-positive cases it was 46.8% (33.3-60.8; 22/47).
The overall specificity for the diagnosis of TB using both culture-negative groups (probable
and non-TB) was 94% (91.4-96.4; 320/339). Using only the non-TB group, Xpert® MTB/RIF
specificity was 95% (91.7-98.4; 132/137; p=0.39). The assay was negative in all 8 cases who
Page 11 of 37
were culture-positive for non-tuberculosis mycobacteria (Figure 1), and negative in 14/15
cases with sputum isolates that were contaminated by bacterial overgrowth. Only 1/496
(0.2%) evaluated samples yielded an indeterminate Xpert® MTB/RIF result.
A single Xpert® MTB/RIF assay outperformed smear microscopy and showed an 18% relative
increase in the rapid (potentially within 24 hours) TB case detection-rate (17 additional cases)
compared to 94 smear positive cases and thus detected significantly more patients than smear
microscopy [78.7% (111/141) vs. 66.7% (94/141); p=0.02].
When a positive smear microscopy or Xpert® MTB/RIF result were combined, the sensitivity
improved further to 82.2% (75.1-87.7;116/141) compared to smear microscopy alone [66.7%
(68.5-73.4; 94/141); p<0.01; Table 2A].
Performance of Xpert® MTB/RIF in HIV-infected patients
Smear microscopy was significantly less sensitive in HIV-infected versus uninfected subjects
[50.0% (23/46) vs. 73.2% (60/82); p=0.01] (Table 2A). Although the sensitivity of Xpert®
MTB/RIF was lower in the HIV-infected group, this did not reach significance [69.6%
(32/46) versus 82.9% (68/82); p=0.09]. The same pattern was seen in those with a CD4 count
above versus below 200 cells/ml. Sensitivity of Xpert® MTB/RIF in the smear-negative group
was unaffected by HIV status or CD4 count (Table 2A and Table 2B). By contrast, the NPV
of the Xpert® MTB/RIF decreased significantly in HIV-infected compared to uninfected
patients [93.3% (195/209) vs. 84.6% (77/91); p=0.02], and was lower in those with a CD4
count < 200 cells/ml vs. those with a CD4 count ≥ 200 cells/ml [83.3% (40/48) vs. 87.5%
(34/40); p=0.60]. The same pattern was seen for smear microscopy.
When the assays were directly compared within patient subgroups, the sensitivity of Xpert®
MTB/RIF was higher than smear microscopy in HIV-uninfected and HIV-infected persons,
Page 12 of 37
and those with a CD4 count < 200 cells/ml but this difference was not significant for all 3
groups (Table 2A and Table 2B). The NPV for Xpert® MTB/RIF did not differ significantly
from that of smear microscopy in any of these groups.
The combination of smear microscopy and Xpert® MTB/RIF had a significantly better
sensitivity than smear microscopy alone in HIV-infected patients [73.9% (34/46) vs. 50
(23/46); p=0.02) and in those with a CD4 count <200 cells/ml [69.6% (16/23) vs. 39.1 (9/23);
p<0.05]. Likelihoods ratios stratified by smear status, HIV status and CD4 count are included
in an online data supplement.
Sample processing and volume, and impact of bacterial burden
Sensitivity (77.9% vs. 81.1%; p=0.70) and specificity (94.3% vs. 94.7%; p>0.99) was similar
in 386 unprocessed compared to 94 liquefied samples (Table 3). Similarly, sputum sample
volume had limited impact on sensitivity [71.2% in samples < than the recommended 1ml
(median 575µl; IQR 300-700µl) vs. 83.2% in samples of 1ml (p=0.10); 40.9% in smear-
negative culture-positive samples < 1 ml (median 500µl; IQR 300-600µl) vs. 52.0% in
samples of 1ml (p=0.47)]. Restricting the analysis to include only unprocessed samples with
1ml of available sputum and those with a culture TTP ≤ 28 days [provided no post-enrolment
samples gave a positive culture (21)] did not significantly improve assay sensitivity (46.8% to
54.6%; p=0.56) for smear-negative TB.
Higher bacterial loads, as determined by both smear grade and MGIT TTP, were associated
with earlier detection by assay and more frequent positive results (Figure 2). The average
cycle threshold value was significantly lower in smear-positive compared to smear-negative
cases (22.0 ± 0.5 vs. 32.0 ± 0.9; p<0.0001). A similar relationship was seen using smear grade
Page 13 of 37
and culture TTP as markers of bacterial load (Figure 2). Using spiked sputum samples the
limit of detection of the MTB/RIF assay was found to be 100 CFU/ml and freeze thaw
experiments showed no decrease in assay sensitivity (data not shown).
Performance of Xpert® MTB/RIF for the detection of rifampicin resistance using the 2nd
Xpert® MTB/RIF identified 6 samples as rifampicin resistant. By contrast, MGIT DST
identified 5 of these isolates as sensitive to rifampicin. 5 out of the 6 samples where
confirmed to be genotypically resistant by sequencing of DNA extracted from the isolate
and/or a GenoType® MTBDRplus test. We were unable to reliably compute sensitivity given
the small number of drug-resistant cases but Xpert® MTB/RIF correctly determined
susceptibility to rifampicin in 151/152 cases, and hence the specificity was 99.4% and the
Discordance between Xpert® MTB/RIF and culture (discrepant analysis)
Xpert® MTB/RIF-positive results in the probable group
As shown in Figure 1, there were 19 Xpert® MTB/RIF-positive patients who were culture-
negative based on their simultaneously obtained paired sputum sample. Of these, 5 (26%)
were found to be culture-positive on a second sputum obtained within 2 weeks of enrolment.
A further 10 (53%) had M. tuberculosis DNA detected in their archived sputum using
sequencing and/or a GenoType® MTBDRplus test. In 1 (1%) additional patient the chest
radiograph was compatible with and suggestive of active TB. Furthermore, in all 16 cases
diagnosed with TB using either culture, sequencing of DNA from sputum and/or sample
sediment, a GenoType® MTBDRplus test, or chest radiography, the sequencing of the Xpert®
MTB/RIF cartridge amplicon confirmed M. tuberculosis. Thus, 16 of the 19 Xpert®
MTB/RIF-positive culture-negative patients were deemed likely to be true positives and
Page 14 of 37
designated as ‘highly likely TB’. In the remaining 3 patients a chest radiograph was
unavailable precluding meaningful classification although they were placed on treatment by
the attending clinician based on clinical suspicion.
When the 16/19 patients were combined with the definite TB group and the data reanalysed
(Table 4), there was no significant change in sensitivity [78.7% (111/141) to 80.9%
(127/157); p=0.64; Table 4]. However, specificity [94.4% (320/339) increased to 99.1%
(320/323); p<0.001] and the PPV improved significantly [85.4% (111/130) to 97.7 (127/130);
p<0.001], and the relative increase in the proportion of patients diagnosed compared to smear
microscopy improved significantly [111/94 (18%) to 127/94 (35%); p<0.01]. The PPV in
individuals with smear-negative TB improved significantly [53.4% (22/41) to 92.7% (38/41);
p<0.001] and the sensitivity increased to 60.3%. Patients in the culture-negative Xpert®
MTB/RIF-positive patients group had a higher mean average cycle threshold value compared
to those who were culture positive, Xpert® MTB/RIF-positive [29.3 (23.9-34.0) vs. 23.6
Xpert® MTB/RIF-negative culture-positive results
30 patients were culture-positive but Xpert® MTB/RIF-negative. There was a higher
proportion of smear-negative individuals in this group compared to culture-positive, Xpert®
MTB/RIF-positive individuals (83% vs. 20%; p<0.001). The median TTP (IQR) was
significantly longer in this group [18 (13-25) vs. 7 (6-12) days; p<0.001]. There was no
significant difference in median sample volume across these groups (0.9 vs. 1.0 ml; p=0.13).
Page 15 of 37