Remarks

For recommendation 2: Sputum includes expectorated and induced sputum. Studies assessing the impact of Xpert MTB/RIF on outcomes in children are lacking. The choice of the specimen will depend on the acceptability (for children, parents, health care workers and other stakeholders) and the feasibility of collecting and preparing specimens in the local context. Regarding Xpert MTB/RIF, the certainty of evidence is higher for sputum and nasopharyngeal aspirates than for other specimen types. The recommendation can be extrapolated for children living with HIV. The direct benefit from testing for rifampicin resistance in sputum (very low certainty of evidence for test accuracy) can be extrapolated to other specimens.

For recommendation 4: The justification for a conditional recommendation is based on:

• low certainty of evidence for test accuracy;
• uncertainty about the interpretation of Xpert Ultra trace results in patients with a prior history of disease and the associated high false-positivity rate; and
• uncertainty about the required resources.

For patients with Xpert Ultra trace results, decisions regarding treatment initiation should include considerations of the clinical presentation and the patient context (including prior treatment history, probability of relapse and other test results).

Remarks

For recommendation 6: This recommendation applies to all patients with signs and symptoms of TB meningitis. The recommendation in children with signs and symptoms of TB meningitis is based on very low certainty of evidence for test accuracy for Xpert MTB/RIF. No data were available on the accuracy of Xpert Ultra for TB meningitis in children.

For recommendation 7: Clinical judgement and pretest probability should guide treatment. In a high pretest probability setting (>5%), a negative test result will not rule out the condition. Available data on Xpert MTB/RIF for children have included lymph node aspirate and lymph node biopsy specimens; given the similarity of the effects, the recommendation for adults is extrapolated for children.

For recommendation 8: The composite reference standard for Xpert Ultra gave similar results when lymph nodes aspirate was compared to lymph nodes biopsy.

For recommendation 9: Clinical judgement and pretest probability should guide treatment. In a high pretest probability setting, a negative test result will not rule out the condition.

For recommendation 10: Blood was only evaluated in people living with HIV (PLHIV) and under particular processing specifications (6), using third-generation Xpert MTB/RIF cartridges, based on one study with a small number of participants. The recommendation applies only to a particular population (HIV-positive adults with signs and symptoms of disseminated TB). The GDG did not feel comfortable extrapolating this recommendation to other patient populations.

Remarks

For recommendation 11: Xpert Ultra trace results will require follow-up, including reassessing clinical symptoms and information on prior history of TB. In the case of suspected rifampicin resistance, repeated testing may provide additional benefit for detection as well as an initial attempt to assess rifampicin resistance.

For recommendation 13: The GDG felt that the implementation of the recommendation depends on the acceptability (for children, parents or caregivers, health care workers and other stakeholders) and the feasibility of conducting repeated testing in the local context. The evidence reviewed evaluated repeating the same test on the same type of specimen. However, from the data reviewed on comparing single tests on different specimen types, there appears to be no difference, regardless of which second specimen is obtained. The recommendation can be extrapolated for children living with HIV (for Xpert MTB/RIF). This includes consideration of the direct benefit from detecting rifampicin resistance in sputum samples (very low certainty of evidence for test accuracy), which the GDG felt can be extrapolated to other samples. The recommendation applies to a moderate or high pretest setting (>5%). If the first test result is positive, the test should not be repeated. In settings with moderate to high pretest probability, the incremental yield of more than two tests is unknown.

For recommendation 15: Desirable and undesirable effects were judged to be moderate, but the GDG felt that testing twice in the moderate and high pretest probability (>5%) settings on balance may provide more benefits than harms. The recommendation is applicable for sputum and nasopharyngeal aspirates. No evidence was identified for stool and gastric aspirates.

Remarks

For recommendation 16: This recommendation was informed by evidence from recent national surveys of TB disease prevalence in four high TB burden countries. Indirectness of the evidence was classified as serious, given that the methods applied in TB prevalence surveys differ from usual programmatic conditions (e.g. symptom screen limited to cough for 14 days or more, and a requirement in surveys to have the results of both symptom screen and chest radiography available). In addition, inconsistency of the evidence was also classified as serious, owing to variability of the data from different countries. As a result, certainty in the estimates of effect was downgraded to low for sensitivity and moderate for specificity. The recommendation applies only to the use of Xpert MTB/RIF or Xpert Ultra for clinical case management in situations where an immediate decision on patient treatment needs to be made and recourse to supplementary tests is not available or would incur delays. It does not apply to scientific studies with other objectives, such as the reliable estimation of the prevalence of TB disease in the community, for which alternative testing algorithms are required (in particular, to address the issue of false positive results, as illustrated in Table 1.17). Recommendations about the screening and diagnostic algorithms to be used in such studies are beyond the scope of this GDG. Recommendations for the diagnostic algorithm(s) to recommend in national TB prevalence surveys specifically are being developed by WHO and are scheduled for release in 2020.

For recommendation 17: There are concerns about losing global and national capacity for culture testing – the current reference standard for identifying active TB disease. An Xpert Ultra trace result was considered as negative in these studies. More false positive results are expected for Xpert Ultra for pulmonary TB. The recommendation applies only to the use of Xpert Ultra for clinical case management. When Xpert Ultra gives a positive result, clinical management should be followed according to national guidelines. When Xpert Ultra gives a negative result, the patient should be re-evaluated clinically. In the case of a culture-positive result, clinical management should be followed according to national guidelines. In the case of a culture-negative result, the patient should be re-evaluated clinically. The recommendation does not apply to scientific studies with other objectives, such as the reliable estimation of the prevalence of TB disease in the community, in which alternative testing algorithms (e.g. using more than one test) may be required. Recommendations for the diagnostic algorithm(s) to be used in such studies are beyond the scope of this GDG. Recommendations for the diagnostic algorithm(s) to recommend in national TB prevalence surveys specifically are being developed by WHO and are scheduled for release in 2020.

PICO 1: Among adults with signs and symptoms of pulmonary TB, seeking care at health care facilities, should Xpert MTB/RIF / Xpert Ultra be used as an initial test for diagnosis of pulmonary TB and rifampicin resistance?

1.1 What is the impact of Xpert MTB/RIF on patient-important outcomes (cure, mortality, time to diagnosis and time to start treatment)?

The aim of the review was to assess the impact on patient-important outcomes of diagnostic strategies using Xpert MTB/RIF compared with strategies using smear microscopy. The following outcomes were considered: all-cause mortality, pretreatment loss to follow-up, cure, time to diagnosis and time to treatment initiation.

For the impact of Xpert MTB/RIF on patient-important outcomes for TB, seven studies were included (16 421 participants): two individually randomized trials (Mupfumi 2014; Theron 2014), four cluster randomized trials (Churchyard 2015; Cox 2014; Ngwira LG 2017; Durovni 2014), and one individual patient data (IPD) meta-analysis (Di Tanna 2019) (see Web Annex 1.1: Xpert MTB/RIF and Xpert Ultra for details of these and other studies). All studies were conducted in high TB burden and high TB/HIV burden countries. There were two trials in South Africa (Churchyard 2015; Cox 2014), one in Zimbabwe (Mupfumi 2014), one in Malawi (Ngwira LG 2017), one in Brazil (Durovni 2014) and two multicountry studies with sites in South Africa, United Republic of Tanzania, Zambia and Zimbabwe (Theron 2014, Di Tanna 2019). All studies were conducted in outpatient settings and enrolled participants aged 18 years or older.

Web Annex 4.1: Impact of diagnostic test Xpert MTB/RIF on patient-important outcomes for tuberculosis: a systematic review.

1.2 What is the diagnostic accuracy of Xpert MTB/RIF for pulmonary TB and rifampicin resistance, as compared with MRS?

The aim of the review was to assess the diagnostic accuracy of Xpert MTB/RIF for pulmonary TB and rifampicin resistance in adults. Randomized trials, cross-sectional studies and cohort studies were included, using respiratory specimens that evaluated Xpert MTB/RIF alone or together with Xpert Ultra against the reference standards of culture for TB detection and culture-based DST or MTBDRplus for rifampicin resistance. Only studies that enrolled adults (aged >15 years) were eligible. For the evaluation of TB detection, studies were included that evaluated the index tests in people with signs and symptoms of pulmonary TB, except for studies in PLHIV, where studies were eligible for inclusion irrespective of signs and symptoms of pulmonary TB (e.g. studies that performed TB screening in PLHIV as part of intensified case finding or before TB preventive therapy).

For detection of pulmonary TB, a total of 94 studies were identified. Of these, 85 studies (40 652 participants) evaluated Xpert MTB/RIF and nine studies (3881 participants) evaluated both Xpert Ultra and Xpert MTB/RIF. Of the 94 studies, 50 (53%) took place in high TB burden and 54 (57%) in high TB/HIV burden countries. Most studies had low risk of bias. Also, most studies had low concern about applicability because participants in these studies were evaluated in primary care facilities, local hospitals or both settings.

For detection of rifampicin resistance, 57 studies (8287 participants) evaluated Xpert MTB/RIF. Of the 57 studies, 27 took place in high MDR-TB burden countries. Most studies were judged as having low risk of bias.

Web Annex 4.2: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in adults with signs and symptoms of pulmonary TB: an updated systematic review.

1.3 What is the diagnostic accuracy of Xpert Ultra for pulmonary TB and rifampicin resistance, as compared with MRS?

For detection of pulmonary TB, a total of nine studies (3881 participants) evaluated both Xpert Ultra and Xpert MTB/RIF. For Xpert Ultra, a composite reference standard was also used that included clinical components as defined by the primary study authors. For detection of rifampicin resistance, eight studies (1039 participants) evaluated Xpert Ultra. The total number of Xpert Ultra studies includes one study that provided data for two cohorts; therefore, we classified these as two distinct studies, Mishra 2019a and Mishra 2019b. Most studies were judged as having high certainty of evidence.

Web Annex 4.2: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in adults with signs and symptoms of pulmonary TB: an updated systematic review.

PICO 2: Among children with signs and symptoms of pulmonary TB, seeking care at health care facilities, should Xpert MTB/RIF / Xpert Ultra be used as an initial test for diagnosis of pulmonary TB and rifampicin resistance?

2.1 What is the diagnostic accuracy of Xpert MTB/RIF for pulmonary TB and rifampicin resistance in children, as compared with MRS and CRS?

The initial search resulted in 835 individual records, with one additional reference identified through other sources, giving a total of 836 records, from which 707 were excluded. Initially, the remaining 129 articles were retrieved. After full-text review, 50 studies were included in the quantitative meta-analysis; of these, 40 (80%) took place in high TB burden countries and 10 in high TB/HIV burden countries. For pulmonary TB detection, 43 studies were included that evaluated the diagnostic accuracy of Xpert MTB/RIF in children, and three that evaluated both Xpert Ultra and Xpert MTB/RIF. Forty-two studies evaluated pulmonary TB using a reference standard of culture, and one study evaluated pulmonary TB using smear microscopy only.

In terms of methodological quality, in the patient selection domain, most studies (83%) evaluating pulmonary TB were judged to have low risk of bias. In the index test domain, all studies were judged to have low risk of bias. In the flow and timing domain, most studies (88%) were judged to have low risk of bias. In the reference standard domain, with respect to the MRS, 47% of studies were judged to have unclear risk of bias because only one culture was used to exclude TB. With respect to the composite reference standard, all studies were judged to have unclear risk of bias because of imperfect accuracy of the composite reference standard and differing definitions of this standard used by the primary study authors. Regarding applicability, in the patient selection domain, 50% of studies were judged as having high or unclear risk of bias, because participants were evaluated exclusively as inpatients at tertiary care centres, or the clinical setting was unclear. With respect to applicability of the index test, most studies (72%) were judged as having low concern owing to standardized application of the index tests. Eleven studies evaluating stool as a specimen for Xpert MTB/RIF or Xpert Ultra were judged to have unclear risk of bias because of the absence of a standardized protocol for stool preparation. Applicability of the reference standard was considered as a low concern for most studies (93%).

To generate evidence about the detection of rifampicin resistance, six studies were included. All of the six studies (223 participants) evaluated only Xpert MTB/RIF and were conducted in high TB burden countries and in high MDR-TB burden countries. Among the studies, 50% had a low risk of bias with respect to patient selection, while all studies had a low risk of bias with respect to the reference standard. Risk of bias was considered low for the reference standard if an automated process was used or it was clear that the reference standard results were interpreted without knowledge of the index tests. For all six studies, there were applicability concerns regarding patient selection because of enrolment exclusively from inpatient or tertiary centres.

For the meta-analysis, a total of 23 studies (6612 participants) evaluated sputum specimens; 14 studies (3468 participants) evaluated gastric specimens; four studies (1125 participants) evaluated nasopharyngeal specimens; and 11 studies (1592 participants) evaluated stool specimens – all of these studies evaluated Xpert MTB/RIF alone. Three studies (753 participants) evaluated both Xpert MTB/RIF and Xpert Ultra on frozen sputum specimens. One study (195 participants) evaluated both Xpert MTB/RIF and Xpert Ultra on nasopharyngeal specimens.

2.2 What is the diagnostic accuracy of Xpert Ultra for pulmonary TB and rifampicin resistance in children, as compared with MRS and CRS?

No studies evaluated Xpert Ultra alone. Three studies (753 participants) evaluated both Xpert MTB/RIF and Xpert Ultra on frozen sputum specimens. One study (195 participants) evaluated both Xpert MTB/RIF and Xpert Ultra on nasopharyngeal specimens.

Web Annex 4.4: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in children: an updated systematic review.

PICO 3: Among adults with signs and symptoms of extrapulmonary TB, seeking care at health care facilities, should Xpert MTB/RIF / Xpert Ultra be used as an initial test for diagnosis of extrapulmonary TB and rifampicin resistance?

3.1 What is the diagnostic accuracy of Xpert MTB/RIF for extrapulmonary TB and rifampicin resistance in adults, as compared with MRS and CRS?

There are difficulties in obtaining extrapulmonary specimens both from children and adults, and technical limitations of conventional bacteriological methods to aid diagnosis. Thus, various non-pulmonary specimens and composite reference standards are often used in evaluating the performance of new diagnostic technologies in extrapulmonary TB.

For detection of extrapulmonary TB, 65 studies were included. A total of 63 studies (13 144 participants) evaluated Xpert MTB/RIF, including five that evaluated both Xpert MTB/RIF and Xpert Ultra. The included studies evaluated Xpert MTB/RIF in cerebrospinal fluid (CSF) specimens comprising lymph node aspirate, lymph node biopsy, pleural fluid, urine, synovial fluid, peritoneal fluid, pericardial fluid and blood.

Of the total of 65 studies, 39 (60%) took place in high TB burden and 41 (63%) in high TB/HIV burden countries. Risk of bias was judged to be low in the domains of patient selection, index test, and flow and timing; and high or unclear in the reference standard domain because many studies decontaminated sterile specimens before culture inoculation. Regarding applicability, in the patient selection domain, high or unclear concern was expressed for most studies because either the participants were evaluated exclusively as inpatients at tertiary care centres, or the clinical settings were unclear.

Annex 4.3: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in adults with signs and symptoms of extrapulmonary TB: an updated systematic review.

3.2 What is the diagnostic accuracy of Xpert Ultra for extrapulmonary TB and rifampicin resistance in adults, as compared with MRS?

Six studies (507 participants) evaluated Xpert Ultra for the detection of extrapulmonary TB. The included studies evaluated the test in CSF specimens comprising lymph node biopsy, pleural fluid, urine and synovial fluid. Serious concerns were expressed regarding the indirectness of the evidence; these concerns related to applicability (i.e. evidence was generated in tertiary referral medical centres), and imprecision of the evidence, related mostly to low numbers of participants included in studies. Certainty of evidence was generally judged as being between low and very low.

Web Annex 4.3: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in adults with signs and symptoms of extrapulmonary TB: an updated systematic review.

PICO 4: Among children with signs and symptoms of extrapulmonary TB and rifampicin resistance, seeking care at health care facilities, should Xpert MTB/RIF / Xpert Ultra be used as an initial test for diagnosis of extrapulmonary TB and rifampicin resistance?

4.1 What is the diagnostic accuracy of Xpert MTB/RIF for extrapulmonary TB and rifampicin resistance in children, as compared with MRS and CRS?

4.2 What is the diagnostic accuracy of Xpert Ultra for extrapulmonary TB and rifampicin resistance in children, as compared with MRS and CRS?

To evaluate detection of extrapulmonary TB, studies that evaluated the diagnostic accuracy of Xpert MTB/RIF in children with signs or symptoms of lymph node TB or TB meningitis were included.

For diagnosis of lymph node TB, six studies (210 participants) evaluated Xpert MTB/RIF against an MRS of smear or culture on lymph node specimens. Two studies (105 participants) evaluated Xpert MTB/RIF against a composite reference standard for lymph node TB. For TB meningitis, six studies (241 participants) evaluated Xpert MTB/RIF against culture on CSF. In addition, two studies (155 participants) assessed Xpert MTB/RIF against a composite reference standard that included a clinical diagnosis of TB meningitis. The certainty of evidence was judged to be very low for sensitivity, and low for specificity of detection of both TB meningitis and lymph node TB.

No studies evaluating the accuracy of Xpert Ultra for detecting lymph node TB or TB meningitis were identified.

Web Annex 4.4: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in children: an updated systematic review.

PICO 5: Among people with signs and symptoms of pulmonary TB, seeking care at health care facilities, do repeated Xpert (Ultra) tests on subsequent samples as an initial test for diagnosis of pulmonary TB and rifampicin resistance increase sensitivity/specificity compared with a single initial test?

5.1 Xpert Ultra repeated test for the diagnosis of pulmonary TB in adults with signs and symptoms of pulmonary TB who have an initial Xpert Ultra trace result, as compared with MRS?

For adults, with initial Xpert Ultra trace results, three studies were identified: Mishra 2019a (4 participants), Piersimoni 2019 (4 participants), and Dorman 2018 (42 participants) (see Web Annex 1.1: Xpert MTB/RIF and Xpert Ultra for details of included studies). Piersimoni 2019 retested the same initial sample, whereas Dorman 2018 retested a separately collected sputum sample. Mishra 2019a retested only those participants with discrepant results (i.e. Ultra trace positive/culture negative), and retested new specimens obtained a median of 444 days (range 245–526 days) after initial testing. Owing to limited data, a meta-analysis was not performed. The evidence was downgraded one level for inconsistency and two levels for imprecision. Serious concerns were expressed for inconsistency, and very serious concerns for imprecision. Certainty of evidence was judged to be very low for both sensitivity and specificity.

Web Annex 4.2: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in adults with signs and symptoms of pulmonary TB: an updated systematic review.

5.2 More than one Xpert MTB/RIF versus one Xpert MTB/RIF to diagnose pulmonary TB in children with signs and symptoms of pulmonary TB, as compared with MRS?

For children, five studies (2119 participants) were included that have evaluated the diagnostic accuracy of multiple Xpert MTB/RIF tests compared with a single test. Serious concerns were expressed for indirectness, because patients were enrolled from inpatient tertiary care settings, which could lead to the enrolment of children with more advanced disease. Also, serious concerns were expressed for imprecision, related to the low number of children with pulmonary TB contributing to this analysis for the observed sensitivity. Overall, the certainty of evidence was judged to be very low for sensitivity and moderate for specificity.

Web Annex 4.4: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in children: an updated systematic review.

5.3 More than one Xpert Ultra versus one Xpert Ultra to diagnose pulmonary TB in children with signs and symptoms of pulmonary TB, as compared with MRS?

For children, one study (163 participants) was included that evaluated the diagnostic accuracy of multiple Xpert Ultra tests in sputum compared with a single test. The certainty of evidence was judged to be very low for sensitivity and low for specificity owing to serious concerns for indirectness and imprecision. In addition, one study (130 participants) was included that evaluated the diagnostic accuracy of multiple Xpert Ultra tests in nasopharyngeal aspirates compared with a single test. Overall, the certainty of evidence was judged to be very low both for sensitivity and specificity, owing to very serious concerns for indirectness and imprecision.

Web Annex 4.4: Xpert MTB/RIF and Xpert Ultra for detecting active tuberculosis in children: an updated systematic review.

PICO 6: Among adults either with signs and symptoms of TB or chest radiograph with lung abnormalities suggestive of pulmonary TB or both, should Xpert MTB/RIF or Xpert Ultra alone be used to define a case of active TB disease (10)?

The aim of the review was to assess the diagnostic accuracy of Xpert MTB/RIF and Xpert Ultra for pulmonary TB in adults (aged ≥15 years) among the general population. Data from four nationally representative and two subnational prevalence surveys for active TB disease, cross-sectional in design, were included. These surveys used sputum samples that evaluated Xpert MTB/RIF or Xpert Ultra against the reference standard of culture for TB. For the evaluation of TB detection, the surveys evaluated the index tests in adults (aged ≥15 years) with chest X-ray abnormalities or symptoms suggestive of pulmonary TB (or both). For detection of pulmonary TB, a total of six surveys were identified.

6.1 Xpert MTB/RIF to diagnose pulmonary TB in adults in the general population with signs and symptoms of pulmonary TB or chest radiograph with lung abnormalities or both, as compared with MRS?

The analysis reported on the results of four surveys, including 49 556 participants. Assessment of the quality of the evidence revealed serious deficiencies in the evidence quality.

Indirectness: the populations in these prevalence surveys differed from the general population with respect to prior testing (e.g. symptom screen was limited to cough for 14 days or more) and the availability of results of both symptom screen and chest radiography in most participants included in the studies. The evidence was downgraded one level for indirectness.

Inconsistency: the sensitivity estimate for Bangladesh was 84%, which was higher than the sensitivity estimates for the other three countries (range, 68–69%). Lower HIV prevalence in Bangladesh could only partly explain the inconsistency. The evidence was downgraded one level for inconsistency. Overall, the certainty of evidence was judged to be low for sensitivity and moderate for specificity.

6.2 Xpert Ultra to diagnose pulmonary TB in adults in the general population with signs and symptoms of pulmonary TB or chest radiograph with lung abnormalities or both, as compared with MRS.

The analysis reported on the results of four surveys, including 11 488 participants. The included countries were Myanmar, South Africa (TREATS project) and Zambia (TREATS project). The average prevalence of TB in these countries was 2.8% (range 1.6–6.7%).

Indirectness: the populations in these prevalence surveys differed from the general population with respect to prior testing (e.g. symptom screen was limited to cough for 14 days or more) and the availability of results of both symptom screen and chest radiography in most participants included in the studies. The evidence was downgraded one level for indirectness.

Imprecision: there were relatively few participants contributing to this analysis, and a wide 95% confidence interval (CI). The 95% CI around true positives and false negatives may lead to different decisions, depending on which limits are assumed. The evidence was downgraded one level for imprecision. Overall, the certainty of evidence was judged to be low for sensitivity and moderate for specificity.

6.3 Two Xpert Ultra versus one Xpert Ultra to diagnose pulmonary TB in adults in the general population with signs and symptoms of TB or chest radiograph with lung abnormalities or both, as compared with MRS.

The analysis reported on the results of three surveys, including 5080 participants. Serious concerns were expressed about the indirectness of the available evidence. This was because most of the data were from Myanmar, and the results may not be applicable to other settings. In addition, very serious concerns were expressed about imprecision because the analysis was based on data for only a small number of individuals. The 95% CIs for two Xpert Ultra assays and one Xpert Ultra assay were wide. Overall, the certainty of evidence was judged to be very low for sensitivity and moderate for specificity.

Summary of the results

1. Xpert is unable to bridge disconnects or lack of capacity in general laboratory services. Participants valued the option to use a specimen other than sputum, but having GeneXpert machines available in the public sector does not necessarily mean that facilities and capacities are available to extract and make use of those specimens. For example, services for histopathology and bacteriology in one country may be disconnected, and sending a specimen to histopathology in the private sector, for instance, may mean that the sample will not return to a public sector GeneXpert machine.
2. Xpert Ultra trace results complicate decision-making. Laboratory and clinical management of trace results was rarely straightforward. Study participants reported challenges with obtaining a second fresh sample when patients had left the facility or had been put on treatment and could not easily produce sputum. If repeat tests are conducted after trace, they cause confusion if the second test has a different result (e.g. is negative). Some laboratory managers are unsure which result to report, and clinicians need expertise and experience to conduct more extensive evaluation for trace patients. This presents challenges in peripheral settings and where turnaround times of confirmatory tests (e.g. phenotypic DST and LPA) slow down clinical decision-making.
3. Discordant results of repeat tests and confirmatory tests can cause confusion around what should be considered the gold standard. This is particularly the case when specimen quality might be poor. Understanding and contextualizing discordant results requires continuous training, experience and expertise.
4. Establishing a thorough TB history of patients is uncommon, and “previously treated” is defined differently. This has implications for potential false positive results through Xpert testing. Clear guidance is needed on how to define previously treated patients, how to handle their Xpert results, and how to capture outcomes in national databases accurately.
5. The lack of trained counsellors and of information provided to patients on diagnostics have negative implications. Patients may be unwilling to accept a diagnosis and invest time and money in clinic visits, follow-up tests and treatment. Patients need better quality counselling by health workers to continue with diagnostic journeys and treatment; such counselling should include information about diagnostic technology and considerations for follow-up testing.
6. Persistent underuse of GeneXpert machines is compounded by the challenges of delays due to sample transport, module breakdown, stock-out of cartridges or complicated diagnostic algorithms. The presence of local Cepheid agents is key for repair. However, high workload and staff turnover, combined with infrastructure and environmental conditions, still cause frequent module breakdown, and repair work can be slow or services deemed insufficient. The challenges of cartridge stock-out lead to important delays and disruption of workflows, leading to underuse.
7. Diagnostic algorithms that are simple to follow in a specific facility (e.g. test all those with presumptive TB) are more feasible and enhance use, but this simplicity depends on cost and supplies. Cartridge stock-outs or prohibitive costs can complicate diagnostic algorithms, making them less feasible to follow and further compounding underuse. In Uganda, Xpert testing eligibility criteria had to be temporarily restricted to particular patient groups because of cartridge shortages that complicated the algorithm.
8. Current donor agreements with governments regarding introduction of new diagnostic technologies are not transparent enough for civil society to be able to hold accountable and follow-up. Involving civil society in negotiating agreements and social contracts at the national level and local facility levels can enhance accountability and the responsiveness of governments, leading to improved implementation processes and access to diagnostics.

Web Annex 4.6: Report on user perspectives on Xpert testing: results from qualitative research.

Recommendations on Truenat MTB, MTB Plus and Truenat MTB-RIF Dx in adults and children with signs and symptoms of pulmonary TB

PICO 7: Among people with signs and symptoms of pulmonary TB, seeking care at health care facilities, should Molbio Truenat MTB, MTB Plus and MTB-RIF Dx be used as an initial test for diagnosis of pulmonary TB and rifampicin resistance?

6.4 What is the diagnostic accuracy of Truenat MTB to diagnose pulmonary TB in adults with signs and symptoms of pulmonary TB, as compared with MRS?

Evidence for the use of Truenat MTB, MTB Plus and MTB-RIF Dx assays to diagnose pulmonary TB and rifampicin resistance in adults was generated through a multicentre prospective clinical evaluation study implemented by FIND. The study was conducted at 19 clinical sites (each with a microscopy centre attached) and seven reference laboratories in four countries. The aim was to determine the diagnostic accuracy of the Truenat assays when performed in the intended settings of use (i.e. microscopy centres), relative to microbiological confirmation (culture) as the reference standard. The performance of the Truenat assays was also compared head-to-head (on the same specimens) to Xpert or Ultra in reference laboratories, as part of this assessment. All sites performed Xpert except for sites in Peru, which performed Ultra. The analysis for Truenat MTB reported on the results for 1336 participants. Serious concerns were expressed for imprecision and inconsistency of evidence related to sensitivity. Overall, the certainty of evidence was judged to be low for sensitivity but high for specificity.

6.5 What is the diagnostic accuracy of Truenat MTB Plus to diagnose pulmonary TB in adults with signs and symptoms of pulmonary TB, as compared with MRS?

The analysis for Truenat MTB Plus reported on the results for 1336 participants. Serious concerns were expressed for imprecision for sensitivity, related to the few participants contributing to the analysis. Overall, the certainty of evidence was judged to be low for sensitivity and high for specificity.

6.6 What is the diagnostic accuracy of Truenat MTB-RIF Dx to diagnose rifampicin resistance in adults with signs and symptoms of pulmonary TB, as compared with MRS?

The analysis for Truenat MTB-RIF Dx reported on the results for 186 participants. For sensitivity, there were serious concerns about indirectness (India and Peru contributed most of the data to the determination of rifampicin resistance) and inconsistency (variable sensitivity estimates: 100% for Peru, based on seven rifampicin-resistant specimens; 100% for Ethiopia, based on one rifampicin-resistant specimen; 100% for Papua New Guinea, based on one rifampicin-resistant specimen; and 81% for India, based on 42 rifampicin-resistant specimens). These results may not be applicable to other settings. In addition, very serious concerns were expressed for imprecision, owing to the small number of participants contributing to this analysis. Overall, the certainty of evidence was judged to be very low for sensitivity. Serious concerns were expressed for indirectness for specificity, related to the low numbers of rifampicin-resistant cases and the fact that most of them were from India and Peru.

Web Annex 4.7: Report on the diagnostic accuracy of the Molbio Truenat tuberculosis and rifampicin-resistance assays in the intended setting of use.

Recommendation

There are several subgroups to be considered for this recommendation:

• The recommendation is based on evidence of diagnostic accuracy in respiratory samples of adults with signs and symptoms of pulmonary TB.
• The recommendation applies to people living with HIV (studies included a varying proportion of such individuals); performance on smear-negative samples was reviewed but was only available for TB detection, not for rifampicin and isoniazid resistance, and data stratified by HIV status were not available.
• The recommendation applies to adolescents and children based on the generalization of data from adults; an increased rate of indeterminate results may be found with paucibacillary TB disease in children.
• The review did not consider extrapolation of the finding for use in people with extrapulmonary TB and testing on non-sputum samples because data on diagnostic accuracy of technologies in the class for non-sputum samples were limited.

Test descriptions

Abbott Molecular has two NAATs for TB, one for detection of Mycobacterium tuberculosis (Mtb) (RealTime MTB test), and one for detection of both rifampicin and isoniazid resistance (RealTime MTB RIF/INH). TB detection targets both the IS6110 genetic element and the pab gene. The rifampicin and isoniazid resistance test uses eight dye-labelled probes to detect variants in the rifampicin resistance determining region (RRDR) of the rpoB gene and four probes to detect isoniazid resistance, with two probes each for the katG and inhA genes. The company reports a limit of detection (LoD) of 17 cfu/mL for the RealTime MTB assay and of 60 colony forming units (cfu)/mL for the RealTime RIF/INH assay (14–16). The test is performed on the m2000 platform, m2000sp for automated DNA extraction and m2000rt for the real-time PCR.

Fig. 2.1.4

m2000sp RealTime system (a) and RealTime MTB Amplification Reagent Kit (b).

Becton Dickinson (BD) has a multiplexed real-time PCR (BD MAX™ MDR-TB) NAAT for the detection of Mtb and resistance to both rifampicin and isoniazid. The test is performed on a platform that uses five-colour detection (14). For Mtb detection, this test targets the multicopy genomic elements IS6110 and IS1081, as well as a single-copy genomic target. To detect resistance to rifampicin, the test targets the RRDR codons 507–533 Escherichia coli nomenclature (426–452 Mtb nomenclature) of the rpoB gene; for detection of resistance to isoniazid, the test targets both the inhA promoter region and the 315 codon of the katG gene. The LoD reported by the company is 0.5 cfu/mL for Mtb detection and 6 cfu/mL for resistance detection. The test is performed on the BD MAX platform, with the DNA automatically extracted and real-time PCR performed.

Fig. 2.1.5

BD MAX™ System (a) and BD MAX PCR Cartridges (b).

Bruker-Hain Diagnostics has two real-time NAATs: the FluoroType^® MTB, which detects Mtb, and the FluoroType MTBDR, which detects Mtb and rifampicin and isoniazid resistance. These platforms are completely independent of the GenoType MTBDR platforms. The FluoroType MTBDR test uses asymmetric excess PCR and light on/off probes. The target genes are rpoB for detection of TB and rifampicin resistance and the inhA promoter and katG gene to detect isoniazid resistance. The LoD reported by the company is 15 cfu/mL for the FluoroType MTB test and 20 cfu/mL for the FluoroType MTBDR assay (14, 17, 18). For DNA extraction, manual (FluoroLyse) and automated (GenoXtract) options are available. The platforms used for amplification and detection are FluoroCycler^® for the MTB assay and FluoroCycler XT for the MTBDR assay.

Fig 2.1.6

the FluoroType^® MTB (a) and FluoroType MTBDR^® (b) test principles.

Roche Diagnostics (Roche) has two NAATs: cobas^® MTB assay to detect Mtb, and cobas^® MTB-RIF/INHassay to detect drug resistance (rifampicin and isoniazid) (14). The cobas MTB assay detects both 16S ribosomal RNA (rRNA) and esx genes as target genes for Mtb detection. The LoD reported by the company for this test was 7.6–8.8 cfu/mL. Rifampicin resistance is detected using RRDR and isoniazid resistance using the inhA promoter region and the katG gene. The tests are run on the cobas 6800/8800 systems, with the DNA automatically extracted and realtime PCR performed.

Figure 2.1.7

cobas^® 6800 or 8800 system (a) and cobas^® MTB Positive Control Kit (b).

Table 2.1.22

Mycobacterium genomic regions targeted by the different assays for TB detection included in the evaluation.

In the moderate complexity class, an automated test is one that has (a) automated DNA extraction, (b) automated PCR preparation and (c) automated result interpretation, with either no pipetting steps or only one pipetting step between (a) and (c). These automated tests may require an initial manual specimen treatment step before the test material is transferred into the sample processing tube. Tests in the moderate complexity category require medical laboratories with biosafety measures in place and test-specific equipment; they also need well-trained, skilled and qualified laboratory staff to set up the tests and carry out the necessary equipment maintenance.

Justification and evidence

The WHO Global TB Programme initiated an update of the current guidelines and commissioned a systematic review on the use of moderate complexity automated NAATs for detection of TB and resistance to rifampicin and isoniazid in people with signs and symptoms of TB.

Three PICO questions were designed to form the basis for the evidence search, retrieval and analysis:

1. Should moderate complexity automated NAATs be used on respiratory samples in people with signs and symptoms of pulmonary TB for detection of pulmonary TB, as compared with culture?
2. Should moderate complexity automated NAATs be used on respiratory samples in people with signs and symptoms of pulmonary TB for detection of resistance to rifampicin, as compared with culture-based phenotypic DST?
3. Should moderate complexity automated NAATs be used on respiratory samples in people with signs and symptoms of pulmonary TB for detection of resistance to isoniazid, as compared with culture-based phenotypic DST?

A comprehensive search of the following databases (PubMed, Embase, BIOSIS, Web of Science, LILACS and Cochrane) for relevant citations was performed. The search was restricted to the period January 2009 to July 2020. Reference lists from included studies were also searched. No language restriction was applied. Because there were few studies for the selected index tests, the diagnostic companies were contacted for reports of their internal validation data. Studies were also included from the WHO public call for submission of data. Mycobacterial culture was used as the reference standard for evaluation of Mtb detection. Resistance detection was compared with a phenotypic DST reference standard and a composite reference standard (that combines phenotypic and genotypic DST results) in studies where both had been performed.

Bivariate random-effects meta-analyses were performed using Stata software, to obtain pooled sensitivity and specificity estimates with 95% CIs for rifampicin resistance, isoniazid resistance and Mtb detection. Where only a limited number of studies were available, descriptive analyses were conducted.

For meta-analysis, studies were first meta-analysed separately for each test. Studies from all the tests were then used to obtain a pooled estimate for all technologies.

To decide whether pooling of all the tests would give meaningful estimates, various criteria for pooling were developed and agreed upon by the GDG panel before they were applied. Data were also evaluated and visualized using head-to-head comparisons of the tests with Xpert^® MTB/RIF or any other WHO-recommended test.

Data for all the index platforms were only pooled to answer PICO questions if they met the preconditions given in Table 2.1.23 and fulfilled either Condition 1 or Condition 2.

Table 2.1.23

Criteria for pooling studies on moderate complexity automated NAATs.

The certainty of the evidence of the pooled studies was assessed systematically through PICO questions, using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach (19, 20). The GRADE approach produces an overall quality assessment (or certainty) of evidence and has a framework for translating evidence into recommendations; also, under this approach, even if diagnostic accuracy studies are of observational design, they start as high-quality evidence.

GRADEpro Guideline Development Tool software (19) was used to generate summary of findings tables. The quality of evidence was rated as high (not downgraded), moderate (downgraded one level), low (downgraded two levels) or very low (downgraded more than two levels), based on five factors: risk of bias, indirectness, inconsistency, imprecision and other considerations. The quality (certainty) of evidence was downgraded by one level when a serious issue was identified and by two levels when a very serious issue was identified in any of the factors used to judge the quality of evidence.

Data synthesis was structured around the three preset PICO questions, as outlined below. Three web annexes⁴ give additional information, as follows:

• details of studies included in the current analysis (Web Annex 1.3: Moderate complexity automated NAATs;
• a summary of the results and details of the evidence quality assessment (Web Annex 2.3: Moderate complexity automated NAATs); and
• a summary of the GDG panel judgements (Web Annex 3.3: Moderate complexity automated NAATs).

PICO 1: Should moderate complexity automated NAATs be used on respiratory samples in people with signs and symptoms of pulmonary TB for detection of pulmonary TB, as compared with culture?

A total of 29 studies with 13 852 specimens provided data for evaluating TB detection from the five index tests (Fig. 2.1.4). Of these 29 studies, 12 were conducted on the Abbott RealTime MTB test, six on FluoroType MTB, four on FluoroType MTBDR, five on BD MAX and two on the cobas MTB test. The reference standard for each of these studies for TB detection was mycobacterial culture.

Of the 29 studies, 16 (55%) had high or unclear risk of bias because they tested specimens before inclusion in the study, used convenience sampling or did not report the method of participant selection. Thus, the evidence was downgraded one level for risk of bias. Overall, the certainty of the evidence was moderate for sensitivity and high for specificity.

Fig. 2.1.8

Forest plot of included studies for TB detection with culture as the reference standard. CI: confidence interval; FN: false negative; FP: false positive; TB: tuberculosis; TN: true negative; TP: true positive.

The overall sensitivity in these 29 studies ranged from 79% to 100%, and the specificity from 60% to 100%. The pooled sensitivity was 93.0% (95% CI: 90.9–94.7%) and the pooled specificity was 97.7% (95% CI: 95.6–98.8%).

PICO 2: Should moderate complexity automated NAATs be used on respiratory samples in people with signs and symptoms of pulmonary TB for detection of resistance to rifampicin, as compared with culture-based phenotypic DST?

A total of 18 studies with 2874 specimens provided data for resistance testing of rifampicin using moderate complexity automated NAATs (Fig. 2.1.5). Of these 18 studies, nine were conducted on the Abbott RealTime RIF/INH test, three on FluoroType MTBDR, four on BD MAX and two on the cobas RIF/INH test. The reference standard for each of these studies for resistance detection was phenotypic DST, using a composite reference standard with both phenotypic DST and sequencing results.

Eight (44%) of the 18 studies had high or unclear risk of bias because they did not report participant selection or tested specimens before inclusion in the study.

Fig. 2.1.9

Forest plot of included studies for rifampicin resistance detection with phenotypic DST as the reference standard. CI: confidence interval; DST: drug susceptibility testing; FN: false negative; FP: false positive; TB: tuberculosis; TN: true negative; (more...)

The overall sensitivity for rifampicin resistance in these 18 studies ranged from 88% to 100% and the specificity from 98% to 100%. The pooled sensitivity was 96.7% (95% CI: 93.1–98.4%) and the pooled specificity was 98.9% (95% CI: 97.5–99.5%).

In determining rifampicin resistance, the results from genetic sequencing (genotypic DST) were obtained where possible, and a composite reference standard was developed that combined the results from phenotypic and genotypic DST. For rifampicin resistance detection, the diagnostic test accuracy of moderate complexity automated NAATs was similar for phenotypic DST and the composite reference standard.

PICO 3: Should moderate complexity automated NAATs be used on respiratory samples in people with signs and symptoms of pulmonary TB for detection of resistance to isoniazid, as compared with culture-based phenotypic DST?

A total of 18 studies with 1758 specimens provided data for resistance testing of isoniazid using moderate complexity automated NAATs (Fig. 2.1.6). Of these 18 studies, nine were conducted on the Abbott RealTime RIF/INH test, three on FluoroType MTBDR, four on BD MAX and two on the cobas MTB-RIF/INH test. The reference standard for each of these studies for resistance detection was phenotypic DST, and a composite reference standard with both phenotypic DST and sequencing results.

Eight (44%) of the 18 studies had high or unclear risk of bias, because participant selection was not reported or prior testing was done on the included specimens.

Fig. 2.1.10

Forest plot of included studies for isoniazid resistance detection with phenotypic DST as the reference standard. CI: confidence interval; DST: drug susceptibility testing; FN: false negative; FP: false positive; RIF: rifampicin; TB: tuberculosis; TN: (more...)

The overall sensitivity for isoniazid resistance in these 18 studies ranged from 58% to 100% and the specificity from 94% to 100%. The pooled sensitivity was 86.4% (95% CI: 82.1–89.8%) and the pooled specificity was 99.8% (95% CI: 98.3–99.8%).

In determining isoniazid resistance, the results from genetic sequencing (genotypic DST) were obtained where possible, and a composite reference standard was developed that combined the results from phenotypic and genotypic DST. For detecting isoniazid resistance, the diagnostic test accuracy of phenotypic DST was similar to that of the composite reference standard.

Cost–effectiveness analysis

This section answers the following additional question:

What is the comparative cost, affordability and cost–effectiveness of implementation of moderate complexity automated NAATs?

A systematic review was conducted, focusing on economic evaluations of moderate complexity automated NAATs. Four online databases (Embase, Medline, Web of Science and Scopus) were searched for new studies published from 1 January 2010 through 17 September 2020. The citations of all eligible articles, guidelines and reviews were reviewed for additional studies. Experts and test manufacturers were also contacted to identify any additional unpublished studies.

The objective of the review was to summarize current economic evidence and further understand the costs, cost–effectiveness and affordability of moderate complexity automated NAATs.

Several commercially available tests were included as eligible tests in the moderate complexity automated NAATs category; however, no published studies were identified assessing the costs or cost–effectiveness of any of those tests. One unpublished study comparing available data on two technologies from moderate complexity automated NAATs class was identified, and the data from that study are described below.

Unpublished data from FIND was provided through direct communication. This costing-only study used time and motion studies combined with a bottom-up, ingredients-based approach to estimate the unit test cost for the two selected technologies.⁵ Time and motion studies were conducted at a reference-level laboratory in South Africa. Several important simplifying assumptions were made that may limit the generalizability of the results; for example, 50% of laboratory operations dedicated to TB, a minimum daily throughput of 24 samples or the equivalent of one BD MAX run (24 tests/run), equipment costs fixed at US$ 100 000 for both platforms, a 5% annual maintenance cost, and the standard 3% discount rate and 10 years expected useful life years.

Additional literature searches conducted to look for economic data using similar platforms from non-TB disease areas identified three additional studies from HIV and hepatitis C virus (HCV) with limited cost data: one (20) using Abbott RealTime HIV and two on HCV (21, 22). Data were limited to cost per unit test kit and are not transferrable to test kit costs for the tests being considered in this review.

How large are the resource requirements (costs)?

Available unit test costs for two moderate complexity automated NAATs ranged from US$ 18.52 (US$ 13.79–40.70) and US$ 15.37 (US$ 9.61–37.40), with one study reporting cheaper per-test kit costs and higher operational costs associated with laboratory processing time. Equipment costs were strong drivers of cost variation and will vary across laboratory networks and operations. If equipment can be optimally placed or multiplexed to ensure high testing volume, the per-test cost can be minimized.

In one-way sensitivity analyses, annual testing volumes varied from fewer than 5000 tests/year to more than 25 000 tests/year. Per-test cost was highly sensitive to testing volume when fewer than 5000 tests were conducted per year; however, unit test costs begin to stabilize between 5000 and 10 000 tests/year, and above 10 000 tests/year, unit cost estimate was robust. When equipment can be multiplexed and used at capacity, per-test cost can be minimized.

What is the certainty of the evidence of resource requirements (costs)?

Available per-test cost data were unpublished but did include overheads, equipment, building, staff and consumable costs; however, complete quality assessment of the study was not possible. Test cost will vary according to testing volume and laboratory operations. There is limited evidence to assess the important variability across sites, countries and implementation approaches.

Does the cost–effectiveness of the intervention favour the intervention or the comparison?

No studies were identified that assessed cost–effectiveness for any of the moderate complexity automated NAATs, and extrapolation was not appropriate given differences in standard of care, care cascades and associated costs, operational conditions, testing volume and diagnostic accuracy. Implementation considerations (e.g. test placement, laboratory network and ability of the programme to initiate treatment quickly) are all likely to affect unit test cost and cost–effectiveness. Economic modelling is needed across various settings to understand the range of cost–effectiveness profiles of moderate complexity automated NAATs, and how they are likely to vary under different operational criteria.

Additional details on economic evidence synthesis and analysis are provided in Web Annex 4.9: Systematic literature review of economic evidence for NAATs to detect TB and DR-TB in adults and children.

User perspective

This section answers the following questions about key informants’ views and perspectives on the use of moderate complexity automated NAATs:

• Is there important uncertainty about or variability in how much end-users value the main outcomes?
• What would be the impact on health equity?
• Is the intervention acceptable to key stakeholders?
• Is the intervention feasible to implement?

User perspectives on the value, feasibility, usability and acceptability of diagnostic technologies are important in the implementation of such technologies. If the perspectives of laboratory personnel, clinicians, patients and TB programme personnel are not considered, the technologies risk being inaccessible to and underused by those for whom they are intended.

To address questions related to user perspective, two activities were undertaken:

• A systematic review of evidence on user perspectives and experiences with NAATs for detection of TB and TB drug resistance (moderate and low complexity automated assays, and high complexity hybridization-based assays) was undertaken from July to November 2020.
• A total of 14 semi-structured interviews with clinicians, programme officers, laboratory staff and patient advocates were conducted in India, Moldova and South Africa from October to November 2020.

The findings from these activities are discussed below.

Interviews

The aim of the interviews was to understand participants’ experiences of using the various technologies (i.e. NAATs for detection of TB and TB drug resistance) and their general TB diagnostic experiences. The three countries – India, Moldova and South Africa – were selected based on them being on WHO’s list of 30 high MDR-TB burden countries (2) and that index tests have been used to some extent in research contexts within these countries. Due to the short time frame, participants were purposively sampled and approached based on convenience through personal contacts and colleagues.

An overview of the participants is given in Table 2.1.24. To mask the identity of study participants they were coded by their country (Moldova [M], India [I] or South Africa [S]), their profession (clinician or medical doctor [M], patient advocate/representative [R], laboratory personnel [L] or programme officers [P]) and a number.

Table 2.1.24

Overview of participants for the end-users’ interviews.

Interviews were conducted using Zoom, Skype or phone. Topics discussed included:

• current approach to diagnosing TB, MDR-TB and extensively drug-resistant TB (XDR-TB), including specific challenges;
• experiences with using molecular TB diagnostics and the index tests specifically, including details on steps taken in the diagnostic process;
• experiences with determining eligibility and treatment initiation, and challenges and benefits of using the index tests;
• overall usefulness of the index tests;
• the feasibility of implementing the index tests;
• the potential impact of the index tests on health equity; and
• how the potential impact of the index tests relates to current policy context.

Several important limitations of this approach were noted. Only a few participants were interviewed per country. Owing to the use of Zoom, Skype or phone for interviews, it was not possible to triangulate interview data with other evidence commonly collected through ethnographic approaches (e.g. multiple interviews and informal conversations at the same facility, observations or site visits). In addition, only some of the participants had personal experience with one or all of the index tests, and those participants who did have experience with the tests had used them in research settings rather than for routine practice.

More details on these interviews are given in Web Annex 4.11: User perspectives on nucleic acid amplification tests for tuberculosis and tuberculosis drug resistance: Interviews study.

Implementation considerations

Factors to consider when implementing moderate complexity automated NAATs for detection of TB and resistance to rifampicin and isoniazid are as follows:

• local epidemiological data on resistance prevalence should guide local testing algorithms, whereas pretest probability is important for the clinical interpretation of test results;

• the cost of a test varies depending on parameters such as the number of samples in a batch and the staff time required; therefore, a local costing exercise should be performed;
• low, moderate and high complexity tests have successive increase in technical competency needs (qualifications and skills) and staff time, which affects planning and budgeting;
• availability and timeliness of local support services and maintenance should be considered when selecting a provider;
• laboratory accreditation and compliance with a robust quality management system (including appropriate quality control) are essential for sustained service excellence and trust;
• training of both laboratory and clinical staff is needed to ensure effective delivery of services and clinical impact;
• use of connectivity solutions for communication of results is encouraged, to improve efficiency of service delivery and reduce time to treatment initiation;
• moderate complexity automated NAATs may already be used programmatically for other diseases – for example, severe acute respiratory syndrome coronavirus 2 (SARS-CoV2), HIV and antimicrobial resistance (AMR) – which could potentially facilitate implementation of TB testing on shared platforms;
• implementation of moderate complexity automated NAATs requires laboratories with the required infrastructure, space and efficient sample referral systems;
• although these are automated tests, well-trained skilled staff are needed to set up assays and complete maintenance requirements; and
• implementation of these tests should be context specific; thus, it should take into account access issues, especially in remote areas, where less centralized WHO-recommended technologies may be more appropriate.

Research priorities

Research priorities for moderate complexity automated NAATs for detection of TB and resistance to rifampicin and isoniazid are as follows:

• diagnostic accuracy in specific patient populations (e.g. children, people living with HIV, and patients with signs and symptoms of extrapulmonary TB) and in non-sputum samples;
• impact of diagnostic technologies on clinical decision-making and outcomes that are important to patients (e.g. cure, mortality, time to diagnosis and time to start treatment) in all patient populations;
• impact of specific mutations on treatment outcomes among people with DR-TB;
• use, integration and optimization of diagnostic technologies in the overall landscape of testing and care, as well as diagnostic pathways and algorithms;
• economic studies evaluating the costs, cost–effectiveness and cost–benefit of different diagnostic technologies;
• qualitative studies evaluating equity, acceptability, feasibility and end-user values of different diagnostic technologies;
• effect of non-actionable results (indeterminate, non-determinate or invalid) on diagnostic accuracy and outcomes that are important to patients;
• operational research on the advantages and disadvantages of individual technologies within the class of moderate complexity automated NAATs;
• effect of moderate complexity automated NAATs in fostering collaboration and integration between disease programmes; and
• the potential utility of detecting katG resistance to identify MDR-TB clones that may be missed because they do not have an RRDR mutation (e.g. the Eswatini MDR-TB clone, which has both the katG S315T and the non-RRDR rpoB I491F mutation).

Remarks

1. These recommendations apply to settings where conventional sputum-smear microscopy can be performed.
2. TB-LAMP should not replace the use of rapid molecular tests that detect TB and resistance to rifampicin, especially among populations at risk of MDR-TB.
3. The test has limited additional diagnostic value over sputum-smear microscopy for testing PLHIV who have signs and symptoms consistent with TB.
4. These recommendations apply only to the use of TB-LAMP in testing sputum specimens from patients with signs and symptoms consistent with pulmonary TB.
5. These recommendations are extrapolated to using TB-LAMP in children, based on the generalization of data from adults, while acknowledging the difficulties of collecting sputum specimens from children.

Remarks

1. The reviewed evidence and recommendations apply to the use of AlereLAM only, because other in-house LAM-based assays have not been adequately validated or used outside limited research settings. Any new or generic LAM-based assay should be subject to adequate validation in the settings of intended use.
2. All patients with signs and symptoms of pulmonary TB who are capable of producing sputum should submit at least one sputum specimen for Xpert MTB/RIF (Ultra) assay, as their initial diagnostic test. This also includes children and adolescents living with HIV who are able to provide a sputum sample.
3. These recommendations also apply to adolescents and children living with HIV, based on generalization of data from adults, while acknowledging that there are very limited data for these population groups.
4. LF-LAM should be used as an add-on to clinical judgement in combination with other tests; it should not be used as a replacement or triage test.

PICO 1: Should low complexity automated NAATs be used on sputum in people with signs and symptoms of pulmonary TB, irrespective of resistance to rifampicin, for detection of resistance to isoniazid, as compared with culture-based phenotypic DST?

Three multinational studies with 1605 participants provided data for evaluating isoniazid resistance detection. The reference standard for each of these studies was culture-based phenotypic DST. Each study centre in the multinational studies was analysed as a separate study (Fig. 2.3.1).

Several concerns were expressed about indirectness in the study populations. First, the median prevalence of isoniazid resistance in the included studies was 67.2% (range, 26.8% [Diagnostics for Multidrug Resistant Tuberculosis in Africa – DIAMA, Benin] to 93.9% [FIND, Moldova]), which is higher than the global estimates for isoniazid resistance. Hence, applicability to settings with a lower prevalence of isoniazid resistance comes with some uncertainty. Second, there are potential differences in the mutations present in isoniazid monoresistant strains and MDR strains; that is, some studies suggest that the mutations found in monoresistant strains are more diverse than the mutations found in MDR strains. Third, although the population for this PICO question is “irrespective of rifampicin resistance”, enrolment criteria in the studies meant that most participants within the included studies had RR-TB. As a result of these concerns, certainty of evidence was downgraded one level for indirectness both for sensitivity and specificity, and the quality (certainty) of evidence was rated moderate both for sensitivity and specificity.

Fig. 2.3.1

Forest plot of included studies for isoniazid resistance detection, irrespective of rifampicin resistance with culture-based phenotypic DST as the reference standard. CI: confidence interval; DIAMA: Diagnostics for Multidrug Resistant Tuberculosis in (more...)

The sensitivity in these three studies ranged from 81% to 100% and the specificity from 87% to 100%. The pooled sensitivity was 94.2% (95% CI: 89.3–97.0%) and the pooled specificity was 98.0% (95% CI: 95.2–99.2%).

PICO 2: Should low complexity automated NAATs be used on sputum in people with signs and symptoms of pulmonary TB, irrespective of resistance to rifampicin, for detection of resistance to fluoroquinolones, as compared with culture-based phenotypic DST?

Three multinational studies with 1337 participants provided data for evaluation of detection of fluoroquinolone resistance. The reference standard for each of these studies was culture-based phenotypic DST. Each study centre in the multinational studies was analysed as a separate study (Fig. 2.3.3).

Specificity estimates were inconsistent, at 84% (FIND, Mumbai), 91% (FIND, New Delhi) and more than 96% for other studies. The heterogeneity in specificity estimates could not be explained. Consequently, the certainty of the evidence was downgraded one level for inconsistency; the quality (certainty) of the evidence was rated high for sensitivity and moderate for specificity.

Fig. 2.3.2

Forest plot of included studies for fluoroquinolone resistance detection, irrespective of rifampicin resistance with culture-based phenotypic DST as the reference standard. CI: confidence interval; DIAMA: Diagnostics for Multidrug Resistant Tuberculosis (more...)

The sensitivity for fluoroquinolone resistance in these three studies ranged from 83% to 100% and the specificity from 84% to 100%. The pooled sensitivity was 93.1% (95% CI: 88.0–96.1%) and the pooled specificity was 98.3% (95% CI: 94.5–99.5%).

PICO 3: Should low complexity automated NAATs be used on culture isolates in people with signs and symptoms of pulmonary TB, and detected resistance to rifampicin, for detection of resistance to ethionamide, as compared with genotypic sequencing of the inhA promoter?

One multinational study with 434 participants provided data for evaluating resistance to ethionamide. The reference standard for this study was DNA sequencing of the inhA promoter. Each study centre in the multinational study was analysed as a separate study (Fig. 2.3.2).

The study was judged to be at very serious risk of bias in the reference standard domain because it did not include all loci (i.e. ethA, ethR and inhA promoter) required for the reference standard to classify the target condition correctly. Against a reference standard of phenotypic DST, the pooled sensitivity was considerably lower, at 51.7% (95% CI: 33.1–69.8%). Consequently, certainty of evidence was downgraded two levels for risk of bias for both sensitivity and specificity. In addition, the 95% CIs were wide for both sensitivity and specificity, which could lead to different decisions, depending on which confidence limits are assumed. Consequently, the certainty of the evidence was downgraded one level for imprecision for both sensitivity and specificity; the quality (certainty) of evidence was rated very low for both sensitivity and specificity.

Fig. 2.3.3

Forest plot of included studies for ethionamide resistance detection with genotypic DST as the reference standard. CI: confidence interval; DST: drug susceptibility testing; FIND: Foundation for Innovative New Diagnostics; FN: false negative; FP: false (more...)

The sensitivity for ethionamide resistance in this study ranged from 78% to 100% and the specificity from 97% to 100%. The pooled sensitivity was 98.0% (95% CI: 74.2–99.9%) and the pooled specificity was 99.7% (95% CI: 83.5–100.0%).

PICO 4: Should low complexity automated NAATs be used on sputum in people with signs and symptoms of pulmonary TB, and detected resistance to rifampicin, for detection of resistance to amikacin, as compared with culture-based phenotypic DST?

One multinational study with 490 participants provided data for evaluating resistance to amikacin. The reference standard for this study was culture-based phenotypic DST. Each study centre in this multinational study was analysed as a separate study (Fig. 2.3.4).

The 95% CI for sensitivity was wide, which could lead to different decisions around true positives and false negatives, depending on which confidence limits are assumed. Also, there were few participants with amikacin resistance contributing to this analysis for the observed sensitivity. Consequently, the certainty of the evidence was downgraded two levels for imprecision. Also, there were few participants with amikacin resistance contributing to this analysis for the observed sensitivity. Consequently, the certainty of the evidence was downgraded two levels for imprecision; the quality (certainty) of evidence was rated low for sensitivity and high for specificity.

Fig. 2.3.4

Forest plot of included studies for amikacin resistance detection with culture-based phenotypic DST as the reference standard. CI: confidence interval; DIAMA: Diagnostics for Multidrug Resistant Tuberculosis in Africa; DST: drug susceptibility testing; (more...)

The sensitivity for amikacin resistance in this study ranged from 75% to 95% and the specificity from 96% to 100%. The pooled sensitivity was 86.1% (95% CI: 75.0–92.7%) and the pooled specificity was 98.9% (95% CI: 93.0–99.8%).

Cost–effectiveness analysis

This section answers the following additional question:

What is the comparative cost, affordability and cost–effectiveness of implementation of low complexity automated NAATs?

A systematic review was conducted, focusing on economic evaluations of low complexity automated NAATs. Four online databases (Embase, Medline, Web of Science and Scopus) were searched for new studies published from 1 January 2010 through 17 September 2020. The citations of all eligible articles, guidelines and reviews were reviewed for additional studies. Experts and test manufacturers were also contacted to identify any additional unpublished studies.

The objective of the review was to summarize current economic evidence and further understand the costs, cost–effectiveness and affordability of low complexity automated NAATs.

Two low complexity automated NAATs were identified: the MeltPro MTB/RIF (Xiamen Zeesan Biotech Co Ltd, China) and the Xpert MTB/XDR assay (Cepheid, Sunnyvale, USA). Only data concerning Xpert MTB/XDR are included in this review. As is the case with Xpert MTB/RIF, the novel XDR assay can be used to test either unprocessed or concentrated sputum. No published studies providing direct evidence on the cost or cost–effectiveness of low complexity automated NAATs were identified.

Through direct communication from the Xpert MTB/XDR manufacturer, Cepheid, the low- and middle-income country (LMIC) cost for the XDR cartridge is expected to be US$ 19.80 ex-works. Shipping and customs costs will be additional and will be borne by the ordering nations or organizations, as is currently the case for Xpert MTB/RIF and Ultra cartridges.

As with the Xpert MTB/RIF and Ultra assays, the test cartridge costs represent just one component of the total unit test costs that must be considered, with equipment being another important consideration. The Xpert MTB/XDR test will not work on existing six-colour modules and will require laboratories to upgrade to 10-colour GeneXpert modules. There will be different upgrade options for the 10-colour system, with different price points depending on the needs and resources available. Upgrade options include:

• a new 10-colour system – this is the most costly option, at US$ 9420 for one module to US$ 72 350 for 16 modules, including the GeneXpert platform, computer and scanner;
• a new 10-colour satellite instrument with the GeneXpert connected to an existing system – this costs from US$ 6495 for one module to US$ 69 525 for 16 modules; and
• converting an existing GeneXpert system from a six-colour to a 10-colour system by replacing modules – a 10-colour module kit costs US$ 3860.

Additional cost considerations for Xpert MTB/XDR include additional testing or repeated testing in the case of indeterminate or non-actionable results (indeterminate, non-determinate or invalid). The potential cost burden of this is likely to vary, depending on the proportion of indeterminate test results across settings and the associated re-testing protocols.

No studies that have directly assessed the cost–effectiveness of the Xpert MTB/XDR cartridge were identified. Although extrapolation from other platforms and testing approaches for costing may be appropriate, extrapolation of cost–effectiveness data from Xpert MTB/RIF (Ultra) or other NAATs is not advised because of differences in diagnostic accuracy, costs associated with XDR treatment, and the different testing and treatment cascade of care.

Several factors are likely to influence the cost–effectiveness of Xpert MTB/XDR; they include diagnostic accuracy, which may lead to more or fewer individuals being diagnosed compared with the standard of care (which in turn will vary, depending on the local standard of care). In addition to diagnostic accuracy associated with the test itself, the diagnostic algorithm and placement of the Xpert MTB/XDR test within the algorithm has important implications.

The novel Xpert MTB/XDR provides results in less than 90 minutes. Thus, introduction of this test is likely to result in faster time to a result for genotypic DST and could affect cost–effectiveness by improving the numbers of patients initiating treatment, reducing loss to follow-up and improving survival rates. Costs associated with XDR treatment are likely to be an important driver of cost and cost–effectiveness because previous work has shown that these costs are high compared to diagnostic and other treatment costs. As larger numbers of XDR-positive individuals requiring treatment are identified, total resources required to treat these individuals will increase.

In the absence of transmission modelling studies, there is no information on the long-term population level impact of introducing Xpert MTB/XDR. Nevertheless, the benefits of identifying more cases earlier could lead to a reduction in ongoing transmission and potential cost-savings over the long term. This requires thorough investigations through transmission modelling.

How large are the resource requirements (costs)?

No published studies provided direct evidence about the total resources required. Resource requirements will include the purchase of cartridges (US$ 19.80/cartridge), upgrading of existing platforms to 10-colour modules (an upgrade that will eventually be required for all Xpert platforms: US$ 3860 to >US$ 72 350) and operational and programmatic costs associated with implementing the novel diagnostic. Resource requirements for XDR treatment (e.g. drugs, hospital capacity and staff) are also likely to increase as the number of people diagnosed increases. Total costs will vary, depending on testing volume and prevalence of XDR in the population; also, the impact on the budget will depend on the current standard of care and associated resource use.

What is the certainty of the evidence of resource requirements (costs)?

Direct costs related to the purchase of cartridges and machinery are provided from the manufacturer; however, several important items related to resource use for implementing Xpert MTB/XDR have not been investigated (e.g. staff time, overhead and operational costs). Differences in resource use between Xpert MTB/XDR and existing approaches will vary across settings using different phenotypic and genotypic DST. There is important variability in costs of staff time and operational costs (e.g. testing volume) across settings.

Does the cost–effectiveness of the intervention favour the intervention or the comparison?

No cost–effectiveness studies using Xpert MTB/XDR were identified. Extrapolation of cost–effectiveness data from Xpert MTB/RIF or other NAATs is not advised because of differences in diagnostic accuracy, and costs associated with XDR treatment and the testing and treatment cascade of care.

More details on economic evidence synthesis and analysis are provided in Web Annex 4.9: Systematic literature review of economic evidence for NAATs to detect TB and DR-TB in adults and children.

User perspective

This section answers the following question about key informants’ views and perspectives on the use of low complexity automated NAATs:

• Is there important uncertainty about or variability in how much end-users value the main outcomes?
• What would be the impact on health equity?
• Is the intervention acceptable to key stakeholders?
• Is the intervention feasible to implement?

The synthesis and analysis of qualitative evidence on end-users’ perspectives are discussed above in the section “User perspective” for moderate complexity automated NAATs (p. 61–65).

Findings of the review and interviews

The main findings of the systematic review and interviews are given below. Where information is from the review, a level of confidence in the QES is given; where it is from interviews, this is indicated with ‘Interviews’.

Is there important uncertainty about or variability in how much end-users value the main outcomes?

• Patients in high burden TB settings value:

  –

  getting an accurate diagnosis and reaching diagnostic closure (finally knowing “what is wrong with me”);

  –

  avoiding diagnostic delays because they exacerbate existing financial hardships and emotional and physical suffering, and make patients feel guilty for infecting others (especially children);

  –

  having accessible facilities; and

  –

  reducing diagnosis-associated costs (e.g. travel, missing work) as important outcomes of the diagnostic.

  QES: moderate confidence

• Low complexity automated NAATs, when compared with existing tests or sputum microscopy, are appreciated by health care professionals because of:

  –

  the rapidity and accuracy of the results;

  –

  the confidence that a result generates to start treatment and motivate patients;

  –

  the diversity of sample types;

  –

  the ability to detect drug resistance earlier or at all, for as many drugs as possible (altering a clinician’s risk perception of drug resistance in children), and the consequence of avoiding costlier investigations or hospital stays.

  QES: high confidence

  –

  Compared with other available diagnostic methods, the cartridge has a quicker turnaround time for first- and second-line DST. Health care professionals value the faster turnaround time, the potential ability to reflex samples from the Xpert MTB/RIF to the Xpert MTB/XDR cartridge, and receiving information on multiple drugs and high-level or low-level resistance simultaneously, because it could enable quicker diagnosis and optimized treatment for patients.

  Interviews

• Laboratory technicians appreciate low complexity automated NAATs for the following reasons:

  –

  Overall, the tests improve laboratory work compared with sputum microscopy in terms of ease of use, ergonomics and biosafety.

  QES: high confidence

  –

  These tests require minimal user steps, and the GeneXpert platform is a familiar system that people feel comfortable running and interpreting.

  Interviews

• Laboratory managers appreciate that monitoring of laboratory work and training is easier than with sputum microscopy, and that use of low complexity automated NAATs eases staff retention because it increases staff satisfaction and is symbolic of progress within the TB world.
  QES: low confidence

What would be the impact on health equity?

The impact on health equity would be similar to that of moderate complexity automated NAATs (p. 63–64).

Is the intervention acceptable to key stakeholders?

The acceptability to key stakeholders is similar to that of moderate complexity automated NAATs (p. 64–65).

• The identified challenges in implementing the use of low complexity automated NAATs and accumulated delays at every step may compromise the added value and benefits identified by the users (e.g. avoiding delays, keeping costs low, accurate results, information on drug resistance and easing laboratory work), ultimately leading to use.
  QES: high confidence
• If these values are not met, it can be assumed that users are less likely to find low complexity automated NAATs acceptable.

Is the intervention feasible to implement?

• Low complexity automated NAATs may decrease the workload in the laboratory in terms of freeing up time for laboratory staff. However, based on experience with Xpert MTB/RIF (Ultra), the introduction of a new class of technologies may increase the workload of laboratory staff if added onto existing work without adjusting staffing arrangements or if the new technology does not replace existing diagnostic tests.
  QES: moderate confidence
• Low complexity automated NAATs require less user training than other DST methods (e.g. LPA and culture), making these tests more feasible to implement than methods with more user steps and those that require significant additional training.
  Interview study
  Implementation of new diagnostics must be accompanied by training for clinicians, to help them interpret results from new molecular tests and understand how this relates to the treatment of a patient. In the past, with the introduction of Xpert MTB/RIF (Ultra), this has been a challenge and has led to underuse.
  QES: high confidence and interview study
  Introduction of Xpert MTB/RIF (Ultra) has also led to overreliance on results of cartridge-based NAATs at the expense of clinical acumen.
  QES: moderate confidence
• Introduction of new diagnostics must also be accompanied by guidelines and algorithms that support clinicians and laboratories in communicating with each other; for example, these resources allow clinicians and laboratories to discuss discordant results, and interpret laboratory results in the context of drug availability, patient history and patient progress on a current drug regimen.
  Interviews
• An efficient sample transportation system, with sustainable funding mechanisms, is crucial for feasibility, especially if an algorithm requires multiple samples at different times from different collection points, as is the case when dealing with DR-TB. If mishandled during preparation, there is a risk that the sample may become contaminated and yield inconclusive results on molecular diagnostics. Participants cited good personnel skills, standardized operating procedures and significant laboratory infrastructure as essential in reducing sample contamination in their laboratory.
  Interviews
• The feasibility of low complexity automated NAATs is challenged if there is an accumulation of diagnostic delays or underuse (or both) at every step in the process, mainly because of health system factors:

  –

  non-adherence to testing algorithms, testing for TB or MDR-TB late in the process, empirical treatment, false negatives due to technology failure, large sample volumes and staff shortages, poor or delayed sample transport and sample quality, poor or delayed communication of results, delays in scheduling follow-up visits and recalling patients, and inconsistent recording of results;

  –

  lack of sufficient resources and maintenance (e.g. stock-outs; unreliable logistics; lack of funding, electricity, space, air conditioners and sputum containers; dusty environment; and delayed or absent local repair option);

  –

  inefficient or unclear workflows and patient flows (e.g. inefficient organizational processes, poor links between providers, and unclear follow-up mechanisms or information on where patients need to go); and

  –

  lack of data-driven and inclusive national implementation processes.

  QES: high confidence

• The feasibility of using low complexity automated NAATs is also challenged by the value of diagnosing MTB over DR-TB at primary care. This situation makes the NAAT less feasible as a baseline test, although it would fit at a district or intermediate level laboratory.

Implementation considerations

Factors to consider when implementing low complexity automated NAATs for detection of resistance to isoniazid and second-line anti-TB agents are as follows:

• local epidemiological data on resistance prevalence should guide local testing algorithms, whereas pretest probability is important for the clinical interpretation of test results;
• the cost of a test varies depending on parameters such as the number of samples in a batch and the staff time required; therefore, a local costing exercise should be performed;
• low, moderate and high complexity tests have successive increase in technical competency needs (qualifications and skills) and staff time, which affects planning and budgeting;
• availability and timeliness of local support services and maintenance should be considered when selecting a provider;
• laboratory accreditation and compliance with a robust quality management system (including appropriate quality control) are essential for sustained service excellence and trust;
• training of both laboratory and clinical staff will ensure effective delivery of services and clinical impact;
• use of connectivity solutions for communication of results is encouraged, to improve efficiency of service delivery and time to treatment initiation;
• rapid and early testing for the detection of fluoroquinolone resistance is essential before starting treatment with the all-oral MDR/RR-TB shorter regimen (i.e. 6–9 months); this may also become relevant (depending on the epidemiological context) if new shorter drug-susceptible TB regimens that include fluoroquinolones are introduced;
• these tests can be used to rule in ethionamide resistance, but not to rule out resistance, because mutations conferring resistance to ethionamide are not limited to the inhA promoter region – they also include ethA, ethR and other genes;
• culture-based phenotypic DST may still be required, particularly among those with a high pretest probability of resistance when the low complexity automated NAATs does not detect drug resistance; in addition, culture-based phenotypic DST:

  –

  remains important to determine resistance to other anti-TB agents, particularly the new and repurposed medicines, and to monitor the emergence of additional drug resistance;

  –

  does not apply to ethionamide because it is unreliable and poorly reproducible;

• for second-line injectable drugs, the panel evaluated the performance in detecting resistance to amikacin only because both kanamycin and capreomycin are no longer recommended for the treatment of DR-TB; and
• culture-based phenotypic DST may be important to confirm amikacin susceptibility in situations where it is appropriate to use this medicine, to balance risk and benefit.

Research priorities

Research priorities for low complexity automated NAATs for detection of resistance to isoniazid and second-line anti-TB agents are as follows:

• diagnostic accuracy, in specific patient populations (e.g. children, people living with HIV, and patients with signs and symptoms of extrapulmonary TB) and in non-sputum samples;
• impact of diagnostic technologies on clinical decision-making and outcomes that are important to patients (e.g. cure, mortality, time to diagnosis and time to start treatment) in all patient populations;
• impact of specific mutations on treatment outcomes among people with DR-TB;
• use, integration and optimization of diagnostic technologies in the overall landscape of testing and care, as well as diagnostic pathways and algorithms;
• economic studies evaluating the costs, cost–effectiveness and cost–benefit of different diagnostic technologies;
• qualitative studies evaluating equity, acceptability, feasibility and end-user values of different diagnostic technologies;
• effect of non-actionable results (indeterminate, non-determinate or invalid) on diagnostic accuracy and outcomes that are important to patients;
• evaluation of low complexity automated NAATs for initial TB detection, in addition to its use as a follow-on test, in all people with signs and symptoms of TB, in children and in people living with HIV; and
• the potential utility of katG resistance detection to identify MDR-TB clones that may be missed because they do not have an RRDR mutation (e.g. the Eswatini MDR-TB clone, which has both the katG S315T and the non-RRDR rpoB I491F mutation).

Remarks

1. These recommendations apply to the use of LPAs for testing sputum smear-positive specimens (direct testing) and cultured isolates of MTBC (indirect testing) from both pulmonary and extrapulmonary sites.
2. LPAs are not recommended for the direct testing of sputum smear-negative specimens.
3. These recommendations apply to the detection of MTBC and the diagnosis of MDR-TB, but acknowledge that the accuracy of detecting resistance to rifampicin and isoniazid differs and, hence, that the accuracy of a diagnosis of MDR-TB is reduced overall.
4. These recommendations do not eliminate the need for conventional culture-based DST, which will be necessary to determine resistance to other anti-TB agents and to monitor the emergence of additional drug resistance.
5. Conventional culture-based DST for isoniazid may still be used to evaluate patients when the LPA result does not detect isoniazid resistance. This is particularly important for populations with a high pretest probability of resistance to isoniazid.
6. These recommendations apply to the use of LPA in children based on the generalization of data from adults.

Remarks

• These recommendations apply to the use of SL-LPA for testing sputum specimens (direct testing) and cultured isolates of M. tuberculosis (indirect testing) from both pulmonary and extrapulmonary sites. Direct testing on sputum specimens allows for the earlier initiation of appropriate treatment.
• These recommendations apply to the direct testing of sputum specimens from MDR/RR-TB, irrespective of the smear status, while acknowledging that the indeterminate rate is higher when testing smear-negative sputum specimens than with smear-positive sputum specimens.
• These recommendations do not eliminate the need for conventional phenotypic DST capacity, which will be necessary to confirm resistance to other drugs and to monitor the emergence of additional drug resistance.
• Conventional phenotypic DST can still be used in the evaluation of patients with negative SL-LPA results, particularly in populations with a high pretest probability for resistance to fluoroquinolones or SLID (or both).
• These recommendations apply to the use of SL-LPA in children with confirmed MDR/RR-TB, based on the generalization of data from adults.
• Resistance-conferring mutations detected by SL-LPA are highly correlated with phenotypic resistance to ofloxacin and levofloxacin.
• Resistance-conferring mutations detected by SL-LPA are highly correlated with phenotypic resistance to SLID.
• Given the high specificity for detecting resistance to fluoroquinolones and SLID, the positive results of SL-LPA could be used to guide the implementation of appropriate infection control precautions.

PICO 1: Should high complexity reverse hybridization-based NAATs on sputum be used to diagnose pyrazinamide resistance in patients with microbiologically confirmed pulmonary TB, irrespective of resistance to rifampicin, as compared with culture-based phenotypic DST or composite reference standard?

Three studies with a total of 122 participants provided data for evaluation of these NAATs for detection of pyrazinamide resistance, including two studies (101 participants) with phenotypic culture-based reference standard and one study (21 participants) with genotypic reference standard. The number of studies and participants were considered insufficient to make a conclusion on a diagnostic accuracy of high complexity reverse hybridization-based NAATs on sputum.

PICO 2: Should high complexity reverse hybridization-based NAATs on isolates be used to diagnose pyrazinamide resistance in patients with microbiologically confirmed pulmonary TB, irrespective of resistance to rifampicin, as compared with culture-based phenotypic DST?

Seven studies with a total of 964 participants provided data for evaluation of these NAATs for detection of pyrazinamide resistance compared with a phenotypic culture-based reference standard (Fig. 2.3.8).

The studies suffered from selection bias because they selected isolates with a wide range of different pncA mutations rather than a representative sample from a population. Thus, the evidence was downgraded by one level for risk of bias. The included studies did not directly address the review question; hence, the evidence was downgraded one level for indirectness. The Burhan trial and the Rienthong study are outliers for their sensitivities compared with the other studies; hence, the evidence was downgraded one level for inconsistency. Taking these judgements together, the quality (certainty) of evidence was rated very low for sensitivity and low for specificity.

Fig. 2.3.8

Forest plot of included studies for pyrazinamide resistance detection, irrespective of rifampicin resistance with culture-based phenotypic DST as the reference standard. CI: confidence interval; DST: drug susceptibility testing; FN: false negative; FP: (more...)

The overall sensitivity for pyrazinamide resistance in these seven studies ranged from 36% to 100% and the specificity from 96% to 100%. The pooled sensitivity was 81.2% (95% CI: 75.4–85.8%) and specificity was 97.8% (95% CI: 96.5–98.6%).

More details on diagnostic accuracy of the high complexity reverse hybridization-based NAATs, including comparison with genotypic and composite reference standards are available in Web Annex 4.17: High complexity reverse hybridization-based NAATs: diagnostic accuracy for detection of resistance to pyrazinamide. A systematic review.

Diagnostic accuracy and health benefits

Two health questions were designed using the PICO approach, to form the basis for the evidence search, retrieval and analysis.

1. Should targeted NGS as the initial test be used to diagnose drug resistance in individuals with bacteriologically confirmed pulmonary TB disease? This question applies to:

   –

   rifampicin, using a composite reference standard of phenotypic DST and whole genome sequencing (WGS), and Xpert MTB/RIF^® or Xpert Ultra^®;

   –

   isoniazid, using phenotypic DST as the reference standard;

   –

   levofloxacin, using phenotypic DST as the reference standard;

   –

   moxifloxacin, using phenotypic DST as the reference standard;

   –

   pyrazinamide, using a composite reference standard of phenotypic DST and WGS; and

   –

   ethambutol, using a composite reference standard of phenotypic DST and WGS.

2. Should targeted NGS be used to diagnose drug resistance in individuals with bacteriologically confirmed rifampicin-resistant pulmonary TB disease? This question applies to:

   –

   isoniazid, using phenotypic DST as the reference standard;

   –

   levofloxacin, using phenotypic DST as the reference standard;

   –

   moxifloxacin, using phenotypic DST as the reference standard;

   –

   pyrazinamide, using a composite reference standard of phenotypic DST and WGS;

   –

   bedaquiline, using phenotypic DST as the reference standard;

   –

   linezolid, using phenotypic DST as the reference standard;

   –

   clofazimine, using phenotypic DST as the reference standard;

   –

   amikacin, using phenotypic DST as the reference standard;

   –

   ethambutol, using a composite reference standard of phenotypic DST and WGS; and

   –

   streptomycin, using phenotypic DST as the reference standard.

A broad search was conducted to find, appraise and synthesize evidence about health benefits and the diagnostic test accuracy of targeted NGS compared with phenotypic drug sensitivity testing for patients with bacteriologically confirmed TB or with bacteriologically confirmed rifampicin-resistant pulmonary TB disease. A comprehensive search of three databases (Medline, Ovid Embase and Scopus) for relevant citations was performed. No date restriction was applied and the search was initially performed on 7 September 2022 and repeated on 17 January 2023. In addition, WHO made a public call for data and contacted well-known experts in the field to ask whether they had, or knew of, unpublished data that could contribute.

No data were found for the impact of targeted NGS on patient-level health effects. For the analysis of diagnostic accuracy, because few data were available in the literature, all data identified from the literature were included after correspondence with the authors. Hence, no manual data extraction from publications was required. A post-hoc decision was made to perform only an individual patient data (IPD) meta-analysis; thus, any study that could not provide IPD was excluded. Two report authors made independent assessments of methodological quality using QUADAS-2. Disagreements were resolved by discussion and uncertainties or disagreements were reviewed by an independent third party.

Subanalyses were performed to assess the diagnostic test accuracy in PLHIV and for semiquantitative results (derived from cycle thresholds) from Xpert MTB/RIF^® or Xpert Ultra^®, where “very low” or “low” concentrations of M. tuberculosis were compared with “medium” or “high” concentrations. The very low or low semiquantitative categories represent paucibacillary disease states, such as those frequently observed in paediatric TB.

Data were included from both published and unpublished prospective, observational clinical studies of targeted NGS platform diagnostic accuracy. All studies where targeted NGS had been performed directly from processed clinical samples were included, whereas those performed exclusively on cultured isolates were excluded. All studies were required to have comparator phenotypic DST data as a reference; in the cases of rifampicin, ethambutol and pyrazinamide, studies were required to also have WGS, to allow a composite reference to be generated. Rifampicin resistance results and semiquantitative results from Xpert MTB/RIF^® or Xpert Ultra^® were requested from all studies.

Given that this was a review of the diagnostic accuracy of a class of diagnostic platforms, all the data from each platform alone were analysed to assess which to include in an analysis to inform a class recommendation. Where the performance of any one platform appeared to be an outlier for sensitivity or specificity, that platform was excluded from subsequent meta-analyses. A platform was considered to be an outlier for a particular drug if the point estimate for sensitivity was more than 10 percentage points worse than the best performing platform, or where the point estimate for specificity was more than 5 percentage points worse.

An IPD meta-analysis was performed instead of a classical meta-analysis, because the studies identified in the literature were generally too small to contribute to a classical meta-analysis, particularly for the new and repurposed drugs. In addition, this type of approach allowed for relevant co-variables to be included in the model; it could also control for repeated testing on the same samples using different platforms, which was the case for much of the available data.

For each dependent variable, a multivariable model included a number of co-variables as fixed effects. These included rifampicin resistance as determined by Xpert MTB/RIF^® or Xpert Ultra^® for all drugs other than rifampicin; semiquantitative cycle threshold (CT) value from Xpert MTB/RIF^® or Xpert Ultra^®; and a co-variable to indicate which samples featured in duplicate, meaning that some samples were sequenced on two different platforms and thus were represented twice in the analysis. For models looking specifically at diagnostic test accuracy in PLHIV, the HIV test result was included as a co-variable. Finally, the study site was included as a random effect. The models were run in Stata (version 17) using the melogit command, and the outputs were transformed using the margins command. Models were run for all PICO questions for sensitivity and specificity.

The certainty of the evidence of the pooled studies was assessed systematically for each of the PICO questions using the GRADE approach, which produces an overall quality assessment (or certainty) of evidence and has a framework for translating evidence into recommendations.

The GRADEpro Guideline Development Tool software (20) was used to generate summary of findings tables for the sensitivity and specificity of each drug. The numbers of samples classified as true, false positive or negative were then calculated across a range of three prevalences of drug resistance, chosen to be representative of different global settings. The quality of evidence was rated as high (not downgraded), moderate (downgraded one level), low (downgraded two levels) or very low (downgraded more than two levels), based on five factors: risk of bias, indirectness, inconsistency, imprecision and other considerations. The quality (certainty) of evidence was downgraded by one level when a serious issue was identified and by two levels when a very serious issue was identified in any of the factors used to judge the quality of evidence.

The data sources for the IPD data analysis are shown in Fig. 2.3.9. The analysis included data from published studies, a large multicountry trial conducted by FIND, and several other studies across multiple countries. Most of the studies only evaluated the Deeplex assay, while the FIND trial evaluated both the Deeplex and the AmPORE-TB. Only one study evaluated TBseq. For each drug, one or two platforms were dropped from the analysis based on the overall number of resistant or susceptible samples available for that platform and drug, or because the accuracy of the platform did not meet the diagnostic test accuracy criteria for inclusion when compared with the best performing platform.

Fig. 2.3.9

Studies included in the IPD meta-analysis for targeted NGS. ERJ: European Respiratory Journal; FIND: Foundation for Innovative New Diagnostics; IPD: individual patient data; NGS: next-generation sequencing; NICD: National Institute for Communicable Diseases; (more...)

Data synthesis was structured around the two preset PICO questions, as outlined below.

PICO 1: Should targeted NGS as the initial test be used to diagnose drug resistance in patients with bacteriologically confirmed pulmonary TB disease?

The available evidence included in the final pooled analysis varied by drug, from 12 studies with 1440 participants for the sensitivity of isoniazid to three studies with 269 participants for the specificity of pyrazinamide (Table 2.3.5). The pooled estimates were determined using a multivariable, mixed-effects model. All drugs were downgraded by one level for indirectness for sensitivity and specificity, because all studies were enriched for rifampicin resistance, leading to applicability concerns. In addition, for rifampicin, levofloxacin and pyrazinamide, specificity was downgraded a further level for imprecision; however, for ethambutol, it was downgraded for risk of bias because different samples were used for the index and reference tests. The overall certainty of the evidence for test accuracy ranged from moderate to very low.

The test performance was determined to be accurate for all drugs included in the assessment, with a pooled sensitivity of at least 95% for isoniazid, moxifloxacin and ethambutol, more than 93% for rifampicin and levofloxacin, and 88% for pyrazinamide. The pooled specificity was at least 96% for all drugs.

The reference standard was culture-based phenotypic DST for isoniazid, levofloxacin and moxifloxacin, and a combination of phenotypic DST and WGS for rifampicin, pyrazinamide and ethambutol. The percentage of tests with indeterminate results ranged from 9% (levofloxacin and moxifloxacin) to 18% (pyrazinamide), with higher indeterminate rates in samples with lower bacterial load (semiquantitative category low or very low).

Table 2.3.5

The accuracy and certainty of evidence of targeted NGS for the detection of resistance to anti-TB drugs among bacteriologically confirmed pulmonary TB.

There were no data on the impact of targeted NGS on patient outcomes such as time to treatment or treatment outcome.

PICO 2: Should targeted NGS be used to diagnose drug resistance in patients with bacteriologically confirmed rifampicin-resistant pulmonary TB disease?

The available evidence varied by drug, from 12 studies with 1440 participants for sensitivity of isoniazid to three studies with 31 participants for sensitivity of bedaquiline (Table 2.3.6). The pooled estimates were determined using a multivariable, mixed-effects model.

The overall certainty was high for some of the drugs. Levofloxacin was downgraded one level for inconsistency. Bedaquiline and linezolid were downgraded by two levels for imprecision in sensitivity because the number of resistant samples was below the threshold set and the confidence intervals were wide. Clofazimine was also downgraded by two levels, one for inconsistency (because two studies were outliers) and another level for imprecision (because the confidence intervals were wide). Amikacin was downgraded by one level for sensitivity and specificity because critical concentrations outside those recommended by WHO were used for a large proportion of samples. Amikacin sensitivity was further downgraded by two more levels, one for inconsistency and the other for imprecision. Ethambutol was downgraded by one level for risk of bias because different samples were used for the index and reference tests. Streptomycin specificity was downgraded by two levels, one for inconsistency and the other for imprecision. The overall certainty of the evidence for test accuracy ranged from high to very low.

The test performance among people with RR-TB was determined to be accurate for isoniazid, levofloxacin, moxifloxacin, ethambutol and streptomycin (pooled sensitivity ≥95%) and acceptable for pyrazinamide (90%), bedaquiline (68%), linezolid (69%), clofazimine (70%) and amikacin (87%). The pooled specificity was 95% or greater for all drugs except streptomycin (75%). The reference standard was culture-based phenotypic DST for all drugs except for ethambutol and pyrazinamide, where a combination of phenotypic DST and WGS was used. The percentage of tests with indeterminate results ranged from 9% (levofloxacin and moxifloxacin) to 21% (ethambutol); indeterminate rates were higher in samples with a lower bacterial load (semiquantitative category low or very low).

Table 2.3.6

The accuracy and certainty of evidence of targeted NGS for the detection of resistance to anti-TB drugs among bacteriologically confirmed rifampicin-resistant pulmonary TB.

There were no data on the impact of targeted NGS on patient outcomes such as time to treatment or treatment outcome.

Three web annexes give additional information, as follows:

• details of studies included in the current analysis (Web Annex 4.18: Review of the diagnostic accuracy of targeted NGS technologies for detection of drug resistance among people diagnosed with TB);
• a summary of the results and details of the evidence quality assessment (Web Annex 2.10: GRADE profiles of targeted next-generation sequencing for detection of TB drug resistance); and
• a summary of the GDG panel judgements (Web Annex 3.10: Evidence to decision tables: targeted next-generation sequencing for detection of TB drug resistance).

Cost–effectiveness analysis

The cost and cost–effectiveness data for targeted NGS were assessed through a systematic review of the published literature and a generalized model-based cost–effectiveness analysis commissioned by WHO.

The systematic review on the cost and cost–effectiveness of using either targeted NGS or WGS to diagnose DR-TB searched three databases: PubMed, Embase and Scopus. The search was run on 30 October 2022 and had no time restriction. All costing data were inflated to 2021 US dollars. Findings were synthesized descriptively, given the considerable degree of heterogeneity in study methodology and outcomes. Among the studies included in the systematic review, three were on targeted NGS only, three were on targeted NGS and WGS, and four were on WGS only. For targeted NGS based on a single study (n=1), the cost per sample was between US$ 69.64 for Illumina MiSeq on 24 samples, and US$ 73.47 for Nanopore MinION on 12 samples; however, this costing was limited to only some components and did not include human resource costs or overhead costs. For WGS (n=5), cost per sample ranged from US$ 63.00 on Nanopore MinION to US$ 277.00 on Illumina MiSeq; given that studies used an inconsistent number of component costs, comparisons were challenging. Based on the review, the most significant cost component was the sequencing step, and the largest component costs were reagents and consumables, including those necessary for sequencing, sample processing and targeted NGS steps library preparation. Study authors identified four major cost drivers: use of different sequencers, depth and breadth of coverage, inefficiencies in initial sample runs, and economies of scale via batching or cross-batching.

The cost data from the systematic review were limited; therefore, an empirical unit costing was performed, in consultation with manufacturers and FIND. At the time of this work, only pricing for Deeplex Myc-TB was available and it was used for estimation of cost for the class. Unit costs included consumables, equipment, staffing and overheads (where available); also, costs assumed targeted NGS testing for all drugs. Based on the empirical analysis, the cost of targeted NGS was estimated to be:

• US$ 134 to US$ 257 in South Africa;
• US$ 120 to US$ 198 in Georgia; and
• US$ 121 to US$ 175 in India.

These costs are dependent on patient volume, batching and negotiated cost per targeted NGS kit.

Recognizing the lack of economic evidence on this topic, a hypothetical cost–effectiveness modelling study was undertaken to assess the cost–effectiveness (Objective 1) and affordability (Objective 2) of these tests for the diagnosis of DR-TB in various high TB burden settings.

Objective 1: To assess the potential cost–effectiveness of introducing the targeted NGS technology for the diagnosis of DR-TB in Georgia, India and South Africa.

This assessment included modelling the cost–effectiveness of targeted NGS in three separate scenarios with distinct comparison options:

• Cost–effectiveness of targeted NGS for DST among individuals with RR-TB after a rapid molecular test for rifampicin resistance as a replacement for phenotypic DST (PICO 2).
• Cost–effectiveness of targeted NGS for DST among individuals with RR-TB after a rapid molecular test for rifampicin resistance as a replacement for current in-country DST practice (PICO 2).
• Cost–effectiveness of targeted NGS as the initial test for TB drug resistance in patients with bacteriologically confirmed TB compared with rapid molecular testing for drug resistance and phenotypic DST in a high DR-TB burden setting (PICO 1).

In the first scenario, targeted NGS was compared with universal phenotypic DST; in the second scenario, targeted NGS was compared with current in-country phenotypic DST practice among individuals with detected rifampicin resistance (PICO 2). This was done across three countries: Georgia, India and South Africa. Current DST practice in Georgia and South Africa includes Xpert XDR^® followed by phenotypic DST; in India it includes LPAs and phenotypic DST done in parallel. A final scenario included targeted NGS compared to rapid molecular testing for drug resistance and phenotypic DST as initial tests for TB drug resistance among all TB patients (PICO 1) but was modelled for only one setting, Georgia – a high DR-TB burden setting. Epidemiological data were sourced from published literature; targeted NGS diagnostic accuracy data were sourced from the systematic review and IDP analysis conducted for this guideline. Economic data were sourced from published literature and a systematic and scoping review done in parallel by our team and supplemented with empirical data collection.

A decision analysis modelling approach was used to estimate the incremental cost–effectiveness of using targeted NGS for the diagnosis of DR-TB compared with various existing DST scenarios. This was done from the perspective of the health care system and accounts only for the health care system costs required to diagnose and treat TB. The estimation did not account for societal costs, or any direct or indirect costs incurred by patients. In addition, costs for sample transportation were not included in this analysis. The primary outcome was the incremental cost–effectiveness ratio (ICER), which was calculated as the incremental cost in US dollars per disability-adjusted life year (DALY) averted.

Main findings for PICO 1: Using targeted NGS as an initial test

Using targeted NGS as an initial test for DST in the high DR-TB burden setting of Georgia led to more health gains (DALYs=0.49) compared with Xpert MTB/RIF or Xpert Ultra, followed by phenotypic DST (DALY=0.51). The ICER per DALY averted was US$ 9261 (95% uncertainty range [UR]: US$ 5258–32 040/DALY averted), which was considered cost effective at a willingness-to-pay (WTP) threshold of three times the country GDP per capita (US$ 15 609), with 80% of simulated iterations falling below the WTP threshold.

Main findings for PICO 2: Using targeted NGS among those with RR-TB

Using targeted NGS as a replacement for universal phenotypic DST among RR-TB patients, targeted NGS was dominated by phenotypic DST, with targeted NGS having higher costs and leading to fewer health gains. This finding was driven by the high diagnostic accuracy of phenotypic DST (which was assumed to be universal in this scenario), and an assumption of no difference in loss to follow-up between targeted NGS and phenotypic DST. When in-country DST practice was used as the comparator (instead of universal phenotypic DST), targeted NGS led to more health gains than in-country DST across all three countries. Targeted NGS was cost effective in South Africa (ICER: US$ 15 619/DALY averted, 95% UR: cost saving −US$ 114 782, at a WTP threshold of US$ 21 165), but was not cost effective in Georgia (ICER: US$ 18,375/DALY averted, UR: cost saving −US$ 158 972/DALY averted, at a WTP threshold of US$ 15 065). In India, where LPA, liquid culture and DST are being used as part of in-country DST, targeted NGS dominated the country’s current DST practice, with lower costs and more health gains (95% UR: cost saving −US$ 60 083).

Main findings: scenario analyses

Several key scenario analyses were investigated. In the base case approach, loss to follow-up was assumed to be equivalent between phenotypic DST and targeted NGS; in a scenario where there was no loss to follow-up in targeted NGS compared with 10% in phenotypic DST, targeted NGS was cost effective in South Africa (ICER: US$ 13 004/DALY averted, WTP: US$ 21 165) and Georgia (ICER: US$ 13 640/DALY averted, WTP: US$ 15 069) and targeted NGS still dominated in-country DST practice in India. In scenarios where sequencing platforms are used for multiple different diseases to reduce the unit test cost of targeted NGS, the cost–effectiveness of targeted NGS improves in all three countries. A batching scenario was investigated, with an assumed 20% fewer samples per targeted NGS run, and led to an increased unit test cost for targeted NGS; in this scenario, the targeted NGS approach retained cost–effectiveness only in South Africa. When a 50% price reduction in targeted NGS test kit cost was assumed, targeted NGS cost–effectiveness further improved in all countries.

Objective 2: To assess the financial impact of introducing targeted NGS as a replacement for existing DST for diagnosis of DR-TB among TB patients across three countries: Georgia, India and South Africa.

A budget impact assessment was undertaken to estimate the financial consequences of adopting targeted NGS for DST for all patients diagnosed with TB, and replacing in-country DST practice in Georgia (PICO 1). The analysis suggested that implementing targeted NGS for all patients diagnosed with TB would be more expensive than testing all patients with Xpert MTB/RIF or Xpert Ultra, followed by phenotypic DST (see Fig. 2.3.10).

Fig. 2.3.10

Budget impact assessment results comparing current standard practice for DST with implementation of targeted NGS for all patients diagnosed with TB in Georgia. DST: drug susceptibility testing; NGS: next-generation sequencing; pDST: phenotypic DST; TB: (more...)

A budget impact assessment was undertaken to estimate the financial consequences of adopting targeted NGS for DST after a rapid molecular test for rifampicin resistance, and replacing in-country DST practice in Georgia, India and South Africa (PICO 2). In-country DST practice included Xpert XDR combined with phenotypic DST in Georgia and South Africa, and Xpert XDR combined with LPA in Georgia over a 1-year and 5-year period. It was assumed that the eligible RR-TB patient populations requiring DST were 58 837, 8200 and 187 in South Africa, India and Georgia, respectively, and that the TB reduction rate over the 5 years was stable (2). To estimate the impact on the country-specific budget, the economic costs generated by the model were multiplied by the number of patients.

Results from a 1-year budget impact assessment for PICO 2 are presented in Fig. 2.3.11. In India, it was estimated that implementing targeted NGS would cost about US$ 57 130 727 – slightly lower than the current practice of LPA combined with phenotypic DST, which has a cost of US$ 57 719 097. In South Africa, it was estimated that implementing targeted NGS would result in a rise in budget to about US$ 27 888 200, slightly more than LPA combined with phenotypic DST, which has a cost of US$ 26 428 600. Finally in Georgia, where there are fewer bacteriologically confirmed patients, it was estimated that implementing targeted NGS would cost about US$ 592 221, slightly more than LPA combined with phenotypic DST, which has a cost of US$ 568 480.

Fig. 2.3.11

Budget impact assessment results comparing current standard practice for DST to implementing targeted NGS for patients with RR-TB in India, South Africa and Georgia. DST: drug susceptibility testing; LPA: line probe assay; NGS: next-generation sequencing; (more...)

User perspective

A rapid review was commissioned to identify and synthesize qualitative evidence on the use of targeted NGS for the detection of TB drug resistance; in particular, the aim was to examine the implementation considerations related to acceptability, feasibility, and values, preferences and equity. The review searched Medline with no year or language limits. The search was run on 19 August 2022, and then rerun on 10 October 2022 to include WGS-related studies for the detection of TB drug resistance. The review did not identify any eligible studies for analysis and synthesis. Based on the systematic search, three records were identified; in addition, based on the open, hand and expert searches, 27 records were found. On full-text review of the 30 records, none were found to be eligible for inclusion. Given that no direct evidence was found, note was made of a Cochrane qualitative evidence synthesis published in 2022 that examined recipient and provider perspectives on rapid molecular tests for TB and drug resistance (52); that study provides relevant (though indirect) evidence on the subject. The authors noted that people with TB valued reaching diagnostic closure with an accurate diagnosis, avoiding diagnostic delays and keeping diagnostic-associated costs low, whereas health care providers valued aspects of accuracy and the resulting confidence in low-complexity NAAT results, rapid turnaround times and low costs to people seeking a diagnosis.

To address the direct evidence gap, WHO commissioned an additional qualitative cross-sectional study comprising semi-structured interviews, primarily with laboratory staff and management personnel directly involved with implementing targeted NGS in the three FIND trial sites, as well as with three global experts involved in TB care and diagnostics. In total, there were 17 respondents, and the work was conducted during September to October 2022. The objective was to explore the perceptions and experiences of those implementing targeted NGS technology, with respect to acceptability, feasibility, and values, preferences and equity. The main findings are summarized below.