Validation of a host blood transcriptomic biomarker for pulmonary tuberculosis in people living with HIV: a prospective diagnostic and prognostic accuracy study

Abstract

Background: A rapid, blood-based triage test that allows targeted investigation for tuberculosis at the point of care could shorten the time to tuberculosis treatment and reduce mortality. We aimed to test the performance of a host blood transcriptomic signature (RISK11) in diagnosing tuberculosis and predicting progression to active pulmonary disease (prognosis) in people with HIV in a community setting.

Methods: In this prospective diagnostic and prognostic accuracy study, adults (aged 18-59 years) with HIV were recruited from five communities in South Africa. Individuals with a history of tuberculosis or household exposure to multidrug-resistant tuberculosis within the past 3 years, comorbid risk factors for tuberculosis, or any condition that would interfere with the study were excluded. RISK11 status was assessed at baseline by real-time PCR; participants and study staff were masked to the result. Participants underwent active surveillance for microbiologically confirmed tuberculosis by providing spontaneously expectorated sputum samples at baseline, if symptomatic during 15 months of follow-up, and at 15 months (the end of the study). The coprimary outcomes were the prevalence and cumulative incidence of tuberculosis disease confirmed by a positive Xpert MTB/RIF, Xpert Ultra, or Mycobacteria Growth Indicator Tube culture, or a combination of such, on at least two separate sputum samples collected within any 30-day period.

Findings: Between March 22, 2017, and May 15, 2018, 963 participants were assessed for eligibility and 861 were enrolled. Among 820 participants with valid RISK11 results, eight (1%) had prevalent tuberculosis at baseline: seven (2·5%; 95% CI 1·2-5·0) of 285 RISK11-positive participants and one (0·2%; 0·0-1·1) of 535 RISK11-negative participants. The relative risk (RR) of prevalent tuberculosis was 13·1 times (95% CI 2·1-81·6) greater in RISK11-positive participants than in RISK11-negative participants. RISK11 had a diagnostic area under the receiver operating characteristic curve (AUC) of 88·2% (95% CI 77·6-96·7), and a sensitivity of 87·5% (58·3-100·0) and specificity of 65·8% (62·5-69·0) at a predefined score threshold (60%). Of those with RISK11 results, eight had primary endpoint incident tuberculosis during 15 months of follow-up. Tuberculosis incidence was 2·5 per 100 person-years (95% CI 0·7-4·4) in the RISK11-positive group and 0·2 per 100 person-years (0·0-0·5) in the RISK11-negative group. The probability of primary endpoint incident tuberculosis was greater in the RISK11-positive group than in the RISK11-negative group (cumulative incidence ratio 16·0 [95% CI 2·0-129·5]). RISK11 had a prognostic AUC of 80·0% (95% CI 70·6-86·9), and a sensitivity of 88·6% (43·5-98·7) and a specificity of 68·9% (65·3-72·3) for incident tuberculosis at the 60% threshold.

Interpretation: RISK11 identified prevalent tuberculosis and predicted risk of progression to incident tuberculosis within 15 months in ambulant people living with HIV. RISK11's performance approached, but did not meet, WHO's target product profile benchmarks for screening and prognostic tests for tuberculosis.

Funding: Bill & Melinda Gates Foundation and the South African Medical Research Council.

Figure 1

Trial profile IGRA=interferon-γ release assay. LAM=lipoarabinomannan. MGIT=Mycobacteria Growth Indicator Tube. *Any medical, surgical, or other condition, including, but not limited to, known diabetes (requiring oral or injectable therapy), liver disease, or alcohol misuse disorder, that in the opinion of the investigator is likely to interfere with RISK11 performance; safety or efficacy of antiretroviral or isoniazid preventive therapy; or adherence to protocol requirements. †One enrolled participant without a PAXgene RNA sample and one with an indeterminate RISK11 result at baseline had primary endpoint prevalent tuberculosis. One participant with an indeterminate RISK11 result at baseline progressed to primary endpoint incident tuberculosis during follow-up. ‡Probably due to inadequate quality of the RNA sample. §Eight participants with one sputum sample-positive prevalent tuberculosis (three in the RISK11-negative group and five in the RISK11-positive group) were included in the primary endpoint (two or more positive samples) prognostic performance analysis, but excluded from the secondary endpoint (one or more positive samples) prognostic performance analysis. ¶Participants who did not complete follow-up per-protocol were included in the RISK11 and IGRA prognostic performance analysis but censored as non-tuberculosis controls at their last study visit.

Figure 2

RISK11 signature score distribution (A) RISK11 signature scores by symptom status. Box-and-whisker plots depicting RISK11 signature scores measured at enrolment (each dot represents a participant) in participants with symptomatic, clinical tuberculosis, asymptomatic, subclinical tuberculosis, or no tuberculosis. (B) RISK11 signature scores measured at enrolment by HIV plasma viral load (copies per mL). Prevalent and incident tuberculosis comprised all microbiologically confirmed secondary endpoint cases. Symptoms were recorded at the time of diagnosis for participants with prevalent and incident tuberculosis, and at enrolment for participants without tuberculosis. p values for comparison of median RISK11 signature scores between groups in box-and-whisker plots were calculated with the Mann-Whitney U test and corrected for multiple comparisons by use of the Benjamini-Hochberg Procedure. Boxes depict the IQR, the midline represents the median, and the whiskers indicate the IQR ± (1·5 × IQR). The dashed line depicts the a priori RISK11 score threshold (60%).

Figure 3

Prevalence and cumulative incidence of primary and secondary endpoint tuberculosis Prevalence and cumulative incidence of primary (A) and secondary (B) endpoint tuberculosis in RISK11-positive and RISK11-negative participants at study enrolment and over 15 months of follow-up. Error bars and shaded areas represent the 95% CIs.

Figure 4

RISK11 diagnostic and prognostic performance for primary and secondary endpoint tuberculosis Receiver operating characteristic curve depicting RISK11 diagnostic performance for the primary (A) and secondary (B) endpoint. The graph shows participants with symptomatic clinical prevalent tuberculosis versus symptomatic controls, and participants with asymptomatic subclinical prevalent tuberculosis versus asymptomatic controls. The shaded areas represent the 95% CIs. The solid box depicts the optimal criteria (95% sensitivity and 80% specificity) and the dashed box depicts the minimal criteria (90% sensitivity and 70% specificity) set out in WHO's target product profile for a triage test. Receiver operating characteristic curve depicting RISK11 prognostic performance for incident tuberculosis for the primary (C) and secondary (D) tuberculosis endpoints. The shaded areas represent the 95% CIs. The solid box depicts the optimal criteria (90% sensitivity and 90% specificity) and the dashed box depicts the minimal criteria (75% sensitivity and 75% specificity) set out in WHO's target product profile for an incipient tuberculosis test. AUC=area under the receiver operating characteristic curve.