Articles

Assessment of the sensitivity and specificity of Xpert

MTB/RIF assay as an early sputum biomarker of response to
tuberculosis treatment
Sven O Friedrich*, Andrea Rachow*, Elmar Saathoff, Kasha Singh, Chacha D Mangu, Rodney Dawson, Patrick P J Phillips, Amour Venter,
Anna Bateson, Catharina C Boehme, Norbert Heinrich, Robert D Hunt, Martin J Boeree, Alimuddin Zumla, Timothy D McHugh, Stephen H Gillespie,
Andreas H Diacon, Michael Hoelscher, on behalf of the Pan African Consortium for the Evaluation of Anti-tuberculosis Antibiotics (PanACEA)

Summary
Lancet Respir Med 2013; Background An accurate biomarker is urgently needed to monitor the response to treatment in patients with
1: 462–70 pulmonary tuberculosis. The Xpert MTB/RIF assay is a commercially available real-time PCR that can be used to
Published Online detect Mycobacterium-tuberculosis-specific DNA sequences in sputum samples. We therefore evaluated this assay with
July 1, 2013
http://dx.doi.org/10.1016/
serial sputum samples obtained over 26 weeks from patients undergoing treatment for tuberculosis.
S2213-2600(13)70119-X
See Comment page 427 Methods We analysed sputum samples from 221 patients with smear-positive tuberculosis enrolled at two sites (Cape
*Both authors contributed
Town, South Africa, and Mbeya, Tanzania) of a multicentre randomised clinical trial REMoxTB of antituberculosis
equally treatment on a weekly basis (weeks 0 to 8), then at weeks 12, 17, 22, and 26 after treatment initiation. The Xpert
Division of Medical MTB/RIF results over time were compared with the results of standard smear microscopy and culture methods.
Physiology, Department of
Biomedical Sciences, Faculty Findings We obtained and analysed 2741 sputum samples from 221 patients. The reduction in positivity rates with
of Medicine and Health
Sciences, Stellenbosch
Xpert MTB/RIF were slower than those with the standard methods. At week 8, positive results were obtained for
University, Cape Town, South 62 (29%) of 212 sputum samples with smear microscopy, 46 (26%) of 175 with solid culture (Löwenstein-Jensen
Africa (S O Friedrich PhD, medium), 77 (42%) of 183 with liquid culture (Bactec MGIT960 system), and 174 (84%) of 207 with Xpert MTB/RIF;
Prof A H Diacon MD); National at 26 weeks, positive results were obtained for ten (5%) of 199, four (3%) of 157, seven (4%) of 169, and 22 (27%) of 83
Institute of Medical
Research–Mbeya Medical
sputum samples, respectively. The reduction in detection of quantitative M tuberculosis DNA with Xpert MTB/RIF
Research Centre, Mbeya, correlated with smear grades (ˮ=–0·74; p<0·0001), solid culture grades (ˮ=–0·73; p<0·0001), and time to liquid
Tanzania (A Rachow MD, culture positivity (ˮ=0·73; p<0·0001). Compared with the combined binary smear and culture results as a reference
C D Mangu MD); Division of
standard, the Xpert MTB/RIF assay had high sensitivity (97·0%, 95% CI 95·8–97·9), but poor specificity (48·6%,
Infectious Diseases and
Tropical Medicine, Medical 45·0–52·2).
Centre of the University of
Munich (LMU), Munich, Interpretation The poor specificity precludes the use of the Xpert MTB/RIF assay as a biomarker for monitoring
Germany (A Rachow,
tuberculosis treatment, and should not replace standard smear microscopy and culture.
E Saathoff PhD, N Heinrich MD,
Prof M Hoelscher FRCP); DZIF
German Centre for Infection Funding Global Alliance for TB Drug Development, Bill & Melinda Gates Foundation, UK Medical Research Council,
Research, Munich, Germany German Ministry of Science and Technology.
(A Rachow, Prof M Hoelscher);
Centre for Clinical
Microbiology, Division of Introduction and non-tuberculosis mycobacteria reliably6 and between
Infection and Immunity, Substantial progress has been made in global tuberculosis live and dead bacilli. Sputum culture takes a long time, is
University College London, control but these gains are threatened by the global spread prone to contamination, is expensive, and is rarely
London, UK (K Singh MD,
of drug-resistant tuberculosis.1 The urgency of this available in high-burden settings. In the absence of an
A Bateson PhD, R D Hunt PhD,
Prof A Zumla FRCP, situation calls for better treatment markers that can be ideal marker to measure treatment success in individuals,
Prof T D McHugh PhD); used to detect insufficient treatment—the main cause of WHO currently recommends smear microscopy at
Radboud University Nijmegen the emergence of resistance—early and easily. Additionally, month 2 to identify individuals at risk of poor treatment
Medical Centre/UCCZ
Dekkerswald, Nijmegen
the clinical development of the emerging new portfolio of outcome. Also, in recent clinical studies of bedaquiline
(M J Boeree MD); Division of tuberculosis drugs2 could be accelerated if phase 2 and 3 (TMC207)7,8 and delamanid (OPC-67683),9 both
Pulmonology, Department of trials could be greatly shortened and reliable surrogate biomarkers have been used to predict the therapeutic
Medicine, University of Cape markers used for the measurement of treatment success.3 value of early drug regimens. Sputum-culture conversion
Town Lung Institute, Groote
Schuur Hospital, Cape Town,
Current biomarkers that have been used as predictors at 8 weeks has some predictive value for the measurement
South Africa of non-relapsing cure are sputum-smear microscopy and of the sterilising activity of a novel regimen.10 Potentially
(Prof R Dawson MD); Medical sputum-culture conversion,4 both of which have more promising, but not yet sufficiently validated, is the
Research Centre (MRC) Clinical shortcomings as markers for the measurement of time to culture positivity measured with semiautomated
Trials Unit, London, UK
(P P J Phillips PhD); MRC Centre
treatment response. Sputum-smear microscopy is liquid culture.11–13
for Molecular and Cellular insensitive, operator dependent, and has been shown to Although several other M tuberculosis and human
Biology, Department of be a poor predictor of treatment outcome.5 It cannot be biomarkers have been studied over the past decade,14,15
Biomedical Sciences, Faculty used to differentiate between Mycobacterium tuberculosis including M tuberculosis DNA-based and RNA-based

462 www.thelancet.com/respiratory Vol 1 August 2013

 Articles

assays,16–18 no new specific and accurate biomarkers of 18 months. To avoid interference with the endpoints of the of Medicine and Health
disease activity and relapse have yet been validated. main treatment trial, the sample analysis with Xpert Sciences, Stellenbosch
University, Cape Town, South
Therefore, an urgent need exists for biomarkers that can MTB/RIF was restricted until week 26 and all investigators Africa (A Venter NatDipMicro);
be used to accurately measure bacillary clearance and were still masked to the treatment allocations. The results Foundation of Innovative
identify individuals receiving insufficient treatment with of the main study will become available in 2014. New Diagnostics, Geneva,
high risk of tuberculosis recurrence. Furthermore, for Ethical and regulatory approval, granted by all local and Switzerland (C C Boehme MD);
and University of St Andrews
innovative trial designs in which the aim is to reduce the national ethics committees and regulatory authorities, School of Medicine, North
size and duration of clinical trials, the use of faster included the permission to undertake studies to improve Haugh, St Andrews, Scotland,
surrogate endpoints for treatment efficacy, disease treatment and diagnosis of tuberculosis. UK (Prof S H Gillespie FRCP)
activity, cure, and relapse will be indispensable.3 Correspondence to:
WHO approved the Xpert MTB/RIF assay (Cepheid, Sputum samples Prof Stephen H Gillespie,
University of St Andrews School
Sunnyvale, CA, USA) for sputum-based rapid diagnosis All participants were invited for regular visits to the study of Medicine, North Haugh
of pulmonary tuberculosis and multidrug-resistant clinics and asked to provide one early morning sputum St Andrews, Fife KY16 9TF,
(MDR) tuberculosis. The assay can be used to accurately sample, which was spontaneously expectorated at home Scotland, UK
measure the M tuberculosis load beyond the detection or at the clinics before treatment (week 0) and then shg3@st-andrews.ac.uk

limit of 131 organisms per mL in an in-vitro suspension.19,20 weekly until week 8, followed by monthly sputum sample
The Xpert MTB/RIF assay in patients with suspected collection (weeks 12, 17, and 22) until week 26 after
tuberculosis and newly diagnosed cases of tuberculosis treatment initiation. Sputum specimens were processed
has been evaluated in several studies.21–24 The results of by two study laboratories (Department of Biomedical
these studies suggested that the Xpert MTB/RIF assay is Sciences, University of Stellenbosch, and the tuberculosis
better than sputum-smear microscopy; its quantitative laboratory at the National Institute for Medical
readouts correlate well with results of conventional solid Research-Mbeya Medical Research Centre) in accordance
and liquid cultures.21–24 Specifically, for the detection of with the standardised REMoxTB laboratory manual at
tuberculosis and MDR-tuberculosis in HIV-positive both laboratories. Briefly, a maximum of 10 mL of sample
individuals, a cost benefit was noted compared with was homogenised and decontaminated with the
conventional smear microscopy.25,26 In this respect, N-acetyl-L-cysteine-sodium hydroxide procedure.27 After
detection of M tuberculosis DNA in serial sputum samples centrifugation for 15 min at 4°C and 3000×g, the resulting
over time with the Xpert MTB/RIF assay could have the pellet was stored at 4°C for further analyses.
potential to replace conventional microbiological tests for 30 μL of the concentrated sputum pellet was used for
monitoring response to tuberculosis treatment regimens. Ziehl-Neelsen staining followed by smear microscopy;
We evaluated the Xpert MTB/RIF assay in serial sputum 1·5 mL phosphate-buffered saline (pH 6·8; Becton
samples obtained over 26 weeks of tuberculosis treatment Dickinson, Sparks, MD, USA) was added and the pellet
from patients enrolled in the multicentre randomised was resuspended. 500 μL of the suspension was used for
clinical trial REMoxTB (ClinicalTrials.gov, NCT00864383). evaluation in liquid culture with the Bactec MGIT960
system (MGIT; Becton Dickinson) and 150 μL in solid
Methods culture on Löwenstein-Jensen medium (Becton
Patients and study design Dickinson) with standard protocols.
In this substudy, we obtained sputum samples from a The remaining sputum pellet was mixed with 2 mL of
subset of 221 participants enrolled in the multicentre, the phosphate-buffered saline and vortexed. 1 mL was
randomised clinical drug trial REMoxTB at two of the processed for evaluation in the Xpert MTB/RIF assay
47 clinical trial sites in Cape Town, South Africa, and according to the manufacturer’s instructions by use of
Mbeya, Tanzania, between January, 2008, and April, 2010. the GeneXpert Dx software (version 2.1 and 4.0) for
In the REMoxTB trial, 1931 patients with previously automation of all PCR processing, reaction, and detection
untreated, drug-sensitive, smear-positive, pulmonary steps. The signals for fluorescence were recorded,
tuberculosis were enrolled between January, 2008, and
January, 2012. HIV-positive patients with a CD4-cell count
of fewer than 2·5 × 10⁸ per L or those already on Both sites (n=221) Mbeya (n=90) Cape Town (n=131) p value*
antiretroviral treatment were excluded to avoid Age (years) 30 (23–40) 30 (25–39) 30 (23–41) 0·43
concomitant antiretroviral treatment, according to current Sex, male 144 (65%) 57 (63%) 87 (66%) 0·64
practice at the time the protocol was prepared. All HIV-positive status 20 (9%) 16 (18%) 4 (3%) <0·0001
participants were treated for tuberculosis either with one CD4-cell count×108 per L in 3·62 (2·91–4·52) 3·59 (2·91–4·96) 3·92 (3·10–4·31) 0·56
of two treatment-shortening regimens of 4 months, in participants who are
HIV positive
which moxifloxacin was substituted for either isoniazid or
ethambutol or with the standard 6 month regimen Data are number (%) or median (IQR), unless otherwise indicated. *For difference between sites, from Wilcoxon
(2 months of isoniazid, rifampicin, ethambutol, and rank-sum test (binary variables) or two-sample test of proportions (continuous variables).
pyrazinamide followed by 4 months of isoniazid and Table 1: Demographic and clinical characteristics of study participants
rifampicin). Every patient was followed up for a total of

www.thelancet.com/respiratory Vol 1 August 2013 463

 Articles

analysed, and reported as semiquantitative readouts CTmin values over time were analysed by use of the non-linear
based on the minimal cycle threshold (CTmin), which is least-squares estimation and non-linear mixed-effects
defined as one of five probes specific for M tuberculosis models to better account for between-patient variability.
complex becoming positive for detection.2 The same biexponential model was used previously to fit
to these type of data.18,29 Stata software (version 12.1) was
Statistical analysis used for all statistical analyses with the exception of the
A negative result for smear microscopy was reported if no non-linear mixed-effects models, which were fit using the
acid-fast bacilli were detected in at least 100 observation nlme package in R (version 2.9.0).
fields. Cultures in MGIT and on Löwenstein-Jensen media
were judged to have a negative result if no mycobacterial Role of the funding source
growth was seen until 6 weeks and 8 weeks after The study was a collaboration between the Global TB
incubation, respectively. A tuberculosis negative result for Alliance and the Pan African Consortium for the
the Xpert MTB/RIF assay was generated automatically Evaluation of Anti-tuberculosis Antibiotics (PanACEA).
following an interpretative algorithm with the GeneXpert This study was undertaken with permission from the
Dx software. Apart from calculating binary (positive, REMox trial sponsor, but the study design, data collection,
negative) variables for all test outcomes, results were analysis, and interpretation, and writing of the report
processed to analyse them quantitatively: the intensity were done independently. The corresponding author had
grades for smear (0, 1+, 2+, 3+, 4+)28 and Löwenstein-Jensen full access to all the data in the study and had final
(negative, <20 colonies, 1+, 2+, 3+) were coded into five responsibility for the decision to submit for publication.
categories from 0 (negative) to 4. Time to positivity of
MGIT and CTmin of the Xpert MTB/RIF assay were used as Results
quantitative outcomes and both were coded as 50 days and 2741 sputum specimens were obtained and analysed from
50 cycles, respectively, if negative. Diagnostic test 221 patients (131 in Cape Town, 90 in Mbeya) from
performance (sensitivity, specificity, negative predictive weeks 0–26 after initiation of antituberculosis treatment.
value, positive predictive value) and Spearman rank Table 1 shows the demographic characteristics of the
correlations were calculated with standard formulae. In patients.
the absence of a perfect and accurate marker for Figure 1 shows the qualitative data and table 2 the
tuberculosis disease activity, treatment response, or quantitative data for all tests at baseline and follow-up
treatment outcome, either the combined result for smear visits. Valid smear results were available for 2732 samples
microscopy, Löwenstein-Jensen culture, and MGIT and valid Xpert MTB/RIF assay results for 2000 samples.
culture, or each individual test was used as reference After exclusion of acid-fast-bacillus-negative, but
standard for the calculation of accuracy of the Xpert contaminated, cultures and those with no speciation
MTB/RIF assay. Differences between groups were tested result, 2324 and 2383 results for Löwenstein-Jensen and
for significance with the two-sample proportion Z test for MGIT cultures, respectively, were included in the
binary variables and the non-parametric Wilcoxon rank qualitative analysis (figure 1). There were 262 (11%) of
sum test for continuous variables because none of the 2383 acid-fast-bacillus-positive MGIT cultures showing
results conformed to a normal distribution. Longitudinal contamination; these were excluded from the calculation

100 Xpert MTB/RIF assay

Bactec MGIT960 system
90 Löwenstein-Jensen medium
Smear microscopy
80

70
Test positive (%)

60

50

40

30

20

10

0
0 1 2 3 4 5 6 7 8 12 17 22 26
Weeks after treatment initiation

Figure 1: Qualitative data for all tests at baseline and follow-up visits

464 www.thelancet.com/respiratory Vol 1 August 2013

 Articles

of the mean time to positivity (table 2). Because only all negative results from the analysis were excluded. In
patients with smear-positive tuberculosis were included model 2, a patient’s first negative result was imputed with
in the trial after screening, 92% and 98% of baseline a CTmin of 50 and the subsequent negative results were
values for smear microscopy and all other diagnostic excluded. In model 3, all negative results were imputed
tests, respectively, were positive for M tuberculosis with a CTmin of 50. The raw means (excluding negative
(figure 1; table 2). Similarly, at week 0, quantitative results) with 95% CIs and raw medians were plotted by
measurements were lowest for Xpert MTB/RIF CTmin and visit week. Figure 2 shows the fitted line for each model
MGIT time to positivity, and highest for smear and with the raw means and medians for CTmin. During the
Löwenstein-Jensen grade (figure 1; table 2). The sputum first 7 weeks, when there were few negative results with
load of M tuberculosis measured with smear and culture the Xpert MTB/RIF assay, the lines for the mean and
tests fell rapidly and consistently after induction of median values were close, indicating that the distribution
treatment (figure 1; table 2). A delayed, less rapid, and of CTmin was symmetrical. Assuming that the underlying
almost linear fall in the mycobacterial load was noted distribution of CTmin remains symmetrical throughout the
with the Xpert MTB/RIF assay (figure 1). For instance, observation period, the median is judged to be a better
84% of the sputum samples were still positive with the summary measure than the mean for calculations with
Xpert MTB/RIF assay at 8 weeks after treatment initiation the symmetrically distributed data for CTmin because no
whereas 29% were smear positive (table 2). Furthermore, assumptions are needed for values greater than the limit
at the end of the observation period (week 26), 27% of the of detection, as is the case for models 1–3. Therefore,
samples were positive with the Xpert MTB/RIF assay but although, the increase in mean CTmin suggested a shallower
only 3% and 4% with the Löwenstein-Jensen and MGIT slope after week 8, the median continued to increase
cultures, respectively (table 2). almost linearly until week 17. The lines for all models had
To further elaborate the findings, we used three linear slopes with slightly flattened curves after week 12,
different statistical models to calculate the increase in depending on the inclusion of the negative results as
CTmin with the Xpert MTB/RIF assay (figure 2). In model 1, determined by the model. Numerical values for the

Week 0 Week 1 Week 2 Week 3 Week 4 Week 5 Week 6 Week 7 Week 8 Week 12 Week 17 Week 22 Week 26
Smear 204/221 198/218 173/217 150/215 127/212 105/207 83/206 74/210 62/212 31/210 13/202 3/203 10/199
microscopy, (92%) (91%) (80%) (70%) (60%) (51%) (40%) (35%) (29%) (15%) (6%) (1%) (5%)
positive
results
Smear 3·3 2·9 2·3 1·8 1·5 1·2 0·8 0·7 0·5 0·3 0·1 0·0 0·1
microscopy, (3·1–3·4) (2·7–3·1) (2·1–2·5) (1·6–2·0) (1·3–1·7) (1·0–1·3) (0·6–1·0) (0·5–0·8) (0·4–0·6) (0·2–0·4) (0·1–0·2) (0·0–0·1) (0·0–0·2)
grade
Löwenstein- 188/191 195/202 176/189 172/194 145/178 122/179 100/184 65/176 46/175 14/179 0/168 3/152 4/157
Jensen (98%) (97%) (93%) (89%) (81%) (68%) (54%) (37%) (26%) (8%) (0%) (2%) (3%)
medium,
positive
results
Löwenstein- 3·6 3·0 2·5 2·0 1·7 1·2 0·9 0·6 0·5 0·2 0·0 0·1 0·1
Jensen (3·5–3·7) (2·8–3·1) (2·3–2·7) (1·8–2·2) (1·5–1·8) (1·1–1·4) (0·8–1·1) (0·4–0·7) (0·3–0·6) (0·1–0·3) (0·0–0·0) (0·0–0·1) (0·0–0·2)
medium,
grade
Bactec 215/219 205/210 186/195 173/187 154/184 128/170 124/174 95/173 77/183 27/181 1/168 3/170 7/169
MGIT960 (98%) (98%) (95%) (93%) (84%) (75%) (71%) (55%) (42%) (15%) (<1%) (2%) (4%)
system,
positive
results
Bactec 6·0 10·2 13·5 16·0 21·3 24·9 29·2 35·0 39·0 46·0 49·7 49·2 48·5
MGIT960 (5·0–7·0) (9·1–11·2) (12·0–15·0) (14·1–17·8) (18·9–23·6) (22·2–27·6) (26·4–32·0) (32·3–37·7) (36·5–41·4) (44·1–47·8) (49·2–50·3) (48·4–50·1) (47·3–49·7)
system, time
to positivity
(days)
Xpert MTB/ 215/219 214/219 85/87 85/88 196/212 85/89 78/86 80/89 174/207 131/198 107/198 72/196 22/83
RIF assay, (98%) (98%) (98%) (97%) (92%) (96%) (91%) (90%) (84%) (66%) (54%) (37%) (27%)
positive
results
Xpert MTB/ 19·0 20·4 20·2 21·7 25·1 24·2 26·1 27·1 29·2 34·6 38·4 42·4 44·2
RIF assay, CTmin (18·1–19·8) (19·6–21·3) (19·0–21·4) (20·3–23·1) (23·9–26·3) (22·7–25·7) (24·3–28·0) (25·2–29·0) (27·9–30·6) (32·9–36·2) (36·9–40·0) (40·9–43·8) (42·1–46·4)

Binary data are n/N (%) and quantitative data are mean (95% CI).

Table 2: Quantitative data for all tests at baseline and follow-up visits

www.thelancet.com/respiratory Vol 1 August 2013 465

 Articles

Negative

Actual values
45 Raw mean (95% CI)
Raw median
Model 1
Model 2
40
Model 3

35
CTmin

30

25

20

15

10

0 1 2 3 4 5 6 7 8 12 17 22 26
Weeks after treatment initiation

Figure 2: Increase in CTmin during tuberculosis treatment fitted with three different statistical models
Note break in y axis. The horizontal line is the positivity cutoff value. CTmin=minimal cycle threshold.

Number of Sensitivity (true positives/false Specificity (true negatives/false Positive predictive Negative predictive
comparisons negatives; %, 95% CI) positives; %, 95% CI) value (%, 95% CI) value (%, 95% CI)
Week 0 219 214/3 (98·6, 96·0–99·7) 1/1 (50·0, 1·3–98·7) 99·5 (97·4–100·0) 25·0 (0·6–80·6)
Week 1 216 210/4 (98·1, 95·3–99·5) 1/1 (50·0, 1·3–98·7) 99·5 (97·4–100·0) 20·0 (0·5–71·6)
Week 2 85 83/1 (98·8, 93·5–100·0) 1/0 (100·0, 2·5–100·0) 100·0 (95·7–100·0) 50·0 (1·3–98·7)
Week 3 87 82/1 (98·8, 93·5–100·0) 2/2 (50·0, 6·8–93·2) 97·6 (91·7–99·7) 66·7 (9·4–99·2)
Week 4 206 178/3 (98·3, 95·2–99·7) 11/14 (44·0, 24·4–65·1) 92·7 (88·1–96·0) 78·6 (49·2–95·3)
Week 5 87 78/1 (98·7, 93·2–100·0) 3/5 (37·5, 8·5–75·5) 94·0 (86·5–98·0) 75·0 (19·4–99·4)
Week 6 86 65/2 (97·0, 89·6–99·6) 6/13 (31·6, 12·6–56·6) 83·3 (73·2–90·8) 75·0 (34·9–96·8)
Week 7 86 44/4 (91·7, 80·0–97·7) 5/33 (13·2, 4·4–28·1) 57·1 (45·4–68·4) 55·6 (21·2–86·3)
Week 8 200 101/6 (94·4, 88·2–97·9) 25/68 (26·9, 18·2–37·1) 59·8 (52·0–67·2) 80·6 (62·5–92·6)
Week 12 196 44/8 (84·6, 71·9–93·1) 57/87 (39·6, 31·5–48·1) 33·6 (25·6–42·4) 87·7 (77·2–94·5)
Week 17 191 12/1 (92·3, 64·0–99·8) 86/92 (48·3, 40·8–55·9) 11·5 (6·1–19·3) 98·9 (93·8–100·0)
Week 22 179 3/1 (75·0, 19·4–99·4) 115/60 (65·7, 58·2–72·7) 4·8 (1·0–13·3) 99·1 (95·3–100·0)
Week 26 80 3/0 (100·0, 29·2–100·0) 59/18 (76·6, 65·6–85·5) 14·3 (3·1–36·3) 100·0 (93·9–100·0)
Overall 1918 1117/35 (97·0, 95·8–97·9) 372/394 (48·6, 45·0–52·2) 73·9 (71·6–76·1) 91·4 (88·2–93·9)

Data for sensitivity, specificity, positive predictive value, and negative predictive value are calculated against the reference standard. Samples with a negative smear result and
missing culture result (Löwenstein-Jensen medium and Bactec MGIT960 system) were excluded from the analysis, and samples with a positive smear result and missing
culture result (Löwenstein-Jensen medium and Bactec MGIT960 system) were included in the analysis.

Table 3: Diagnostic performance of Xpert MTB/RIF assay compared with the combined reference standard (smear microscopy, Löwenstein-Jensen
medium, and Bactec MGIT960 system)

medians at weeks 22 and 26 could not be calculated coefficient was calculated for each study timepoint. The
because more than 50% of patients had negative results magnitudes of the overall correlation coefficients (ˮ) over
in the Xpert MTB/RIF assay (figure 2).1 the entire 26 weeks were almost identical for smear
To investigate the relation between CTmin and the (–0·74; p<0·0001), Löwenstein-Jensen (–0·73; p<0·0001),
quantitative readouts for the smear, Löwenstein-Jensen, and MGIT (0·73; p<0·0001). Correlations were variable at
and MGIT assays, the Spearman rank correlation discrete timepoints. CTmin values correlated moderately

466 www.thelancet.com/respiratory Vol 1 August 2013

 Articles

with smear microscopy, Löwenstein-Jensen solid culture, We also investigated whether any change in CTmin over
and MGIT liquid culture in the first 8 weeks but showed time or absolute CTmin could be used to predict a patient’s
poorer correlation thereafter. conversion to a negative smear or culture result during
We analysed the Xpert MTB/RIF assay for prediction treatment and did not find it to be predictive (data not
of the binary smear and culture results at consecutive shown).
timepoints during treatment. Although the overall
sensitivity of the assay was 97%, specificity was low Discussion
compared with the reference standard, which combined The important findings of our study are that
smear microscopy, MGIT, and Löwenstein-Jensen quantitative Xpert MTB/RIF assay readouts for sputum
results (table 3). At baseline and early follow-up visits, M tuberculosis DNA as a biomarker correlate well with
the putative poor specificity of Xpert MTB/RIF assay was M tuberculosis smear and culture results during
misleading because it was caused by false-negative antituberculosis treatment. Positivity rates for sputum
microbiological results with the reference methods and M tuberculosis DNA detection with Xpert MTB/RIF
a very low number of patients. Of greater relevance was assay decline more slowly than do those with
the low specificity at weeks 6–12, indicating a delayed conventional sputum smear microscopy, solid culture,
Xpert MTB/RIF assay conversion compared with smear and liquid culture and seemed to be linear throughout
and culture methods (figure 1; table 3). The diagnostic treatment. Compared with a combined reference
performance of the Xpert MTB/RIF assay versus smear standard of smear and culture methods, the Xpert
or culture methods alone did not show substantially MTB/RIF assay had high sensitivity (97%) but poor
different values (table 4). Notably, compared with MGIT, specificity (49%; table 3). To our knowledge, this study
the Xpert MTB/RIF assay sensitivity and negative is the first evaluation of the capacity of the
predictive value for culture positivity were both 100% Xpert MTB/RIF assay as an early sputum biomarker of
from week 17 onwards (table 4; appendix). response for monitoring tuberculosis treatment, and See Online for appendix

Smear microscopy Löwenstein-Jensen medium Bactec MGIT 960 system

Sensitivity Specificity Positive Negative Sensitivity Specificity Positive Negative Sensitivity Specificity Positive Negative
predictive predictive predictive predictive predictive predictive
value value value value value value
Week 0 98·5% 5·9% 92·6% 25·0% 98·4% 33·3% 98·9% 25·0% 98·6% 25·0% 98·6% 25·0%
(95·7–99·7) (0·2–28·7) (88·2–95·7) (0·6–80·6) (95·4–99·7) (0·8–90·6) (96·2–99·9) (0·6–80·6) (95·9–99·7) (0·6–80·6) (95·9–99·7) (0·6–80·6)
Week 1 100·0% 25·0% 92·9% 100·0% 100·0% 57·1% 98·5% 100·0% 98·0% 20·0% 98·0% 20·0%
(98·1–100·0) (8·7–49·1) (88·6–96·0) (47·8–100·0) (98·1–100·0) (18·4–90·1) (95·6–99·7) (39·8–100·0) (95·0–99·5) (0·5–71·6) (95·0–99·5) (0·5–71·6)
Week 2 98·5% 5·3% 78·8% 50·0% 98·7% 33·3% 97·3% 50·0% 98·7% 25·0% 96·3% 50·0%
(92·1–100·0) (0·1–26·0) (68·6–86·9) (1·3–98·7) (92·7–100·0) (0·8–90·6) (90·7–99·7) (1·3–98·7) (93·2–100·0) (0·6–80·6) (89·6–99·2) (1·3–98·7)
Week 3 98·2% 6·3% 64·7% 66·7% 98·6% 20·0% 89·7% 66·7% 98·6% 28·6% 93·5% 66·7%
(90·5–100·0) (0·8–20·8) (53·6–74·8) (9·4–99·2) (92·4–100·0) (2·5–55·6) (80·8–95·5) (9·4–99·2) (92·6–100·0) (3·7–71·0) (85·5–97·9) (9·4–99·2)
Week 4 99·2% 18·1% 64·8% 93·8% 98·6% 30·3% 85·9% 83·3% 98·7% 30·0% 87·7% 81·8%
(95·7–100·0) (10·5–28·1) (57·6–71·5) (69·8–99·8) (95·0–99·8) (15·6–48·7) (79·6–90·8) (51·6–97·9) (95·3–99·8) (14·7–49·4) (81·8–92·2) (48·2–97·7)
Week 5 97·0% 5·4% 37·7% 75·0% 98·3% 12·5% 73·8% 75·0% 98·3% 17·7% 80·6% 75·0%
(84·2 –99·9) (1·1–14·9) (27·4–48·8) (19·4–99·4) (91·1–100·0) (2·7–32·4) (62·7–83·0) (19·4–99·4) (90·9–100·0) (3·8–43·4) (69·5–88·9) (19·4–99·4)
Week 6 100·0% 12·7% 29·5% 100·0% 95·7% 17·7% 61·6% 75·0% 98·2% 23·8% 77·5% 83·3%
(85·2–100·0) (5·7–23·5) (19·7–40·9) (63·1–100·0) (85·5–99·5) (6·8–34·5) (49·5–72·8) (34·9–96·8) (90·5–100·0) (8·2–47·2) (66·0–86·5) (35·9–99·6)
Week 7 95·8% 12·3% 28·8% 88·9% 96·0% 14·0% 35·8% 87·5% 90·9% 12·9% 59·7% 50·0%
(78·9 –99·9) (5·5–22·8) (19·2–40·0) (51·8–99·7) (79·7–99·9) (5·8–26·7) (24·5–48·5) (47·4–99·7) (78·3–97·5) (3·6–29·8) (47·0–71·5) (15·7–84·3)
Week 8 98·4% 21·4% 34·5% 96·9% 95·6% 16·8% 29·3% 91·3% 94·7% 22·6% 47·3% 85·2%
(91·2–100·0) (15·0–29·0) (27·5–42·1) (83·8–99·9) (84·9–99·5) (10·7–24·5) (22·1–37·3) (72·0–98·9) (86·9–98·5) (14·9–31·9) (39·1–55·6) (66·3–95·8)
Week 12 100·0% 39·6% 22·1% 100·0% 71·4% 35·9% 9·1% 93·3% 76·9% 34·5% 17·1% 89·5%
(88·1–100·0) (32·2–47·4) (15·4–30·2) (94·6–100·0) (41·9–91·6) (28·4–44·0) (4·5–16·1) (83·8–98·2) (56·4–91·0) (26·9–42·7) (10·8–25·2) (78·5–96·0)
Week 17 91·7% 48·9% 10·5% 98·9% ·· ·· ·· ·· 100·0% 45·1% 1·1% 100·0%
(61·5–99·8) (41·5–56·4) (5·4–18·0) (94·0–100·0) (2·5–100·0) (37·4–53·1) (0·0–6·0) (95·1–100·0)
Week 22 66·7% 63·7% 2·8% 99·2% 100·0% 68·1% 6·1% 100·0% 100·0% 64·8% 5·0% 100·0%
(9·4–99·2) (56·5–70·5) (0·3–9·7) (95·6–100·0) (29·2–100·0) (59·8–75·6) (1·3–16·9) (96·3–100·0) (29·2–100·0) (56·9–72·1) (1·0–13·9) (96·6–100·0)
Week 26 100·0% 75·3% 9·1% 100·0% 100·0% 77·1% 11·1% 100·0% 100·0% 76·8% 11·1% 100·0%
(15·8–100·0) (64·5–84·2) (1·1–29·2) (94·1–100·0) (15·8–100·0) (65·6–86·3) (1·4–34·7) (93·4–100·0) (15·8–100·0) (65·1–86·1) (1·4–34·7) (93·3–100·0)
Overall 98·7% 36·8% 53·7% 97·4% 98·0% 40·1% 63·5% 95·2% 97·3% 43·5% 68·9% 92·5%
(97·7–99·3) (34·0–39·7) (51·1–56·2) (95·4–98·7) (96·9–98·9) (37·6–44·4) (60·8–66·1) (92·4–97·2) (96·0–98·2) (39·9–47·1) (66·4–71·4) (89·3–95·0)

Data are percentage (95% CI). Data for sensitivity, specificity, positive predictive value, and negative predictive value were calculated against individual reference standards

Table 4: Diagnostic performance of Xpert MTB/RIF assay compared with smear microscopy, Löwenstein-Jensen medium, and Bactec MGIT960 system

www.thelancet.com/respiratory Vol 1 August 2013 467

 Articles

as a alternative for standard smear microscopy and and cannot replace conventional smear culture for the
culture (panel). monitoring of patients undergoing treatment for
The Xpert MTB/RIF assay is assumed to be specific; it tuberculosis. This is in agreement with the results of a
only detects DNA from intact M tuberculosis bacilli and recently reported study in which patients taking part in
contamination from free DNA is thought to be removed an early bactericidal activity trial were monitored by use
by a washing step.20,30 The assay, however, does not of the Xpert MTB/RIF assay, and liquid and solid
differentiate between viable, dormant,31 and non-viable cultures; the Xpert MTB/RIF assay was the least suitable
intact M tuberculosis bacilli that are shed during effective for the assessment of drug efficacy.8 Hence, several
antituberculosis treatment. Furthermore, our findings— aspects of the capacity of the assay for use in monitoring
high rates of positive results at the end of the 6 month tuberculosis treatment require further study: whether
treatment—suggest that even DNA fragments from the specificity of the Xpert MTB/RIF assay can be
lysed or damaged bacteria could have been detected by improved without affecting sensitivity through the
use of Xpert MTB/RIF. Detection of these bacteria would prevention of DNA amplification from dead bacteria;
explain the lengthy and constant slope, indicating a whether a positive or negative Xpert MTB/RIF assay
linear decay of DNA and dead mycobacteria, and is in result at the end of 6 months of tuberculosis treatment
contrast to the biphasic curves reported for culture or can be used to provide prognostic information about
RNA-based assays.12,18 By contrast with DNA, the treatment failure and relapse; and the prognostic
presence and amount of M tuberculosis RNA, particularly relevance of baseline quantitative values or the
mRNA, correlated well with culture positivity and viable timepoint of conversion to negativity, which was shown
colony counts, respectively,17,18,32 whereas DNA-based for time to positivity with MGIT.13
assays were also positive beyond culture conversion or This study has several advantages because it was nested
even the end of successful treatment.16,33 The results of a in a clinical trial. A large number of samples (n=2741)
recent study suggest that viable and non-viable were obtained and processed under standardised and
M tuberculosis bacilli can be distinguished by use of highly monitored conditions, as shown by the robust and
propidiummonoazide, which enters damaged or high-quality data generated at both study sites for the
non-viable mycobacteria, covalently binds to DNA after Xpert MTB/RIF assay, and smear and culture methods.
light-exposure, and prevents DNA amplification, such as The similarity of the data between different sites suggests
in the Xpert MTB/RIF assay.34 that our findings could not be attributed to technical
The data show that detection of M tuberculosis DNA in factors. The number of patients lost to follow-up was very
sputum with the Xpert MTB/RIF assay in its current low and the falling numbers of available results during
format cannot be used as a biomarker of disease activity follow-up are usually the result of patients being unable
to produce sputum samples at the end of an effective
Panel: Research in context treatment.
A limitation of this study is that the results of the
Systematic review Xpert MTB/RIF assay could not be linked with clinical
We did an online search of the PubMed database on Jan 31, 2011, to identify scientific outcomes such as cure and relapse. Since the REMoxTB
articles about the Xpert MTB/RIF assay published since Jan 1, 2010, using the terms “Xpert” trial is still in progress, information about the allocated
or “GeneXpert” and “tuberculosis”. Only articles published in English were selected. Six tuberculosis treatment or the presence of drug
articles were reviewed. The results of these studies showed the ability of this testing resistance in relation to the patients’ responses to
platform as a primary diagnostic method for detection of Mycobacterium tuberculosis mainly treatment was not available for data analysis in this
in expectorated sputum samples from patients with untreated tuberculosis.20–24 Although study. However, because the Xpert MTB/RIF assay is
several studies have been planned, there are no reports of the use of the Xpert MTB/RIF widely available in many countries, clinicians might be
assay to monitor tuberculosis treatment in the context of a clinical trial. starting to use this assay speculatively to monitor
Interpretation patients, as is commonly done with microscopy.
Previous research has shown that the Xpert MTB/RIF assay has a high specificity in the Therefore, the results of our study are important
diagnosis of M tuberculosis with improved sensitivity compared with smear microscopy because they show that the Xpert MTB/RIF assay is less
and culture. Thus, the available evidence shows that the Xpert MTB/RIF assay is useful as a responsive for the monitoring of tuberculosis treatment
primary diagnostic aid. In view of the need for improved biomarkers, researchers have than is smear microscopy.
speculated that the assay might be used to monitor treatment response and the samples Notably, however, the absence of an adequate reference
available to us provided a unique opportunity to test this hypothesis. As a result, the Xpert standard, and the consecutive use of smear microscopy
MTB/RIF assay was compared with the standard method to assess its usefulness for and culture as imperfect surrogates, might have distorted
treatment monitoring and its potential as a biomarker. We concluded that, in its current the results of the diagnostic performance of the
format, the Xpert MTB/RIF assay cannot be used to monitor antituberculosis treatment Xpert MTB/RIF assay. The results are further limited by
due to its slow decline in positivity compared with smear microscopy and culture. Future our study being undertaken in only two sites in Africa
technological change that would eliminate the signal from dead bacilli could overcome and in a population of patients who were not severely ill.
this problem and provide a useful biomarker for use in resource-poor settings. Our data indicate that the Xpert MTB/RIF assay in its
current format is not suitable for the monitoring of

468 www.thelancet.com/respiratory Vol 1 August 2013

 Articles

tuberculosis treatment and, thus, cannot replace 4 Wallis RS, Perkins MD, Phillips M, et al. Predicting the outcome of
standard smear microscopy and culture. Therefore, therapy for pulmonary tuberculosis. Am J Respir Crit Care Med
2000; 161 (4 pt 1): 1076–80.
technical improvements in the Xpert MTB/RIF assay 5 Yeager H LJ, Smith LR, LeMaistre CA. Quantitative studies of
that prevent the assay from detecting DNA from mycobacterial populations in sputum and saliva. Am Rev Respir Dis
non-viable or even damaged bacteria will be needed 1967; 95: 998–1004.
6 Wright PW, Wallace RJ Jr, Wright NW, Brown BA, Griffith DE.
before the test is suitable for monitoring patients in Sensitivity of fluorochrome microscopy for detection of
settings where it has already replaced smear microscopy Mycobacterium tuberculosis versus nontuberculous mycobacteria.
as a diagnostic test. Larger studies with longer follow-up J Clin Microbiol 1998; 36: 1046–49.
7 Diacon AH, Dawson R, Von Groote-Bidlingmaier F, et al.
are needed to clarify the prognostic relevance of the Randomized dose-ranging study of the 14-day early bactericidal
Xpert MTB/RIF assay results for the prediction of activity of bedaquiline (TMC207) in patients with sputum
treatment failure or relapse. microscopy smear-positive pulmonary tuberculosis.
Antimicrob Agents Chemother 2013; 57: 2199–203.
Contributors 8 Kayigire XA, Friedrich SO, Venter A, et al. Direct comparison of
All authors are part of PanACEA and the REMoxTB trial and contributed Xpert MTB/RIF assay with liquid and solid mycobacterial culture
to the Article. TDM, SHG, AHD, and MH designed the study and for quantification of early bactericidal activity. J Clin Microbiol 2013;
protocols. Data were gathered by SOF, AR, AV, CDM, RD, CCB, and NH. 51: 1894–98.
SOF, AR, ES, PPJP, RDH, AHD, and MH undertook the management 9 Gler MT, Skripconoka V, Sanchez-Garavito E, et al. Delamanid for
and the analyses of the data. SOF, AR, MH, AHD, SHG, TDM, MJB, multidrug-resistant pulmonary tuberculosis. N Engl J Med 2012;
and AZ wrote the first and final drafts. KS and AB contributed to the 366: 2151–60.
work through management and analysis of the data. All authors 10 Mitchison DA. Assessment of new sterilizing drugs for treating
participated in writing the Article. pulmonary tuberculosis by culture at 2 months. Am Rev Respir Dis
1993; 147: 1062–63.
PanACEA
11 Pheiffer C, Carroll NM, Beyers N, et al. Time to detection of
Medical Centre of the University of Munich, Munich, Germany Mycobacterium tuberculosis in BACTEC systems as a viable alternative
(Sonja Henne, Anna Maria Mekota, Norbert Heinrich, Andrea Rachow, to colony counting. Int J Tuberc Lung Dis 2008; 12: 792–98.
Anke Kohlenberg, Elmar Saathoff, Michael Hoelscher); University of 12 Diacon AH, Maritz JS, Venter A, van Helden PD, Dawson R,
St Andrews, St Andrews, UK (Stephen Gillespie); Radboud University Donald PR. Time to liquid culture positivity can substitute for
Nijmegen Medical Centre, Nijmegen, Netherlands colony counting on agar plates in early bactericidal activity studies
(Georgette Plemper van Balen, Marloes Weijers, Rob Aarnoutse, of antituberculosis agents. Clin Microbiol Infect 2012; 18: 711–17.
Martin J Boeree); University College of London, London, UK 13 Hesseling AC, Walzl G, Enarson DA, et al. Baseline sputum time to
(Anna Bateson, Timothy McHugh, Kasha Singh, Robert Hunt, detection predicts month two culture conversion and relapse in
Alimuddin Zumla); Medical Research Council Clinical Trials Unit, non-HIV-infected patients. Int J Tuberc Lung Dis 2010; 14: 560–70.
London, UK (Andrew Nunn, Patrick Phillips); University of Cape Town, 14 Wallis RS, Pai M, Menzies D, et al. Biomarkers and diagnostics for
Cape Town, South Africa (Rodney Dawson, Kim Narunsky); University tuberculosis: progress, needs, and translation into practice. Lancet
of Stellenbosch, Cape Town, South Africa (Andreas Diacon, 2010; 375: 1920–37.
Jeannine du Bois, Amour Venter, Sven Friedrich); University of the 15 Walzl G, Ronacher K, Hanekom W, Scriba TJ, Zumla A.
Witswatersrand, Johannesburg, South Africa (Ian Sanne, Karla Mellet, Immunological biomarkers of tuberculosis. Nat Rev Immunol 2011;
Eefje de Jong); Aurum Institute, Johannesburg, South Africa 11: 343–54.
(Gavin Churchyard, Salome Charalambous); University of Zambia, 16 Kennedy N, Gillespie SH, Saruni AO, et al. Polymerase chain
Lusaka, Zambia (Peter Mwaba); National Institute of Medical reaction for assessing treatment response in patients with
Research-Mbeya Medical Research Centre, Mbeya, Tanzania pulmonary tuberculosis. J Infect Dis 1994; 170: 713–16.
(Nyanda Elias, Chacha Mangu, Gabriel Rojas-Ponce, Bariki Mtafya, 17 Desjardin LE, Perkins MD, Wolski K, et al. Measurement of sputum
Leonard Maboko); Ifakara Health Institute-Bagamoyo Research and Mycobacterium tuberculosis messenger RNA as a surrogate for
response to chemotherapy. Am J Respir Crit Care Med 1999;
Training Centre, Bagamoyo, Tanzania (Klaus Reither, Levan Jugheli);
160: 203–10.
Kilimanjaro Clinical Research Institute, Moshi, Tanzania (Noel Sam,
18 Honeyborne I, McHugh TD, Phillips PP, et al. Molecular bacterial
Gibson Kibiki, Hadija Semvua, Stellah Mpagama); Medical Research
load assay, a culture-free biomarker for rapid and accurate
Unit–Albert Schweitzer Hospital, Lambarene, Gabon (Abraham Alabi,
quantification of sputum Mycobacterium tuberculosis bacillary load
Ayola Akim Adegnika); Kenya Medical Research Institute, Nairobi, during treatment. J Clin Microbiol 2011; 49: 3905–11.
Kenya (Evans Amukoye); Makerere University, Kampala, Uganda
19 van Zyl-Smit RN, Binder A, Meldau R, et al. Comparison of
(Alphonse Okwera). quantitative techniques including Xpert MTB/RIF to evaluate
Conflicts of interest mycobacterial burden. PLoS One 2011; 6: e28815.
We declare that we have no conflicts of interest. 20 Helb D, Jones M, Story E, et al. Rapid detection of Mycobacterium
tuberculosis and rifampin resistance by use of on-demand,
Acknowledgments near-patient technology. J Clin Microbiol 2010; 48: 229–37.
PanACEA is funded by the European and Developing Countries Clinical 21 Theron G, Peter J, van Zyl-Smit R, et al. Evaluation of the Xpert MTB/
Trials Partnership (grants numbers CT.2004.32011.001, IP.2007.32011.011, RIF assay for the diagnosis of pulmonary tuberculosis in a high HIV
IP.2007.32011.012, IP.2007.32011.013), Global Alliance for TB Drug prevalence setting. Am J Respir Crit Care Med 2011; 184: 132–40.
Development, Bill & Melinda Gates Foundation, UK Medical Research 22 Blakemore R, Nabeta P, Davidow AL, et al. A multi-site assessment
Council, and German Ministry of Science and Technology (grant of the quantitative capabilities of the Xpert(R) MTB/RIF assay.
number 01KA0901). We thank all patients who took part in this study. Am J Respir Crit Care Med 2011; 184: 1076–84.
23 Boehme CC, Nabeta P, Hillemann D, et al. Rapid molecular
References
detection of tuberculosis and rifampin resistance. N Engl J Med
1 Raviglione M, Marais B, Floyd K, et al. Scaling up interventions to
2010; 363: 1005–15.
achieve global tuberculosis control: progress and new
developments. Lancet 2012; 379: 1902–13. 24 Rachow A, Zumla A, Heinrich N, et al. Rapid and accurate detection
of Mycobacterium tuberculosis in sputum samples by Cepheid Xpert
2 Zumla A, Hafner R, Lienhardt C, Hoelscher M, Nunn A. Advancing
MTB/RIF assay—a clinical validation study. PLoS One 2011; 6: e20458.
the development of tuberculosis therapy. Nat Rev Drug Discov 2012;
11: 171–72. 25 Weyer K, Mirzayev F, Migliori G, et al. Rapid molecular TB
diagnosis: evidence, policy-making and global implementation of
3 Phillips PP, Gillespie SH, Boeree M, et al. Innovative trial designs
Xpert(R)MTB/RIF. Eur Respir J 2012; published online Nov 22.
are practical solutions for improving the treatment of tuberculosis.
DOI:10.1183/09031936.00157212.
J Infect Dis 2012; 205 (suppl 2): S250–57.

www.thelancet.com/respiratory Vol 1 August 2013 469

 Articles

26 Pantoja A, Fitzpatrick C, Vassall A, Weyer K, Floyd K. Xpert 31 Garton NJ, Waddell SJ, Sherratt AL, et al. Cytological and transcript
MTB/RIF for diagnosis of TB and drug-resistant TB: a cost and analyses reveal fat and lazy persister-like bacilli in tuberculous
affordability analysis. Eur Respir J 2012; published online Dec 20. sputum. PLoS Med 2008; 5: e75.
DOI:10.1183/09031936.00147912. 32 Li L, Mahan CS, Palaci M, et al. Sputum Mycobacterium tuberculosis
27 Kent PT, Kubica GP. Public health mycobacteriology: a guide for the mRNA as a marker of bacteriologic clearance in response to
level III laboratory. Atlanta, GA: US Department of Public Health antituberculosis therapy. J Clin Microbiol 2010; 48: 46–51.
and Human Services, 1985. 33 Hellyer TJ, Fletcher TW, Bates JH, et al. Strand displacement
28 American Thoracic Society. Diagnostic standards and classification amplification and the polymerase chain reaction for monitoring
of tuberculosis in adults and children. Am J Respir Crit Care Med response to treatment in patients with pulmonary tuberculosis.
2000; 4: 1376–95. J Infect Dis 1996; 173: 934–41.
29 Rustomjee R, Lienhardt C, Kanyok T, et al. A phase II study of the 34 Miotto P, Bigoni S, Migliori GB, Matteelli A, Cirillo DM. Early
sterilising activities of ofloxacin, gatifloxacin and moxifloxacin in tuberculosis treatment monitoring by Xpert(R) MTB/RIF.
pulmonary tuberculosis. Int J Tuberc Lung Dis 2008; 12: 128–38. Eur Respir J 2012; 39: 1269–71.
30 Blakemore R, Story E, Helb D, et al. Evaluation of the analytical
performance of the Xpert MTB/RIF assay. J Clin Microbiol 2010;
48: 2495–501.

470 www.thelancet.com/respiratory Vol 1 August 2013