key: cord-0755533-6q7bmipu
authors: Nathavitharana, Ruvandhi R; Lederer, Philip; Chaplin, Marty; Bjerrum, Stephanie; Steingart, Karen R; Shah, Maunank
title: Impact of diagnostic strategies for tuberculosis using lateral flow urine lipoarabinomannan assay in people living with HIV
date: 2021-08-20
journal: Cochrane Database Syst Rev
DOI: 10.1002/14651858.cd014641
sha: 5a6fae16dcca321013e23443883e78d1e85e1c00
doc_id: 755533
cord_uid: 6q7bmipu

BACKGROUND: Tuberculosis is the primary cause of hospital admission in people living with HIV, and the likelihood of death in the hospital is unacceptably high. The Alere Determine TB LAM Ag test (AlereLAM) is a point‐of‐care test and the only lateral flow lipoarabinomannan assay (LF‐LAM) assay currently commercially available and recommended by the World Health Organization (WHO). A 2019 Cochrane Review summarised the diagnostic accuracy of LF‐LAM for tuberculosis in people living with HIV. This systematic review assesses the impact of the use of LF‐LAM (AlereLAM) on mortality and other patient‐important outcomes. OBJECTIVES: To assess the impact of the use of LF‐LAM (AlereLAM) on mortality in adults living with HIV in inpatient and outpatient settings. To assess the impact of the use of LF‐LAM (AlereLAM) on other patient‐important outcomes in adults living with HIV, including time to diagnosis of tuberculosis, and time to initiation of tuberculosis treatment. SEARCH METHODS: We searched the Cochrane Infectious Diseases Group Specialized Register; the Cochrane Central Register of Controlled Trials (CENTRAL); MEDLINE (PubMed); Embase (Ovid); Science Citation Index Expanded (Web of Science), BIOSIS Previews, Scopus, LILACS; ProQuest Dissertations and Theses; ClinicalTrials.gov; and the WHO ICTRP up to 12 March 2021. SELECTION CRITERIA: Randomized controlled trials that compared a diagnostic intervention including LF‐LAM with diagnostic strategies that used smear microscopy, mycobacterial culture, a nucleic acid amplification test such as Xpert MTB/RIF, or a combination of these tests. We included adults (≥ 15 years) living with HIV. DATA COLLECTION AND ANALYSIS: Two review authors independently assessed trials for eligibility, extracted data, and analysed risk of bias using the Cochrane tool for assessing risk of bias in randomized studies. We contacted study authors for clarification as needed. We used risk ratio (RR) with 95% confidence intervals (CI). We used a fixed‐effect model except in the presence of clinical or statistical heterogeneity, in which case we used a random‐effects model. We assessed the certainty of the evidence using GRADE. MAIN RESULTS: We included three trials, two in inpatient settings and one in outpatient settings. All trials were conducted in sub‐Saharan Africa and assessed the impact of diagnostic strategies that included LF‐LAM on mortality when the test was used in conjunction with other tuberculosis diagnostic tests or clinical assessment for clinical decision‐making in adults living with HIV. Inpatient settings In inpatient settings, the use of LF‐LAM testing as part of a tuberculosis diagnostic strategy likely reduces mortality in people living with HIV at eight weeks compared to routine tuberculosis diagnostic testing without LF‐LAM (pooled RR 0.85, 95% CI 0.76 to 0.94; 5102 participants, 2 trials; moderate‐certainty evidence). That is, people living with HIV who received LF‐LAM had 15% lower risk of mortality. The absolute effect was 34 fewer deaths per 1000 (from 14 fewer to 55 fewer). In inpatient settings, the use of LF‐LAM testing as part of a tuberculosis diagnostic strategy probably results in a slight increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF‐LAM (pooled RR 1.26, 95% CI 0.94 to 1.69; 5102 participants, 2 trials; moderate‐certainty evidence). Outpatient settings In outpatient settings, the use of LF‐LAM testing as part of a tuberculosis diagnostic strategy may reduce mortality in people living with HIV at six months compared to routine tuberculosis diagnostic testing without LF‐LAM (RR 0.89, 95% CI 0.71 to 1.11; 2972 participants, 1 trial; low‐certainty evidence). Although this trial did not detect a difference in mortality, the direction of effect was towards a mortality reduction, and the effect size was similar to that in inpatient settings. In outpatient settings, the use of LF‐LAM testing as part of a tuberculosis diagnostic strategy may result in a large increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF‐LAM (RR 5.44, 95% CI 4.70 to 6.29, 3022 participants, 1 trial; low‐certainty evidence). Other patient‐important outcomes Assessment of other patient‐important and implementation outcomes in the trials varied. The included trials demonstrated that a higher proportion of people living with HIV were able to produce urine compared to sputum for tuberculosis diagnostic testing; a higher proportion of people living with HIV were diagnosed with tuberculosis in the group that received LF‐LAM; and the incremental diagnostic yield was higher for LF‐LAM than for urine or sputum Xpert MTB/RIF. AUTHORS' CONCLUSIONS: In inpatient settings, the use of LF‐LAM as part of a tuberculosis diagnostic testing strategy likely reduces mortality and probably results in a slight increase in tuberculosis treatment initiation in people living with HIV. The reduction in mortality may be due to earlier diagnosis, which facilitates prompt treatment initiation. In outpatient settings, the use of LF‐LAM testing as part of a tuberculosis diagnostic strategy may reduce mortality and may result in a large increase in tuberculosis treatment initiation in people living with HIV. Our results support the implementation of LF‐LAM to be used in conjunction with other WHO‐recommended tuberculosis diagnostic tests to assist in the rapid diagnosis of tuberculosis in people living with HIV.

To assess the impact of the use of LF-LAM (AlereLAM) on mortality in adults living with HIV in inpatient and outpatient settings.

To assess the impact of the use of LF-LAM (AlereLAM) on other patient-important outcomes in adults living with HIV, including time to diagnosis of tuberculosis, and time to initiation of tuberculosis treatment.

We searched the Cochrane Infectious Diseases Group Specialized Register; the Cochrane Central Register of Controlled Trials (CENTRAL); MEDLINE (PubMed); Embase (Ovid); Science Citation Index Expanded (Web of Science), BIOSIS Previews, Scopus, LILACS; ProQuest Dissertations and Theses; ClinicalTrials.gov; and the WHO ICTRP up to 12 March 2021.

Randomized controlled trials that compared a diagnostic intervention including LF-LAM with diagnostic strategies that used smear microscopy, mycobacterial culture, a nucleic acid amplification test such as Xpert MTB/RIF, or a combination of these tests. We included adults (≥ 15 years) living with HIV.

Trusted evidence. Informed decisions. Better health.

Two review authors independently assessed trials for eligibility, extracted data, and analysed risk of bias using the Cochrane tool for assessing risk of bias in randomized studies. We contacted study authors for clarification as needed. We used risk ratio (RR) with 95% confidence intervals (CI). We used a fixed-e ect model except in the presence of clinical or statistical heterogeneity, in which case we used a random-e ects model. We assessed the certainty of the evidence using GRADE.

We included three trials, two in inpatient settings and one in outpatient settings. All trials were conducted in sub-Saharan Africa and assessed the impact of diagnostic strategies that included LF-LAM on mortality when the test was used in conjunction with other tuberculosis diagnostic tests or clinical assessment for clinical decision-making in adults living with HIV.

In inpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy likely reduces mortality in people living with HIV at eight weeks compared to routine tuberculosis diagnostic testing without LF-LAM (pooled RR 0.85, 95% CI 0.76 to 0.94; 5102 participants, 2 trials; moderate-certainty evidence). That is, people living with HIV who received LF-LAM had 15% lower risk of mortality. The absolute e ect was 34 fewer deaths per 1000 (from 14 fewer to 55 fewer).

In inpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy probably results in a slight increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF-LAM (pooled RR 1.26, 95% CI 0.94 to 1.69; 5102 participants, 2 trials; moderate-certainty evidence).

In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may reduce mortality in people living with HIV at six months compared to routine tuberculosis diagnostic testing without LF-LAM (RR 0.89, 95% CI 0.71 to 1.11; 2972 participants, 1 trial; low-certainty evidence). Although this trial did not detect a di erence in mortality, the direction of e ect was towards a mortality reduction, and the e ect size was similar to that in inpatient settings.

In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may result in a large increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF-LAM (RR 5.44, 95% CI 4.70 to 6.29, 3022 participants, 1 trial; low-certainty evidence).

Assessment of other patient-important and implementation outcomes in the trials varied. The included trials demonstrated that a higher proportion of people living with HIV were able to produce urine compared to sputum for tuberculosis diagnostic testing; a higher proportion of people living with HIV were diagnosed with tuberculosis in the group that received LF-LAM; and the incremental diagnostic yield was higher for LF-LAM than for urine or sputum Xpert MTB/RIF.

In inpatient settings, the use of LF-LAM as part of a tuberculosis diagnostic testing strategy likely reduces mortality and probably results in a slight increase in tuberculosis treatment initiation in people living with HIV. The reduction in mortality may be due to earlier diagnosis, which facilitates prompt treatment initiation. In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may reduce mortality and may result in a large increase in tuberculosis treatment initiation in people living with HIV. Our results support the implementation of LF-LAM to be used in conjunction with other WHO-recommended tuberculosis diagnostic tests to assist in the rapid diagnosis of tuberculosis in people living with HIV.

Cochrane Database of Systematic Reviews

In inpatient settings, the use of LF-LAM as part of a tuberculosis diagnostic testing strategy likely reduces deaths and probably results in a slight increase in tuberculosis treatment initiation in people living with HIV.

In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may reduce deaths and may result in a large increase in tuberculosis treatment initiation in people living with HIV.

What was studied in the review?

We searched for trials in adults (15 years and older) that evaluated the e ect of a tuberculosis diagnostic strategy that included the LF-LAM test compared to standard care using other WHO-recommended diagnostic tests in adults living with HIV.

What were the main results of the review?

We identified three trials, two in inpatient settings and one in outpatient settings.

In inpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy likely reduces mortality in people living with HIV at eight weeks compared to routine tuberculosis diagnostic testing without LF-LAM (2 trials, 5102 participants, moderate-certainty evidence).

In inpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy probably results in a slight increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF-LAM (2 trials, 5102 participants moderate-certainty evidence).

In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may reduce mortality in people living with HIV at six months compared to routine tuberculosis diagnostic testing without LF-LAM (1 trial, 2972 participants, low-certainty evidence).

In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may result in a large increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF-LAM (1 trial, 3022 participants, low-certainty evidence).

The included studies assessed other patient-important outcomes in di erent ways. The studies demonstrated that more people living with HIV were able to produce urine compared to sputum for tuberculosis diagnostic testing, and more people living with HIV were diagnosed with tuberculosis in the group that received LF-LAM.

We searched for relevant trials up to 12 March 2021. In inpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy probably results in a slight increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF-LAM.

*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

CI: confidence interval; LF-LAM: lateral flow urine lipoarabinomannan assay; RCT: randomized controlled trial; RR: risk ratio. 5 a We did not downgrade. In Gupta-Wright 2018, investigators, all study sta (other than the laboratory technician and statistician), hospital-attending clinical teams, and participants were masked to the study group allocation. In Peter 2016, neither participants nor research nurses were masked to either allocation or test results. However, we doubt that the test results were biased in light of this. b We downgraded one level for indirectness. The two trials were conducted in African countries. We do not have direct evidence of the applicability of the findings to other settings outside of Africa, although we note that Peter 2016 took place at 10 sites across four countries, and Gupta-Wright 2018 took place in two sites across two countries. In Gupta-Wright 2018, the test was conducted in the laboratory, not at the point of care. In addition, the intervention in Gupta-Wright 2018 was a combination of urine LAM and urine Xpert. In Peter 2016, the intervention was urine LAM plus a "nurse-informed" treatment decision. These additional considerations may not reflect how the test would be performed in routine practice. In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may result in a large increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF-LAM.

Trusted evidence. Informed decisions. Better health.

Cochrane Database of Systematic Reviews

The same or similar text appears in the Background section in several other Cochrane protocols and reviews on tests for diagnosing tuberculosis (Bjerrum 2019; Kay 2020; Kohli 2021; Shapiro 2021).

Tuberculosis (TB) is an airborne infection caused by the bacterium Mycobacterium tuberculosis. Although pulmonary tuberculosis (infection in the lungs) is the most common form of the disease, tuberculosis can a ect almost any other site in the body (extrapulmonary tuberculosis). Globally in 2019, 10 million people were estimated to become sick due to tuberculosis, of whom 1.4 million died (including 208,000 people living with HIV), making tuberculosis a leading cause of death due to an infectious disease in adults (WHO Global Tuberculosis Report 2020). Tuberculosis is the primary cause of hospital admission and death in people living with HIV, with cohort studies demonstrating an in-hospital attributable mortality of 25% in HIV-positive adults and 30% in HIVpositive children across regions (Ford 2016). A systematic review of health facility-based autopsy studies demonstrated that in-hospital mortality may be even higher, with tuberculosis accounting for approximately 40% of deaths in HIV-positive people (Gupta 2015). Furthermore, it is likely that current estimates of mortality that rely on verbal autopsies underestimate mortality due to HIV-associated tuberculosis (Karat 2017). Importantly, in almost half of those with tuberculosis, the disease remained undiagnosed at the time of death (Gupta 2015). Tuberculosis is a curable infection, and global policy emphasizes the importance of early diagnosis and initiation of e ective treatment to improve individual outcomes and decrease transmission (Barrera 2015; WHO Global Tuberculosis Report 2020). The World Health Organization (WHO) estimates that, from 2000 to 2019, more than 60 million lives were saved by diagnosing and treating tuberculosis. The COVID-19 pandemic threatens to reverse the gains made in recent years. A modelling study by the WHO suggests that there could be between 200,000 and 400,000 additional tuberculosis deaths in 2020 if, over a period of three months, 25% to 50% fewer people were detected and treated with tuberculosis (WHO Global Tuberculosis Report 2020).

In 2019, there was a substantial gap (2.9 million) between the 10 million estimated cases of tuberculosis and the number of people newly diagnosed and reported to national programmes (WHO Global Tuberculosis Report 2020). Even those who are diagnosed o en face extensive delays (Hanson 2017; Sreeramareddy 2014). The agreed-upon best reference standard for pulmonary tuberculosis is sputum culture (Lewinsohn 2017). However, culture is a relatively complex and slow procedure, and test results may not be available for two to eight weeks. Xpert MTB/RIF (Cepheid, Sunnyvale, USA), Xpert Ultra (Cepheid, Sunnyvale, USA), and Truenat (Molbio Diagnostics/Bigtec Labs, Goa/Bengaluru, India) are molecular diagnostic assays that use nucleic acid amplification to determine the presence of M tuberculosis and resistance to rifampicin (one of the core first-line antibiotics used to treat tuberculosis). They are recommended by the WHO as initial tests for any person being evaluated for tuberculosis (WHO Consolidated Guidelines (Module 3) 2020). Xpert MTB/RIF and Xpert Ultra can be used to test extrapulmonary specimens; however, for most non-respiratory specimens, the sensitivity is lower than for sputum specimens (Kohli 2021). Individuals with tuberculosis and HIV co-infection face particular challenges since they are o en unable to produce sputum specimens and are more likely than HIV-negative individuals to have extrapulmonary disease, which is harder to diagnose (Pai 2016; Shivakoti 2017). The WHO recommends that people living with HIV be systematically screened for active tuberculosis at each visit to a health facility (WHO Compendium of WHO Guidelines 2018). Typically, screening consists of evaluation for four symptoms of tuberculosis: cough, fever, night sweats, and weight loss, followed, for those who screen positive, by microbiological testing, which should include a WHO-recommended rapid diagnostic test in all individuals being evaluated for tuberculosis (WHO Consolidated Guidelines (Module 3) 2020). However, increasing data from prevalence surveys demonstrate high rates of tuberculosis in asymptomatic individuals (WHO Global Tuberculosis Report 2020), highlighting the gap in current symptom-based screening approaches, which is particularly pertinent for those at higher risk of disease progression, such as people living with HIV.

There is an urgent need for a rapid non-sputum-based diagnostic test for tuberculosis (WHO Target Product Profile 2014). Such a biomarker-based diagnostic test should ideally be highly sensitive, enable a short time to diagnosis (less than one hour), have minimal maintenance and ideally no calibration, and be low cost (< USD 6) (WHO Target Product Profile 2014). There has been growing interest in the detection of mycobacterial antigens such as lipoarabinomannan (LAM) in urine, a specimen that is easy to collect and process without the infection control risks associated with the collection of sputum. Despite increasing interest in the development of biomarker-based tuberculosis diagnostics (MacLean 2019), only one biomarker-based test has been recommended by the WHO for tuberculosis diagnosis: the lateral flow lipoarabinomannan assay (LF-LAM) (WHO Consolidated Guidelines (Module 3) 2020).

The LF-LAM assay is a commercially available point-of-care test for active tuberculosis. This review is limited to studies that used the Alere Determine TB LAM Ag (AlereLAM) (Abbott, Palatine, IL, USA, previously Alere Inc, Waltham, MA, USA), which is the only LF-LAM test currently recommended by the WHO; thus in the context of this review, LF-LAM refers only to AlereLAM. LF-LAM is an immunocapture assay that detects LAM antigen in urine. LAM is a lipopolysaccharide present in mycobacterial cell walls (Brennan 2003) , which is released from metabolically active or degenerating bacterial cells during tuberculosis disease (Briken 2004 A Cochrane Review update on the diagnostic accuracy of LF-LAM for the detection of tuberculosis in HIV-positive adults found that LF-LAM has a sensitivity of 42% and specificity of 91% to diagnose tuberculosis in HIV-positive individuals with tuberculosis symptoms and sensitivity of 35% and specificity of 95% in HIVpositive individuals not assessed for tuberculosis symptoms. LF-LAM sensitivity is higher in inpatients compared to outpatients and those with lower CD4 cell counts compared to those with higher CD4 counts, whereas specificity is lower in both of these subgroups (Bjerrum 2019).

In 2015, informed by the original Cochrane Review (Shah 2016) , the WHO made a conditional recommendation for using LF-LAM to assist with the diagnosis of tuberculosis in HIV-positive people with advanced disease, and a strong recommendation against using the test "as a screening test for tuberculosis" based on the data among unselected participants (WHO Consolidated Guidelines (Module 3) 2020). Based on evidence from randomized trials and an updated Cochrane Review (Bjerrum 2019), the WHO currently recommends that LF-LAM should be used to assist in the diagnosis of active tuberculosis in HIV-positive adults, adolescents, and children (WHO Consolidated Guidelines (Module 3) 2020). The key change from the WHO 2015 guidelines is a broadening of the indication for use of the LF-LAM assay among HIV-positive inpatients with signs and symptoms of active tuberculosis (pulmonary and extrapulmonary) irrespective of their CD4 count or inpatients with advanced HIV or who are seriously ill or irrespective of signs and symptoms of active tuberculosis if they have a CD4 count of less than 200 cells/µL. The updated guidelines recommend the use of LF-LAM in HIV-positive outpatients and children with signs and symptoms of tuberculosis (pulmonary or extrapulmonary tuberculosis, or both forms) or irrespective of signs and symptoms of active tuberculosis if they have a CD4 count of less than 100 cells/ µL, based on the generalization of data from adult inpatients, whilst acknowledging the limitation of the available data. The WHO recommends that LF-LAM should not be used for general tuberculosis screening in people without HIV "owing to sub-optimal sensitivity". The guidelines further suggest that LF-LAM should be used in combination with existing tests, and not as a replacement test (to existing tests) (WHO Consolidated Guidelines (Module 3) 2020). 

LF-LAM has lower sensitivity to detect tuberculosis in adults living with HIV than the internationally suggested minimum target of 65% for non-sputum based tuberculosis tests, recommended for trained microscopy technicians (WHO Target Product Profile 2014) . However, since it is a rapid point-of-care assay, it is expected that obtaining a positive LF-LAM result will enable the prompt initiation of tuberculosis therapy in people living with HIV who are more likely to have advanced tuberculosis disease and be at higher risk for adverse outcomes, including death. Additionally, since people living with HIV are less likely to be able to produce sputum, which is needed to perform other WHO-recommended rapid tests such as Xpert MTB/RIF, Xpert Ultra, and Truenat, the use of LF-LAM may increase the yield of microbiologically confirmed tuberculosis.

This review is important for several reasons. To our knowledge, evidence of the e ects of diagnostic tests for tuberculosis on patient outcomes has o en been inferred from diagnostic accuracy studies rather than randomized trials. An exception is the Xpert MTB/RIF assay, described below. The impact of a diagnostic test relies on the test results being used to guide clinical management (di Ru ano 2012). The best way to evaluate the e ect of a test strategy is by using a test-and-treat randomized controlled trial design. In this design, researchers assign patients to an experimental versus a standard testing approach and measure the (test strategy) e ect on mortality and other patient outcomes (di Ru ano 2012; Schünemann 2016).

Regarding the LF-LAM assay, a multisite randomized controlled trial published in 2016 demonstrated that bedside urine LAMguided initiation of antituberculosis treatment in HIV-positive adult inpatients with presumptive tuberculosis was associated with decreased eight-week mortality (Peter 2016). A second large multisite randomized controlled trial has subsequently been published (Gupta-Wright 2018), which did not demonstrate Cochrane Database of Systematic Reviews a significant decrease in mortality across the overall study population, but found a reduction in mortality in three predefined subgroups (participants with signs and symptoms of tuberculosis, participants with CD4 < 100 cells/µL, and participants with anaemia defined as haemoglobin < 8 g/dL). In addition, several observational cohort studies have suggested an association between urine LAM positivity and higher disease severity and mortality ( The aim of our systematic review is to determine the impact of the use of LF-LAM on mortality and other patient-important outcomes. This review, along with the Cochrane Review update on the diagnostic test accuracy of LF-LAM (Bjerrum 2019), will have important implications for the further scale-up of this diagnostic test and the use of other similar tests in high tuberculosis burden countries in the future. In addition, this review may contribute to implications for future research concerning the design of trials of new diagnostic tools.

To assess the impact of the use of LF-LAM (AlereLAM) on mortality in adults living with HIV in inpatient and outpatient settings.

To assess the impact of the use of LF-LAM (AlereLAM) on other patient-important outcomes in adults living with HIV, including time to diagnosis of tuberculosis, and time to initiation of tuberculosis treatment.

We included only randomized controlled trials (RCTs) and cluster-RCTs. We excluded other study designs and data reported only in abstracts, reviews, comments, and editorial notes.

We included participants who were adults (15 years and older is considered 'adult' for purpose of tuberculosis surveillance) living with HIV.

We included studies in which there was a suspicion of tuberculosis amongst study participants based on the presence of signs and symptoms compatible with tuberculosis (studies with symptomatic participants), as well as studies that included participants who presented for medical care irrespective of signs and symptoms of tuberculosis (studies with unselected participants). Signs and symptoms of tuberculosis included cough, fever, weight loss, and night sweats.

Diagnostic strategies that used LF-LAM either alone or in combination with other tests.

Diagnostic strategies that used smear microscopy, mycobacterial culture, or Xpert MTB/RIF, or a combination of these tests, which did not include LF-LAM.

• All-cause mortality during study follow-up time.

• Tuberculosis-related mortality during study follow-up time.

• Time to diagnosis of tuberculosis.

• Time to tuberculosis treatment initiation.

• Time from diagnosis to tuberculosis treatment initiation.

• Proportion of study participants who were diagnosed with tuberculosis. • Proportion of study participants who were treated for tuberculosis. • Proportion of study participants who were treated for tuberculosis but did not have tuberculosis. • Proportion of study participants who were able to produce a specimen for diagnostic testing. • Incremental diagnostic yield due to addition of LF-LAM to the diagnostic algorithm. • Tuberculosis-related treatment outcomes (treatment success or failure, relapse or cure).

Trusted evidence. Informed decisions. Better health.

The search for this review was conducted in conjunction with the search for a Cochrane Review update on the diagnostic accuracy of AlereLAM (Bjerrum 2019). That literature search was conducted up to 11 May 2018. In addition, specifically for this intervention review, we conducted a search up to 12 March 2021 in the following databases, using the search terms reported in Appendix 1:

• We also included search results from the original Cochrane Review (Shah 2016) . We performed the searches with no language restriction.

We further examined reference lists of relevant reviews and studies and searched the WHO website.

We used Covidence systematic review so ware to manage the selection of studies (Covidence). The search for this review was conducted in parallel with the search for the Cochrane Review update on the diagnostic accuracy of LF-LAM for tuberculosis (Bjerrum 2019). Two review authors (MS and SB) working independently screened titles and abstracts (including those that were identified in the original Cochrane Review on the diagnostic accuracy of LF-LAM for tuberculosis (Shah 2016) ) to identify citations that included data on health impact. We retrieved the article of any citation identified by either review author for fulltext review. Then, two other review authors (RRN and PL) working independently assessed articles for inclusion using predefined inclusion and exclusion criteria. In the study selection process, we assessed the full-texts of the 15 studies included in the Cochrane Review update, as well as any studies excluded during the fulltext screening (Bjerrum 2019). Any di erences in opinion were resolved through discussion. We listed studies excluded a er fulltext assessment and their reasons for exclusion in a PRISMA flow diagram (Moher 2009).

We developed and piloted a standardized data extraction form. Subsequently, two review authors (RRN and PL) independently extracted data from each included study. Any disagreements were resolved through discussion or by consulting a third review author (KRS). We extracted data on the following:

• Author, publication year, study design, country(ies), clinical setting (outpatient or inpatient), number enrolled and analysed. • Participants: age, HIV status and whether they are taking antiretroviral therapy, presence of symptoms (symptoms versus unselected). • Mode of mortality assessment, type of mortality (all-cause versus tuberculosis-related), timing of mortality assessment. • LF-LAM grade and use of old versus new reference card, timing of LF-LAM. • Mortality analysis metrics used (risk ratio, absolute risk reduction, (adjusted) hazard ratio or Kaplan-Meier, (adjusted) odds ratio). • Comparator groups analysed.

• Mortality in the intervention group, mortality in the control group. • Mortality data stratified by CD4 count. • Time to diagnosis; time to treatment initiation; time from diagnosis to treatment initiation. • Proportion of study participants who were diagnosed with tuberculosis. • Proportion of study participants who were treated for tuberculosis. • Proportion of study participants who were treated for tuberculosis and did not have tuberculosis. • Proportion of study participants who were able to produce a specimen for diagnostic testing. • Other outcomes assessed in the study (e.g. incremental diagnostic yield due to addition of LF-LAM to the diagnostic algorithm). • Other tuberculosis-related outcomes (e.g. treatment success or failure, relapse or cure).

We contacted study authors for clarification as needed.

Two review authors (RRN and PL) independently assessed risk of bias using the Cochrane tool for assessing risk of bias in randomized studies (Higgins 2011), employing Review Manager 5 (Review Manager 2020). We contacted the corresponding study authors for clarification or more information if data were missing or unclear. Any discrepancies were resolved through discussion or by consulting a third review author (KRS) if required. We assessed the included studies for the method of allocation, sequence generation and allocation concealment, blinding, missing outcome data, outcome measurement and reporting, and selective reporting bias. We assessed the risk of bias as low, high, or unclear.

We presented our findings of the e ect of the use of LF-LAM on dichotomous outcomes using risk ratios or hazard ratios (when Cochrane Database of Systematic Reviews e ect measure was time-to-event) with respective 95% confidence intervals.

We decided that if we identified cluster-RCTs, we would extract adjusted measures of e ect where possible. If the study authors did not perform any adjustment for clustering, we would adjust the raw data using an intraclass correlation coe icient (ICC) value. If an ICC value was not reported in the study, we would contact the study authors for this information, obtain it from similar studies, or estimate the ICC. We would not present results from cluster-RCTs that were not adjusted for clustering. If we identified cluster-RCTs, we would estimate the ICC, and perform sensitivity analyses to investigate the robustness of our analyses. If we identified studies for inclusion that had multiple intervention arms, we would include data from these studies by either combining intervention arms, or by splitting the control group so that participants would only be included in the meta-analysis once.

Before extracting data from the studies, we determined the reasons for missing data by attempting to contact the respective corresponding study author. We investigated whether the missingness of data may have introduced attrition bias. We would carry out an available-case analysis if we considered the missing data to be missing at random. If we suspected that missing data were not missing at random, we would perform sensitivity analyses in which we imputed the data using specific assumptions, such as assuming all missing participants experienced or did not experience the event.

We assessed the existence of clinical heterogeneity by examining di erences in study characteristics and participant demographic factors in order to inform decisions regarding the appropriateness of pooling data from studies in meta-analysis. We assessed statistical heterogeneity by visually inspecting forest plots and using a Chi test for heterogeneity (with a P value of 0.10 for significance), and the I statistic as a measure of inconsistency across studies, with an I value of 50% higher representing substantial heterogeneity (Deeks 2021).

We planned to assess reporting biases if more than 10 studies were included in the meta-analysis by examining the funnel plots visually for symmetry or asymmetry and interpreting test results in the context of visual inspection of funnel plots. In the case of asymmetry, we would follow recommendations for interpretation as recommended by Sterne 2011.

We conducted analyses using Review Manager 5 (Review Manager 2020). For dichotomous outcomes, we performed meta-analyses to estimate the pooled risk ratio (95% confidence interval). We used a fixed-e ect model unless we identified clinical or statistical heterogeneity, in which case we used a random-e ects model. For time-to-event outcomes, we would perform meta-analyses to estimate the pooled hazard ratio (95% confidence interval).

Based on data availability, we performed subgroup analyses in participants with varying CD4 levels, in particular, CD4 > 200 cells/ µL versus CD4 ≤ 200 cells/µL, and CD4 > 100 cells/µL versus CD4 ≤ 100 cells/µL. The rationale for subgroup analyses stratified by CD4 count is because we hypothesized that the use of urine LAM would have a larger e ect on mortality in participants with lower CD4 counts, since the diagnostic accuracy of urine LAM is higher with lower CD4 counts (Bjerrum 2019).

We planned to perform sensitivity analyses in the case of circumstances that we thought were likely to influence outcomes, including the following.

• Excluding studies with missing data that were likely to influence the outcome. • Excluding studies with outliers that were suspected to influence the outcome. • Excluding studies with high risk of bias that were likely to a ect the outcome.

Outliers would have been identified on visual examination of forest plots as extreme values that we judged important to investigate the e ect of excluding. None of the included studies had missing data, outliers, or a high risk of bias that was likely to influence the outcome. However, we performed sensitivity analyses in which we imputed the data using specific assumptions. Based on prior literature informing the expected tuberculosis mortality at eight weeks (Ford 2016; Nliwasa 2018), we varied the mortality event rate in missing participants between 0% and 25%.

We summarized our findings in the summary of findings tables.

We assessed the certainty of evidence using the GRADE approach (Hultcrantz 2017), employing GRADEpro GDT so ware (GRADEpro GDT). We rated each important outcome (primary outcomes) as described by Balshem 2011, as follows.

• High: we are very confident that the true e ect lies close to that of the estimate of the e ect. • Moderate: we are moderately confident in the e ect estimate:

the true e ect is likely to be close to the estimate of the e ect. • Low: our confidence in the e ect estimate is limited: the true e ect may be substantially di erent from the estimate of the e ect. • Very low: we have very little confidence in the e ect estimate:

the true e ect is likely to be substantially di erent from the estimate of e ect.

Randomized controlled trials start as high-certainty evidence, but can be downgraded if there are valid concerns within the following five domains: risk of bias, imprecision, inconsistency, indirectness, and publication bias. Studies can also be upgraded if there is a large e ect, a dose-response e ect, and if all plausible residual confounding would reduce a demonstrated e ect or would suggest a spurious e ect if no e ect was observed (Balshem 2011).

As recommended, we reported the findings in simple, standardized statements (Santesso 2020). Cochrane Database of Systematic Reviews

We provided descriptions of the included studies in the Characteristics of included studies and Table 1 , 'Characteristics of studies that evaluated a diagnostic intervention that included LF-LAM in adults living with HIV'. We provided descriptions of the excluded studies in Characteristics of excluded studies.

The literature search for this review was conducted in conjunction with the search for a Cochrane Review update on the diagnostic accuracy of LF-LAM (Bjerrum 2019). Two studies that were included by Bjerrum 2019 were identified as meeting the inclusion criteria for this review (Gupta-Wright 2018; Peter 2016). Our updated searches identified a further 218 unique studies for title and abstract screening, two of which met the criteria for full-text review (Blanc 2020; Grant 2020), and one of which met our inclusion criteria to undergo data extraction (Grant 2020). A flow diagram of the study selection process is shown in Figure 1 .

We included three studies that assessed the impact of LF-LAM on mortality when the test was used for clinical decisionmaking; two of these took place in inpatient settings (Gupta-Wright 2018; Peter 2016), and one in outpatient settings (Grant 2020). Both inpatient studies were multisite RCTs conducted in sub-Saharan African countries that evaluated the impact of a diagnostic strategy that included LF-LAM as a tuberculosis diagnostic test to guide treatment initiation in HIV-positive adults, comparing all-cause mortality at 56 days (eight weeks) between the LF-LAM intervention arm and standard-of-care control arm (Gupta-Wright 2018; Peter 2016). The outpatient study was an open-label cluster-RCT conducted in South Africa that evaluated the impact of a diagnostic strategy that included LF-LAM as a tuberculosis diagnostic test to guide treatment initiation in people living with HIV, comparing all-cause mortality at six months between the LF-LAM intervention arm and standard-of-care control arm (Grant 2020). None of the trials reported the cause of death, including tuberculosis-related mortality.

There were several di erences between the trials. in no LF-LAM arms). The percentage of participants on antiretroviral therapy was lower in Peter 2016 than in Gupta-Wright 2018 (48% versus 72%). A greater proportion of participants were started on tuberculosis treatment in Peter 2016 compared to Gupta-Wright 2018 (50% versus 17%), likely reflecting di erent inclusion criteria (clinical suspicion of tuberculosis in the former compared with an unselected population irrespective of symptoms in the latter) and a consequent di erence in the severity of illness in the study populations between the two trials. Despite these di erences, we considered the interventions implemented, target population, and study setting to be su iciently similar in both inpatient trials to permit pooling in a meta-analysis.

The outpatient trial was conducted at 24 primary healthcare clinics in South Africa (Grant 2020). In Grant 2020, study nurses at intervention clinics assessed patients using a multifactorial assessment, consisting of tuberculosis symptoms, BMI, point-of-care haemoglobin concentrations, and urine LF-LAM results. A positive urine LF-LAM result was one of the high probability criteria defined in the study algorithm which categorized participants as high probability of tuberculosis, for which initiation of tuberculosis treatment was recommended immediately followed by antiretroviral treatment two weeks later, versus medium probability of tuberculosis, for which symptomguided investigation was recommended, versus low probability of tuberculosis, for which initiation of antiretroviral treatment was recommended. Although this trial was conducted in outpatient settings, where participants are typically expected to be less sick than in inpatient settings, the eligibility criteria included having a CD4 count of 150 cells/µL or less, and the median CD4 count was lower than in the two inpatient trials (72 cells/µL). Most participants (69%) were symptomatic, although this was not part of the eligibility criteria. Individuals with clinical signs necessitating urgent referral to secondary care were excluded.

We excluded eight full-text articles from the literature searches performed up to 12 March 2021. We excluded seven of these studies because they were not RCTs (Cummings 2019; Gupta-Wright 2019; Huerga 2019; Kubiak 2018; Mathabire Rucker 2019; Mthiyane 2019; Naidoo 2019). We excluded one study because the standard-ofcare arm utilized a diagnostic strategy that did not include any of the tuberculosis tests that we described in our protocol criteria for inclusion, and was not used to guide clinical decision-making, precluding comparison of the impact of LF-LAM to other diagnostic strategies (Blanc 2020).

The assessment of risk of bias in the included studies is shown in Figure 2 and Cochrane Database of Systematic Reviews Random sequence generation (selection bias) Allocation concealment (selection bias) Blinding of participants and personnel (performance bias): All outcomes Blinding of outcome assessment (detection bias): All outcomes Incomplete outcome data (attrition bias): All outcomes Selective reporting (reporting bias) Other bias

Grant 2020 + + ? ? + + + Gupta-Wright 2018 + + + + + + ?

Peter 2016 + + ? ? + + ?

Gupta-Wright 2018 and Peter 2016 used random sequence generation and were thus assessed as at low risk of selection bias in this domain, and were also judged to be at low risk of selection bias with respect to allocation concealment. In Peter 2016, participants were assigned to the intervention arm which included LF-LAM versus standard-of-care using computer-generated allocation lists.

Research nurses who did not have access to these lists either contacted a data manager by telephone or used a text message system to assign each participant. In Gupta-Wright 2018, study nurses or clinicians took a consecutive sealed envelope which contained the unique patient identifier but did not reveal the study group, to which they remained masked. In Grant 2020, randomization at the clinic level was performed by a study Cochrane Database of Systematic Reviews statistician to achieve reasonable balance, taking into account mean CD4 count, peri-urban versus rural clinic location, and total monthly antiretroviral therapy initiations.

We judged two studies to have an unclear risk of both performance and detection bias because participants and research sta in Peter 2016 and Grant 2020, and clinic sta in Grant 2020, were not masked to either allocation or test results, and it is unclear how this lack of blinding may have a ected clinical decision-making. Since investigators, all study sta (other than the laboratory technician and statistician), hospital attending clinical teams, and participants were masked to the study group allocation in Gupta-Wright 2018, we judged this trial to be at low risk of bias.

We judged all three studies to have a low risk of attrition bias due to incomplete outcome data, reporting a 5% loss to follow-up rate (Peter 2016); 1% loss to follow-up rate (Gupta-Wright 2018); and 0% loss to follow-up rate (Grant 2020), respectively.

We judged all three studies to have a low risk of reporting bias, as they prespecified their primary and secondary outcome measures 

We judged the two studies in inpatient settings to have an unclear risk of other bias (Peter 2016; Gupta-Wright 2018). The informed consent requirements for both inpatient trials may have resulted in the exclusion of people who were critically ill, who would have been part of the target population for the intervention in realworld practice settings. Similarly, limiting enrolment to standard working hours and for less than 48 hours a er admission may have also introduced other bias. Peter 2016 also noted a heterogeneous mortality e ect across the four countries in which the study was conducted. We judged Grant 2020 to have a low risk of other bias because although there were some di erences between the intervention and standard-of-care groups, notably presence of tuberculosis symptoms, prior receipt of isoniazid preventive therapy, and receipt of tuberculosis tests in the preceding six months, we did not think these di erences were likely to introduce bias.

See: Summary of findings 1 E ect of LF-LAM compared to no LF-LAM on mortality and tuberculosis treatment initiation in people living with HIV, inpatient settings; Summary of findings 2 E ect of LF-LAM compared to no LF-LAM on mortality and tuberculosis treatment initiation in people living with HIV, outpatient settings

Both included trials assessed mortality at eight weeks. The pooled risk ratio (RR) was 0.85 (95% confidence interval (CI) 0.76 to 0.94; 5102 participants, 2 trials, Analysis 1.1, Figure 4 ). That is, people living with HIV who received LF-LAM had 15% lower risk of mortality. The absolute e ect was 34 fewer deaths per 1000 (from 14 fewer to 55 fewer) (moderate-certainty evidence) (Summary of findings 1). Owing to a 5% loss to follow-up rate in Peter 2016 with the possibility of potentially unbalanced study arms, we performed sensitivity analyses to investigate the robustness of the availablecase analyses. The pooled RR was 0.87 (95% CI 0.78 to 0.97) with a 25% mortality event rate in the intervention compared to 0% event rate in the standard-of-care group (Analysis 1.2, Figure 5 ), and 0.82 (95% CI 0.74 to 0.91) with a 0% mortality event rate in the intervention compared to 0% event rate in the standard-ofcare group (Analysis 1.3, Figure 6 ). We thus found that the overall meta-analysis result remained statistically significant regardless of varying the hypothetical event rate amongst missing data, and we can therefore be confident in our overall finding. Cochrane Database of Systematic Reviews 

The one included trial assessed mortality at six months. The RR was 0.89 (95% CI 0.71 to 1.11; 2972 participants, 1 trial, Analysis 1.4, Figure 7 ; Summary of findings 2). Although this trial did not detect a di erence in mortality at six months in people living with HIV who had received LF-LAM testing as part of a tuberculosis diagnostic strategy compared to routine tuberculosis diagnostic testing without LF-LAM, the direction of e ect was towards a mortality reduction, and similar to the e ect in people evaluated in inpatient settings.

Grant 2020 

Analysis 2.1 examines the impact of LF-LAM on mortality in individuals in inpatient settings stratified by CD4 count with a threshold of 100 cells/µL. Table 2 demonstrates the heterogeneity in e ect estimates across CD4 strata that were seen within and between the two included trials.

For individuals with a CD4 count of ≤ 100 cells/µL, the pooled RR was 0.88 (95% CI 0.77 to 1.01; 2016 participants, 2 trials, Analysis 2.1.1, subgroup 1, Figure 8 ).

Gupta-Wright 2018 Peter 2016 Subtotal (95% CI) Total events: Heterogeneity: Tau² = 0.00; Chi² = 1.03, df = 1 (P = 0.31); I² = 3% Test for overall effect: Z = 1.85 (P = 0.06) For individuals with a CD4 count of > 100 cells/µL, the pooled RR was 0.80 (95% CI 0.55 to 1.17; 2867 participants, 2 trials, Analysis 2.1.2, subgroup 2, Figure 8 ).

Given that the I statistic for Analysis 2.1.2 demonstrated substantial heterogeneity, we used a random-e ects model for these meta-analyses. 

Analysis 2.2 examines the impact of LF-LAM on mortality in individuals in inpatient settings stratified by CD4 count with a threshold of 200 cells/µL. Table 2 demonstrates the heterogeneity in e ect estimates across CD4 strata that were seen within and between the two included trials.

For individuals with a CD4 count of ≤ 200 cells/µL, the pooled RR was 0.87 (95% CI 0.77 to 0.98; 2886 participants, 2 trials, Analysis 2.2.1, subgroup 1, Figure 9 ). Figure 9 . Forest plot comparing the risk ratios for mortality in inpatients who received a diagnostic intervention including LAM compared to those who received standard of care (SoC), mortality at eight weeks, by CD4 strata ≤ 200 cells/µL versus > 200 cells/µL. Between brackets are the 95% confidence intervals (CI) for the risk ratios. The figure shows the estimated mortality risk ratio for each study (blue square) and its 95% CI (black horizontal line) and the pooled estimated mortality risk ratio combining results from both studies (black diamond). Figure 9 ).

Given that the I statistic for Analysis 2.2.2 demonstrated substantial heterogeneity, we used a random-e ects model for these meta-analyses.

Analysis 2.3 examines the impact of LF-LAM on mortality in outpatient settings stratified by CD4 count with a threshold of 50 cells/µL.

For individuals with a CD4 count of < 50 cells/µL, the RR was 1.01 (95% CI 0.78 to 1.31, 460 person-years, 1 trial, Analysis 2.3.1, subgroup 1), that is a detectable di erence in mortality was not seen in study participants with a CD4 count of < 50 cells/µL who received a diagnostic assessment that included LF-LAM testing compared to those undergoing standard of care without LF-LAM.

For individuals with a CD4 count of ≥ 50 cells/µL, the RR was 0.76 (95% CI 0.55 to 1.05, 942 person-years, 1 trial, Analysis 2.3.2, subgroup 2), that is the direction of e ect was towards a decrease in mortality in study participants with a CD4 count of ≥ 50 cells/µL who received a diagnostic assessment that included LF-LAM testing compared to those undergoing standard of care without LF-LAM.

Although mortality was the primary patient-important outcome of interest, we also recorded data on other patient-important outcomes. However, data for our prespecified secondary outcomes were o en not reported or were analysed variably between studies, which limited our ability to perform meta-analyses.

Gupta-Wright 2018 found that the time from randomization to diagnosis was marginally shorter in the LF-LAM intervention group compared to the standard-of-care group (median 0 days (interquartile range (IQR) 0 to 1)) versus 1 day (IQR 0 to 6), adjusted hazard ratio 1. Cochrane Database of Systematic Reviews treatment was short (median 1 day (IQR 0 to 1)) and did not di er between the group that received LF-LAM and the standard-of-care group (adjusted hazard ratio 0.83, 95% CI 0.69 to 1.01).

Peter 2016 reported that time-to-treatment initiation in the group that received LF-LAM was shorter than in the group that did not receive LF-LAM, with a higher proportion of those treated having antituberculosis treatment initiated by the end of days 1 to 4 of hospitalization (84% versus 76%, P < 0.001).

Gupta-Wright 2018 reported that increases in tuberculosis diagnoses in the intervention group that received LF-LAM (22% of tuberculosis diagnoses) compared to the standard-of-care group (15% of tuberculosis diagnoses) were not confined to high-risk subgroups, with an overall adjusted absolute risk di erence of 7.3% (95% CI 4.4 to 10.2). These results were adjusted for study sites; unadjusted data were not provided.

In inpatient settings, both included trials assessed tuberculosis treatment initiation. The pooled RR was 1.26 (95% CI 0.94 to 1.69; 5102 participants, 2 trials, Analysis 3.1, Figure 10 ). Of note the, I statistic (90%) suggested substantial heterogeneity, thus we used random-e ects meta-analysis. We hypothesize that this heterogeneity was due to the smaller di erence between the proportion of participants initiated on treatment for the group that received the diagnostic intervention that included LF-LAM compared to the standard-of-care group that did not receive LF-LAM in Peter 2016 compared to Gupta-Wright 2018. This di erence is likely reflective of the population in Peter 2016 being sicker, resulting in a higher likelihood of empiric treatment even in the standard-of-care group, compared to Gupta-Wright 2018, in which participants were enrolled irrespective of symptoms and signs of tuberculosis and were less sick (Table 1 ). In outpatient settings, the included trial assessed tuberculosis treatment initiation. The RR was 5.44 (95% CI 4.70 to 6.29, 3022 participants, 1 trial, Analysis 3.2, Figure 11 ).

Grant 2020 

None of the included trials evaluated the proportion of study participants who were treated for tuberculosis but did not have tuberculosis.

Both inpatient trials reported that a higher proportion of study participants were able to provide urine as a diagnostic specimen compared to sputum. Gupta-Wright 2018 reported that urine was provided by 99% (2548/2574) of participants, whereas only 57% (1464/2574) were able to produce sputum. Of note, the proportion of participants able to produce sputum in Malawi (39%, 518/1316) was lower than in South Africa (75%, 946/1258). Peter 2016 similarly reported that 99% (2507/2528) were able to produce a urine specimen; however, in this study 93% (2355/2528) were able to produce a sputum specimen for diagnosis. We believe that this heterogeneity between studies may reflect the enrolment of a population with a higher burden of pulmonary disease in Peter 2016, or study operational factors such as improved counselling regarding specimen collection or the availability and use of sputum induction for participants who were unable to produce sputum spontaneously in Peter 2016. In contrast, only a single spontaneous sputum specimen was collected in Gupta-Wright 2018.

Gupta-Wright 2018 reported that increases in tuberculosis diagnoses in the intervention group that received LF-LAM (22% of tuberculosis diagnoses) compared to the standard-of-care group (15% of tuberculosis diagnoses) were not confined to high-risk subgroups, with an overall adjusted absolute risk di erence of 7.3% (95% CI 4.4 to 10.2). These results were adjusted for study sites; unadjusted data were not provided. Gupta-Wright 2018 reported that of participants with microbiologically confirmed tuberculosis who had only one positive LF-LAM test, LF-LAM had an incremental diagnostic yield of 41% (87/210), compared to 6% (13/210) for urine Xpert MTB/RIF and 14% (30/210) for sputum Xpert MTB/RIF.

None of the included trials evaluated other tuberculosis-related treatment outcomes (treatment success or failure, relapse or cure).

Two trials in inpatient settings and one trial in outpatient settings assessed the impact of diagnostic strategies that included LF-LAM on mortality when the test was used in conjunction with other tuberculosis diagnostic tests or clinical assessment for clinical decision-making in people living with HIV (see Summary of findings 1 and Summary of findings 2). In inpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy likely reduces mortality in people living with HIV at eight weeks compared to routine tuberculosis diagnostic testing without LF-LAM (moderate-certainty evidence). Although the certainty of the evidence is moderate, this finding represents a moderate e ect on a patient important outcome (mortality) and in an important population (people living with HIV) that is aligned with the decision of the WHO Guideline Development Group, which issued a "strong recommendation for the intervention" in the inpatient context (WHO Consolidated Guidelines (Module 3) 2020). In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may reduce mortality in people living with HIV at six months compared to routine tuberculosis diagnostic testing without LF-LAM (low-certainty evidence). Although this trial did not detect a di erence in mortality, the direction of e ect was towards a mortality reduction, and the e ect size was similar to that in inpatient settings.

Assessment of other patient-important and implementation outcomes in the trials varied. In inpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy probably results in a slight increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF-LAM. In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may result in a large increase in the proportion of people living with HIV who were started on tuberculosis treatment compared to routine tuberculosis diagnostic testing without LF-LAM. The included trials demonstrated that a higher proportion of people living with HIV were able to produce urine compared to sputum for tuberculosis diagnostic testing; a higher proportion of people living with HIV were diagnosed with tuberculosis in the group that received LF-LAM; and the incremental diagnostic yield was higher for LF-LAM than for urine or sputum Xpert MTB/RIF.

The three included trials were conducted in sub-Saharan Africa, thus we do not have direct evidence of the applicability of the findings to other settings outside of Africa. Although the majority of people with HIV-associated tuberculosis live in sub-Saharan Africa, caution is required with respect to inferences to policy from the limited number of trials conducted to date. Although the only outpatient trial, Grant 2020, was conducted at multiple clinics in one country (South Africa), both inpatient trials included more than one country, and Peter 2016 included four countries, which increases the applicability of the study findings to settings outside of South Africa, which is important because other countries in sub-Saharan Africa are more resource-constrained than South Africa.

We emphasize the importance of the finding that despite di erences in study design, two large, multicentre trials demonstrated that the use of LF-LAM reduced mortality in inpatient settings, which increases generalizability. In Gupta-Wright 2018, the test was performed in the laboratory, not at the point of care. The intervention in Gupta-Wright 2018 was a combination of LF-LAM and urine Xpert MTB/RIF, which could impact generalizability, and introduces heterogeneity since it is not possible to assess the impact of LF-LAM alone in this study. In Peter 2016, the intervention was bedside LF-LAM plus a "nurse-informed" treatment decision. In Grant 2020, the intervention was a study nurse-led diagnostic evaluation that included assessment of tuberculosis symptoms, BMI, fingerprick haemoglobin testing, and point-of-care LF-LAM testing. There was heterogeneity in the inpatient trial populations in that Gupta-Wright 2018 screened patients admitted to medical wards, and Peter 2016 screened patients admitted to emergency units, short-stay wards, and medical wards. This may have biased the study populations to being sicker, particularly if this led to Library Trusted evidence. Informed decisions. Better health.

Cochrane Database of Systematic Reviews an earlier time of enrolment in Peter 2016, although both trials only enrolled patients admitted for less than 48 hours. We note that the informed consent requirements for both inpatient trials likely resulted in the exclusion of patients who were critically ill, and that enrolment being limited to standard working hours and for less than 48 hours a er admission may have a ected the representativeness of the population studied. In Grant 2020, patients who met criteria for urgent referral to secondary care were excluded, as were patients with a higher risk of adverse events due to tuberculosis treatment, specifically those with chronic liver disease or high weekly alcohol intake.

These additional considerations may not reflect how the test will be performed in routine practice in other clinical settings. Important questions related to implementation remain, such as how diagnostic algorithms will be implemented to identify patients to undergo LF-LAM testing; what is the impact of LF-LAM being performed at the bedside by sta who may be less familiar with the challenges of interpreting the results based on the reference scale card versus in the laboratory by trained sta ; and how will LF-LAM results be acted upon, particularly in the clinical scenario where the LF-LAM test is positive, and another test such as Xpert is negative. The relatively low accuracy of the current LF-LAM test may compromise its impact, depending upon how it is interpreted by the end-users of the test, for example if a negative LF-LAM test is interpreted as the absence of tuberculosis. We await studies that evaluate the impact of more accurate urine LAM tests, such as FujiLAM (Broger 2020), on mortality and patient-important outcomes.

The only trial evaluating the impact of LF-LAM in outpatient settings did not demonstrate an e ect on mortality, although the direction of e ect was towards a mortality reduction, and the e ect size was similar to that in inpatient settings. However, we do not consider that inpatient status should necessarily be a limiting factor for test use. We note that people with HIV presenting in outpatient settings with tuberculosis may be very sick, which aligns with the WHO's conditional recommendation for the use of LF-LAM in this population. We also emphasize that mortality is not the only patient-important outcome to be considered. Results from all three trials included in this review demonstrated an increase or direction of e ect towards an increase in tuberculosis treatment initiation amongst people who received a diagnostic strategy that included LF-LAM, which is an important implementation outcome as part of e orts to improve the quality of tuberculosis care.

We hypothesized that e ect estimates may vary between CD4 strata (summarized in Table 2 for the inpatient trials Gupta-Wright 2018 and Peter 2016) owing to factors that could lead to either a reduced or improved e ect size for participants with CD4 ≤ 100 cells/µL or CD4 ≤ 200 cells/µL, compared to CD4 > 100 cells/µL or CD4 > 200 cells/µL, respectively. Similarly, the results for Grant 2020 summarized in Analysis 2.3 demonstrate that CD4 count may be associated with both LF-LAM positivity, and mortality (i.e. it can serve simultaneously as a covariate of interest and outcome). CD4 count could also be associated with antiretroviral therapy (ART) usage, and empiric tuberculosis treatment initiation. Consequently, these factors could impact the e ect of any benefit from earlier diagnosis (e.g. through LF-LAM implementation) and lead to variable e ects (of LF-LAM implementation). For example, if in both study arms participants with a lower CD4 were more likely to start tuberculosis treatment (which would be done empirically in the standard-of-care arm), then the benefit of confirmed diagnosis by LF-LAM could be blunted. Conversely, individuals with lower CD4 are more likely to have LF-LAM positivity leading to earlier treatment diagnosis and initiation. LF-LAM compared to standard of care could therefore have reduced mortality. LF-LAM positivity is also associated with mortality (Gupta-Wright 2016; Shah 2016). Consequently, earlier diagnosis through LF-LAM may not lead to a modifiable outcome, since having a lower CD4 count is associated with an increased risk of mortality (Kaplan 2018). LF-LAM positivity may also serve as a surrogate for a point-of-care CD4 test, thus LAM positivity could have stimulated earlier ART initiation (which may have a mortality benefit) if it is interpreted as a CD4 proxy. LF-LAM may also be a marker of bacterial load (Shah 2010) , and treatment may thus have di erential impact amongst people with tuberculosis who have a positive LF-LAM test compared to people with tuberculosis who have a negative LF-LAM test, but neither included study compared mortality in participants with a positive versus negative LF-LAM test.

Understanding whether the use of LF-LAM testing has a di erential impact on mortality at lower CD4 counts is clinically relevant given the higher overall risk of tuberculosis-related mortality in these subgroups. Since the results from individual studies were heterogeneous within CD4 strata, and neither inpatient trial reported mortality rates by CD4 strata, evidence to explain how the e ectiveness of the use of LF-LAM may vary according to CD4 count is uncertain. However, the subgroup analyses demonstrated that the direction of e ect was towards decreased mortality for all CD4 strata examined, and that the e ect did not vary significantly between di erent CD4 subgroups, aside from CD4 count < 50 cells/ µL in the trial that was conducted in outpatient settings (Grant 2020).

Analysis of subgroup data, particularly those stratified by CD4 count, is limited, since the trials were not powered to detect changes at the subgroup level. Consequently, these results must be interpreted with caution when considering how the impact of the use of LF-LAM may vary according to CD4 count. Despite some di erences in the pooled e ect estimates across CD4 strata, these di erences were not statistically significant, and the direction of the pooled e ect estimates was towards a decrease in mortality for each CD4 stratum, aside from CD4 count < 50 cells/µL in Grant 2020. CD4 subgroup analysis results may not have reached statistical significance owing to the association of lower CD4 count with both LF-LAM positivity and mortality. LF-LAM positivity is also associated with increased mortality, thus LF-LAM-directed treatment may have been insu icient to avert the increased mortality associated with LF-LAM positivity. Understanding whether there is a di erence in treatment outcomes amongst those who have a positive LF-LAM result compared to a negative LF-LAM result, adjusting for time of tuberculosis treatment, ART usage, and CD4 count may be important for future LAM implementation.

Analysis of secondary outcome data should also be interpreted with caution, given the substantial heterogeneity between the three trials. The di erences in these secondary outcomes, including time to treatment initiation and likelihood of being able to produce a specimen for tuberculosis diagnostic testing, may reflect Cochrane Database of Systematic Reviews the di erent severity of illness in the study populations. Peter 2016 included people with signs and symptoms of tuberculosis, whereas Gupta-Wright 2018 included people irrespective of signs and symptoms of tuberculosis. It is likely that enrolment of a sicker population in Peter 2016 resulted in more individuals in both arms being started on tuberculosis treatment. It may be that sicker individuals in Peter 2016 had a higher burden of respiratory disease that resulted in a higher proportion being able to produce a sputum specimen, or other operational factors such as potentially improved counselling related to specimen production may have increased the feasibility and yield of sputum-based testing. Similarly, although Grant 2020 was performed in outpatient settings, the median CD4 count was lower than for the two trials performed in inpatient settings, which likely resulted in more people being started on tuberculosis treatment.

This review evaluated the impact of LF-LAM to guide treatment decisions in the context of clinical trials. Further implementation research is needed to assess the reach, e ectiveness, costs, and other operational considerations during programmatic implementation.

Details regarding the downgrading of the certainty of the evidence using the GRADE approach are included in the summary of findings tables. Our main reason for downgrading the certainty of the evidence for the analysis of trials in inpatient settings was indirectness, a concern about the generalizability to other settings due to heterogeneity in the diagnostic strategy used (aside from the use of LF-LAM) owing to the use of a pragmatic trial design. In both trials, patients who could not give informed consent were ineligible to participate, which may have biased the e ect of the intervention towards the null if sicker patients were less likely to be able to give informed consent. Our reasons for downgrading the certainty of the evidence for the analysis of trials in outpatient settings were concerns about indirectness, given that only one trial, conducted in South Africa, was included, and imprecision, based on the wide 95% confidence interval that was reported.

We were careful to limit bias in the review process by strict adherence to Cochrane methods. Since policymakers, clinicians, and members of the wider tuberculosis community considered this review to be critical to decision-making, we included subgroup analyses by CD4 count, even though only three trials were identified for inclusion in the review, and we specified these analyses a priori in our protocol. We advise that results from these subgroup analyses be interpreted with caution given the small number of studies and residual heterogeneity between results of individual studies analysed according to CD4 count. However, the selection of CD4 count was also motivated by biological and clinical considerations, and we note that both inpatient trials contributed more than 2000 participants to each CD4 strata analysed.

We are not aware of other systematic reviews on this topic. A systematic review that sought to evaluate the prognostic value of LF-LAM to determine mortality risk demonstrated that the detection of LF-LAM in urine was independently associated with increased risk of mortality during treatment, a er adjusting for other factors associated with mortality (Gupta-Wright 2016). Another recent study evaluating the implementation of LF-LAM (excluded from this review due its observational study design) found that LAM-positive patients not diagnosed through other tools and not treated for tuberculosis had a significantly higher risk of mortality compared to LAM-positive patients who received treatment (Huerga 2019). This is consistent with our explanation of the heterogeneity in e ect estimates observed across CD4 strata, which we hypothesized may reflect the increased mortality associated with LF-LAM positivity, although the studies included in our review did not specifically evaluate mortality in patients who had a LF-LAM positive test compared to those who had a LF-LAM negative test. Our review is the only systematic review to assess the impact of using a diagnostic testing strategy that includes the use of LF-LAM to guide management decisions on mortality and other patient-important outcomes.

In inpatient settings, the use of LF-LAM as part of a tuberculosis diagnostic testing strategy likely reduces mortality and probably results in a slight increase in tuberculosis treatment initiation in people living with HIV. The reduction in mortality is likely due to earlier diagnosis, which facilitates prompt treatment initiation.

In outpatient settings, the use of LF-LAM testing as part of a tuberculosis diagnostic strategy may reduce mortality and may result in a large increase in tuberculosis treatment initiation in people living with HIV. Our results support the implementation of LF-LAM to be used in conjunction with other WHO-recommended tuberculosis diagnostic tests to assist in the rapid diagnosis of tuberculosis in people living with HIV.

Given the limited number of trials on this topic, there is less certainty in the e ect estimates for the subgroup analyses stratified by CD4 count, which reflects a patient population of clinical importance. Additional studies will help to understand the heterogeneity demonstrated in e ect estimates across CD4 strata and to determine the impact of LF-LAM testing on mortality and patient-important outcomes in outpatient populations. There is also an urgent need for studies evaluating the e ect of LF-LAM on patient-important outcomes in children. Further studies are needed to understand the impact of new, urine-based, pointof-care tests that detect urine LAM for tuberculosis diagnosis, such as the FujiLAM assay, on mortality and patient-important outcomes that inform the use of LAM assays on both diagnostic and treatment pathways. The use of implementation science research can strengthen the evidence base to inform the adoption of interventions such as LF-LAM by health systems.

The Academic Editor is Dr Nathan Ford.

The editorial base of the Cochrane Infectious Diseases Group (CIDG) is funded by UK aid from the UK government for the benefit of low-and middle-income countries (project number 300342-104).

Theron 

World Health Organization. WHO consolidated guidelines on tuberculosis. Module 3: diagnosis -rapid diagnostics for tuberculosis detection. Licence: CC BY-NC-SA 3.0 IGO. who.int/publications/i/item/who-consolidated-guidelineson-tuberculosis-module-3-diagnosis---rapid-diagnostics-fortuberculosis-detection (accessed 2 July 2020).

World Health Organization. Global Tuberculosis Report 2020. apps.who.int/iris/bitstream/ handle/10665/336069/9789240013131-eng.pdf (accessed 20 June 2021).

Methods RCT 

Random sequence generation (selection bias) Low risk Quote: "A statistician randomised primary health-care clinics (1:1), using computer-generated random numbers, to the intervention or standard of care (control)."

Allocation concealment (selection bias) Low risk Quote: "A statistician randomised primary health-care clinics (1:1), using computer-generated random numbers, to the intervention or standard of care (control), based on restriction to achieve reasonable balance, separately, for mean CD4 count, peri-urban versus rural clinic location, and total monthly ART initiations."

Blinding of participants and personnel (performance bias) All outcomes Unclear risk Quote: "Due to the nature of the intervention being examined, participants, research sta , and clinic sta were aware of group allocation."

Blinding of outcome assessment (detection bias) All outcomes Unclear risk Quote: "To ascertain possible serious and severe adverse events... research nurses and research assistants enquired about symptoms and history suggesting possible adverse events at every study visit, including early study review visits for participants in the intervention group, and sought relevant data in case note reviews. Due to the pragmatic trial design, no equivalent early study visits were done for participants in the control group; thus we anticipat- Cochrane Database of Systematic Reviews ed that adverse events would be more completely ascertained in the intervention group."

Incomplete outcome data (attrition bias) All outcomes Low risk Quote: "Vital status at 6 months was determined for 1487 (98.7%) of 1507 participants in the intervention group and 1485 (98.0%) of 1515 participants in the control group."

Selective reporting (reporting bias)

Low risk Comment: the authors prespecified primary and secondary outcomes for analysis, which they reported.

Other bias Low risk Although there were some differences between the intervention and standard-of-care groups, notably the presence of tuberculosis symptoms, prior receipt of isoniazid preventive therapy, and receipt of tuberculosis tests in the preceding 6 months, we did not consider these differences to be likely to introduce bias.

Methods RCT 

Random sequence generation (selection bias) Low risk Quote: "a randomisation list of unique patient identifiers was generated by the study statistician using a computer-generated random block size."

Allocation concealment (selection bias) Low risk Quote: "On enrolment, study nurses or clinicians took a consecutive sealed opaque envelope containing the unique patient identifier but not the study group, to which they remained masked. A paired set of sealed envelopes were kept in a locked cabinet in the study laboratory, labelled with the unique patient identifier and containing the study group allocation. These were opened by the laboratory technician on receipt of study tuberculosis screening specimens."

Blinding of participants and personnel (performance bias) All outcomes results, which were combined in an algorithm and interpreted as high, medium, or low probability of TB, which was used to guide treatment decision-making versus SoC which did not specify the testing approach but did not include urine LAM Study nurses performed LF-LAM testing in the intervention clinics and interpreted the results in conjunction with the other clinical assessment findings to determine the probability of TB, and issued treatment based on a study algorithm stratified by the probability of TB.

72 cells/μL NR 21.4 Primary outcome:

• Mortality at 6 months Secondary outcome:

• Proportion initiated on TB treatment 

TOPIC: (tuberculosis or TB) AND TOPIC: (lipoarabinomannan or LAM) AND TITLE: (test OR assay OR antigen OR Ag OR lateral flow assay*OR urine antigen OR point of care)

( ( TITLE-ABS-KEY ( tuberculosis OR tb ) ) AND ( lam OR lipoarabinomannan ) ) AND ( TITLE ( test OR assay OR antigen ) ) 

RRN and KRS dra ed the manuscript, with specific input from MC for the sections on data analysis. SB, PL, and MS provided critical revisions to the manuscript. All review authors read and approved the final manuscript dra .

The impact of Xpert MTB/ RIF -do we have a final answer

GRADE Working Group. GRADE Guidelines: 16. GRADE evidence to decision frameworks for tests in clinical practice and public health

Quantitative analysis of a urine-based assay for detection of lipoarabinomannan in patients with tuberculosis

Lateral flow urine lipoarabinomannan assay for detecting active tuberculosis in HIV-positive adults

Xpert MTB/RIF and Xpert Ultra assays for screening for pulmonary tuberculosis and rifampicin resistance in adults, irrespective of signs or symptoms

Association of HIV infection with extrapulmonary tuberculosis: a systematic review

SILVAMP TB LAM" rapid urine tuberculosis test predicts mortality in patients hospitalized with human immunodeficiency virus in South Africa

Delays in diagnosis and treatment of pulmonary tuberculosis in India: a systematic review

Recommendations for examining and interpreting funnel plot asymmetry in meta-analyses of randomised controlled trials

Feasibility, accuracy, and clinical e ect of point-of-care Xpert MTB/RIF testing for tuberculosis in primary-care settings in Africa: a multicentre, randomised, controlled trial

Impact of diagnostic strategies using lateral flow urine lipoarabinomannan assay on health outcomes for tuberculosis in people living with HIV. PROSPERO 2019 CRD42019153471

She serves voluntarily as Chair of a tuberculosis advocacy organisation, TB Proof, based in South Africa. Although TB Proof has not directly led advocacy e orts related to urine LAM, the organisation has supported calls to improve access to urine LAM based on WHO guidance

Canada; and USAID, USA, administered by the World Health Organization Global TB Programme, Switzerland for the preparation of systematic reviews and educational materials

• National Institutes of Health, USA RRN is supported by an NIH/NIAID Career Development Award

• United States Agency for International Development (USAID) administered by the World Health Organization (WHO) Global TB Programme, Switzerland Development of the systematic review was in part made possible with financial support from the USAID administered by the WHO Global TB Programme

UK Project number

Primary stratification by clinical setting is consistent with the clinical pathway suggested by the current World Health Organization guidelines (WHO Consolidated Guidelines (Module 3) 2020), since a priori there is likely to be clinical heterogeneity with respect to inpatient and outpatient participant groups. We were only able to perform meta-analyses for one of our prespecified secondary outcomes (proportion of participants started on tuberculosis treatment), since data on the other outcomes were either not reported or only

The views expressed do not necessarily reflect the UK government's o icial policies.We wish to thank Vittoria Lutje, CIDG Information Specialist, who dra ed the search strategy. We also thank Claudia Denkinger (University Hospital Heidelberg) Eleanor Ochodo (Stellenbosch University), Samuel Schumacher (WHO Global TB Programme, formerly with Foundation for Innovative New Diagnostics), and Madhukar Pai (McGill University), who provided guidance regarding how to assess the impact of randomized trials for new tuberculosis diagnostic tests. 

Random sequence generation (selection bias) Low risk Quote: "We randomly assigned eligible patients..using computer generated allocation lists."Comment: issue arose with duplicate randomization numbers, but this was resolved.

Low risk Quote: "Once an eligible patient was identified, the research nurse at each site (who did not have access to these lists) either contacted a centrally located data manager by telephone to obtain assignment or used a text-message system, which automatically accessed the allocation list used to assign each patient."Blinding of participants and personnel (performance bias) All outcomes 

Blanc 2020 Standard-of-care arm utilized a diagnostic strategy that did not include any of the tuberculosis tests described in our protocol as criteria for inclusion, and study was not used to guide clinical decision-making, precluding comparison of the impact of LF-LAM compared to other diagnostic strategies.

Not an RCT