Articles

Diagnostic accuracy of a three-gene Mycobacterium

tuberculosis host response cartridge using fingerstick blood
for childhood tuberculosis: a multicentre prospective study
in low-income and middle-income countries
Laura Olbrich*, Valsan P Verghese*, Zoe Franckling-Smith, Issa Sabi, Nyanda E Ntinginya, Alfred Mfinanga, Denise Banze, Sofia Viegas,
Celso Khosa, Robina Semphere, Marriott Nliwasa, Timothy D McHugh, Leyla Larsson, Alia Razid, Rinn Song, Elizabeth L Corbett, Pamela Nabeta,
Andre Trollip, Stephen M Graham, Michael Hoelscher, Christof Geldmacher, Heather J Zar, Joy Sarojini Michael*, Norbert Heinrich*, on behalf of
the RaPaed-TB consortium†

Summary
Lancet Infect Dis 2024; Background Childhood tuberculosis remains a major cause of morbidity and mortality in part due to missed diagnosis.
24: 140–49 Diagnostic methods with enhanced sensitivity using easy-to-obtain specimens are needed. We aimed to assess the
Published Online diagnostic accuracy of the Cepheid Mycobacterium tuberculosis Host Response prototype cartridge (MTB-HR), a
October 30, 2023
candidate test measuring a three-gene transcriptomic signature from fingerstick blood, in children with presumptive
https://doi.org/10.1016/
S1473-3099(23)00491-7 tuberculosis disease.
See Comment page 110
*Contributed equally
Methods RaPaed-TB was a prospective diagnostic accuracy study conducted at four sites in African countries (Malawi,
†Consortium members are listed
Mozambique, South Africa, and Tanzania) and one site in India. Children younger than 15 years with presumptive
at the end of the Article pulmonary or extrapulmonary tuberculosis were enrolled between Jan 21, 2019, and June 30, 2021. MTB-HR was
Division of Infectious Diseases performed at baseline and at 1 month in all children and was repeated at 3 months and 6 months in children on
and Tropical Medicine tuberculosis treatment. Accuracy was compared with tuberculosis status based on standardised microbiological,
(L Olbrich MD DPhil, radiological, and clinical data.
L Larsson MSc, A Razid BSc,
Prof M Hoelscher MD,
C Geldmacher PhD, Findings 5313 potentially eligible children were screened, of whom 975 were eligible. 784 children had MTB-HR test
N Heinrich MD PhD) and results, of whom 639 had a diagnostic classification and were included in the analysis. MTB-HR differentiated
CIHLMU Center for children with culture-confirmed tuberculosis from those with unlikely tuberculosis with a sensitivity of 59·8%
International Health
(Prof M Hoelscher, N Heinrich),
(95% CI 50·8–68·4). Using any microbiological confirmation (culture, Xpert MTB/RIF Ultra, or both), sensitivity
LMU University Hospital, LMU was 41·6% (34·7–48·7), and using a composite clinical reference standard, sensitivity was 29·6% (25·4–34·2).
Munich, Munich, Germany; Specificity for all three reference standards was 90·3% (95% CI 85·5–94·0). Performance was similar in different age
German Centre for Infection groups and by malnutrition status. Among children living with HIV, accuracy against the strict reference standard
Research (DZIF), Partner Site
Munich, Munich, Germany
tended to be lower (sensitivity 50·0%, 15·7–84·3) compared with those without HIV (61·0%, 51·6–69·9), although
(L Olbrich, L Larsson, A Razid, the difference did not reach statistical significance. Combining baseline MTB-HR result with one Ultra result
Prof M Hoelscher, C Geldmacher, identified 71·2% of children with microbiologically confirmed tuberculosis.
N Heinrich); Fraunhofer
Institute ITMP, Immunology,
Infection and Pandemic
Interpretation MTB-HR showed promising diagnostic accuracy for culture-confirmed tuberculosis in this large,
Research, Munich, Germany geographically diverse, paediatric cohort and hard-to-diagnose subgroups.
(L Olbrich, Prof M Hoelscher,
C Geldmacher, N Heinrich); Funding European and Developing Countries Clinical Trials Partnership, UK Medical Research Council, Swedish
Oxford Vaccine Group,
Department of Paediatrics and
International Development Cooperation Agency, Bundesministerium für Bildung und Forschung; German Center
the NIHR Oxford Biomedical for Infection Research (DZIF).
Research Centre, University of
Oxford, Oxford, UK (L Olbrich,
Copyright © 2023 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY
R Song MD PhD); Pediatric
Infectious Diseases, 4.0 license.
Department of Pediatrics,
Christian Medical College, Introduction countries where challenges and delays, particularly in
Vellore, India
Approximately 12% of people with tuberculosis tuberculosis case detection, contribute to poor outcomes.
(Prof V P Verghese MD);
Department of Paediatrics and worldwide are children younger than 15 years, with Tuberculosis diagnosis in children remains difficult.
Child Health, SA-MRC Unit on 1·2 million cases of tuberculosis disease among children Signs and symptoms are non-specific and confirmation
Child and Adolescent Health, each year.1 An estimated 240 000 children are projected to of disease by a microbiological reference standard is
University of Cape Town,
die annually, accounting for 16% of all tuberculosis- infrequent.3 The reported sensitivity of direct pathogen
Cape Town, South Africa
(Z Franckling-Smith MD, related deaths, with 80% of these occurring in those detection ranges from 2% to 40%, even in high-resource
Prof H J Zar MD PhD); Mbeya younger than 2 years.1,2 The burden of disease and settings.4,5 WHO has approved both tuberculosis culture
Medical Research Centre, mortality is greatest in low-income and middle-income and molecular rapid diagnostic tests, such as Xpert

140 www.thelancet.com/infection Vol 24 February 2024

 Articles

National Institute for Medical

Research in context Research, Mbeya, Tanzania
(I Sabi MD PhD,
Evidence before this study search terms “Child”, “Children”, “Paediatric”, or “Pediatric” N E Ntinginya MD PhD,
Novel diagnostic tests using non-sputum specimens for rapid yielded no results. A Mfinanga MD); Instituto
and accurate tuberculosis diagnosis are urgently needed, Nacional de Saúde,
Added value of this study Marracuene, Mozambique
especially for children. Host blood transcriptomic signatures
To the best of our knowledge, this is the first study to assess the (D Banze MD, S Viegas PhD,
can distinguish tuberculosis from other diseases and latent C Khosa MD PhD); Helse Nord
performance of the three-gene signature using the MTB-HR in a
tuberculosis infection by using blood as an easy-to-collect, Tuberculosis Initiative,
prospectively recruited cohort of children with presumptive Department of Pathology,
non-respiratory specimen type. Based on a multicohort
pulmonary or extrapulmonary tuberculosis. Applying the Kamuzu University of Health
analysis from 14 published datasets with more than
previously published cutoff (TB-score of 1·5), MTB-HR was Sciences, Blantyre, Malawi
2500 samples originating from ten countries, Sweeney and (R Semphere MD,
reliably able to differentiate children with culture-confirmed
colleagues derived a three-gene mRNA-signature, which was M Nliwasa MD PhD,
tuberculosis from those unlikely to have tuberculosis, both in Prof E L Corbett MD); Centre for
translated into the Mycobacterium tuberculosis Host Response
the overall cohort and in key subgroups of interest, including Clinical Microbiology,
prototype cartridge (MTB-HR) to be performed on the
infants younger than 1 year, children living with HIV, children University College London,
GeneXpert platform. The first prospective evaluation of London, UK
with severe acute malnutrition, or children with
MTB-HR in an adult cohort (75 patients with tuberculosis and (Prof T D McHugh PhD); Clinical
extrapulmonary tuberculosis manifestations. Accuracy was Research Department, London
120 patients with other respiratory diseases) indicated a
lower when using a composite microbiological reference School of Hygiene & Tropical
promising diagnostic accuracy, with an area under the receiver
standard (ie, culture positive, Ultra positive, or both) or a Medicine, London, UK
operating characteristic curve of 0·88 (95% CI 0·83–0·94), (Prof E L Corbett); Foundation
composite clinical reference standard, and there was a positive
80% sensitivity (95% CI 76–85), and 94% specificity for Innovative New Diagnostics
association between strength of microbiological detection and (FIND), Geneva, Switzerland
(95% CI 91–96) compared with a composite microbiological
MTB-HR accuracy. Our results suggest a potential diagnostic (P Nabeta MD, A Trollip PhD);
reference standard. Department of Paediatrics,
value by combining one Ultra on a respiratory sample with one
We searched PubMed for articles published in any language University of Melbourne and
MTB-HR. Additionally, the MTB-HR has potential value in
Murdoch Children’s Research
between database inception in 1996 and Feb 27, 2023, using monitoring treatment response, addressing an important gap Institute, Melbourne, VIC,
the search terms “TB”, “Transcriptomics”, “Transcriptome”, among available paediatric tuberculosis diagnostic tests. Lastly, Australia
“Host blood transcriptional signatures”, “Host response we explored the use of optimised cutoffs, suggesting a benefit (Prof S M Graham MD PhD); Unit
cartridge”, and “Accuracy”, in various combinations. From this Global Health, Helmholtz
from identifying a more suitable child-specific MTB-HR cutoff. Zentrum München, German
search, as well as cross-referencing within the articles, we were Research Center for
able to include studies that assessed the performance of a Implications of all the available evidence
Environmental Health (HMGU),
three-gene (GBP5, DUSP3, and KLF2) signature in diagnosing Given the widespread availability of the GeneXpert platform, Neuherberg, Germany
tuberculosis disease compared with uninfected controls, the short time to result, ease of using capillary blood, and the (Prof M Hoelscher); Department

latent tuberculosis, and other respiratory diseases, using a reported diagnostic yield, MTB-HR is a promising tool to of Clinical Microbiology,
Christian Medical College,
microbiological reference standard of a positive M tuberculosis substantially improve tuberculosis diagnosis for children,
Vellore, India
culture, positive Xpert MTB/RIF Ultra assay, or both. These which is currently the main obstacle to reducing child (Prof J S Michael MD)
studies were largely conducted in adults and only a few were tuberculosis mortality. Correspondence to:
prospective. A repeat search on the same date with the added Dr Norbert Heinrich, Division of
Infectious Diseases and Tropical
Medicine, University Hospital,
Ludwig Maximilian University of
MTB/RIF Ultra (Cepheid, Sunnyvale, CA, USA; hereafter optimal sensitivity of at least 66%.9 However, currently
Munich, Munich 80802,
referred to as Ultra), but these are suboptimally sensitive available WHO-endorsed tests using child-friendly Germany
because paucibacillary and extrapulmonary presentations samples do not meet this target, with sensitivities of [email protected]
are common.3,5–7 Collection methods of recommended 56% reported for Ultra on stool and 44% on muenchen.de
respiratory specimens are invasive and might be nasopharyngeal aspirate compared with culture, and even
demanding in terms of infrastructure and skilled staff.6,8 candidate tests such as the urine FujiLAM (Fujifilm,
This deficiency of diagnostic reference standards renders Tokyo, Japan) showing sensitivities of 42–63% compared
paediatric diagnostic validation studies particularly with composite microbiological reference standards.10,11
challenging. Host gene transcriptional signatures can distinguish
To address the need for better diagnostics, non-sputum- between tuberculosis disease and other illnesses and are
based assays with enhanced sensitivity are needed, measured in blood, a sample easier to obtain than sputum
especially for key subgroups, including infants younger in children.12–14 The Cepheid Mycobacterium tuberculosis
than 1 year, children with severe acute malnutrition, Host Response prototype cartridge (MTB-HR) is a
children living with HIV, and children in resource-limited GeneXpert-based RT-PCR test assessing relative
settings. In these subgroups, conventional diagnosis is messenger RNA levels from fingerstick whole-blood
particularly challenging, as highlighted by WHO’s high- samples.14,15 It distinguishes tuberculosis from other
priority target product profiles for new tuberculosis diseases based on the expression of GBP5 and DUSP3
diagnostics, which do not specify a minimum sensitivity (upregulated in tuberculosis), and KLF2 (downregulated
for a paediatric tuberculosis diagnostic test, but an in tuberculosis). The platform auto­matically calculates an

www.thelancet.com/infection Vol 24 February 2024 141

 Articles

MTB-HR TB-score based on cycle threshold (Ct) values. A sites, or gastric lavage at the Indian site) obtained from
previously published cutoff (1·5) was derived by applying each participant, solid and liquid media cultures
Youden’s index to determine optimal sensitivity and (Mycobacteria Growth Indicator Tube; Becton Dickinson,
specificity to meet the WHO threshold for a triage test Sparks, NJ, USA) were inoculated and one Ultra was
(90% sensitivity and 70% specificity).15 The first prospective performed. Children younger than 5 years underwent
evaluation in an adult cohort suggested an area under the nasopharyngeal aspirate for a second Ultra. Children
receiver operating characteristic (ROC) curve (AUC) were followed up at 1 month and 3 months, and at
of 0·88 (95% CI 0·83–0·94) against a reference standard 6 months if on tuberculosis treatment or still unwell at
of Ultra and culture, with 80% sensitivity (95% CI 76–85) 3 months.
and 94% specificity (95% CI 91–96), and an AUC of 0·94 MTB-HR was done at baseline and repeated at 1 month,
(95% CI 0·91–0·97) against Ultra alone.15 In this study, we and additionally at 3 months and 6 months if on
present the first diagnostic accuracy assessment of tuberculosis treatment. Fingerstick blood samples of
MTB-HR in a large, prospective cohort of children with 100 μL were taken, instilled into the MTB-HR cartridge
presumptive tuberculosis disease recruited in five low- within 15 min, and loaded onto the GeneXpert platform
income and middle-income countries. within 1 h of blood draw. Capillary samples taken before
the MTB-HR cartridge became available on site were
Methods transferred into PAXgene buffer (Qiagen, Hilden,
Study design and population Germany) within 15 min of collection and stored at –80°C.
RaPaed-TB was a prospective, multicentre diagnostic Ct values for individual genes (DUSP, GBP5, and KLF2)
accuracy study aiming to evaluate novel diagnostic tests were obtained. In case of an invalid result, the test was
and testing approaches for tuberculosis in children with repeated on the same day or the next day, if possible.
presumptive tuberculosis disease. The study protocol and Laboratory staff were masked to the clinical findings and
methods are published elsewhere.16 The study was clinicians were masked to MTB-HR results. An MTB-HR
performed in accordance with the Declaration of TB-score was provided by the GeneXpert instrument
Helsinki.17 The protocol and informed consent documents ([Ct GBP5 + Ct DUSP3]/2 – Ct KLF2). A tentative previously
were approved by institutional review boards at all sites published cutoff was used, with less than 1·5 representing
and the coordinating site (Division of Infectious Diseases a positive result for tuberculosis disease.15 To account for
and Tropical Medicine, Ludwig Maximilian University of variations due to storage, a correction factor based on the
Munich, Munich, Germany). The study is registered on strict reference standard was derived by determining the
ClinicalTrials.gov, NCT03734172. difference of mean MTB-HR TB-score between samples
Briefly, children younger than 15 years were recruited stored in PAXgene buffer (hereafter referred to as
in participating health facilities between Jan 21, 2019, biobanked samples) and prospectively tested samples, and
and June 30, 2021. These centres included three tertiary then applied to biobanked samples.
hospitals in South Africa, Malawi, and India, and two Diagnostic classifications were adapted from the US
urban health facilities in Tanzania and Mozambique. National Institutes of Health consensus statement for
Children were recruited in both inpatient (South Africa, paediatric tuberculosis diagnostic evaluation studies.2,16
Malawi, India, and Mozambique) and outpatient Clinical case definitions were used to derive reference
(Malawi, Tanzania, and Mozambique) settings. The standards for diagnostic accuracy estimation, which are
See Online for appendix inclusion and exclusion criteria are listed in the appendix described in detail in the appendix (p 3). Briefly, we
(p 2); children were eligible for recruitment if they had at defined a strict reference standard (SRS) including only
least one of: microbiological confirmation of tuberculosis children with culture or cultures positive for M tuberculosis
disease, signs or symptoms of pulmonary tuberculosis as confirmed tuberculosis, compared with children
disease, or signs or symptoms of extrapulmonary with unlikely tuberculosis. Additionally, we evaluated
tuberculosis disease. Children were excluded if they diagnostic accuracy using a composite microbiological
weighed less than 2 kg, had received three or more doses reference standard (MRS), including children with a
of tuberculosis medication, or if study procedures were positive Ultra, culture, or both. Lastly, a composite
considered an undue risk to the child.16 Parents or clinical reference standard (CRS) was defined, whereby
guardians provided written informed consent, and children classified as confirmed tuberculosis or
participants provided written assent when they were unconfirmed tuberculosis were regarded as positive.
older than the locally specified age for doing so.
Statistical analysis
Procedures Diagnostic accuracy was tested using the SRS, MRS, and
After extensive training of all site staff, a standardised CRS. Differences in MTB-HR readout (gene transcripts
clinical, radiological, and laboratory workup was and MTB-HR TB-score) between diagnostic categories
performed; monitoring and retraining, if necessary, was were assessed by Mann-Whitney U tests. AUC was used
performed every 6 months.16 For the two respiratory to determine the accuracy of the MTB-HR TB-score and
specimens (spontaneous or induced sputum at African 95% CIs were calculated. Between-group AUCs were

142 www.thelancet.com/infection Vol 24 February 2024

 Articles

5313 potentially eligible children screened

4338 not recruited

4091 did not meet inclusion criteria
124 did not provide consent
123 out of catchment area

975 eligible children

191 no MTB-HR result (test not

performed)

784 children with MTB-HR result

609 MTB-HR negative 168 MTB-HR positive 7 MTB-HR inconclusive

6 error
1 invalid result

118 no diagnostic classification 20 no diagnostic classification

6 no valid microbiological result 2 no valid microbiological result
68 unclear clinical trajectory 10 unclear clinical trajectory
44 empirical tuberculosis treatment started 8 empirical tuberculosis treatment started

491 children with diagnostic classification 148 children with diagnostic classification
118 confirmed tuberculosis 84 confirmed tuberculosis
186 unconfirmed tuberculosis 44 unconfirmed tuberculosis
187 unlikely tuberculosis 20 unlikely tuberculosis

Figure 1: Flow diagram of participants in RaPaed-TB

Definitions of clinical case classification are presented in the appendix (p 3). All children with defined clinical case definition were included in the analysis. Of those
excluded, most had an unclear clinical trajectory (ie, were lost to follow-up). MTB-HR=Cepheid Mycobacterium tuberculosis Host Response prototype cartridge.

compared using the DeLong method. Sensitivity and interpretation, or writing of this report. Cepheid, which
specificity were calculated using the previously proposed supplied testing kits at no cost, was given the opportunity
cutoff of 1·5 and shown together with AUC for predefined to comment on the manuscript before submission.
subgroups of interest.15 To explore different cutoffs, we
calculated the maximum Youden’s index of the AUC and Results
generated cutoffs with fixed sensitivity at 65% and 5313 potentially eligible children were screened, of whom
at 90%.18 When plotting MTB-HR TB-scores over time, a 975 children were eligible and provided consent. Of
one-way repeated measures ANOVA with pair-wise these, 191 did not have an MTB-HR test result available at
paired t tests was used to evaluate the difference between enrolment, most frequently due to a lost sample
timepoints (appendix p 4). To account for heterogeneity shipment from the Malawian site (n=116). Of
between sites, a random-effects meta-analysis was 784 MTB-HR tests, 609 (78%) were negative, 168 (21%)
conducted. Finally, the incremental yield of combining were positive, and seven (1%) were inconclusive.
MTB-HR and one Ultra on respiratory samples was 138 children with an MTB-HR result had no diagnostic
computed, including all microbiologically confirmed classification. 202 (32%) of 639 children included in the
(positive culture, Ultra, or both) as the overall population. analyses had confirmed tuberculosis, 230 (36%) had
Data were cleaned and analysed using Stata v16.1 unconfirmed tuberculosis, and 207 (32%) were classified
and R v4.1.3. ROC curves and DeLong tests were as unlikely tuberculosis (figure 1).
computed using the pROC package and the roc.test Among the study population, median age was 5·4 years
function; random-effect meta-analysis used the mada (IQR 1·8–9·2) across categories and was higher in
package and the reitsma function. children with confirmed tuberculosis (6·5 years, 1·8–11·8)
than in children with unconfirmed tuberculosis
Role of the funding source (4·4 years, 1·7–8·2) or unlikely tuberculosis (5·5 years,
The funders of the study (EDCTP2 and DZIF) had no 2·3–7·8). Of all children, 89 (14%) were living with HIV,
role in study design, data collection, data analysis, data 71 (11%) had severe acute malnutrition, and 13 (20%) were

www.thelancet.com/infection Vol 24 February 2024 143

 Articles

Confirmed tuberculosis Unconfirmed Unlikely tuberculosis All (n=639)

(n=202) tuberculosis (n=230) (n=207)
Site
South Africa 69 (34%) 78 (34%) 29 (14%) 176 (28%)
Tanzania 43 (21%) 57 (25%) 78 (38%) 178 (28%)
Mozambique 20 (10%) 72 (31%) 59 (29%) 151 (24%)
Malawi 11 (5%) 11 (5%) 28 (14%) 50 (8%)
India 59 (29%) 12 (5%) 13 (6%) 84 (13%)
Sex
Female 96 (48%) 101 (44%) 109 (53%) 306 (48%)
Male 106 (52%) 129 (56%) 98 (47%) 333 (52%)
Age, years 6·5 (1·8–11·8) 4·4 (1·7–8·2) 5·5 (2·3–7·8) 5·4 (1·8–9·2)
<1 32 (16%) 32 (14%) 18 (9%) 82 (13%)
1–4 23 (11%) 38 (17%) 29 (14%) 90 (14%)
5–9 32 (16%) 51 (22%) 42 (20%) 125 (20%)
10–14 43 (21%) 73 (32%) 86 (42%) 202 (32%)
Comorbidities
Severe acute malnutrition 31/195 (16%) 23/228 (10%) 17/205 (8%) 71/628 (11%)
HIV 16/202 (8%) 55/230 (24%) 18/207 (9%) 89/639 (14%)
Tuberculosis laboratory findings
Culture positive only 31 (15%) ·· ·· 31 (15%)
Ultra positive only 75 (37%) ·· ·· 75 (37%)
Culture and Ultra positive 96 (48%) ·· ·· 96 (48%)
Tuberculosis disease
Pulmonary tuberculosis only 111/196 (57%) 174/222 (78%) ·· 285/418 (68%)
Extrapulmonary tuberculosis only 39/196 (20%) 20/222 (9%) ·· 59/418 (14%)
Pulmonary and extrapulmonary tuberculosis 46/196 (23%) 28/222 (13%) ·· 74/418 (18%)
Tuberculosis-related clinical findings
Tuberculin skin test positive 113/182 (63%) 105/199 (53%) 81/192 (42%) 299/573 (52%)
Non-severe tuberculosis disease* 41/196 (21%) 85/222 (38%) ·· 126/418 (30%)
Hospitalised at enrolment 105/202 (52%) 76/230 (33%) 44/207 (21%) 225/639 (35%)
Chest x-ray findings consistent with tuberculosis† 94/202 (47%) 73/230 (32%) 29/207 (14%) 196/639 (31%)
Data are n (%), median (IQR), or n/N (%). Stratification of clinical and demographic characteristics was done by clinical case definition. Ultra=Xpert MTB/RIF Ultra diagnostic
test. *According to WHO definition,18 outlined in the appendix (p 3). †According to masked expert reading.

Table 1: Demographic and clinical characteristics

living with HIV and had severe acute malnutrition figure 2A) with a sensitivity of 59·8% (95% CI 50·8–68·4).
(table 1). Using MRS (n=409), the AUC was 0·71 (0·66–0·76) and
MTB-HR performance was calculated using three sensitivity was 41·6% (34·7–48·7). Using CRS (n=639),
distinct reference standards: SRS, MRS, and CRS the AUC was 0·65 (0·61–0·69) and sensitivity was
(appendix p 3). Median TB-scores differed between 29·6% (25·4–34·2; figure 2B). Using all three reference
MTB-HR conducted on prospective (fresh) versus standards, specificity was 90·3% (95% CI 85·5–94·0)
biobanked (stored) samples, with prospective samples Diagnostic accuracy of MTB-HR TB-score in key
showing overall higher median scores and Ct values subgroups was assessed applying the SRS (figure 3,
for DUSP3, GBP5, and KLF2 (figure 2). Recentred scores table 2; appendix pp 8–9), the MRS (appendix pp 8–10),
for biobanked samples were used for the rest of the and the CRS (appendix pp 8, 10). Performance differed
analysis by applying the derived correction factor (0·533). slightly between subgroups without reaching statistical
AUCs and median Ct values for all three genes stratified significance for any of the reference standards, including
by prospective versus biobanked samples and by reference between age groups (DeLong p=0·055), HIV status
standards are presented in the appendix (pp 5–6). (DeLong p=0·31), and malnutrition status (DeLong
Across the cohort, using SRS (n=334), the p=0·052). Stratification by country suggested similar
MTB-HR TB-score discriminated between children with results for most sites except Malawi, which contributed
culture-confirmed tuberculosis (median TB-score 1·1, only three culture-confirmed patients. The sensitivity
IQR 0·3–1·9) and children with unlikely tuberculosis analysis using random-effects meta-analysis for site
(2·8, 2·1–3·2) at an AUC of 0·85 (95% CI 0·80–0·89; heterogeneity resulted in a pooled sensitivity and

144 www.thelancet.com/infection Vol 24 February 2024

 Articles

A B
6 Biobanked 1·0 TB-score
p<0·0001
Prospective KLF2
GBP5
0·8 DUSP3
4

0·6

Sensitivity
TB-score

2
0·4

0 AUC 0·846
0·2 AUC 0·616
AUC 0·834
AUC 0·758
–2 0
Confirmed TB Unlikely TB 1·0 0·8 0·6 0·4 0·2 0
SRS

6 1·0
p<0·0001

0·8
4

0·6
Sensitivity
TB-score

2
0·4

0 AUC 0·708
0·2 AUC 0·640
AUC 0·759
AUC 0·720
–2 0
Confirmed TB Unlikely TB 1·0 0·8 0·6 0·4 0·2 0
MRS

6 1·0
p<0·0001

0·8
4

0·6
Sensitivity
TB-score

2
0·4

0 AUC 0·652
0·2 AUC 0·589
AUC 0·685
AUC 0·644
–2 0
Confirmed TB Unlikely TB 1·0 0·8 0·6 0·4 0·2 0
CRS Specificity

Figure 2: MTB-HR TB-score distribution of participants classified as having confirmed tuberculosis or unlikely tuberculosis by SRS, MRS, or CRS
(A) Raincloud plots show the distribution of the TB-score for the SRS (top panel), MRS (middle panel), and CRS (bottom panel). The bars indicate the
10–90% percentile, and the box indicates the median and IQR. The points are individual data points and show the spread. The data are displayed stratified by whether
the sample was prospective or biobanked (before recentring), but the statistical test was conducted on the entire sample, comparing confirmed tuberculosis with
unlikely tuberculosis. Data were analysed using Mann-Whitney U tests. (B) AUC for the MTB-HR TB-score of all participants and gene transcripts for the SRS
(top panel), MRS (middle panel), and CRS (bottom panel). For the SRS, DeLong tests between TB-score and gene transcripts were: KLF2 p<0·0001, DUSP3 p=0·0018,
and GBP5 p=0·53. For the MRS, DeLong tests between TB-score and gene transcripts were: KLF2 p=0·094, DUSP3 p=0·69, and GBP5 p=0·018. For the CRS, DeLong
tests between TB-score and gene transcripts were: KLF2 p=0·067, DUSP3 p=0·76, and GBP5 p=0·12. AUC=area under the receiver operating characteristic curve.
CRS=clinical reference standard. MRS=microbiological reference standard. MTB-HR=Cepheid Mycobacterium tuberculosis Host Response prototype cartridge.
SRS=strict reference standard. TB=tuberculosis.

specificity of 57·9% (95% CI 42·7–71·7) and 89·1% MTB-HR TB-score performance was stratified by disease
(95% CI 83·2–93·0), respectively, which was similar severity, as per WHO definition,18 and organ manifestation
to the fixed-effects analysis, with a sensitivity and of tuberculosis disease in children with culture-confirmed
specificity of 59·8% (50·8–68·4) and 90·3% (85·5–94·0), tuberculosis. We found a higher sensitivity in children
respectively (appendix p 11). with severe (64·7%, 95% CI 54·6–73·9) versus non-severe

www.thelancet.com/infection Vol 24 February 2024 145

 Articles

specificity of 76·4%. Setting the sensitivity at 65%

A B
1·0
resulted in a cutoff at 1·62 with a specificity of 87·9%,
and setting the sensitivity at 90% resulted in a cutoff
at 2·76 with a specificity of 51·2%.
0·8
To inform potential sampling and testing strategies,
we assessed the incremental yield of MTB-HR and
0·6 AUC 0·816 one Ultra on respiratory samples. Children were
Sensitivity

AUC 0·859
AUC 0·779
included if confirmed by Ultra or culture (or both) from
0·4 AUC 0·884 sputum, gastric lavage, or nasopharyngeal aspirate. In
Age group (years) AUC 0·848 older children aged 5–14 years, 50 (46%) of 109 were
AUC 0·701
<1 (n=37) identified by a positive MTB-HR, with an additional
0·2 1–4 (n=99) HIV status
5–9 (n=116) Positive (n=26)
20 (18%) testing positive on sputum or gastric lavage
10–14 (n=82) Negative (n=302) Ultra. Tuberculosis culture identified five (5%) additional
0 children (appendix p 12). Among young children
(<5 years), 19 (32%) of 60 tested positive by MTB-HR,
C D
with an additional seven (12%) identified by one positive
1·0
Ultra on nasopharyngeal aspirate, and an additional
20 (33%) by a positive Ultra on a respiratory sample.
0·8 Thus, if MTB-HR were combined with one Ultra on
AUC 0·883 sputum or gastric lavage, 43 (71%) of 60 children would
AUC 0·789
0·6 AUC 0·892 have been identified, whereas adding a subsequent
Sensitivity

AUC 0·658 nasopharyngeal aspirate Ultra would have identified

AUC 0·908 AUC 0·812
AUC 0·927
three (5%) additional children. The additional yield of a
0·4 Site
AUC 0·821 tuberculosis culture in addition to Ultra was low
South Africa (n=77)
Malnutrition status Tanzania (n=109) (one child, 2%; appendix p 12).
0·2 None (n=265) Mozambique (n=69) Lastly, longitudinal MTB-HR TB-scores were visualised
Mild or moderate (n=34) Malawi (n=34)
Severe (n=35) India (n=45) for children with culture-confirmed tuberculosis (n=127)
0 and children with unlikely tuberculosis (n=207; appendix
1·0 0·8 0·6 0·4 0·2 0 1·0 0·8 0·6 0·4 0·2 0
p 12). MTB-HR TB-score increased in children treated for
Specificity Specificity
tuberculosis from baseline (median 1·1, IQR 0·3–1·9) to
Figure 3: MTB-HR TB-score to estimate diagnostic accuracy of MTB-HR in subgroups of interest according to 1 month (1·9, 1·3–2·6), 3 months (2·4, 1·7–2·9), and
the strict reference standard 6 months (3·0, 2·5–3·3; ANOVA p<0·0001) but did not
(A) Diagnostic accuracy by age group (n=334). DeLong test between age <1 year and other age groups: 1–5 years
show significant changes over time in children with
p=0·60, 5–10 years p=0·68, >10 years p=0·63. (B) Diagnostic accuracy by HIV status (n=328). DeLong test between
HIV negative and HIV positive: p=0·31. (C) Diagnostic accuracy by malnutrition status (n=334). DeLong test unlikely tuberculosis (baseline median TB-score: 2·8,
between no malnutrition and other malnutrition statuses: severe p=0·15, mild or moderate p=0·052. 2·1–3·2; 1 month: 2·9, 2·4–3·2; 3 months: 3·0, 2·6–3·23;
(D) Diagnostic accuracy by site (n=334). DeLong test between South Africa and other sites: Tanzania p=0·14, ANOVA p=0·079). Pairwise paired t tests showed
Mozambique p=0·92, Malawi p=0·047, India p=0·42. AUC=area under the receiver operating characteristic curve.
significant differences in MTB-HR T-score among
MTB-HR=Cepheid Mycobacterium tuberculosis Host Response prototype cartridge.
children who were culture positive between each
visit (p≤0·05) and no significant differences between the
tuberculosis disease (42·9%, 21·8–66·0), and for visits of those who were culture negative. Children were
disseminated disease—ie, both pulmonary and extra­ then grouped by clinical characteristics and time to
pulmonary tuberculosis (75·0%, 57·8–87·9) compared MTB-HR TB-score of more than 1·5 was calculated; most
with only pulmonary (56·5%, 44·0–68·4) or only children younger than 1 year and with non-severe
extrapulmonary disease (50·0%, 26·0–74·0). tuberculosis changed from a positive to a negative
MTB-HR TB-score was stratified by means of MTB-HR result within 1 month (appendix p 13).
microbiological detection and strength of signal
(appendix p 11). MTB-HR TB-scores did not change Discussion
substantially between Mycobacteria Growth Indicator Evidence presented in this study on the first prospective
Tube time-to-positivity tertiles, but Ultra readouts and evaluation of the MTB-HR cartridge in children using
median MTB-HR TB-scores in children showed the proposed TB-score of 1·5 suggests an AUC of 0·85,
consistency, with very low or trace Ultra results being which is similar to that previously reported in adults
above the proposed cutoff of 1·5 (ie, the majority would (AUCs of 0·88 and 0·89).15,20 Additionally, despite small
have had a negative MTB-HR result). differences, diagnostic accuracy was similar across age
We explored other cutoffs based on the MTB-HR categories and in children with malnutrition, HIV, and
results and diagnostic classifications. Based on the extrapulmonary tuberculosis.
MTB-HR results in the SRS, the Youden’s index proposed Currently available diagnostic approaches in children
a cutoff at 1·99 with a sensitivity of 82·1% and a are either less sensitive (bacteriological) or less specific

146 www.thelancet.com/infection Vol 24 February 2024

 Articles

Samples, n Sensitivity, % Specificity, %

(95% CI) (95% CI)
All True False False True
positive positive negative negative
Overall 334 76 20 51 187 59·8% (50·8–68·4) 90·3% (85·5–94·0)
Age, years
<1 37 8 1 11 17 42·1% (20·3–66·5) 94·4% (72·7–99·9)
1–4 99 19 6 9 65 67·9% (47·6–84·1) 91·5% (82·5–96·8)
5–9 116 14 9 16 77 46·7% (28·3–65·7) 89·5% (81·1–95·1)
10–14 82 35 4 15 28 70·0% (55·4–82·1) 87·5% (71·0–96·5)
HIV status
Negative* 302 72 17 46 167 61·0% (51·6–69·9) 90·8% (85·6–94·5)
Positive 26 4 2 4 16 50·0% (15·7–84·3) 88·9% (65·3–98·6)
Malnutrition†
None 265 51 18 41 155 55·4% (44·7–65·8) 89·6% (84·1–93·7)
Mild or moderate 34 13 1 4 16 76·5% (50·1–93·2) 94·1% (71·3–99·9)
Severe 35 12 1 6 16 66·7% (41·0–86·7) 94·1% (71·3–99·9)
Site
South Africa 77 35 3 13 26 72·9% (58·2–84·7) 89·7% (72·6–97·8)
Tanzania 109 15 6 16 72 48·4% (30·2–66·9) 92·3% (84·0–97·1)
Mozambique 69 7 5 3 54 70·0% (34·8–93·3) 91·5% (81·3–97·2)
Malawi 34 1 4 5 24 16·7% (0·4–64·1) 85·7% (67·3–96·0)
India 45 18 2 14 11 56·3% (37·7–73·6) 84·6% (54·6–98·1)
Tuberculosis severity‡
Non severe§ 21 9 ·· 12 ·· 42·9% (21·8–66·0) ··
Severe¶ 102 66 ·· 36 ·· 64·7% (54·6–73·9) ··
Tuberculosis location‡
Pulmonary tuberculosis 69 39 ·· 30 ·· 56·5% (44·0–68·4) ··
Extrapulmonary tuberculosis 18 9 ·· 9 ·· 50·0% (26·0–74·0) ··
Pulmonary and extrapulmonary tuberculosis 36 27 ·· 9 ·· 75·0% (57·8–87·9) ··
Lymph node tuberculosis 20 14 ·· 6 ·· 70·0% (45·7–88·1) ··
Tuberculosis meningitis 12 8 ·· 4 ·· 66·7% (34·9–90·1) ··
Totals might differ due to missing data. MTB-HR=Cepheid Mycobacterium tuberculosis Host Response prototype cartridge. *Includes children who are HIV-exposed but not
positive. †According to WHO child growth standards.19 ‡Tuberculosis disease severity and location imply that the child already has tuberculosis; therefore, calculating a
specificity was not possible. §According to WHO definition,18 outlined in the appendix (p 3). ¶Children not meeting non-severe tuberculosis disease definition.

Table 2: Diagnostic accuracy of MTB-HR at a cutoff of 1·5 against the strict reference standard using culture for different subgroups

(clinical or radiological) than in adults.3,5 This is results of 0 days and almost 90% treatment initiation.21
underlined by the fact that WHO target product profiles However, despite upward trends, countries have been
for a paediatric test do not define a minimal sensitivity, slow to roll out and scale up implementation of
but an optimal one of at least 66%.9 In our study, automated testing platforms, mostly due to purchase and
MTB-HR either meets or is close to this target in some distribution costs.22,23 Poor sensitisation and training of
subgroups, underlining the potential of rapid diagnosis. clinical and laboratory staff result in underutilisation and
This result holds promise for addressing the diagnostic inadequate exploitation, highlighting the need to improve
gap in children investigated for tuberculosis. the overall cascade of paediatric tuberculosis care.24 Even
Semi-automated testing platforms such as GeneXpert if these platforms are available, specimen collection
facilitate a rapid turnaround time of complex assays in remains a major hurdle.6,8
settings with limited resources. Paediatric care is often Fingerstick blood is an attractive specimen type for a
highly specialised in settings with a high burden of paediatric tuberculosis test, being much simpler and
tuberculosis and is therefore centralised at secondary or quicker to obtain than traditional respiratory specimens.
tertiary levels of health services, although most children It is generally well accepted by children and health-care
present at primary health care.2 Xpert MTB/RIF has been workers, even at the primary health care level.25 Our
shown to be an efficient first-line test for paediatric results suggest that more than two-thirds of children
tuberculosis by a large study in India, with a median with confirmed tuberculosis can be identified by the
duration between specimen collection and reporting of combination of one MTB-HR with one respiratory

www.thelancet.com/infection Vol 24 February 2024 147

 Articles

specimen tested on Ultra. Considering the operational proportion of children defined as unconfirmed
characteristics of the platform, the rapid sample-to-result tuberculosis did not have tuberculosis in our study,
turnaround time of less than 1 h, and the diagnostic offering one explanation for the inferior diagnostic
performance using the current cutoff, MTB-HR might be performance found using the composite reference
best used as a diagnostic instead of as a triage test. standards.
In 2022, WHO recommended treatment decision This study had several limitations. Because we
algorithms for childhood tuberculosis at low levels of performed a rigorous diagnostic study in a well defined
health care, such as primary health care clinics, in population of children investigated for tuberculosis
countries with high tuberculosis burden, and a point-of- disease, large proportions had confirmed and unconfirmed
care test with a high specificity, such as MTB-HR, could tuberculosis. This over-representation of true tuberculosis
improve the diagnostic accuracy for clinical diagnosis.18 cases might overestimate MTB-HR performance in the
Additionally, we found that diagnostic accuracy was target population of children presenting at primary health
generally similar across subgroups of interest and in care. Additionally, some subgroups were small, such as
those difficult to diagnose, such as children who are children living with HIV, limiting the interpretability of
severely malnourished. Thus, MTB-HR might be accuracy estimates. Furthermore, to stringently assign the
valuable as a treatment decision and monitoring tool, clinical case definitions, several children were not
because a positive MTB-HR might be indicative of a included, mostly due to unclear clinical trajectory (ie, lost
more severe form of tuberculosis and could potentially to follow-up or lack of treatment response). One site was
inform treatment duration. also heavily under-represented (Malawi) due to a lost
The three-gene mRNA signature used by MTB-HR is shipment. Because the MTB-HR only became available in
derived from a multicohort analysis from 14 published late 2019, our study dataset is composed of samples
datasets with more than 2500 (mostly adult) samples.14 analysed immediately after collection but also following
Similarly, the calculated MTB-HR TB-score was proposed biobanking. We observed that there was a difference in
based on an adult cohort.15 Exploratory analyses in our median scores between the stored and the fresh samples,
study suggest that an adaptation of the cutoff might probably due to RNA degradation in biobanked samples.29
be beneficial to optimise diagnostic performance, However, by recentring the means of the biobanked
warranting further analyses. samples to the prospective samples, we found a similar
Diagnostic accuracy studies in diseases with imperfect distribution of scores between the two groups and the
and unreliable reference standards, such as paediatric calculated AUCs were almost identical. This batch effect
tuberculosis, might be suboptimally accurate, especially difference between stored and fresh samples could have
when composite reference standards are applied.26 Thus, been suboptimal for analysis of diagnostic accuracy. Lastly,
we conducted our analysis reflecting the different levels MTB-HR was conducted by trained study laboratory staff
of certainty: culture confirmation, composite micro­ and not routine clinical staff.
biological confirmation, and composite clinical reference Being the first study on the use of this assay in children,
standard, all of which have limitations. Although culture future studies are required to replicate our findings. The
might have a lower and heterogeneous yield compared use of an age-adjusted or child-specific cutoff needs to be
with other methods, such as clinical diagnosis, it is widely explored, which should be prospectively tested, as well as
accepted to be rarely false positive and has its own place the optimal placement within the health-care cascade,
in rigorous diagnostic accuracy evaluations.27 Relying on including within recently recommended treatment
culture alone might skew the results towards decision algorithms published by WHO.30 Imple­mentation
multibacillary disease, and although Ultra has a reported studies will be needed to assess performance at different
higher sensitivity, uncertainties around specificity levels of the care cascade and to test performance by
remain. In an updated Cochrane review, Ultra specificity health-care workers rather than laboratory staff.
in children ranged from 94·1% to 98·0%, depending on In summary, our study presents the first estimates on
specimen type.10 Translated to our study where we diagnostic accuracy of the MTB-HR on a fingerstick
conducted a total of 1467 Ultras, between 29 and 87 would blood sample within a large, well defined, and
be false positive. Similarly, the use of a composite clinical geographically diverse paediatric cohort. Considering its
reference standard has considerable limitations. Although operational characteristics, MTB-HR has the potential to
it is important to estimate the accuracy of a novel test in provide an urgently needed new approach to improve
the group of children in most need for a new diagnostic tuberculosis diagnosis for children globally.
approach (ie, unconfirmed tuberculosis), ascertaining a The RaPaed-TB consortium
tuberculosis diagnosis is per definition impossible by Cynthia Biddle Baard, Jacinta Diane Munro, Margaretha Prins,
combining sets of existing imperfect tests.26 A study in Nolufefe Benzi, Linda Claire Bateman, Ashleigh Ryan, Kutala Booi,
Nezisa Paulo, Anthenette Heydenrych, Wonita Petersen,
Lima, Peru, followed up 35 children meeting the criteria Raquel Brookes, Michele Mento, Chad Centner, Craig Dalgarno,
for unconfirmed tuberculosis that were not treated for a Friedrich Rieß, Sarah Mutuku, Elmar Saathoff, Kathrin Held,
median time of 16 months, with only two developing Marilyn Mary Ninan, Anila Chacko, Ramya Kumari, R Dhanabhagyam,
tuberculosis disease.28 It is therefore likely that a Nithya Muniswamy, Marc P Nicol, Bariki Mtafya, Harieth Mwambola,

148 www.thelancet.com/infection Vol 24 February 2024

 Articles

Christina Manyama, Hellen Mahiga, Emanuel Sichone, Lwitiho Sudi, 8 Thomas TA, Heysell SK, Moodley P, et al. Intensified specimen
Cremildo Maueia, Carla Madeira, Justina Cambuie, Jorge Ribeiro, collection to improve tuberculosis diagnosis in children from Rural
Lingstone Chiume, Alice Mnyanga, Tionge Sikwese, Happy Masakasa, South Africa, an observational study. BMC Infect Dis 2014; 14: 11.
Diana Kachere, Masheck Kosaka, Stefan Niemann, Novel Chegou, 9 WHO. High priority target product profiles for new tuberculosis
Lyn Horn. diagnostics: report of a consensus meeting, 28–29 April 2014,
Geneva, Switzerland. Geneva: World Health Organization, 2014.
Contributors 10 Kay AW, Ness T, Verkuijl SE, et al. Xpert MTB/RIF Ultra assay for
NH, LO, and MH acquired study funding. LO, RSo, SMG, HJZ, PN, AT, tuberculosis disease and rifampicin resistance in children.
IS, CG, and NH designed the study. ZF-S, DB, SV, CK, IS, NEN, AM, Cochrane Database Syst Rev 2022; 9: CD013359.
RSe, MN, HJZ, VPV, JSM, ELC, TDM, CG, and AR conducted the study 11 Olbrich L, Khambati N, Bijker EM, Ruhwald M, Heinrich N,
and collected data. LO, AR, LL, and NH curated the data. LO, LL, and Song R. FujiLAM for the diagnosis of childhood tuberculosis:
NH did the formal analysis. LO, NH, AR, and MH were responsible for a systematic review. BMJ Paediatr Open 2022; 6: e001447.
project administration. LO, AR, and NH were responsible for 12 Zak DE, Penn-Nicholson A, Scriba TJ, et al. A blood RNA signature
supervision and validation. LL prepared the figures. LO, LL, NH, VPV, for tuberculosis disease risk: a prospective cohort study. Lancet 2016;
and JSM wrote the original draft of the manuscript. All authors 387: 2312–22.
contributed to review and editing of the manuscript. LL and NH 13 Anderson ST, Kaforou M, Brent AJ, et al. Diagnosis of childhood
accessed and verified the data. All authors had full access to all the data tuberculosis and host RNA expression in Africa. N Engl J Med 2014;
in the study and had final responsibility for the decision to submit for 370: 1712–23.
publication. 14 Sweeney TE, Braviak L, Tato CM, Khatri P. Genome-wide
expression for diagnosis of pulmonary tuberculosis: a multicohort
Declaration of interests analysis. Lancet Respir Med 2016; 4: 213–24.
All authors declare receiving grant funding for this work from the 15 Sutherland JS, van der Spuy G, Gindeh A, et al. Diagnostic accuracy
second European and Developing Countries Clinical Trials Partnership of the Cepheid 3-gene host response fingerstick blood test in a
programme (EDCTP2), the German Center for Infection Research prospective, multi-site study: interim results. Clin Infect Dis 2022;
(DZIF), and Beckman Coulter to their respective institutions. Cepheid 74: 2136–41.
provided testing kits at no cost. ZF-S and HJZ declare funding from the 16 Olbrich L, Nliwasa M, Sabi I, et al. Rapid and accurate diagnosis of
SA-MRC Unit on Child and Adolescent Health to their institution. pediatric tuberculosis disease: a diagnostic accuracy study for
TDM received funding to his institution from Médicins Sans Frontières pediatric tuberculosis. Pediatr Infect Dis J 2023; 42: 353–60.
2017–23, GOSH Charity Intramural COVID-19 Rapid Response Funding, 17 World Medical Association. World Medical Association Declaration
Global Alliance Against Tuberculosis, and EU Innovative Medicines of Helsinki: ethical principles for medical research involving
Initiative for other research activities. TDM receives personal payment human subjects. JAMA 2013; 310: 2191–94.
in his function as Editor in Chief of Annals of Clinical Microbiology and 18 WHO. WHO consolidated guidelines on tuberculosis: module 5:
Antimicrobials. management of tuberculosis in children and adolescents. Geneva:
World Health Organization, 2022.
Data sharing
19 WHO. WHO child growth standards and the identification of
Data collected for the study include individual participant data and a data severe acute malnutrition in infants and children. Geneva: World
dictionary defining each field in the set. These will be made available in Health Organization, 2009.
the form of de-identified data upon reasonable request made to the 20 Södersten E, Ongarello S, Mantsoki A, et al. Diagnostic accuracy
corresponding author. Therefore, approval of a proposal by the study of a novel blood-based assay for identification of tuberculosis
RaPaed-TB consortium will need to be sought and a data access or in people living with HIV. J Clin Microbiol 2021; 59: e01643-20.
sharing agreement, following the guidelines of the EDCTP and the EU 21 Kalra A, Parija D, Raizada N, et al. Upfront Xpert MTB/RIF for
General Data Protection Regulation, will need to be signed. diagnosis of pediatric TB—does it work? Experience from India.
PLoS One 2020; 15: e0236057.
Acknowledgments
This study was funded by the European and Developing Countries 22 Adam P, Pai M. Implementation of Xpert MTB/RIF in high-burden
countries: voices from the field matter. Public Health Action 2019;
Clinical Trials Partnership (EDCTP2 programme supported by the EU;
9: 78–79.
RIA2016MC-1623), with assistance from the UK Medical Research
23 Ponnudurai N, Denkinger CM, Van Gemert W, Pai M. New TB tools
Council, Swedish International Development Cooperation Agency, and
need to be affordable in the private sector: the case study of Xpert
Bundesministerium für Bildung und Forschung. Further funding is MTB/RIF. J Epidemiol Glob Health 2018; 8: 103–05.
contributed by the German Center for Infection Research (DZIF) and
24 Cazabon D, Pande T, Kik S, et al. Market penetration of Xpert MTB/
Beckman Coulter. Cepheid provided test kits and GeneXpert platforms RIF in high tuberculosis burden countries: a trend analysis from
at no cost to the RaPaed-TB consortium. 2014–2016. Gates Open Res 2018; 2: 35.
References 25 Sutcliffe CG, Palamountain KM, Maunga S, et al. The feasibility of
1 WHO. Global tuberculosis report 2022. Geneva: World Health fingerstick blood collection for point-of-care HIV-1 viral load
Organization, 2022. monitoring in rural Zambia. Glob Health Innov 2018; 1: 3.
2 WHO. Roadmap towards ending TB in children and adolescents. 26 Dendukuri N, Schiller I, De Groot J, et al. Concerns about
Geneva: World Health Organization, 2018. composite reference standards in diagnostic research. BMJ 2018;
3 Thomas TA. Tuberculosis in children. Pediatr Clin North Am 2017; 360: j5779.
64: 893–909. 27 DiNardo AR, Detjen A, Ustero P, Ngo K, Bacha J,
4 Tebruegge M, Ritz N, Curtis N, Shingadia D. Diagnostic tests for Mandalakas AM. Culture is an imperfect and heterogeneous
childhood tuberculosis: past imperfect, present tense and future reference standard in pediatric tuberculosis. Tuberculosis (Edinb)
perfect? Pediatr Infect Dis J 2015; 34: 1014–19. 2016; 101S: S105–08.
5 Detjen AK, DiNardo AR, Leyden J, et al. Xpert MTB/RIF assay for 28 Wong M, Coit JM, Mendoza M, et al. Incident tuberculosis
the diagnosis of pulmonary tuberculosis in children: a systematic diagnoses in children at high risk for disease. Open Forum Infect Dis
review and meta-analysis. Lancet Respir Med 2015; 3: 451–61. 2021; 8: ofab075.
6 Zar HJ, Hanslo D, Apolles P, Swingler G, Hussey G. Induced 29 Wilson C, Dias NW, Pancini S, Mercadante V, Biase FH. Delayed
sputum versus gastric lavage for microbiological confirmation of processing of blood samples impairs the accuracy of mRNA-based
pulmonary tuberculosis in infants and young children: biomarkers. Sci Rep 2022; 12: 8196.
a prospective study. Lancet 2005; 365: 130–34. 30 WHO. WHO operational handbook on tuberculosis: module 5:
7 WHO. Automated real-time nucleic acid amplification technology management of tuberculosis in children and adolescents. Geneva:
for rapid and simultaneous detection of tuberculosis and rifampicin World Health Organization, 2022.
resistance: Xpert MTB/RIF assay for the diagnosis of pulmonary
and extrapulmonary TB in adults and children: policy update.
Geneva: World Health Organization, 2013.

www.thelancet.com/infection Vol 24 February 2024 149