Diagnostic accuracy, incremental yield and prognostic value of Determine TB-LAM for routine diagnostic testing for tuberculosis in HIV-infected patients requiring acute hospital admission in South Africa: a prospective cohort

Background We previously reported that one-third of HIV-positive adults requiring medical admission to a South African district hospital had laboratory-confirmed tuberculosis (TB) and that almost two-thirds of cases could be rapidly diagnosed using Xpert MTB/RIF-testing of concentrated urine samples obtained on the first day of admission. Implementation of urine-based, routine, point-of-care TB screening is an attractive intervention that might be facilitated by use of a simple, low-cost diagnostic tool, such as the Determine TB-LAM lateral-flow rapid test for HIV-associated TB. Methods Sputum, urine and blood samples were systematically obtained from unselected HIV-positive adults within 24 hours of admission to a South African township hospital. Additional clinical samples were obtained during hospitalization as clinically indicated. TB was defined by the detection of Mycobacterium tuberculosis in any sample using Xpert MTB/RIF or liquid culture. The diagnostic yield, accuracy and prognostic value of urine-lipoarabinomannan (LAM) testing were determined, but urine-LAM results did not inform treatment decisions. Results Consecutive HIV-positive adult acute medical admissions not already receiving TB treatment (n = 427) were enrolled regardless of clinical presentation or symptoms. TB was diagnosed in 139 patients (TB prevalence 32.6%; median CD4 count 80 cells/μL). In the first 24 hours of admission, sputum (spot and/or induced) samples were obtained from 37.0% of patients and urine samples from 99.5% of patients (P < 0.001). The diagnostic yields from these specimens were 19.4% (n = 27/139) for sputum-microscopy, 26.6% (n = 37/139) for sputum-Xpert, 38.1% (n = 53/139) for urine-LAM and 52.5% (n = 73/139) for sputum-Xpert/urine-LAM combined (P < 0.01). Corresponding yields among patients with CD4 counts <100 cells/μL were 18.9%, 24.3%, 55.4% and 63.5%, respectively (P < 0.01). The diagnostic yield of urine-LAM was unrelated to respiratory symptoms, and LAM assay specificity (using a grade-2 cut-off) was 98.9% (274/277; 95% confidence interval [CI] 96.9–99.8). Among TB cases, positive urine-LAM status was strongly associated with mortality at 90 days (adjusted hazard ratio 4.20; 95% CI 1.50–11.75). Conclusions Routine testing for TB in newly admitted HIV-positive adults using Determine TB-LAM to test urine provides major incremental diagnostic yield with very high specificity when used in combination with sputum testing and has important utility among those without respiratory TB symptoms and/or unable to produce sputum. The assay also rapidly identifies individuals with a poor prognosis. Electronic supplementary material The online version of this article (doi:10.1186/s12916-017-0822-8) contains supplementary material, which is available to authorized users.


(Continued from previous page)
Conclusions: Routine testing for TB in newly admitted HIV-positive adults using Determine TB-LAM to test urine provides major incremental diagnostic yield with very high specificity when used in combination with sputum testing and has important utility among those without respiratory TB symptoms and/or unable to produce sputum. The assay also rapidly identifies individuals with a poor prognosis.
Keywords: Tuberculosis, HIV, Diagnosis, Urine, Xpert, Lipoarabinomannan, LAM, Africa, Screening, Hospital, Determine TB-LAM Background Tuberculosis (TB) is the leading cause of HIV/AIDS-related mortality globally, accounting for an estimated 0.4 million HIV-related deaths worldwide in 2015 [1,2]. A large majority of these deaths occur in sub-Saharan Africa. Postmortem studies conducted in African hospitals over the past 20 years have revealed that between 32% and 67% (pooled summary prevalence estimate 39.7%; 95% confidence interval [CI] 32.4-47.0) of HIV-positive adult patients had evidence of TB disease at autopsy [3][4][5] and that this was the primary cause of death in over 90% of these cases [6]. Disease was frequently disseminated and approximately half of these cases remained undetected prior to death [6]. These findings suggest the need for new approaches to routine diagnostic screening for TB in HIVinfected patients newly admitted to hospital in resourcelimited settings.
Using an intensive clinical sampling approach, we have previously demonstrated that one in three unselected HIVinfected acute medical adult admissions to a South African township district hospital had TB that could be diagnosed microbiologically [7]. Presentation was so non-specific that neither symptom screening nor risk profiling were sufficient to reliably identify a low-risk group that did not warrant further laboratory investigation for TB. We therefore concluded that to maximize TB case ascertainment, all HIV-infected patients newly admitted to hospital in settings with high TB burden should be investigated for TB using microbiological tests regardless of their clinical presentation. Unfortunately, laboratory capacity is often severely limited in resource-limited settings. Moreover, it is frequently challenging to obtain sputum samples from sick medical inpatients, even from those suspected of having pulmonary TB [8,9]. Conversely, urine can be readily collected from the vast majority of patients, and we [7] and others [10] have previously demonstrated that Xpert MTB/ RIF testing of urine samples concentrated by centrifugation assay provides a substantial diagnostic yield of TB in this patient population. However, implementing such a strategy in resource-limited settings can be hindered by the cost, and the need for laboratory infrastructure with a reliable electricity supply to allow urine centrifugation and Xpert MTB/RIF testing. In the present study, we therefore explored the utility of the Determine TB-LAM © (Alere Inc., Waltham, MA, USA) lateral-flow urine assay-a technically much simpler, instrumentation-free, low-cost, point-of-care diagnostic tool [11]. We reasoned that this could be used to test urine samples within the emergency admission unit or ward environment without the need for laboratory infrastructure, and might therefore be more feasibly implemented at more remote health facilities in resource-limited settings. In this study, we report on the diagnostic accuracy of the Determine TB-LAM test and its incremental diagnostic yield when used in combination with sputum-based diagnosis. We also report on the strong association between urine-lipoarabinomannan (LAM) status and mortality during 90 days of follow-up.

Setting and patients
This study forms part of a larger study of rapid urine-based approaches to diagnosis of HIV-associated TB that has previously been reported [7]. The prospective, observational cohort reported herein was enrolled at G.F. Jooste Hospital in Cape Town, South Africa. This 200-bed adult district hospital serves township communities in which there is a high seroprevalence of HIV and incidence of TB. This study was approved by the Human Research Ethics Committees of the University of Cape Town and the London School of Hygiene & Tropical Medicine. Patients provided written informed consent in their first language. The study is reported according to STARD initiative guidelines for reporting studies of diagnostic accuracy [12].
Adult patients aged ≥18 years were recruited on 4 days of the week from medical wards. On recruitment days, the study nurse coordinator ascertained from the ward register all medical admissions in the previous 24hour period and recorded these in the study register.
All patients who either previously tested negative or who had undocumented HIV status were offered HIV testing using two approved rapid tests. All patients who had a current TB diagnosis and/or were receiving treatment for TB at the time of hospital admission were excluded from the present study. Otherwise, all other adult patients with documented positive HIV status were eligible for the study and were invited to participate regardless of presenting symptoms or reason for hospital admission.

Procedures and samples
Demographic and clinical details (including the World Health Organization [WHO] four-symptom screen for HIV-associated TB [13]) were recorded. The research team systematically investigated patients by obtaining sputum, urine and blood specimens for TB investigations between 9.00 a.m. on the day of enrolment and 9.00 a.m. the following day. Numerous additional samples (including both sputum and non-respiratory samples) for mycobacteriology were obtained by the routine medical team according to clinical indication throughout the admission period (Additional file 1: Table S1).
In the initial 24-hour period of hospital admission, two sputum samples were requested from each patient with careful instruction and supervision by the study nurse coordinator (an experienced respiratory nurse) as previously described [7]. A spot specimen was obtained first, followed by a second sample, which was induced using nebulized 3% hypertonic saline. When necessary, both specimens were induced. Alternatively, for those either too unwell to leave the ward to attend the sputum induction facility or too unwell for the sputum induction procedure, two spot specimens were requested. Urine samples were systematically collected in single-use disposable bedpans (Litha Healthcare Group, Johannesburg, South Africa). Fresh aliquots (2.0 mL) were sent for Xpert testing and the remaining urine volume (30-40 mL) was stored at -20°C. Venous blood (5.0 mL) was inoculated into BACTEC™ Myco F/Lytic culture vials (Becton Dickinson, Franklin Lakes, NJ, USA).

Laboratory procedures
Specimens were processed using standardized protocols in centralized accredited laboratories of the South African National Health Laboratory Service as described previously [14,15]. Decontaminated, centrifuged deposits of sputum samples were re-suspended in phosphate buffer. Smears stained with auramine O stain were examined with fluorescence microscopy. Equal volumes were tested using Xpert and culture in mycobacterial growth indicator tubes (MGIT; Becton Dickinson, Sparks, MD, USA), which were incubated for up to 6 weeks. Culture isolates were identified as M. tuberculosis complex with the MTBDRplus assay (Hain Lifesciences, Nehren, Germany). Non-respiratory samples were also tested using MGIT liquid culture, with the exception of venous blood, which was tested using Myco F/Lytic culture.
Urine was tested using Xpert MTB/RIF in two different ways as described previously [7]. Fresh urine samples (2.0 ml) were centrifuged and the pellet re-suspended in 0.75 mL of phosphate buffer before testing using the Xpert MTB/RIF assay [7]. Each stored urine sample of between 30 and 40 mL was thawed and centrifuged at 3000 × g for 15 min. Following removal of the supernatant, the pellet was re-suspended in the residual urine volume and 0.75 mL was tested using Xpert. This second sample derived from centrifugation of a large urine volume was referred to as the 'concentrated' urine sample and these samples were batch processed on a weekly basis due to study-related logistical considerations and laboratory workflow.
The same frozen urine samples (unconcentrated and unprocessed) were also retrospectively tested for the presence of LAM using the Determine TB-LAM test (Alere Inc.) following the manufacturer's instructions. Samples were thawed to ambient temperature and for each sample, 60 μL of unprocessed urine was applied to the sample pad at the bottom of the test strip. After 25 min, test strips were independently read by two investigators (SDL and MV) blinded to patient status and all other test results. Having checked for the presence of the positive control band on each strip, test results were scored as grade 1, 2, 3, 4 or 5 when compared to the reference card as described elsewhere [13]. For example, for the test result to be scored as grade 2, the test band had to be at least as intense as the corresponding grade 2 band on the reference card but not as intense as the grade 3 reference band. Any discrepancies in scoring were reviewed by both readers and the result resolved and agreed within 35 min of initial urine application. After resolution of results, the grade scores were categorized as being 'positive' (grade 2 or higher) or 'negative' (no test band at least as intense as grade 2).
Results of all microbiological tests were returned to the clinical team to inform treatment decisions with the exception of the results of the Determine TB-LAM assay, which was not endorsed for clinical use in South Africa or by the WHO at the time of the study.

Clinical outcomes
Patient and ward records as well as district, regional and national electronic data systems were used to ascertain deaths during the first 90 days following enrolment as fully described elsewhere [16]. Patients with an unconfirmed vital status at 90 days (<5%) were considered to be lost to follow-up and data were censored on the last date of contact with the health system.

Data analysis
Diagnostic yield of Determine TB-LAM and relationship with patient symptoms and other characteristics The diagnostic reference standard for a confirmed new TB diagnosis was the detection of M. tuberculosis from at least one clinical sample of any type using MGIT culture and/or Xpert MTB/RIF. The total number of confirmed cases (n = 139) was used as the denominator to calculate the comparative diagnostic yield obtained from different sample types and different assays, including Determine TB-LAM. The yield of TB diagnoses from Determine TB-LAM testing of urine samples obtained during the initial 24 hours was compared with the yield from sputum samples collected in the same period. We included Xpert results in our diagnostic reference standard. While we calculated the yield of diagnoses made by Xpert in different specimen types compared with other methods, this does not represent a measure of diagnostic sensitivity (calculation of sensitivity was not performed). Diagnostic yield in our study reflected both the performance of the diagnostic test and the ability to obtain samples for that test in a real-world clinical setting. The associations between the yield of Determine TB-LAM and patient symptoms and other patient characteristics were explored and multivariable logistic regression was used to identify factors independently associated with positive urine-LAM status among patients with confirmed TB. One or more positive reference standard tests on a non-respiratory sample was taken as evidence of extrapulmonary TB (EPTB).

Diagnostic accuracy of Determine TB-LAM
For the purposes of assessing the diagnostic accuracy (sensitivity and specificity) of Determine TB-LAM, we identified a cohort of patients for whom sufficient microbiological data and the results of Determine TB-LAM urine testing were available. For inclusion in this analysis, patients were required to have the results of reference standard tests (liquid culture and/or Xpert MTB/RIF) from two or more samples obtained from two or more different anatomic sites, such as blood, sputum, urine or other non-respiratory samples. The rationale for this was to reduce the risk of patients with disseminated TB having false-negative reference standard tests and being inappropriately designated as 'TBfree'. By doing so, we enhanced the methodological rigour with which the specificity of Determine TB-LAM was assessed [17]. From this cohort, those testing positive for M. tuberculosis were used as the denominator to calculate the sensitivity of Determine TB-LAM (n = 136). In contrast, patients whose clinical samples all tested negative for M. tuberculosis were defined as 'TB-free' and were used as the denominator to calculate the specificity of Determine TB-LAM (n = 277).

Patient characterization
Patients were characterized using simple descriptive statistics. Anaemia severity was defined using WHO criteria: no anaemia (haemoglobin concentration ≥13.0 g/dL for males, ≥12.0 g/dL for females), mild anaemia (11.0-12.9 g/dL for males, 11.0-11.9 g/dL for females), moderate anaemia (8.0-10.9 g/dL for males and females) or severe anaemia (<8.0 g/dL for males and females) [18]. TB prevalence with 95% exact confidence intervals (95% CI) was calculated. Medians were compared using either Wilcoxon rank-sum tests or Kruskal-Wallis tests as appropriate and means were compared using unpaired t tests. Chi-squared, Fisher's exact and McNemar's tests were used as appropriate to compare proportions. Kaplan-Meier plots were used to examine mortality risk among confirmed TB cases stratified by urine-LAM status and a Cox proportional hazards model was used to determine factors associated with mortality. Person-time was accrued from the date of study enrolment until death, loss to follow-up or censorship 90 days after study entry. All variables in the univariable model meeting a cut-off of P ≤ 0.1 were included in the multivariable model. Statistical tests were two-sided at α = 0.05. Those already receiving treatment for an existing diagnosis of TB at the time of admission (n = 158, 27.0%) were excluded, leaving 427 eligible patients who formed the study cohort ( Fig. 1).

Patients enrolled and their characteristics
Patients were typically young adults; a majority knew their positive HIV status before admission and more than half of all patients had previously started antiretroviral therapy (ART) ( Table 1). Immunodeficiency was typically advanced and overall median CD4 cell counts were lower among those who were ART-naive compared to those currently receiving ART (115 cells/μL versus 212 cells/μL; P < 0.001). Nearly half of all patients had previously received treatment for TB and two-thirds had moderate or severe anaemia (Table 1).

Samples tested and laboratory-confirmed TB diagnoses made
During hospital admission, clinical samples for mycobacteriology were obtained from all 427 eligible study participants. The total of 1745 samples (mean 4.1 samples per patient) were derived from a median of three anatomic locations (Additional file 1: Table S1). These yielded 2391 reference standard test results (mycobacterial liquid culture and Xpert MTB/RIF). Results were missing for 4.3% of sputum cultures (n = 12), which were contaminated, and 4.3% of sputum Xpert tests (n = 12) and 3.1% of urine Xpert tests (n = 26), which had indeterminate results. However, all urine samples yielded interpretable LAM results (n = 418). TB was diagnosed in 139 patients with a mean of 3.2 positive samples per case diagnosed. Thus, the overall TB prevalence was 32.6% (95% CI 28.1-37.2). The mean number of tests done was similar for patients in whom TB was or was not diagnosed (5.8 tests versus 5.5 tests, respectively; P = 0.154, indicating a comparable intensity of investigation). The characteristics of those with and without TB are summarized in Table 1. Compared with those who were TB-free, those with a new TB diagnosis had lower CD4 counts and lower haemoglobin levels. The vast majority of patients had a positive WHO symptom screen regardless of TB status, but only onethird of patients (with or without TB) reported sputum production prior to admission.

Diagnostic accuracy of Determine TB-LAM urine testing
Having carefully defined which patients in the cohort had TB and which were TB-free, the diagnostic accuracy of urine-LAM was next assessed using data from the subset of 413 patients (96.7%) for whom reference standard test results on samples obtained from at least two anatomic sites plus urine-LAM results were available and complete (Fig. 1). In this subset of patients, TB was diagnosed in 136 patients and this was used as the denominator to calculate the sensitivity of Determine TB-LAM. Culture and Xpert tests were negative on all samples obtained from the remaining patients (n = 277), who were therefore designated as 'TB-free' and used to assess the specificity of Determine TB-LAM. Among these 277 patients, the mean number of negative reference standard tests per patient was 5.5.
Using the grade 2 reference band as the cut-off to designate urine-LAM results as 'positive' or 'negative' [13], the Determine TB-LAM results of the two independent readers of the lateral-flow tests were concordant in 407 of 413 tests, giving a proportionate agreement of 98.5% (95% CI 96.7-99.5) and a kappa statistic of 0.937 (95% CI 0.887-0.987). Following review of the discordant results from the remaining six tests, the two readers readily agreed upon a consensus result for each. Thus, TB-LAM results in agreement were available for all 413 patients.
Urine-LAM was positive in 53 of 136 TB cases, giving an overall sensitivity of 39.0% (95% CI 30.7-47.7). Urine-LAM was negative in 274 of the 277 patients who were TB-free, giving a specificity of 98.9% (95% CI 96.9-99.8). Two of three patients with apparent falsepositive urine-LAM results had clinico-radiological diagnoses of TB made by the routine medical team, although diagnoses were never confirmed microbiologically. Of note, neither had a full set of investigations available, with sputum and blood culture results missing. One patient had a 3-week history of weakness, diarrhoea and constitutional symptoms, was pancytopaenic (grade 2 LAM-positive), and subsequently died without any diagnosis being made. The other patient had a 3-month history of abdominal pain and weight loss, and an abdominal ultrasound demonstrated splenomegaly and splenic micro-abscesses with paraaortic and mesenteric lymphadenopathy suggestive of abdominal TB (grade 5 LAM-positive). Thus, had a composite reference standard for TB diagnosis (all microbiological and clinico-radiological diagnoses) been used in a post hoc analysis, the specificity would have been 279 out of 280 TB-free patients (99.6%; 95% CI 98.0-100). Using a grade 1 reference band cut-off  Table  S2). Both grade 1 and 2 reference band cut-offs had improved sensitivity when restricted to those with CD4 counts <100 cells/μL (56.9% and 68.1% using grade 2 and grade 1 cut-offs, respectively).

Comparative diagnostic yield of urine and sputum samples obtained in the first 24 hours
Given the very high specificity of urine-LAM, a positive Determine TB-LAM result was indicative of TB (positive predictive value [95% CI 85.1-98.9]). We therefore next compared the diagnostic yield of urine-LAM, The potential yield of sputum-based diagnoses was substantially limited by the high proportion of patients who were unable to produce a sputum sample. In the first 24 hours of admission only 158 of 427 (37.0%) patients were able to produce at least one sputum sample (spot and/or induced samples) despite assistance from an experienced respiratory nurse. Of these sputum samples, 36 (23.1%) were only obtained following sputum induction. The remaining patients who did not produce sputum were typically too sick or uncooperative to tolerate sputum induction or to be taken to the sputum induction room at the hospital, or it was attempted but was simply unsuccessful. In marked contrast, urine samples were readily obtained from 425 of 427 (99.5%) patients, although seven samples were misplaced in transit to the laboratory, leaving 418 urine samples available for testing (Fig. 1).
Sputum samples were eventually produced by a total of 245 patients (57.4%) at some point during their hospital admission. Of the 161 patients who self-reported sputum production prior to admission, 143 (88.8%) were able to produce a sputum sample for testing. Additionally, of the 201 patients reporting current cough at admission, 159 (79.1%) produced a sputum sample. The incremental diagnostic yield of Determine TB-LAM used in combination with sputum testing is shown in Fig. 2a. The addition of urine-LAM testing to sputum microscopy increased the overall diagnostic yield from 19.4% to 46.8% (n = 65/139: a 2.4-fold increase; P < 0.001) of all TB cases. The addition of urine-LAM testing to sputum-Xpert increased the diagnostic yield from 26.6% to 52.5% (n = 73/139: a 2.0-fold increase; P < 0.001). Figure 2b shows the diagnostic yields obtained from sputum microscopy, sputum-Xpert and urine-LAM among those patients with CD4 cell counts <100 cells/μL (n = 74) for specimens obtained in the first 24 hours of admission. The diagnostic yield from sputum microscopy, sputum-Xpert and urine-LAM were 18.9% (n = 14/74),  Thus, the incremental diagnostic yield of urine-LAM was greater among patients with lower CD4 counts (Fig. 2a, b).

Relationship between patient characteristics and respiratory symptoms and the diagnostic yield of Determine TB-LAM
The diagnostic yield of Determine TB-LAM was substantially greater among patients with lower CD4 cell counts, with a maximum yield of 56.6% (95% CI 45.4-67.2) observed among those with CD4 counts of <50 cells/μL (Fig. 3a). Yield was also much higher among those with more severe anaemia, with the diagnostic yield rising to 53.0% (95% CI 43.0-62.8) among those classified as having 'severe' anaemia (Fig. 3b). In contrast, the yield did not vary according to whether patients complained of cough or sputum production (Fig. 3c, d). Of the 53 patients with confirmed TB who tested urine LAM-positive, 22 (41.5%) reported no cough at the time of admission and 31 (58.5%) had no history of sputum production during their illness. Thus, it was notable that urine-LAM identified an important subgroup of TB cases who had no respiratory symptoms. Moreover, 36 of 53 urine-LAM-positive cases (67.9%) were patients who were unable to produce a sputum sample within the first 24 hours of admission.

Diagnostic yield among patients with pulmonary and extrapulmonary forms of HIV-associated tuberculosis
Of the patients whose TB was diagnosed using Xpert MTB/RIF testing of sputum obtained in the first 24 hours, 43.6% (n = 17) could be diagnosed by urine TB-LAM instead (Table 2). Overall, 113 patients (81.3%) had EPTB and of these 51 (45.1%) tested positive for urine-LAM, although the yield varied substantially by the type of EPTB. The proportion testing positive by TB-LAM was highest among those with positive blood cultures (65.9%) and those with positive lymph node aspirates (66.7%) ( Table 2). The yield of TB-LAM tended to be lower among those with positive cultures of cerebrospinal fluid (12.8%) and pleural fluid (30.8%) ( Table 2).

Comparative diagnostic yield of urine-Xpert and Determine TB-LAM
The greatest yield of any diagnostic approach in this study was the Xpert testing of urine that had been concentrated by centrifugation. A single Xpert test on concentrated urine yielded 59.0% of total diagnoses (n = 82), compared with 39.1% (n = 53) for urine-LAM (  Comparing the clinical phenotype of patients with confirmed TB stratified according to urine-LAM status revealed that those testing LAM-positive had lower CD4 cell counts and a higher prevalence and severity of anaemia (Table 3). In contrast, symptoms (respiratory or otherwise) were not associated with urine-LAM status. In multivariable analysis (Table 4), the only independent predictors of a positive urine-LAM status among confirmed TB patients (n = 136) were lower CD4 cell counts and more severe anaemia.
Having demonstrated that a positive urine-LAM status was associated with the severity of illness, we next compared survival among patients who tested urine LAMpositive or LAM-negative (Fig. 4). Among the 136 TB patients with urine-LAM results available, 13 of 53 testing LAM-positive died (24.5%; 95% CI 13.8-38.3) compared with 6 of 83 testing LAM-negative (7.2%; 95% CI 2.7-15.1; log-rank P = 0.006). Therefore, urine-LAM detected TB in 13 of 19 patients (68.4%) with HIVassociated TB who died within 90 days of study entry. In unadjusted analyses, a greater mortality risk was also associated with male sex and rifampicin-resistant TB but not with CD4 cell count (Table 5). In adjusted analyses, mortality remained strongly associated with urine-LAM status (adjusted hazards ratio 4.20; 95% CI 1.50-11.75) and with rifampicin-resistant disease (Table 5).

Discussion
As reported in the parent study, a microbiological diagnosis of TB was made in one in three unselected HIVpositive adult patients admitted to medical wards in this South African district hospital. The prevalence is so high in this vulnerable patient group that a simple, rapid screening test for TB that can be immediately performed at the time of hospital admission is highly desirable. In the present study, we found that routine urine-LAM testing could detect 39% of all HIV-associated TB diagnoses and the large majority of TB diagnoses among patients dying within 90 days of admission to hospital. We previously showed the remarkable utility of using Xpert MTB/RIF to test urine samples concentrated by centrifugation, with 64% of total diagnoses being made using samples obtained within the first 24 hours of admission (1.6-fold more diagnoses than LAM and 2.3-fold more than sputum Xpert testing). We are currently conducting a clinical trial in sub-Saharan Africa to assess whether implementation of systematic testing of urine  [19].
Should implementation of such a screening strategy (routine urine Xpert testing) be attempted in the countries of southern Africa worst hit by the HIV and TB epidemics, several potential barriers to the implementation would exist. These include the cost to purchase and maintain the GeneXpert platform, as well as the cost and supply chain of Xpert MTB/RIF cartridges (especially if cartridges are required to test both urine and sputum samples from each patient). Our screening strategy employing urine-based testing with Xpert MTB/RIF is also dependent upon the availability of laboratory facilities with sufficient capacity and biosafety equipment to undertake centrifugation of urine samples, including an uninterrupted electricity supply. All of these requirements may potentially undermine routine screening using Xpert testing of concentrated urine. It was for these reasons that we sought to explore the utility of the Determine TB-LAM lateral flow urine assay for LAM. The potential advantages of this diagnostic tool are self-evident: it currently costs approximately $2.66 per test (nearly one-quarter the cost of an Xpert MTB/RIF cartridge) and it requires no laboratory infrastructure, no laboratory personnel or expertise, and no instrumentation or electricity supply. The main perceived limitation to the use of Determine TB-LAM is that it has only modest overall sensitivity [13,20]. However, sensitivity is highest among the sickest patients with advanced immunodeficiency and worst prognostic characteristics, as seen in the present study and other previous observational studies [21]. Thus, the urine-LAM point-of-care test is ideal for use in HIVpositive inpatients requiring acute medical admission, as we enrolled in this study. Moreover, the limited sensitivity is greatly offset by the ease with which urine samples can be obtained from such patients compared to sputum. Whereas urine samples were readily obtained from almost all patients within the first 24 hours, just one in three patients could produce a sputum sample in the same time period despite the availability of sputum induction facilities. Thus, although the overall diagnostic yield from urine-LAM was modest (39%), it greatly exceeded the yield derived from sputum-based screening using fluorescence microscopy or Xpert MTB/RIF, and it doubled or more than doubled the diagnostic yield when used in addition to sputum-based screening.
Previous studies of the diagnostic accuracy of urine-LAM have restricted their evaluation to testing patients selected on the basis of their ability to produce sputum samples [8,15,22]. However, in the present study, just 51 (37.5%) of the 139 patients found to have a microbiological diagnosis of TB reported sputum production as a symptom during their illness, and only 37.0% could produce a sputum sample within the first 24 hours of admission. Of the true-positive urine-LAM results obtained, 31 of 53 (58.5%) were from patients who reported no sputum production (Fig. 3d). Thus, had urine-LAM screening been restricted to those with respiratory symptoms suggestive of pulmonary TB, almost three in five of the TB diagnoses that could be made using urine-LAM would have been missed. Among those with TB, we found no association between reported symptoms  [23]. Many researchers have previously suggested that urine-LAM assays lack sufficient specificity to be used as a 'rulein' diagnostic test. Indeed studies have reported specificity to be as low as 75% [20]. In marked contrast, one of the most important findings of this study was that assay specificity was extremely high (98.9%), indicating that this assay has utility as a rule-in diagnostic test for TB and that the results could be used to confidently inform the decision to start TB treatment in this patient population. When reading the lateral-flow strips, we used the agreed consensus cut-off grade 2 band on the reference card [18]. The strips were easy to read with 98.5% agreement between the two independent, blinded readers. Previous studies have reported variable specificity (75-99%) [20] and this may not only relate to differences in interpretation of the test-strip bands, but also to the rigour of the microbiological reference standard [13]. We designed this study to optimize assessment of LAM specificity, with an average of 5.6 reference standard tests being done for each patient. Moreover, the clinical samples were obtained from a median of three different anatomic compartments. This very comprehensive microbiological reference standard likely underlies the high specificity we observed.
The diagnostic yield varied substantially with CD4 cell count, with no useful yield among those with counts >200 cells/μL, but with a gradually increasing yield within lower CD4 cell count strata, reaching a maximum of 56.6% among those with counts of 0-49 cells/μL. The yield from urine-LAM screening was independent of ART status. This study confirms that assay utility is restricted to HIVinfected patients with CD4 cell counts of <200 cells/μL as previously shown among HIV-infected medical inpatients [8,22] and HIV-infected outpatients starting ART [15]. We also found a strong relationship between higher diagnostic yield and increasing severity of anaemia with a very poor overall yield among those with no or mild anaemia, as we have previously reported among outpatients [24]. Thus, patients with moderate or severe HIV-associated anaemia may especially benefit from TB screening using urine-LAM. In settings that do not permit universal TB screening due to limited resources, moderate or severe anaemia (in addition to advanced immunodeficiency) may serve as a simple marker for entry into the TB diagnostic algorithm.
During prospective follow-up, mortality was also found to be substantially greater among those testing urine-LAM positive in both crude and adjusted analyses. Other studies too have found a strong association between positive LAM status and mortality, including a recent meta-analysis that found a more than two-fold increased odds of mortality associated with LAM-positivity [21,[25][26][27][28][29]. Recently, a multi-country randomized controlled trial based in sub-Saharan Africa found that urine-LAM testing in addition to standard-of-care TB diagnostics was associated with a 17% (95% CI 4-28) relative risk reduction in 8-week all-cause mortality among HIV-infected patients requiring acute medical admission; in those with CD4 counts <50 cells/μL, the implementation of LAM was associated with a 29% (95% CI 10-44) relative risk reduction in 8-week all-cause mortality [30]. This was the first randomized controlled trial to demonstrate a mortality reduction associated with the implementation of a TB detection assay [31].
The Determine TB-LAM assay underwent expert review by the WHO in 2015. On the basis of a Cochrane review of all available data (that included unpublished data from the present study) [20] as well as the randomized trial by Peter et al. demonstrating a mortality benefit associated with LAM [30], the WHO now conditionally recommends that urine-LAM testing be used for all HIV-positive patients with CD4 counts <100 cells/μL who require hospitalization and have signs and symptoms of TB, as well as for HIV-positive inpatients who are seriously ill, irrespective of their CD4 cell count. While the sensitivity of the assay is generally modest, various factors favour implementation of this assay as a rule-in test for TB among HIV-positive inpatients by national TB programmes in resource-limited settings [32]. These include a proven mortality benefit, extremely high specificity and positive predictive value when a grade 2 cut-off is used, low-cost (consumable cost of $2.66 per test strip), and that it can be done at the point of care without laboratory expertise or instrumentation [22,33].
Strengths of this study include the recruitment of consecutive admissions without selection according to presenting symptoms or other criteria. Patients were very thoroughly investigated, yielding a large number of TB diagnoses and providing a comprehensive reference standard to assess the diagnostic accuracy of urine-LAM. We believe that this has provided the most rigorous assessment of the diagnostic accuracy of Determine TB-LAM to date. Patients were well characterized and are likely to be representative of HIV-infected medical admissions in countries with a high TB burden. The consistency in findings of post-mortem studies reporting on the burden of undiagnosed disseminated TB in HIV-infected inpatient deaths across the African continent [6] suggests that these findings may be of relevance throughout Africa. Outcome data at 90 days (~95% available) allowed the prognostic value of urine-LAM to be carefully assessed. Weaknesses include the fact that sputum could only be obtained from 57.4% of participants during their admission despite access to sputum induction facilities. This may have resulted in under-ascertainment of pulmonary TB cases. However, the large numbers of extrapulmonary specimens obtained (Additional file 1: Table S1) likely minimized the underdiagnosis of TB. The low proportion of patients who could provide a sputum sample despite the availability of sputum induction facilities was an important reason for the relatively poor yield of sputum diagnostics in this study. This in part reflects the design of the study, which involved screening all HIV-infected patients whether they reported a cough or not (approximately 80% of those who self-reported a cough were able to produce a sputum sample). Had the proportion producing sputum been higher (e.g. through enhanced sputum induction procedures and transporting sicker patients to the induction facility with medical support), then the proportion of participants diagnosed using urine-LAM alone (rather than sputum Xpert or smear microscopy) may have been lower. However, sputum induction facilities and resources for additional medical support in hospital services are for the most part not available in sub-Saharan Africa. Another weakness is that the study was done at a single site and urine-LAM testing was done retrospectively on frozen samples. However, the feasibility of prospectively testing urine samples at the point of care has been demonstrated in other studies [22,30]. Our results cannot be generalized to the testing of HIV-infected children, in whom the assay has unfortunately not been found to have utility [34]. Ongoing research on pathogen and host diagnostic urine biomarkers for TB may yield urine assays with improved performance characteristics in the future [35][36][37].