Point-of-care C-reactive protein and Xpert MTB/RIF Ultra for tuberculosis screening and diagnosis in unselected antiretroviral therapy initiators: a prospective, cross-sectional, diagnostic accuracy study

Abstract

Background: Tuberculosis, a major cause of death in people living with HIV, remains challenging to diagnose. Diagnostic accuracy data are scarce for promising triage and confirmatory tests such as C-reactive protein (CRP), sputum and urine Xpert MTB/RIF Ultra (Xpert Ultra), and urine Determine TB LAM Ag (a lateral flow lipoarabinomannan [LF-LAM] test), without symptom selection. We evaluated novel triage and confirmatory tests in ambulatory people with HIV initiating antiretroviral therapy (ART).

Methods: 897 ART-initiators were recruited irrespective of symptoms and sputum induction offered. For triage (n=800), we evaluated point-of-care blood-based CRP testing, compared with the WHO-recommended four-symptom screen (W4SS). For sputum-based confirmatory testing (n=787), we evaluated Xpert Ultra versus Xpert MTB/RIF (Xpert). For urine-based confirmatory testing (n=732), we evaluated Xpert Ultra and LF-LAM. We used a sputum culture reference standard.

Findings: 463 (52%) of 897 participants were female. The areas under the receiver operator characteristic curves for CRP was 0·78 (95% CI 0·73-0·83) and for number of W4SS symptoms was 0·70 (0·64-0·75). CRP (≥10 mg/L) had similar sensitivity to W4SS (77% [95% CI 68-85; 80/104] vs 77% [68-85; 80/104]; p>0·99] but higher specificity (64% [61-68; 445/696] vs 48% [45-52; 334/696]; p<0·0001]; reducing unnecessary confirmatory testing by 138 (95% CI 117-160) per 1000 people and number-needed-to-test from 6·91 (95% CI 6·25-7·81) to 4·87 (4·41-5·51). Sputum samples with Xpert Ultra, which required induction in 49 (31%) of 158 of people (95% CI 24-39), had higher sensitivity than Xpert (71% [95% CI 61-80; 74/104] vs 56% [46-66; 58/104]; p<0·0001). Of the people with one or more confirmatory sputum or urine test results that were positive, the proportion detected by Xpert Ultra increased from 45% (26-64) to 66% (46-82) with induction. Programmatically done haemoglobin, triage test combinations, and urine tests showed comparatively worse results.

Interpretation: CRP is a more specific triage test than W4SS in those initiating ART. Sputum induction improves diagnostic yield. Sputum samples with Xpert Ultra is a more accurate confirmatory test than with Xpert.

Funding: South African Medical Research Council, EDCTP2, US National Institutes of Health-National Institute of Allergy and Infectious Diseases.

Figure 1:. Flowchart showing participant enrolment, specimen collection and processing, and tests done

ART=antiretroviral treatment. LF-LAM=lateral flow lipoarabinomannan (Determine TB LAM Ag test). MGIT960=mycobacterial growth indicator tube. NaOH-NALC=sodium hydroxide-N-acetyl-L-cystein. Xpert=Xpert MTB/RIF. Xpert Ultra=Xpert MTB/RIF Ultra. ZN=Ziehl-Neelsen. *Haemoglobin was done when programmatically indicated outside of the study. †Unconcentrated Xpert Ultra done if concentrated Xpert Ultra non-actionable or positive.

Figure 2:. Flow diagrams showing the number of people in head-to-head comparisons of triage (A), confirmatory (B), and urine (C) tests

Triage tests are CRP and W4SS. Confirmatory tests are Xpert and Xpert Ultra on sputum. Urine tests are concentrated Xpert Ultra and LF-LAM. CRP correctly classified more people without tuberculosis compared with W4SS, Xpert Ultra detected more tuberculosis than Xpert on sputum, and Xpert Ultra detected more tuberculosis than LF-LAM on concentrated urine. Reasons for people excluded from each head-to-head analysis: (A–C) no culture (n=42: 22 no sputum and 20 contaminated); (A) no CRP results (n=55); (B) no actionable Xpert (n=91: 4 no result, 1 error, and 86 no or under volume specimen) and Xpert Ultra results (n=16: 2 error and 14 no sputum); and (C) no actionable concentrated urine Xpert Ultra (n=118: 3 no result, 23 invalid, and 92 error), insufficient urine (n=7: 4 for both LF-LAM and concentrated Xpert Ultra, 3 for concentrated Xpert Ultra only, and 1 for LF-LAM only), and LF-LAM unavailable (n=1). CRP=C-reactive protein. LF-LAM=lateral flow lipoarabinomannan (Determine TB LAM Ag test). RIF=rifampicin. Xpert=Xpert MTB/RIF. Xpert Ultra=Xpert MTB/RIF Ultra. W4SS=WHO-recommended four-symptom screen. *69 (66% of 104 people were culture-positive for both sputa; 35 (34%) of 104 were positive for one sputum and negative for the other (30 [86%]), contaminated for the other (4 [11%]), or the other not done (1 [3%]). †Xpert detected three Xpert Ultra-negative cases (3 [10%] of 30 [95% CI 2–27]). Xpert Ultra detected 19 Xpert-negative cases (19 [41%] of 46 [95% CI 27–57]).

Figure 3:. Summaries of triage and confirmatory test performance and effect on NNT

(A) Forrest plots comparing sensitivity and specificity point estimates (with 95% CIs) of triage tests and algorithms as well as sputum and urine confirmatory tests. Black dashed vertical lines indicate WHO target product profile estimates. (B) Effect of different triage tests and algorithms (table 2) on participant classification, showing CRP₁₀ alone or in combination with W4SS (Algorithms 3 and 4) to result in fewer onward unnecessary referrals (false positives). CRP₁₀ had 138 fewer false positives per 1000 people than W4SS (452 vs 314). The number needed to test to detect one culture-positive case when different triage methods are used in our cohort (C) and for Xpert Ultra only (D), modelled against different tuberculosis prevalences versus an Xpert Ultra-in-all scenario. Use of CRP₁₀ would result in a tuberculosis case correctly detected every five rather than every seven people as for W4SS. Other triage methods had lower NNTs but would be offset by diminished sensitivities. The grey column in D shows our prevalence. CRP=C-reactive protein. LF-LAM=lateral flow lipoarabinomannan (Determine TB LAM Ag test). Xpert=Xpert MTB/RIF. Xpert Ultra=Xpert MTB/RIF Ultra. W4SS=WHO-recommended four-symptom screen.