Review

Reimagining the status quo: How close are we to rapid

sputum-free tuberculosis diagnostics for all?
Ruvandhi R. Nathavitharana,a Alberto L. Garcia-Basteiro,b,c Morten Ruhwald,d Frank Cobelens,e and Grant Theron f*
a
Division of Infectious Diseases, Beth Israel Deaconess Medical Center & Harvard Medical School, Boston, USA
b
ISGlobal, Hospital Clínic - Universitat de Barcelona, Barcelona, Spain
c
Centro de Investigaç ~ao em Saude de Manhiça, Maputo, Mozambique.
d
FIND, the global alliance for diagnostics, Geneva, Switzerland
e
Department of Global Health and Amsterdam Institute for Global Health and Development, Amsterdam University Medical
Centers, Amsterdam, Netherlands
f
DSI-NRF Centre of Excellence for Biomedical Tuberculosis Research, South African Medical Research Council Centre for
Tuberculosis Research, Division of Molecular Biology and Human Genetics, Faculty of Medicine and Health Sciences, Stellen-
bosch University, South Africa

Summary
Rapid, accurate, sputum-free tests for tuberculosis (TB) triage and confirmation are urgently needed to close the wid- eBioMedicine 2022;00:
ening diagnostic gap. We summarise key technologies and review programmatic, systems, and resource issues that 103939
Published online xxx
could affect the impact of diagnostics. Mid-to-early-stage technologies like artificial intelligence-based automated dig-
https://doi.org/10.1016/j.
ital chest X-radiography and capillary blood point-of-care assays are particularly promising. Pitfalls in the diagnostic ebiom.2022.103939
pipeline, included a lack of community-based tools. We outline how these technologies may complement one
another within the context of the TB care cascade, help overturn current paradigms (eg, reducing syndromic triage
reliance, permitting subclinical TB to be diagnosed), and expand options for extra-pulmonary TB. We review chal-
lenges such as the difficulty of detecting paucibacillary TB and the limitations of current reference standards, and
discuss how researchers and developers can better design and evaluate assays to optimise programmatic uptake.
Finally, we outline how leveraging the urgency and innovation applied to COVID-19 is critical to improving TB
patients’ diagnostic quality-of-care.

Copyright Ó 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/)
Keywords: Tuberculosis; Diagnosis; Active disease; Non-sputum; Point-of-care

Introduction Even before COVID-19, patients evaluated for TB

Tuberculosis (TB) remains a leading cause of death often experienced long delays exacerbated by circuitous
worldwide.1 TB care, which already existed in fragile pathways to care with many missed opportunities for
and overextended healthcare systems, has been nega- diagnosis. The quality of TB symptom screening is
tively impacted by the diversion of human resources often poor and primary care facilities do not often have
and laboratory capacity for COVID-19, resulting in the access to tests, requiring patients to travel elsewhere.4
number of specimens submitted for TB diagnosis plum- These individuals, often already attending decentralized
meting from 7.1 million in 2019 to 5.8 million in 2020.2 health facilities, collectively represent an obvious first
41% of the 10 million people estimated to develop TB step in closing the diagnostic gap. Importantly, how-
globally each year remain undiagnosed, and TB deaths ever, many people with TB never enter health facilities
have risen for the first time in a decade.1 Yet COVID-19 and will require identification in communities; which is
has also demonstrated that large-scale investments often diagnostically challenging due to early-stage dis-
(orders of magnitude higher than that seen for TB) ease and lower pre-test probabilities, which mean tests
could facilitate rapid diagnostic technology develop- require very high sensitivities, and specificities are a
ment, including the use of novel specimen types at major determinant of cost efficacy. Technologies first
point-of-care (POC). A similar approach for TB could need to demonstrate good performance and feasibility
yield major innovative advances.3 in local health facilities before they can be considered
for community-based active case finding (conditions are
less challenging in facilities than in communities and
people with TB presenting to facilities are easier to diag-
*Corresponding author.
nose due to increased pre-test probability). Having good
E-mail address: [email protected] (G. Theron).

www.thelancet.com Vol 00 Month , 2022 1

 Review

performance near POC in facilities is a key criterion development, assessment in large-scale multi-centre
prior to exploring community suitability. field evaluations, or which are judged to show high
Closing the diagnostic gap is not just a technological potential for TB triage, diagnosis, or the assessment of
challenge: it also requires ensuring that high-quality infectiousness, are prioritised for inclusion in the
and modern tests are available, in a manner that enables review. Our review is not exhaustive nor a review of bio-
patients to be promptly linked to care. New technologies markers, in-house assays, centralised platforms for ref-
should be implementable at point-of-care by health care erence laboratories, tests to diagnose infection or drug-
workers with minimal training, with results available in resistance, or the optimal way to do important commu-
a single patient encounter.5 Yet, despite the advent of nity active case finding (high-quality reviews on these
World Health Organization (WHO)-recommended topics exist).18-23 Rather, we focus on sputum-free diag-
rapid molecular tests (mWRDs) such as Xpert MTB/ nostic technologies and tests with demonstrated poten-
RIF Ultra (Ultra) (Cepheid, Sunnyvale, USA) and True- tial for decentralised deployment, ideally at the point-of-
nat MTB/RIF (Molbio Diagnostics, Verna, India), much care at the peripheral health system level (including
of the world still relies on sputum smear microscopy rural settings) to facilitate prompt clinical decision mak-
(developed in the 1880s) as the initial and often only ing for all forms of active TB.
diagnostic test6 despite poor sensitivity. The cost and
infrastructure requirements of mWRDs remain prohibi-
tive to scale-up and even when mWRDs are available, Overview of diagnostic technologies and tests
capacity is often not efficiently utilised or is by specimen type
misaligned.7,8 Furthermore, the use of mWRDs is typi-
Figure 1 illustrates the role of triage (which includes
cally limited to sputum-based testing. Cascade of care
patients with symptoms or risk factors as well as screen-
studies, although setting-specific, indicate that approxi-
ing of unselected populations) and confirmatory tests in
mately 15% of patients diagnosed with TB are lost-to-fol-
a typical population of people who may have TB and
low-up and do not initiate treatment.4,9,10 Together, this
move from community to clinic and hospital settings
picture shows the urgent need for decentralised testing
for diagnosis, the types of non-sputum specimens tested
for TB, so that the majority of patients can be tested and
by some novel technologies under development, and a
start treatment in one visit (ideally irrespective of their
selection of these technologies, their developmental
reason for initial presentation to a facility). Such tests
stage, likely positioning within the health system, and
should be sputum-free, as high-risk groups, such as
where gaps in the pipeline exist. Briefly, several non-
people living with HIV (PLHIV)11 and children12 can
invasive and easily accessible specimens hold promise
often not naturally expectorate sputum and are at higher
as triage and confirmatory tests, however, artificial intel-
risk of extra-pulmonary TB.
ligence (AI)-based digital chest X-ray (dCXR) is the only
Critically, there is growing recognition of the high
design-locked triage test suitable for primary care
proportion of cases identified in prevalence surveys who
(design-locked refers to tests where manufacturers have
are a- or pre-symptomatic (presumed subclinical TB)13,
fixed the individual subcomponents of the assay and the
underscoring how large numbers will be missed by cur-
core of the technology should not change).24 Table 1
rent symptom screening approaches.14,15 Furthermore,
lists various current and upcoming technologies,
since patients who screen positive but have early disease
including their underlying mechanisms and principles,
may not expectorate sputum, sputum confirmatory test-
alignment with TPP criteria, commercial or in-develop-
ing may not be possible. Earlier diagnosis in such peo-
ment examples, diagnostic accuracy, pros and cons, and
ple with minimal symptoms can, on an individual-level,
open questions. This table is summarized below, organ-
prevent disease progression, subsequent morbidity and
ised from, potentially, the least to most invasive speci-
mortality and, on a population-level, reduce as much as
mens.
50% of transmission (importantly, community testing
Tables 1a and 1b
can enable earlier diagnoses than facility testing, result-
ing in even greater benefits).16 Moreover, as for COVID-
19, such tests can also play a critical public health role if Audio
they have the capacity to identify which patients are The growing interest in AI has opened the possibility of
infectious (eg, via Mycobacterium tuberculosis (Mtb) mea- non-invasively detecting cough changes or lung sounds
surement in aerosol) and for how long infectiousness that differentiate people with and without TB, with
lasts. This reinforces the urgent need for rapid, accu- potential for use as a triage test.25,26 This could be done
rate, non-invasive, and sputum-free tests for triage and using portable digital recording and signal processing
confirmatory diagnosis, which have been identified by mobile phone enabled applications27 or digital stetho-
the WHO as priorities for new TB diagnostics, with the scopes.28 Such specimen free technologies, could provide
Target Product Profiles (TPP) developed in 2014.17 a more objective measure of symptoms such as cough,
Tests and technologies that have undergone (or are in contrast to subjective and challenging syndromic
undergoing) external validation, commercial screening, and are potentially scalable given a

2 www.thelancet.com Vol 00 Month , 2022

 Review

Figure 1. Approaches to diagnosing TB. (a) An overview of a typical facility-based TB diagnostic algorithm. People in a community
(without risk factors for TB, white; with symptoms or risk factors for TB, grey; with TB, red) attend a health facility. After screening, all
at-risk individuals are ideally identified and receive a triage test (note in some very high burden settings, all clinical attendees may
be considered at risk; the definition of at-risk is setting-specific), which is done to exclude unnecessary confirmatory testing. Patients
who triage positive then receive typically expensive (yet critical) confirmatory testing, which is used to inform treatment. Impor-
tantly, screening (and potentially testing) could occur in the community, however, this is not shown as most new technologies need
to first demonstrate potential in facilities. (b) Some of the novel materials under investigation for triage or confirmation are shown
(some applicable to both use cases), and (c) a selection of products and technologies that use these materials, their developmental
stage (if known to be under commercialisation), and potential health system-level of deployment. Notably, there are insufficient
late-stage and design-locked triage tests, as well as early and design-locked confirmatory tests useful for facility-based point-of-care
testing (this deficit is even more serious for community-based testing, which is diagnostically and operationally more challenging).
Abbreviations: AI: artificial intelligence, dCXR: digital chest X-ray.

www.thelancet.com Vol 00 Month , 2022 3

4
Review

Specimen Biomarker Principle and Likely TPP for Select studies and Level of confidence, and accuracy Strengths and challenges of Open questions and key
mechanism active TB assays5 years away from estimatesLow, medium, high, technology class considerations
potential implementation, 2- unclear
5 years, 2 years, unclear

Breath and Bacterial or human Compounds like volatile organ- Triage Aeonose (eNose)32 - Sufficient data exists to generate - Potentially implementable at - Lack of independent validation
aerosol metabolites or ics or Mtb antigens are pooled sensitivity (92%) and scale by non-technical per- data, which would be facilitated
antigens expelled in patient aerosol TB Breathalyser (Rapid Biosen- specificity (93%) estimates33 for sonnel by large public data sets for dis-
and may indicate TB-associ- sor Systems)100 this technology class but individ- - Biological (e.g., smoking) covery and training of electronic
ated lung inflammation and ual products are not extensively and environmental (e.g., pol- nose algorithms
disease. Such compounds evaluated. lution) factors can confound - Detection of some products
are detectable using elec- - Aeonose, a volatile organic readouts33 (metabolites) must be closely
tronic noses. compound test, has shown the - Hardware is expensive and timed with collection due to
most promise: in a large South not yet capable of providing degradation34
African cohort of presumptive TB real-time results - These technologies have diag-
patients it had 90% sensitivity nostic promise in people who
and 59% specificity at a rule-out cannot expectorate sputum, but
threshold98 little data exists
Mtb DNA Transmission occurs when Mtb Confirmatory Face masks with filters or absor- - Promising but too early to tell - Sampling over long periods - Could directly diagnose infec-
is exhaled in aerosols, and bent materials capture bac- - One study found a sensitivity of possible to improve yield tiousness, creating a new use
detection could diagnose teria. This is a collection 87% when Ultra was applied to - Specimen processing likely test case37,99 (distinct from diag-
infectious TB. method and different assays face masks worn by pulmonary simpler than for sputum nosis) to identify potential trans-
could be applied. TB cases37 - Potential to detect early mission hotspots or “super-
Blow tubes with filters are an disease37 spreading” individuals for tar-
alternate capture method. - Collection matrices require geted case finding (pooled aero-
optimisation and standard- sol from a congregate setting
isation could also be tested)
- No performance data for
assays other than Ultra like
Truenat (Molbio)
Tongue swabs Mtb DNA Tongue papillae filter and con- Triage Collection method and not a - Studies are few and heterogenous, - FLOQSwabs (Copan Italia) pre- - Performance of novel assays (e.g.,
centrate biomass from respi- test itself no pooled performance esti- ferred41 next-generation LAM and NAATs
ratory secretions, permitting mates exist - Self-swabbing, comparable unknown) may overcome sensi-
formation of Mtb-containing - Ultra applied to a single swab to health worker-adminis- tivity limitations associated
biofilms. had a sensitivity of 88% in outpa- tered swabs for other patho- - Optimal number of swabs,
tients41 but only 43% sensitivity gens44, appears feasible swab design, and the processing
when used for active case finding - Potential for paediatric TB43 method are under evaluation
in a prison40 - Insufficient Mtb may be and may improve the release of
- TB-LAMP42 applied to oral recovered from swabs in material from swabs
swabs had sensitivities ranging patients with low sputum -No tests purpose-built for
from 33-50% bacillary load40 tongue swabs yet exist
Blood Host mRNA blood signatures Triage Xpert Host Response (Cepheid) - A multicentre study showed 90% - Limited data with small num- - Signatures (including Sweeney3 in
transcriptome associated with the immune sensitivity and 86% specificity64 bers of cases, however, mul- Xpert HR) measured using ultra-
system’s response to Mtb -Other studies have shown lower ticentre studies are sensitive methods (sequencing,
have shown promise for specificities (26%101, 53%102) at emerging Nanostring)68 struggle to meet
diagnosis.68 >90% sensitivity - Xpert HR has the most data WHO TPPs65,66, suggesting a
available performance ceiling for tests
RISK6 signature (QuantumDx) - No product-specific data, but sig- - RNA is labile and, for Xpert based on these signatures
nature performance measured by HR, time from blood collec- - Signatures insufficiently tested
real-time PCR is 91% sensitivity tion to testing must be in PLHIV, EPTB, children, con-
and 56% specificity (increasing to <30 min and stabilisation tacts, and for subclinical disease,
agents may be required nor head-to-head with cheap

Table 1a (Continued)

www.thelancet.com Vol 00 Month , 2022

 Specimen Biomarker Principle and Likely TPP for Select studies and Level of confidence, and accuracy Strengths and challenges of Open questions and key
mechanism active TB assays5 years away from estimatesLow, medium, high, technology class considerations
potential implementation, 2- unclear
5 years, 2 years, unclear

75% in patients without - Cost unclear, but likely high markers like CRP
previous TB)67 - Potential utility treatment - Optimal placement in algo-
response monitoring, man- rithms remains unclear
agement of diseases other
than TB (signature-positive
patients without TB could
have other infections95), and
false-positive TB PCR
results65, which are frequent

www.thelancet.com Vol 00 Month , 2022

in people with previous
TB68,84,103
cfDNA Extracellular Mtb DNA frag- Confirmatory No prototypes with public data - A systematic review and meta- - Technical collection parame- - Still at proof-of-concept stage
ments can be detected in analysis of different cfDNA in- ters influence performance81 - Collection and detection
accessible body fluids like house assays found a sensitivity - Tests on plasma appear to approaches unstandardised
blood and specificity of 78% and 97% perform better than on other - Product pipeline uncertain
respectively but large heteroge- fluids61,81
neity noted62
Host markers Markers of inflammation made Triage CRP (near-POC instruments like - Robust meta-analysis data show - CRP outperforms WHO 4- - More CRP data needed in HIV-
in response to disease may the iChroma platform (Bodi- sensitivity and specificity of 77% symptom based screening in negatives and PLHIV (some stud-
assist in screening and tech)) and other instrument- and 74% in PLHIV104. PLHIV104 ies report large specificity differ-
triage. free rapid diagnostic tests -At >90% sensitivity (10 mg/L - Instrument free CRP tests ences by HIV status76) and EPTB
are commercially-available) threshold), specificities of 62% do not yet exist - Programmes lack guidance on
and 43% have been reported in how to implement CRP-based
HIV-negative and -positive screening, given recent policy
patients, respectively75 updates48
- CRP should be included as a
comparator for all biomarker
studies
Antigens, cytokines, Altered molecular signatures Triage MBT assay - A LFA five marker prototype had - High POC potential - Other than CRP, all tests are proto-
and antibodies are detectable in blood 94% sensitivity and 96% -Regional differences in per- types
specificity72 formance may make it diffi- - Most assays have sensitivity
SeroSelectTB - A LFA protype had 84% sensitivity cult to derive standardised limitations and require indepen-
and 97% specificity73 panels for global use dent validation, especially in
Simoa array panel - 86% sensitivity and 69% specificity - Potential utility on saliva105, high burden, HIV-endemic set-
in a multinational global valida- for progression to incident tings (trials underway108-111)
tion cohort74 TB106 and treatment moni- - Stakeholders should plan how
Mtb peptides Triage, NanoDisk-MS (Nanopin) - 88% sensitivity and 96% specificity toring107 a TPP-compliant test could be
confirmatory in a large Chinese cohort79 - Signatures should be con- implemented
cise (max. three targets) to
improve POC feasibility
Immune cell profiling T-cell activation markers can Confirmatory TAM-TB (Beckman Coulter) - A prospective study with non-TB - Demonstrated potential in - Validations required
discriminate active disease infected controls showed 82% children69 and treatment - Hardware requires simplifica-
from other forms sensitivity and 93% specificity, response 71 tion
which were unaffected by HIV70 - Hardware (and expertise) - Related methods that detect
not POC amenable and may Mtb in immune cells require
need laboratories investigation80

Table 1a (Continued)
Review

5
6

Review
Specimen Biomarker Principle and Likely TPP for Select studies and Level of confidence, and accuracy Strengths and challenges of Open questions and key
mechanism active TB assays5 years away from estimatesLow, medium, high, technology class considerations
potential implementation, 2- unclear
5 years, 2 years, unclear

Stool Mtb DNA Swallowed or disseminated Mtb Confirmatory Ultra - A Tanzanian study in presumptive - Evidence to support use in - Data in adults (including PLHIV)
may be detectable in stool TB patients found sensitivity to children46 and sputum- scarce
range from 63-84% and specific- scarce TB - Specimen provision challeng-
ity from 76-93% depending of - Added complexity due to ing in outpatients, but rectal
the type of laboratory45 specimen processing swabs may have utility
- A systematic review and meta- requirements
analysis of Xpert for paediatric TB
found similar performance46
Urine LAM Urine is easy to collect and will Confirmatory FujiLAM (FujiFILM) - In a large multicentre head-to- - Utility for EPTB58 - Data in important patient groups
have high diagnostic yield. head evaluation in HIV-negative - Significant technological like HIV-negatives are lacking
Mtb cellular components outpatients, sensitivity was 53% development still needed for - Comparative performance data
(nucleic acids, complex mol- and specificity 99% and highly some assays (e.g., FLOW-TB, of different LAM assays is
ecules, or even intact cells) variable between settings57 Mologic) compared to Deter- needed (in one study in HIV-
can filter through the kidney - Another study in HIV-positive mine TB-LAM (Alere) and negatives EclLAM, which is labo-
barrier. The Mtb cell wall and inpatients, sensitivity was 70% FujiLAM ratory-based, outperformed
virulence factor LAM are the and specificity 91%56 - Sensitivity inversely propor- other LAM tests that were more
most promising urine FLOW-TB (Salus Discovery) A second-generation version of the tion to the degree of immu- suitable to POC)57
markers. assay had a sensitivity of 86% nosuppression (e.g., CD4 - For urine cfDNA, many of the
and specificity 89% in count), and this should be same considerations as for blood
inpatients59 used to guide roll-out apply
TB-LAM (Biopromic) Unclear (in development)60 - For cfDNA,61 specialised - Enrichment prior to testing
TB-LAM (Mologic) Unclear (in development)112 specimen processing meth- may improve diagnostic yield
cfDNA No prototypes with public data A proof-of-concept study with an ods may overcome sensitiv-
enrichment step had a sensitivity ity limitations to meet TPP
of 84% and specificity 100%61 criteria but will detract from
POC potential

Table 1a: Current and upcoming non-invasive specimen-based technologies for the rapid non-sputum-based diagnosis of TB. Technologies are listed by specimen and biomarker type (least to most
invasive). Examples with best available sensitivity and specificity estimates (as well as the likely WHO target product profiles tests use case) are listed, together with known limitations. Strengths and
challenges for POC deployment in high burden settings as well as open questions and considerations for researchers and implementers are discussed. Rankings of the technological level of maturity
and level of confidence of available performance estimates are given.
Abbreviations: LFA: lateral flow antigen, POC: point of care
www.thelancet.com Vol 00 Month , 2022
www.thelancet.com Vol 00 Month , 2022

Specimen Biomarker Principle and mechanism Likely TPP Select studies and Level of confidence, and Strengths and Open questions
for active assays5 years away from accuracy estimatesLow, challenges of and key considerations
TB potential implementation, medium, high, unclear technology class
2-5 years, 2 years, unclear

Audio Cough TB disease distorts lung archi- Triage - Only proof-of-concept studies - Promising but too early to - Able to rapidly screen all - Early-stage technology
sounds tecture, affecting chest exist25 tell. facility entrants - Standardised large-
sounds in a way detectable - Smartphone applications are - One study with non-TB - Specimen free scale data collection and
by portable digital signal undergoing evaluation (Tim- infected controls obtained a - Rapidly scalable on mobile curation required for
processing.25,26 Classifica- Bre,114 Hyfe)27,108 sensitivity of 93% and speci- phones algorithm development
tion technologies are dis- ficity of 95%.29 - Abnormal sounds in the and testing
Stethoscopes tinct from cough counting, - Stethee28,108 and others under- - Unknown absence of TB could be - Performance likely
which may have treatment going evaluation used to diagnose other highly setting-specific,
monitoring utility.113 - No published studies respiratory diseases especially when other
- No external validation data lung diseases are fre-
quent
- Potential utility for
evaluating post TB lung
damage
Imaging Portable CXR highly sensitive and Triage Software products include: qXR - In a very large evaluation in - Automated reading systems - Challenges in identifying
dCXR increasingly feasible to be (QURE.ai Technologies), Bangladesh, qXR and CAD4- can overcome radiologist which settings for priori-
near POC due to advances CAD4TB (Delft Imaging Sys- TB have the highest specif- shortages and are WHO- tisation, including case
in low-dose portable instru- tems), and Lunit (FujiFILM), icities (both 73%) at >90% endorsed48 finding scenarios where
ments and automated CAD4Good (EPCON) (see sensitivity50 - Expensive equipment and pre-symptomatic TB can
image reading. AI4HLTH for a complete infrastructure (including be detected115
overview)30 high-speed internet) - Achieving adequate
required52, however, exist- throughput to maximise
ing equipment may be ret- utility is challenging
rofitted such that analogue - Unclear how these
images can be converted to technologies will benefit
digital ones diagnosis of other dis-
- Market rapidly evolving eases
and selection of the right - Open source social
solutions challenging impact programs such
as CAD4Good may
reduce costs30
POCUS Improving affordability and Triage Many (most standard point-of- - Systematic review reported Potential to expand diagnosis - Limited data, currently
portability of ultrasound care ultrasound machines can sensitivities ranging from of extrapulmonary TB and majority of studies in
devices has led to interest in be used) 73-100% for subpleural increase diagnostic yield in adults and for pulmo-
the use of POCUS. nodules detected and 47- populations such as chil- nary TB
80% for lung dren or PLHIV - Effects of operator and
consolidation54 machine variability
unclear
- Optimal imaging pro-
tocols unclear

Table 1b: Current and upcoming non-invasive non-specimen-based digital and/or AI-based technologies for the rapid non-sputum-based diagnosis of TB. Technologies are listed by specimen and
biomarker type (least to most invasive). Examples with best available sensitivity and specificity estimates (as well as the likely WHO target product profiles tests use case) are listed, together with
known limitations. Strengths and challenges for POC deployment in high burden settings as well open questions and considerations for researchers and implementers are discussed. Rankings of the

Review
technological level of maturity and level of confidence of available performance estimates are given.
7
 Review

Table 2: Key challenges that need to be overcome for new diagnostic technologies to have a transformative impact on the TB epidemic.
These issues are complex, multifactorial, and interdisciplinary, however, together they create a significant barrier to the adoption of
promising new non-invasive TB tests at the point-of-care. When developing and planning to implement a new test, all factors require
consideration, otherwise potential impact is undermined.
Colour theme key: Red- technical; orange- programmatic; purple- policy.
Abbreviations: POC: point of care, DST: drug susceptibility testing.

theoretical ease-of-use and rapidity of turnaround (sec- Compounds in breath

onds). Nonetheless, these technologies are early stage, Differences in signatures of volatile organic compounds
with signatures yet to be externally validated and limited expelled in breath can be analysed to facilitate detection
clinical data to support use (one study reported a sensi- of pulmonary diseases like TB, most likely as a triage
tivity of 93% and specificity of 95% of AI-based cough test. Such compounds can be detected using electronic
classification when comparing the coughs of people nose devices or captured and concentrated using a col-
with and without TB).25,29 Initiatives to assemble large lection bag and detected by methods like gas chromatog-
collections of cough audio recordings that can be used raphy.32 Although a recent systematic review suggested
to train and test algorithms are needed, similar to what that electronic nose diagnostic tests may have high accu-
was successfully done for dCXR, where a global image racy (pooled sensitivity and specificity both 93%),33
archive from diverse settings and populations was made these tests were often evaluated in case-controlled stud-
available to developers.30 Performance of audio analyses ies comparing people with TB to healthy controls and
is likely to be setting-specific and impacted by the preva- not in the intended setting of use (potential spectrum
lence of other lung diseases in the population but, as a bias leading to accuracy overestimation). Other tests in
new field, acoustic epidemiology31 holds tremendous this review had sensitivities from 62-100% and specific-
promise for identifying patients for confirmatory test- ities from 11-84%.33 Notably, however, data were insuffi-
ing, monitoring treatment response, and assessing pop- cient to obtain pooled estimates for a single type of test.
ulation-level lung health. While the non-invasive nature of such tests is appealing,

8 www.thelancet.com Vol 00 Month , 2022

 Review

large public datasets are, like for audio, needed for algo- sensitivity limitations and automated extraction and
rithm generation and training. However, methods of wash steps were designed for sputum not tongue swabs.
collection (which can lead to the combination of volatile If increased sensitivity could be achieved, tongue swab-
organic compounds with sample bag material), analyti- based approaches could be particularly useful as a diag-
cal approaches, or confounders (like pulmonary comor- nostic test in children, where feasibility with Ultra from
bidities or smoking), can all introduce variation. tongue swab has been demonstrated43, as well as other
Important operational challenges, including the need sputum-scarce populations. Lastly, tongue swab (self)-
for prompt (ideally real-time) volatile organic compound collection has been done reliably by patients for SARS-
detection post-collection (samples undergo degradation CoV-2 testing44, which is an important consideration
after collection) and hardware complexity remain.34 for scale-up and potential community- or home-based
Tests that require specimen shipment off to a central TB testing in under-resourced settings.
laboratory hold less appeal.

Stool
Mtb in aerosol Tests on stool currently have the most utility in diagnos-
Tests to diagnose subclinical (asymptomatic) TB are ing childhood TB, given the challenges in obtaining
lacking. This state (bacteriologically-positive TB in those sputum; although their suboptimal sensitivity (pooled
reporting no symptoms and typically not seeking sensitivity from nine studies was 67%)45,46 remains a
healthcare) is more prevalent than previously estimated, barrier. In adults, stool-based tests may play a role in
with highly variable duration (six months to beyond five the diagnosis of extrapulmonary TB, particularly in
years)35,36 and may be responsible for more than half of groups such as PLHIV who are more likely to have dis-
TB transmission.16 The detection of Mtb bacilli or DNA seminated disease. Rectal swabs may facilitate specimen
in aerosols has been facilitated by the development of collection and processing, making the approach more
face masks, with capture filters or absorbent materials, amenable to POC during a single encounter, although
or blow tubes with a capture filter.37-39 Although capture accuracy data are limited (one study demonstrated the
methods are still under development, early results have sensitivity of stool Xpert testing using FLOQswabs was
been promising, suggesting high diagnostic yield (one 47%).47 Generally, stool-based approaches are ham-
study demonstrated 87% sensitivity when Ultra was pered by the need for non-POC specimen processing.
used to analyse samples collected from face masks), and
may identify patients with subclinical disease earlier
than sputum-based sampling.37 Furthermore, these Imaging
methods can be applied to people before they cough: Several AI computer-aided detection (CAD) software
sampling aerosol from tidal breathing has yielded cul- platforms for TB triage30 are recommended by WHO as
turable Mtb bacilli, suggesting cough is not a transmis- an alternative to human reading, given their potential to
sion prerequisite.39 Detection of Mtb in expelled overcome the lack of qualified readers and the over-
aerosols may thus play a role in identifying early dis- whelming need for improved triage methods given lim-
ease, thought to be important for transmission,41 and ited resources for confirmatory testing.48 An evaluation
assessment of treatment response (important for infec- of 1032 images demonstrated that six out of 12 CAD
tion control); however, this technology remains early platforms (Qure.ai, DeepTek, Delft Imaging, JF Health-
stage. care, OXIPIT, Lunit) performed similarly to an expert
reader, only three of which (Qure.ai, Delft Imaging and
Lunit) performed significantly better than an intermedi-
Tongue swabs ate reader.49 A large evaluation of almost 24 000 outpa-
Mtb in tongue papillae biofilms may be detected using tients, the majority of whom had symptoms,
existing molecular technologies such as Ultra40 with demonstrated that all five of the algorithms evaluated
high sensitivity: 88% relative to sputum Ultra41 based reduced the number of molecular tests required by 50%
on single flocked swab sampling in symptomatic outpa- while maintaining an overall sensitivity of 90%. Two
tients; however, sensitivity may be heavily setting products: qXR (qure.ai, India) and CAD4TB (Delft
dependent (a prison-based active case finding study Imaging Systems, Netherlands) met the triage TPP cri-
reported a sensitivity of Ultra on a single swab of teria.50 Ongoing challenges include the need to adapt
43%).40 Tongue swab-based diagnoses are also possible score thresholds to local epidemiology and use case sce-
using TB-LAMP.42 In general, however, the optimal narios (e.g., triage versus mass population screening,
number of swabs and approach needed to maximize different clinical settings, pre-symptomatic patients), as
DNA recovery during processing remains the subject of software performance, particularly with respect to speci-
active investigation. Next-generation ultra-sensitive tests ficity, can be variable and precludes it from serving as a
may further increase the feasibility of tongue swab- diagnostic test.51 While the equipment cost remains pro-
based methods as current generation tests still have hibitive52, there are advances in the use of POC imaging

www.thelancet.com Vol 00 Month , 2022 9

 Review

devices (including smartphones) that may bridge this other molecular biomarkers also holds great potential
barrier as the market rapidly evolves, particularly given for scale-up in POC tests. For example, despite being a
the potential of this technology class to be used as part non-specific biomarker of inflammation, C-reactive pro-
of integrated management of other diseases such as tein (CRP) is recommended by the WHO as a screening
COVID-19, which may ensure adequate throughput to test in PLHIV given its superior accuracy for TB com-
justify costs.53 Other imaging technologies of interest pared to symptom screening75,76; however improved
include point-of-care ultrasound (POCUS), for which a implementation guidance is needed.77 Other assays to
systematic review revealed sensitivities from 73-100% detect inflammatory biomarkers78 or TB peptides79
for subpleural nodules and 47-80% for lung consolida- remain in the prototype phase but have POC deploy-
tion; however, data are limited with high risk of bias.54 ment potential. Proof-of-concept data has been gener-
ated for Mtb DNA detection in peripheral blood
mononuclear cells80 and cell-free DNA but no proto-
Urine types with public data are available.62,81
Urine is appealing given ease-of-collection, limited
infection control requirements, and potential for extrap-
ulmonary and pulmonary TB. Mtb cellular components
(nucleic acids, molecules, cells) can filter through the Outstanding questions: considerations for test
kidney barrier into urine. Urine lipoarabinomannan developers and evaluators
(LAM) is the only WHO-recommended biomarker for Select considerations for developers and evaluators are
TB diagnosis (specifically the AlereLAM test).55 Other in Table 2, together with issues for implementers and
next-generation LAM assays from FujiFilm, SD Biosen- policymakers, which are interlinked and fall across mul-
sor, Biopromic, Salus, and others are at different devel- tiple themes, are discussed in the following sections.
opmental stages. Studies comparing SILVAMP TB From a technical perspective, developing POC tests
LAM (FujiFilm) to AlereLAM demonstrate a higher sen- is challenging due to disease complexity and the slow
sitivity (70% vs. 42% in PLHIV, 67% vs. 53% in people growth of Mtb. While ultra-sensitive tests are still
without HIV).56-58 Several ultra-sensitive 3rd generation needed to detect early disease (especially in the context
LAM assays59,60 to be used irrespective of HIV status of active case finding),36,82 as tests slowly increase in
will enter trials in 2023. Lastly, emerging data points to sensitivity, diagnostic accuracy assessments may be
a potential role for urine cell-free DNA for TB diagno- compromised by reference standard limitations, espe-
sis;61 however, the need for specialised processing cially as bacteriological tests like liquid culture are often
methods must be overcome for wide use. only applied to sputum.83 Thus, positive results from
highly sensitive tests could be classified as false posi-
tives, representing potentially missed treatment oppor-
Blood tunities. This false-positivity issue may be of less
While there are no validated blood tests for active TB, concern for tests that use antigens or DNA rather than
there is increasing optimism regarding the detection of host biosignatures such as mRNA, however, challenges
antigens, immune cell profiling, host transcriptomics, remain: for example, sputum Ultra trace semi-quantita-
or cell-free Mtb DNA.62 Several studies have demon- tion results are often culture-negative, especially in set-
strated associations between host mRNA signatures tings where patients have high rates of previous TB or a
and TB risk, although preventive therapy guided by one where there is a high background intensity of TB trans-
such mRNA biosignature failed to reduce TB incidence missiont.84 Biomarkers like host RNA and CRP have
in a large randomised controlled trial.63 Despite com- potential for resolving Ultra false-positive results (if bio-
mercial progress to develop mRNA biosignature assays, marker levels are low),65,75 thus Ultra traces could, for
foremost of which is the Xpert Host Response (Xpert example, be reflexed to a test like Xpert HR. Aside from
HR) cartridge that detects a 3-gene signature in capillary limit of detection differences, other reasons for discrep-
blood directly using the widely-deployed GeneXpert ant results include variations in specimen quality and
platform,64 independent validations65,66 demonstrate processing, extra-pulmonary TB, and non-culturable
most transcriptomic biomarkers will not meet the Mtb (the clinical relevance of which is not well under-
WHO diagnostic accuracy criteria for triage or confirma- stood). Composite reference standards that incorporate
tion.65-67 Furthermore, tests will still likely require additional information (e.g., imaging and treatment
costly processing methods that preclude scale-up.68 response) may be useful, however, these substantially
Tests that measure T-cell activation such as TAM-TB increase expense. Extra-pulmonary TB, which accounts
have demonstrated potential for active TB in children for around 16% of all global TB cases,1 presents addi-
(83% sensitivity vs. culture)69,70 and may have a role in tional diagnostic challenges due to invasive sampling to
treatment response monitoring,71 however, TAM-TB’s obtain site-specific tissue or fluid, on which the sensitiv-
reliance on flow cytometry may restrict deployment to ity of existing tests is generally lower than on pulmo-
clinical laboratories. Detection of TB antigens72-74 and nary specimens.85 In addition, for several tests there are

10 www.thelancet.com Vol 00 Month , 2022

 Review

data or potential for extra-pulmonary TB, indicating that ensure tests are being used, interpreted, and acted upon
invasive sampling could be obviated (see Table 1). correctly and promptly.91
Although rapid POC biosignature tests may contrib- Improving TB diagnosis, particularly given the
ute substantially to reducing those patients lost to follow renewed focus on active case finding, needs to be viewed
up between diagnosis and treatment, one limitation within the broader context of health systems strengthen-
(especially for host biomarker tests) is that they are ing. Despite the release of the WHO’s first Essential
unlikely to provide drug susceptibility information to Diagnostics List in 2019,92 country-level investment,
guide appropriate treatment, in contrast to tests that particularly to strengthen primary health care in high
directly detect Mtb. The latter should be given higher TB incidence settings, falls short of ensuring availability
priority by developers and implementers in high DR-TB of essential tests and infrastructure, such as dCXR with
burden settings. CAD in peripheral healthcare settings.93 Syndromic
Consideration of the feasibility of obtaining alterna- rather than test-guided management remains highly
tive specimen types (such as urine or stool) in routine prevalent.94 It is critical that health care providers are
settings, although non-invasive, is important to avoid trained to act rapidly on test results to link patients to
undermining potential benefits. Blood volumes for bio- care, as well as to interpret negative tests to determine
marker tests could also be a concern and these should appropriate follow-up investigations and avoid patients
ideally require only a finger prick. During development, being lost to care without a working diagnosis.
any platform-based assay should consider potential Even the most promising triage tests (CRP, host
maintenance and quality assurance needs for field use transcriptome signatures) have specificity limitations,
and address these during validation or early field evalua- since other conditions (viral infections, cancers) can
tion studies, rather than leaving this for post-market produce readouts that may resemble TB,95 which
evaluation. Furthermore, local communities’ percep- impact implementation potential given that triage test
tions about specific specimen types (e.g., stool) or tests specificity is the primary cost driver.96 Hence, with the
(e.g., audio) should be carefully studied. TB diagnostics introduction of novel tests, a negative test in a triage-
research should utilise interdisciplinary quantitative positive person should inform further non-TB diagnos-
and qualitative approaches to maximise the high quality tic decision making to justify these tools' cost. Most
uptake of testing algorithms by patients and health care modelling exercises do not factor in potential benefits of
providers. non-TB diagnoses, which may underestimate overall
impact. These diagnostic algorithms, including alterna-
tive diagnoses to be considered in patients without TB
would need to be setting- and population-specific.
Outstanding questions: Key issues beyond test The COVID-19 pandemic has prompted the interna-
performance that need to be addressed by tional scientific community to develop highly accurate
programmes and policy makers for implementation SARS-CoV-2 diagnostic tests in record time, with advan-
and scale-up ces in sample collection, including specimens such as
Systematic reviews evaluating the effect of Xpert MTB/ saliva, oral swabs and absorbent strips in facemasks, and
RIF and LAM on patient-important outcomes like mor- decentralised testing, including the use of home-based
tality demonstrate the challenges of moving from diag- testing kits.14 The TB diagnostic community needs to fol-
nostic test accuracy assessment to proving impact.53,86 low similar development pathways, including integrated
Careful evaluation of intended and actual test placement testing for respiratory diseases such as TB and COVID-
and use within the wider TB care cascade is needed to 19, utilizing multi-disease molecular diagnostic plat-
understand potential implementation barriers,87,88 forms (for example, GeneXpert and Truenat), which lev-
such as gaps in linkage and engagement with care. The erages the setting-specific feasibility of automatically
global roll-out of Xpert MTB/RIF has provided key les- testing a specimen for multiple diseases. Digital connec-
sons for policy makers and program implementers,89 tivity tools can facilitate community-based testing, refer-
which include recognising that high costs and unful- rals, contact tracing, linkage to care and/or indicate the
filled quality assurance and maintenance needs in need for alternative clinical evaluation.3,97
chronically underfunded programmes will continue to
result in underutilisation.90 Technologies can mitigate
these barriers if a POC test can be electricity-free, Conclusions
deployable at wide temperature ranges, have long shelf Most upcoming technologies for sputum-free TB tests
life, and the potential for online and offline use. are at an early developmental phase. Furthermore,
National TB programmes must facilitate less restrictive many methods that use novel specimens are still reliant
use of new tests to increase utilisation volumes, requir- on existing molecular tests not originally designed for
ing price negotiations with test manufacturers, and that purpose (e.g., Ultra on tongue swabs) or require
comprehensive health worker training. Furthermore, infrastructure unavailable at primary care. Despite seri-
continuous quality assurance programmes should ous gaps in the diagnostic pipeline, exploring the

www.thelancet.com Vol 00 Month , 2022 11

 Review

potential for several prototype tests that show promise DIAGNOSIS; grant RIA2020I-3305, CAGE-TB) and the
for POC deployment across different use case scenarios National Institute of Allergy and Infection Diseases of
(TPPs) is critical. Nonetheless, given growing recogni- the National Institutes of Health (U01AI152087). GT’s
tion of the high proportion of people with subclinical research group has received funding and/or in-kind
TB and the nuances of use case scenarios for screening donations in the last five years via his employer from
versus triage, revised TPPs are needed. Key needs to Bruker Hain Lifesciences, Cepheid, LumiraDx, FIND,
accelarate test development include publicly available Biopromic, Newmark Diagnostics, Hemocue, Boditech,
standardised datasets and large-scale validation cohorts and Copan.
that facilitate test design and evaluation from the outset None of the funders had any role in paper design,
in populations and settings that reflect their final data collection, data analysis, interpretation, writing of
intended field of use. To overcome these barriers, it is the paper.
critical to leverage the resources and diagnostic innova-
tion and momentum behind the COVID-19 response,
including scale-up of SARS-CoV-2 test capacity and con- References
1 World Health Organization. Global Tuberculosis Report 2021.
nectivity tools. Only then can the promise of novel spu- Geneva, Switzerland.
tum-free TB tests be realised. 2 World Health Organization. Tuberculosis deaths rise for the first
time in more than a decade due to the COVID-19 pandemic.
https://www.who.int/news/item/14-10-2021-tuberculosis-deaths-
rise-for-the-first-time-in-more-than-a-decade-due-to-the-covid-19-
Search strategy and selection criteria pandemic. 20212021).
Data for this Review were identified by searches of the 3 Ruhwald M, Carmona S, Pai M. Learning from COVID-19 to
PubMed database, and references from relevant articles, reimagine tuberculosis diagnosis. Lancet Microbe. 2021;2(5):e169–
e270.
using the search terms (Tuberculosis or TB) AND (diag- 4 Hanson C, Osberg M, Brown J, Durham G, Chin DP. Finding the
nosis) without restrictions based on language, date, Missing Patients With Tuberculosis: Lessons Learned From
Patient-Pathway Analyses in 5 Countries. J Infect Dis. 2017;216
study type, or setting. Given the focus on novel technol- (suppl_7):S686–SS95.
ogy, additional references were obtained from abstracts, 5 Garcia-Basteiro AL, DiNardo A, Saavedra B, et al. Point of care
meeting reports, and clinical trial registration websites, diagnostics for tuberculosis. Pulmonology. 2018;24(2):73–85.
6 Cazabon D, Pande T, Kik S, et al. Market penetration of Xpert
where relevant. MTB/RIF in high tuberculosis burden countries: A trend analysis
from 2014 - 2016. Gates Open Res. 2018;2:35.
7 Brown S, Leavy JE, Jancey J. Implementation of GeneXpert for TB
testing in low-and middle-income countries: A systematic review.
Contributors Global Health: Science and Practice. 2021;9(3):698–710.
RRN, ALGB, and GT drafted the initial manuscript with 8 Nalugwa T, Shete PB, Nantale M, et al. Challenges with scale-up of
GeneXpert MTB/RIFÒ in Uganda: a health systems perspective.
critical input from MR and FC. All authors read and BMC health services research. 2020;20(1):1–7.
approved the final version of the manuscript. 9 Subbaraman R NR, Satyanarayana S, Pai M, Thomas BE, Chadha
VK, Rade K, Swaminathan S, Mayer KH. The Tuberculosis Cascade
of Care in India’s Public Sector: A Systematic Review and Meta-
Analysis. PLoS Med. 2016;13:(10) e1002149. PMID: 27780217.
10 Naidoo P, Theron G, Rangaka MX, et al. The South African Tuber-
Declaration of interests culosis Care Cascade: Estimated Losses and Methodological Chal-
M. Ruhwald reports working for FIND. FIND conducts lenges. J Infect Dis. 2017;216(suppl_7):S702–S713.
multiple clinical research projects to evaluate new diag- 11 Peter JG, Theron G, Singh N, Singh A, Dheda K. Sputum induc-
tion to aid the diagnosis of smear-negative or sputum-scarce TB in
nostic tests against published target product profiles adults from a HIV-endemic setting. European Respiratory Journal.
that have been defined through consensus processes. 2013.
These include studies of diagnostic products developed 12 Drain PK, Gardiner J, Hannah H, et al. Guidance for Studies Evalu-
ating the Accuracy of Biomarker-Based Nonsputum Tests to Diag-
by private sector companies who provide access to nose Tuberculosis. J Infect Dis. 2019;220(220 Suppl 3):S108–S115.
know-how, equipment/reagents, and may contribute 13 Frascella B, Richards AS, Sossen B, et al. Subclinical tuberculosis
disease - a review and analysis of prevalence surveys to inform defi-
through unrestricted donations according to FIND poli- nitions, burden, associations and screening methodology. Clin
cies and in line with guidance from the organisation's Infect Dis. 2020.
external scientific advisory council. FIND does not attri- 14 Richards AS SB, Emery JC, Horton KC, Heinsohn T, Frascella B,
Balzarini F, Oradini-Alacreu A, Hacker B, Odone A, McCreesh N,
bute any financial value to such access. The other Grant AD, Kranzer K, Cobelens F, Esmail H, Houben RMJH. The
authors have no competing interests to declare. natural history of TB disease-a synthesis of data to quantify pro-
gression and regression across the spectrum. medRxiv. 2021.
https://doiorg/101101/2021091321263499.
15 Stuck L, van Haaster AC, Kapata-Chanda P, Klinkenberg E, Kapata
Acknowledgements N, Cobelens F. How 'subclinical' is subclinical tuberculosis? An
analysis of national prevalence survey data from Zambia. Clin Infect
RRN is supported by National Institutes of Health Dis. 2022.
Career Development Award (NIAID K23 AI132648-04) 16 Emery JC, et al. Estimating the contribution of sub-clinical TB dis-
and an American Society of Tropical Medicine and ease to transmission. Union Conference on Lung Health, October
2021 (virtual).
Hygiene Burroughs Wellcome Fellowship. GT acknowl- 17 Kik SV, Denkinger CM, Casenghi M, Vadnais C, Pai M. Tuberculo-
edges funding from the EDCTP2 programme supported sis diagnostics: which target product profiles should be prioritised?
The European respiratory journal. 2014;44(2):537–540.
by the European Union (grant SF1401, OPTIMAL

12 www.thelancet.com Vol 00 Month , 2022

 Review

18 MacLean E, Broger T, Yerlikaya S, Fernandez-Carballo BL, Pai M, 43 Nicol MP, Wood RC, Workman L, et al. Microbiological diagnosis
Denkinger CMJNm. A systematic review of biomarkers to detect active of pulmonary tuberculosis in children by oral swab polymerase
tuberculosis. 2019;4(5):748–758. chain reaction. Scientific reports. 2019;9(1):1–5.
19 Nguyen TNA, Berre A-L, Ba~ nuls A-L, Nguyen TVA. Molecular diag- 44 Tu Y-P, Jennings R, Hart B, et al. Swabs collected by patients or
nosis of drug-resistant tuberculosis; a literature review. Frontiers in health care workers for SARS-CoV-2 testing. New England Journal
microbiology. 2019;10:794. of Medicine. 2020;383(5):494–496.
20 Wei Z, Zhang X, Wei C, et al. Diagnostic accuracy of in-house real- 45 Ngadaya E, Kimaro G, Sandi E, et al. Evaluation of stool GeneXpert
time PCR assay for Mycobacterium tuberculosis: a systematic MTB/RIF for the diagnosis of pulmonary tuberculosis among pre-
review and meta-analysis. BMC infectious diseases. 2019;19(1):1–11. sumptive patients in Tanzania. Journal of Clinical Tuberculosis and
21 Hamada Y, Cirillo DM, Matteelli A, Penn-Nicholson A, Rangaka Other Mycobacterial Diseases. 2020;21: 100195.
MX, Ruhwald M. Tests for tuberculosis infection: landscape analy- 46 MacLean E, Sulis G, Denkinger CM, Johnston JC, Pai M, Ahmad
sis. European Respiratory Journal. 2021. Khan F. Diagnostic accuracy of stool Xpert MTB/RIF for detection
22 Kohli M, MacLean E, Pai M, Schumacher SG, Denkinger CM. of pulmonary tuberculosis in children: a systematic review and
Diagnostic accuracy of centralised assays for TB detection and meta-analysis. Journal of clinical microbiology. 2019;57(6). e02057-
detection of resistance to rifampicin and isoniazid: a systematic 18.
review and meta-analysis. European Respiratory Journal. 2021;57(2). 47 Nicol MP, Spiers K, Workman L, et al. Xpert MTB/RIF testing of
23 Burke RM, Nliwasa M, Feasey HRA, et al. Community-based active stool samples for the diagnosis of pulmonary tuberculosis in chil-
case-finding interventions for tuberculosis: a systematic review. dren. Clin Infect Dis. 2013;57(3):e18–e21.
Lancet Public Health. 2021;6(5):e283–e299. 48 Organization WH. Rapid communication on systematic screening
24 Engel N, Wolffs PFG. Aligning diagnostics to the point-of-care: les- for tuberculosis. 2020.
sons for innovators, evaluators and decision-makers from tubercu- 49 Codlin AJ, Dao TP, Vo LNQ, et al. Independent evaluation of 12
losis and HIV. BMJ Glob Health. 2020;5(11). artificial intelligence solutions for the detection of tuberculosis. Sci
25 Botha G, Theron G, Warren R, et al. Detection of tuberculosis by Rep. 2021;11(1):23895.
automatic cough sound analysis. Physiological measurement. 50 Qin ZZ, Ahmed S, Sarker MS, et al. Tuberculosis detection from
2018;39:(4) 045005. chest x-rays for triaging in a high tuberculosis-burden setting: an
26 Becker K, Scheffer C, Blanckenberg M, Diacon A. Analysis of evaluation of five artificial intelligence algorithms. The Lancet Digi-
adventitious lung sounds originating from pulmonary tuberculo- tal Health. 2021;3(9):e543–e554.
sis. Annu Int Conf IEEE Eng Med Biol Soc. 2013;2013:4334–4337. 51 Tavaziva G, Harris M, Abidi SK, et al. Chest X-ray analysis with
27 Hyfe. AI-powered cough counting: an objective and scalable end- deep learning-based software as a triage test for pulmonary tuber-
point. https://uploads-ssl.webflow.com/601331581ba868154325 culosis: an individual patient data meta-analysis of diagnostic accu-
e525/608957a21c175c2e8f10d7fe_WP3-Clinical-EndPoints.pdf. racy. Clinical Infectious Diseases: an Official Publication of the
28 Stethee Pro. https://www.m3dicine.com/#fresh_air. Infectious Diseases Society of America. 2021.
29 Pahar M, Klopper M, Reeve B, Warren R, Theron G, Niesler TR. 52 Digital Chest Radiography and Computer-Aided. Detection (CAD)
Automatic cough classification for tuberculosis screening in a real- Solutions for Tuberculosis Diagnostics - Technology Landscape
world environment. Physiological Measurement. 2021. Analysis. FIND. 2021.
30 AI4Health Initiative. https://www.ai4hlth.org/. Accessed Oct 7, 53 Nathavitharana RR, Lederer P, Chaplin M, Bjerrum S, Steingart
2021. KR, Shah M. Impact of diagnostic strategies for tuberculosis using
31 The Hyfe Team. A coustic AI as a scalable tool for diagnostics & lateral flow urine lipoarabinomannan assay in people living with
monitoring: Hyfe’s impact vision. https://global-uploads.webflow. HIV. Cochrane Database Syst Rev. 2021;8: CD014641.
com/601331581ba868154325e525/604284b73f6cc81c541b7fe6_ 54 Bigio J, Kohli M, Klinton JS, et al. Diagnostic accuracy of point-of-
Hyfe-Vision-Paper-v2.2.pdf. 2021). care ultrasound for pulmonary tuberculosis: A systematic review.
32 Nakhleh MK, Jeries R, Binder A, et al. Detecting active pulmonary PLoS One. 2021;16:(5) e0251236.
tuberculosis with a breath test using nanomaterial-based sensors. 55 World Health Organization. The use of lateral flow urine lipoarabino-
European Respiratory Journal. 2014;43(5):1522–1525. mannan assay (LF-LAM) for the diagnosis and screening of active
33 Saktiawati AM, Putera DD, Setyawan A, Mahendradhata Y, van der tuberculosis in people living with HIV TB. 2015. Geneva, Switzerland.
Werf TS. Diagnosis of tuberculosis through breath test: a system- 56 Broger T, Sossen B, du Toit E, et al. Novel lipoarabinomannan
atic review. EBioMedicine. 2019;46:202–214. point-of-care tuberculosis test for people with HIV: a diagnostic
34 Mochalski P, King J, Unterkofler K, Amann A. Stability of selected accuracy study. The Lancet Infectious Diseases. 2019;19(8):852–861.
volatile breath constituents in Tedlar, Kynar and Flexfilm sampling 57 Broger T, Nicol MP, Sigal GB, et al. Diagnostic accuracy of 3 urine
bags. Analyst. 2013;138(5):1405–1418. lipoarabinomannan tuberculosis assays in HIV-negative outpa-
35 Ku C-C, MacPherson P, Khundi M, et al. Durations of asymptom- tients. The Journal of clinical investigation. 2020;130(11).
atic, symptomatic, and care-seeking phases of tuberculosis disease 58 Kerkhoff AD, Sossen B, Schutz C, et al. Diagnostic sensitivity of
with a Bayesian analysis of prevalence survey and notification data. SILVAMP TB-LAM (FujiLAM) point-of-care urine assay for extra-
BMC medicine. 2021;19(1):1–13. pulmonary tuberculosis in people living with HIV. European Respi-
36 Richards AS, Sossen B, Emery JC, et al. The natural history of ratory Journal. 2020;55(2).
TB disease-a synthesis of data to quantify progression and 59 Shapiro AE NM, Magcaba ZP, Boyle D, Cantera J, Lillis L, Mullins
regression across the spectrum. medRxiv. 2021. B, Wittwer T, Hannah H, Warrick J, Berry S, Beebe DJ, Wilson
2021.09.13.21263499. DPK, Drain PK. Validation of a laboratory-based reference test for
37 Williams CM, Abdulwhhab M, Birring SS, et al. Exhaled Mycobac- TB-LAM and FLOW-TB, a novel point-of-care TB diagnostic assay,
terium tuberculosis output and detection of subclinical disease by in a high-HIV-prevalence clinical cohort from KwaZulu-Natal,
face-mask sampling: prospective observational studies. The Lancet South Africa. International AIDS Conference. July 2020.
Infectious Diseases. 2020;20(5):607–617. 60 Wood A, Barizuddin S, Darr CM, et al. Ultrasensitive detection of
38 Williams CM, Cheah ES, Malkin J, et al. Face mask sampling for lipoarabinomannan with plasmonic grating biosensors in clinical
the detection of Mycobacterium tuberculosis in expelled aerosols. samples of HIV negative patients with tuberculosis. PloS one.
PloS one. 2014;9:(8) e104921. 2019;14:(3) e0214161.
39 Patterson B, Bryden W, Call C, et al. Cough-independent produc- 61 Oreskovic A, Panpradist N, Marangu D, et al. Diagnosing pulmo-
tion of viable Mycobacterium tuberculosis in bioaerosol. Tuberculo- nary tuberculosis using sequence-specific purification of urine cell-
sis. 2021;126: 102038. free DNA. Journal of Clinical Microbiology. 2021. JCM. 00074-21.
40 Lima F, Santos AS, Oliveira RD, et al. Oral swab testing by XpertÒ 62 Yu G, Shen Y, Ye B, Shi Y. Diagnostic accuracy of Mycobacterium
MTB/RIF Ultra for mass tuberculosis screening in prisons. Journal tuberculosis cell-free DNA for tuberculosis: A systematic review
of clinical tuberculosis and other mycobacterial diseases. 2020;19: and meta-analysis. PLOS ONE. 2021;16:(6) e0253658.
100148. 63 Scriba TJ, Fiore-Gartland A, Penn-Nicholson A, et al. Biomarker-
41 Wood RC, Andama A, Hermansky G, et al. Characterization of oral guided tuberculosis preventive therapy (CORTIS): a randomised
swab samples for diagnosis of pulmonary tuberculosis. PloS one. controlled trial. The Lancet Infectious Diseases. 2021;21(3):354–365.
2021;16:(5) e0251422. 64 Sutherland JS, van der Spuy G, Gindeh A, et al. Diagnostic accu-
42 Song Y, Ma Y, Liu R, et al. Diagnostic Yield of Oral Swab Testing by racy of the Cepheid 3-gene host response fingerstick blood test in a
TB-LAMP for Diagnosis of Pulmonary Tuberculosis. Infection and prospective, multi-site study: interim results. Clinical Infectious Dis-
Drug Resistance. 2021;14:89. eases. 2021.

www.thelancet.com Vol 00 Month , 2022 13

 Review

65 Turner CT, Gupta RK, Tsaliki E, et al. Blood transcriptional bio- 88 Pai M, Schumacher SG, Abimbola S. Surrogate endpoints in global
markers for active pulmonary tuberculosis in a high-burden set- health research: still searching for killer apps and silver bullets?
ting: a prospective, observational, diagnostic accuracy study. 2020. BMJ Glob Health. 2018;3:(2) e000755.
66 Hoang LT, Jain P, Pillay TD, et al. Transcriptomic signatures for 89 Albert H, Nathavitharana RR, Isaacs C, Pai M, Denkinger CM,
diagnosing tuberculosis in clinical practice: a prospective, multi- Boehme CC. Development, roll-out and impact of Xpert MTB/RIF
centre cohort study. The Lancet Infectious Diseases. 2021;21(3):366– for tuberculosis: what lessons have we learnt and how can we do
375. better? The European respiratory journal. 2016;48(2):516–525.
67 Penn-Nicholson A, Mbandi SK, Thompson E, et al. RISK6, a 6-gene 90 Nalugwa T, Shete PB, Nantale M, et al. Challenges with scale-up of
transcriptomic signature of TB disease risk, diagnosis and treatment GeneXpert MTB/RIF(R) in Uganda: a health systems perspective.
response. 2020;10(1):1–21. BMC Health Serv Res. 2020;20(1):162.
68 Gupta RK, Turner CT, Venturini C, et al. Concise whole blood tran- 91 Cattamanchi A, Berger CA, Shete PB, et al. Implementation sci-
scriptional signatures for incipient tuberculosis: a systematic ence to improve the quality of tuberculosis diagnostic services in
review and patient-level pooled meta-analysis. 2020. Uganda. J Clin Tuberc Other Mycobact Dis. 2020;18: 100136.
69 Portevin D, Moukambi F, Clowes P, et al. Assessment of the novel 92 Organization WH. The selection and use of essential in vitro diag-
T-cell activation marker tuberculosis assay for diagnosis of active nostics: report of the second meeting of the WHO Strategic Advi-
tuberculosis in children: a prospective proof-of-concept study. The sory Group of Experts on In Vitro Diagnostics (including the
Lancet infectious diseases. 2014;14(10):931–938. second WHO model list of essential in vitro diagnostics). 2019.
70 Hiza H, Hella J, Arbues A, et al. Case control diagnostic accuracy 93 Huddart S, MacLean E, Pai M. Location, location, location: tubercu-
study of a non-sputum CD38-based TAM-TB test from a single mil- losis services in highest burden countries. Lancet Glob Health.
liliter of blood. Scientific Reports. 2021;11(1):1–9. 2016;4(12):e907–e908.
71 Ahmed MI, Ntinginya NE, Kibiki G, et al. Phenotypic changes on 94 Pai M, Walia K, Boehme CC. Essential medicines and essential
Mycobacterium tuberculosis-specific CD4 T cells as surrogate diagnostics: a package deal. Lancet Public Health. 2019;4(10):e492.
markers for tuberculosis treatment efficacy. Frontiers in immunol- 95 Mulenga H, Musvosvi M, Mendelsohn SC, et al. Longitudinal
ogy. 2018;9:2247. Dynamics of a Blood Transcriptomic Signature of Tuberculosis.
72 Corstjens P, Hooij AV, Tjon Kon Fat E, et al. OC 8435 Multi-bio- American Journal of Respiratory and Critical Care Medicine. 2021.
marker test strip for point-of-care screening for active tuberculosis: (ja).
a five-country multi-centre test evaluation. BMJ Global Health. 96 van’t Hoog AH, Cobelens F, Vassall A, et al. Optimal triage test
2019;4(Suppl 3). A6-A. characteristics to improve the cost-effectiveness of the Xpert MTB/
73 Borain N, Petersen L, Du Plessis J, Theron G, Holm-Hansen C. A RIF assay for TB diagnosis: a decision analysis. PloS one. 2013;8
rapid serological triage test for detecting active tuberculosis. BMJ (12):e82786.
Global Health. 2017;2(Suppl 2). 97 Bigio J, Sassi A, Temesgen Z, Pai M. Improving the quality of
74 Ahmad R, Xie L, Pyle M, et al. A rapid triage test for active pulmo- tuberculosis care in the post-pandemic world. J Clin Tuberc Other
nary tuberculosis in adult patients with persistent cough. Science Mycobact Dis. 2021;23: 100212.
translational medicine. 2019;11(515). 98 Esmail A, Dheda, K, Dhansay, M. TBScience conference. An
75 Calderwood CJ, Reeve BW, Mann T, et al. C-reactive protein as a tri- exhaled-breath screening test for the detection of active pulmonary
age tool for adults with presumptive pulmonary tuberculosis in tuberculosis in patients presenting to primary health care clinics in
South Africa: a prospective cohort study. medRxiv. 2021. Cape Town, South Africa. https://www.enose.nl/clinical-results/
76 Samuels TH, Wyss R, Ongarello S, Moore DA, Schumacher SG, tuberculosis/. 2021).
Denkinger CM. Evaluation of the diagnostic performance of labora- 99 Bunyasi EW, Middelkoop K, Koch A, et al. Molecular Detection of
tory-based c-reactive protein as a triage test for active pulmonary Airborne Mycobacterium tuberculosis in South African High
tuberculosis. PloS one. 2021;16:(7) e0254002. Schools. American Journal of Respiratory and Critical Care Medicine.
77 WHO consolidated guidelines on tuberculosis. Module 2: Screening 2021.
systematic screening for tuberculosis disease. Geneva: World Health 100 Rapid Biosensor Systems. TB Breathalyser - TB Breath Test. http://
Organization; 2021. Licence: CC BY-NC-SA 3.0 IGO. www.rapidbiosensor.com/tbbreathalyser. 2021).
78 Santos VS, Goletti D, Kontogianni K, et al. Acute phase proteins 101 S€odersten E, Ongarello S, Mantsoki A, et al. Diagnostic Accuracy
and IP-10 as triage tests for the diagnosis of tuberculosis: system- Study of a Novel Blood-Based Assay for Identification of Tuberculo-
atic review and meta-analysis. Clin Microbiol Infect. 2019;25 sis in People Living with HIV. Journal of clinical microbiology.
(2):169–177. 2021;59(3). e01643-20.
79 Liu C, Lyon CJ, Bu Y, et al. Clinical evaluation of a blood assay to 102 Moreira FMF, Verma R, Dos Santos PCP, et al. Blood-based host
diagnose paucibacillary tuberculosis via bacterial antigens. Clinical biomarker diagnostics in active case finding for pulmonary tuber-
chemistry. 2018;64(5):791–800. culosis: A diagnostic case-control study. EClinicalMedicine. 2021;33:
80 Belay M, Tulu B, Younis S, et al. Detection of Mycobacterium 100776.
tuberculosis complex DNA in CD34-positive peripheral blood 103 Theron G, Venter R, Smith L, et al. False positive Xpert MTB/RIF
mononuclear cells of asymptomatic tuberculosis contacts: an obser- results in re-tested patients with previous tuberculosis: frequency,
vational study. The Lancet Microbe. 2021;2(6):e267–e275. profile, and prospective clinical outcomes. Journal of clinical micro-
81 Murugesan K, Hogan CA, Palmer Z, et al. Investigation of preana- biology. 2018. JCM. 01696-17.
lytical variables impacting pathogen cell-free DNA in blood and 104 Dhana A, Hamada Y, Kengne AP, et al. Tuberculosis screening
urine. Journal of clinical microbiology. 2019;57(11). e00782-19. among ambulatory people living with HIV: a systematic review and
82 Kendall EA, Kitonsa PJ, Nalutaaya A, et al. The spectrum of tuber- individual participant data meta-analysis. Lancet Infect Dis. 2021.
culosis disease in an urban Ugandan community and its health 105 Jacobs R, Tshehla E, Malherbe S, et al. Host biomarkers detected in
facilities. Clinical infectious diseases : an official publication of the saliva show promise as markers for the diagnosis of pulmonary
Infectious Diseases Society of America. 2020. tuberculosis disease and monitoring of the response to tuberculosis
83 Saavedra B, Mambuque E, Nguenha D, et al. Performance of Xpert treatment. Cytokine. 2016;81:50–56.
MTB/RIF Ultra for tuberculosis diagnosis in the context of passive 106 Penn-Nicholson A, Hraha T, Thompson EG, et al. Discovery and
and active case finding. The European respiratory journal. 2021. validation of a prognostic proteomic signature for tuberculosis pro-
84 Mishra H, Reeve BWP, Palmer Z, et al. Xpert MTB/RIF Ultra and gression: a prospective cohort study. PLoS medicine. 2019;16:(4)
Xpert MTB/RIF for diagnosis of tuberculosis in an HIV-endemic e1002781.
setting with a high burden of previous tuberculosis: a two-cohort 107 Sigal G, Segal M, Mathew A, et al. Biomarkers of tuberculosis
diagnostic accuracy study. Lancet Respir Med. 2020;8(4):368–382. severity and treatment effect: a directed screen of 70 host markers
85 Kohli M, Schiller I, Dendukuri N, et al. Xpert((R)) MTB/RIF assay in a randomized clinical trial. EBioMedicine. 2017;25:112–121.
for extrapulmonary tuberculosis and rifampicin resistance. 108 Rapid Research in Diagnostics Development for TB Network
Cochrane Database Syst Rev. 2018;8: CD012768. (R2D2TB Network). https://clinicaltrials.gov/ct2/show/
86 Haraka F, Kakolwa M, Schumacher SG, et al. Impact of the diag- NCT04923958.
nostic test Xpert MTB/RIF on patient outcomes for tuberculosis. 109 Evaluation of a Rapid Point-of-Care Serological Triage Test for
Cochrane Database Syst Rev. 2021;5: CD012972. Active TB (SeroSelectTB). https://clinicaltrials.gov/ct2/show/
87 Subbaraman R, Jhaveri T, Nathavitharana RR. Closing gaps in the NCT04752592.
tuberculosis care cascade: an action-oriented research agenda. J 110 Evaluation of Host Biomarker-based Point-of-care Tests for Tar-
Clin Tuberc Other Mycobact Dis. 2020;19: 100144. geted Screening for Active TB (ScreenTB). https://clinicaltrials.

14 www.thelancet.com Vol 00 Month , 2022

 Review

gov/ct2/show/NCT03350048?term=walzl&cond=tuberculosis&dra 113 Lee GO, Comina G, Hernandez-Cordova G, et al. Cough dynamics

w=2&rank=1. in adults receiving tuberculosis treatment. PloS one. 2020;15:(6)
111 Point-of-care Triage Test for Active Tuberculosis (TriageTB). e0231167.
https://clinicaltrials.gov/ct2/show/NCT04232618?term=walzl& 114 Pathri R. TimBre: acoustic-based non-invasive point of care screening of tuber-
cond=tuberculosis&draw=2&rank=2. culosis. SOA-01-1005-31 Union World Congress on Lung Health; 2019.
112 Bill and Melinda Gates Foundation. Mologic Limited LAM RDT 115 Fehr J, Konigorski S, Olivier S, et al. Computer-aided interpretation
Award. https://www.gatesfoundation.org/about/committed- of chest radiography reveals the spectrum of tuberculosis in rural
grants/2019/03/opp1193635. 2021). South Africa. npj Digital Medicine. 2021;4(1):1–10.

www.thelancet.com Vol 00 Month , 2022 15