Biomarker tests of progression from tuberculosis infection to disease: a systematic review

Abstract

Background

Tuberculosis (TB) remains a global health threat, with millions of new infections annually. While most infected individuals remain asymptomatic, a subset progresses to active disease. Effective tests predicting TB progression are urgently needed to enhance prevention strategies and reduce the global TB burden.

Objective

To evaluate biomarker-based tests designed to predict progression from TB infection to active disease.

Methods

We conducted a systematic literature search of four electronic databases for studies published between 2016, when the first trial evaluating biomarker-based tests for TB progression was published, and 2024. Quality assessment was performed using QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2). Data on study design, population, biomarkers, and diagnostic accuracy were extracted and synthesised narratively.

Results

19 studies were included, reporting 70 biomarker-based tests for predicting TB progression. These included 23 gene signatures, 14 protein signatures and 33 cytokines/chemokines. Heterogeneity in study populations, methodologies and outcome definitions precluded direct comparisons. Many studies lacked transparency in reporting key population characteristics, reference standards and diagnostic accuracy outcomes, limiting clinical applicability. Most tests demonstrated only moderate predictive accuracy, with no single biomarker approach emerging as a definitive tool for clinical use.

Conclusions

No biomarker-based test to predict TB progression is currently ready for clinical implementation. Standardised methodologies, larger validation studies and improved reporting transparency are necessary to advance TB diagnostics and improve early detection efforts. Future research should prioritise refining tests for TB progression by improving their predictive accuracy, identifying appropriate target populations and evaluating their cost-effectiveness and feasibility for integration into TB prevention programmes, particularly in high-burden settings.

Shareable abstract

Current biomarker-based tests for predicting tuberculosis progression show promise but lack consistency and clinical readiness. Standardised methodologies, transparent reporting and large-scale validation studies are crucial to advancing TB diagnostics. https://bit.ly/3NFuokV

Background

In 2024, tuberculosis (TB) stood once again as the leading cause of death from a single infectious agent globally, with 10.7 million new cases and 1.23 million fatalities annually [1]. Despite the alarming global burden of TB infection (TBI), with an estimated 1.7 billion individuals infected with Mycobacterium tuberculosis [2], of whom only a small percentage (5% to 15%) will develop active TB disease during their lifetime [3, 4]. Most people with TBI remain asymptomatic and do not progress to active disease – a condition previously referred to as “latent TB infection” [4]. However, for those who do progress to active TB, the consequences can be severe, leading to illness, disability and even death if not promptly diagnosed and treated. In recognition of the persistent threat posed by TB, the World Health Organization (WHO) launched the End TB Strategy in 2015, which aims to reduce TB incidence and mortality [2, 5, 6]. Additionally, WHO guidelines on TB preventive treatment (TPT) emphasise identifying and treating people who have TBI at highest risk of disease progression, such as individuals living with HIV, close contacts of TB cases and other vulnerable groups, given that the safety profile of current TPT regimens does not support indiscriminate treatment of all infected individuals [7]. In this context, biomarker-based tests that reliably predict progression could play a pivotal role in refining risk stratification and targeting preventive therapy to those most likely to benefit, thereby optimising both clinical outcomes and programmatic resources.

Accurate diagnosis of TBI is critical for TB prevention. However, currently available tests, tuberculin skin tests (TSTs) and interferon-gamma release assays (IGRAs), exhibit a remarkably low positive predictive value (PPV) (below 10%) for future disease development [8]. This limitation, coupled with the potential toxicity of preventive treatment regimens [9], restricts programmatic management of TBI to less than 2% of the overall reservoir [10]. To address this gap, research has increasingly focused on identifying biomarkers that can distinguish individuals with TBI who are likely to progress to active TB, in the absence of clinical signs/symptoms or microbiological evidence of disease. Biomarker-based tests hold promise for improved risk stratification and more targeted delivery of TPT [11–13]. In 2016, Zak et al. [14] described the first prototype of a biomarker-based test for TB progression, namely a whole blood RNA signature including 16 genes. The authors demonstrated that this signature predicted TB progression with 66.1% sensitivity and 80.6% specificity [14], a performance very close to the characteristics of a WHO-recommended target product profile for a test of progression [15]. Despite nearly a decade having passed since the publication of this landmark study [14], none of the numerous candidate tests developed since then have reached the market. The inadequacies of existing immunological tests, combined with the absence of commercially available biomarker-based tools, underscores a critical gap in this field.

In recent years, several reviews have explored the diagnostic and predictive capabilities of various biomarker-based tests for TB. However, most have focused on detecting active TB rather than predicting progression from TBI to active disease [16]. While a few studies have attempted to synthesise evidence on the prediction of progression to active TB, the available literature remains limited. For instance, in 2020, Mulenga et al. [17] conducted a systematic review comparing the performance of host blood transcriptomic signatures for diagnosing and predicting progression to TB disease in HIV-negative adults and adolescents. Their analysis included 20 studies evaluating 25 signatures for diagnosis or prediction of progression. However, the authors concluded that the evidence was insufficient to determine if host blood mRNA signatures could be used as standalone diagnostic or predictive tests [17]. Since the publication of this review over 5 years ago, additional evidence has emerged regarding biomarker-based tests for predicting TB progression.

We conducted a systematic review of the literature published since 2016 to 1) identify and describe biomarker-based tests with potential for market entry and 2) evaluate the performance and limitations of the identified tests in predicting progression from TBI to active TB.

Methods

Protocol and registration

The protocol for this systematic review was registered in the International Prospective Register of Systematic Reviews (PROSPERO ID CRD42024554184) and adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Diagnostic Test Accuracy Studies (PRISMA-DTA) guidelines [18]. A completed PRISMA checklist is provided in appendix 1.

Search strategy

An experienced information specialist (B. Skidmore) developed the search strategy in collaboration with the research team. The Medline strategy was peer-reviewed by a senior information specialist (K. Campbell) using the PRESS Checklist [19]. Searches were conducted across multiple databases, including Ovid Medline® ALL, Embase Classic+Embase, Cochrane Central Register of Controlled Trials and Web of Science (core collection). Search terms included controlled vocabulary such as “latent tuberculosis”, “immunologic tests” and “predictive value of tests”, alongside relevant free text such as “LTBI”, “biomarker” and “forecast.” Animal-only records and case studies were excluded. There were no language restrictions. Searches were restricted to publications from 1 January 2016 to 6 April 2024. We applied this date restriction because the first study unveiling a prototype test for predicting TB progression dates to approximately 9 years ago [14]. EndNote (version 9.3.3, Clarivate Analytics) was used for deduplication and records were managed and screened using Covidence (Veritas Health Innovation Ltd.). The full search strategy is provided in appendix 2.

In addition to bibliographic database searches, we hand-searched reference lists of included studies and reviewed grey literature by searching clinical trial registries (e.g. ClinicalTrials.gov) and abstracts from various relevant conferences (e.g. the World Union Conference) from 2020 to 2024. We also explored selected Grey Matters [20] sources to ensure broad coverage of relevant literature.

Eligibility criteria

The review included experimental and observational studies reporting on the features, performance and challenges of biomarker-based tests aimed at evaluating the risk of progression from TBI to active TB. Progression from TBI to active TB was defined as the transition from an asymptomatic infection, where the bacteria are present but inactive in the body and undetectable through conventional microbiological tests, to an active disease state characterised by clinical symptoms, bacterial replication, positive microbiological tests and the potential to transmit the infection to others. This progression is often confirmed through clinical, microbiological or radiological evidence of active TB, such as positive sputum culture, chest radiograph findings consistent with TB or other diagnostic tests demonstrating active disease (e.g. Xpert MTB/Rif or Ultra).

Eligible study designs were randomised and nonrandomised trials, cohort, case–control, and cross-sectional studies. Reviews, commentaries and perspective pieces discussing or reporting on the challenges encountered in the development, evaluation or future implementation of one or more eligible tests were retained to gather further contextual information but were not included in the results of this review. Protocols, economic analyses, modelling studies and case reports were excluded.

The inclusion and exclusion criteria according to the PICO (population, intervention, comparison, outcome) framework were as follows:

• Population: no restrictions were applied based on population demographics.

• Intervention: biomarker-based or immunological tests aimed at predicting progression to active TB disease.

• Comparison: no exclusion criteria were applied regarding comparators, recognising the absence of standard reference tests for predicting progression to active TB disease.

• Outcomes: the primary outcomes included diagnostic accuracy measures, such as sensitivity, specificity, PPV, negative predictive value (NPV) and area-under-the-curve (AUC), for biomarker-based tests predicting progression to active TB.

Screening and selection process

Records retrieved from database searches were imported into Covidence, while grey literature was catalogued separately. Screening involved an initial review of titles and abstracts, followed by full-text evaluation conducted independently by two reviewers among K. Tingley, A. Li, N. Maredia and R. Ravikumar. To expedite the title and abstract screening process, a liberal accelerated approach [21] was employed, whereby any citation marked as “include” or “unsure” by either reviewer advanced to full-text screening. Pilot testing for title and abstract screening was conducted using 25 randomly selected titles and abstracts to ensure high inter-rater reliability. Similarly, full-text screening was piloted on 10 randomly selected full-text articles to maintain reliability. Discrepancies between reviewers were resolved through discussion or, if needed, adjudication by a senior team member. All full-text exclusions were systematically documented and justified to ensure reproducibility.

Assessment of methodological quality

The methodological quality of included studies was evaluated using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool [22]. One reviewer performed the quality assessment, and a second reviewer verified the results. We conducted more than one QUADAS-2 evaluation if different reference standards or distinct patient selection procedures were used for the same test within the same study. Assessments were summarised graphically by domain (patient selection, index test, reference standard, flow and timing). Any disagreements were resolved through discussion or by a third reviewer.

Data extraction

A single reviewer among K. Tingley, A. Li, N. Maredia and R. Ravikumar extracted data for each study using a customised form built in Airtable (Formagrid Inc.), an online collaborative spreadsheeting platform and tailored to the specific objectives of the review. A second reviewer verified all extracted data. The form was pilot tested on four randomly selected studies ensuring its consistency and reliability. Extracted data included publication details, study setting, design, sample size and participant characteristics, alongside details on diagnostic criteria, test types and performance measures. Contextual information regarding test development and implementation challenges was also gathered. Appendix 3 includes the data extraction form.

Data synthesis

Although a meta-analysis was initially planned, the studies included in this review demonstrated substantial clinical and methodological heterogeneity, as well as incomplete and nontransparent reporting, which precluded data pooling. We therefore conducted a descriptive and narrative synthesis to summarise the key findings.

For the descriptive synthesis, data were extracted on the main study characteristics and diagnostic performance measures as described above. Studies were grouped by biomarker test type, with findings on sensitivity, specificity, PPV, NPV and AUC where available. Due to heterogeneity in study design, patient populations, and methodologies, direct comparisons were not feasible; however, we organised data into tables to illustrate the performance range of each test.

Given the variability and limited transparency in reporting, we did not evaluate the biomarker tests against the ASSURED criteria [23] or WHO target product profiles [24]. Nonetheless, the narrative synthesis provides insights into the strengths and limitations of identified tests and highlights key evidence gaps and areas for further research.

Results

Study selection

A total of 5582 studies were identified from electronic database searches, with an additional 28 studies retrieved from grey literature sources (figure 1). After removing duplicates, 5517 articles remained for title and abstract screening. Of these, 5339 studies were determined to be irrelevant to our review questions, leaving 178 studies for full-text review. During the full-text screening, 159 studies were excluded for specific reasons: 48 lacked relevant outcome data, 76 focused on the wrong stage of the TB spectrum (e.g. diagnosis of TBI) or on discrimination of TB stages rather than progression to active TB and 25 were excluded due to ineligible study design (figure 1). Ultimately, 19 studies met all eligibility criteria and were included in the review. Of these, two studies [25, 26] were excluded from the quantitative data synthesis due to incomplete reporting of key test characteristics. A complete list of excluded studies with reasons for exclusion is provided in appendix 4.

FIGURE 1.

Study selection. TB: tuberculosis.

Study characteristics

Our review included 19 studies, the majority of which (16/19, 85%) were published after 2019 (table 1). Nearly half of the corresponding authors were based in the USA (9/19, 47%), followed by South Africa (2/19, 11%), China (2/19, 11%), the United Kingdom (2/19, 11%) and a handful of other countries (table 1). Most studies (16/19, 84%) were supported by nonindustry funding sources. Reported conflicts of interest were relatively evenly split, with 42% (8/19) of studies disclosing potential conflicts (table 1).

TABLE 1.

Descriptive characteristics of included studies (n=19)

Study characteristic	Frequency
	n	%
Year of publication
2016–2018	3	16
2019–2021	10	53
2022–2024	6	32
Country of corresponding author
China	2	11
Germany	1	5
India	1	5
Korea	1	5
Mexico	1	5
South Africa	2	11
UK	9	47
USA	2	11
Study sponsor#
Industry	1	5
Nonindustry	16	84
Not reported	2	11
Conflicts of interest
Authors report conflicts	8	42
Authors report no conflicts	11	58
Study design
Case–control study	3	16
Cohort study	13	68
Cross-sectional study	1	5
Other (e.g. nested case–control study, quasi-experimental study)	2	11
Data/sample collection
Prospective	12	63
Retrospective	6	32
Not reported	1	5
Sampling method
Consecutive	3	16
Convenient	1	5
Random	1	5
Not reported	14	74
Sample size
50–99	4	21
100–499	5	26
500–999	2	11
≥1000	7	37
Not reported	1	5
Country for participant recruitment#
Single country	11	58
Multi-country	8	42
Brazil	3	16
Haiti	1	5
India	1	5
UK	3	16
China	3	16
South Africa	10	53
Gambia	5	26
Ethiopia	3	16
Moldova	1	5
Uganda	2	11
USA	2	11
Venezuela	1	5
Study population#
Confirmed TB	8	42
Healthy controls	8	42
LTBI	12	63
Household contacts of TB cases	7	37
Other special population (e.g. HIV co-infection)	2	11
Not reported	2	11
Number of biomarker tests per study
1	7	37
2	5	26
3	1	5
4+	6	32
Types of test included#
Host: RNA (e.g. transcriptomic signatures)	11	58
Host: proteins	5	26
Pathogen: cytokines/chemokines	3	16

LTBI: latent tuberculosis infection; TB: tuberculosis. ^#: Categories are not mutually exclusive.

Cohort studies were the predominant study design (13/19, 68%). Sample collection was prospective in most studies (12/19, 63%), though sampling methods were frequently not reported, with only 26% specifying methods like consecutive or random sampling (table 1). Study sample sizes varied substantially, with 7/19 (37%) including over 1000 participants. Studies were conducted in both single-country (11/19, 58%) and multi-country settings (8/19, 42%), with common recruitment sites including South Africa (10/19, 53%), Gambia (5/19, 26%) and the UK (3/19, 16%) (table 1). Study populations typically included confirmed TB cases (8/19, 42%) and individuals with TBI (12/19, 63%) (table 1). Studies often reported on more than one unique biomarker-based test (12/19, 63%) (table 1).

Quality of included studies

We assessed the quality of 17 out of 19 eligible studies. The two studies excluded from data synthesis due to incomplete reporting on key test characteristics were also excluded from quality assessment. For two of the 17 assessed studies, we conducted more than one QUADAS-2 evaluation, as each study used different reference standards or had distinct patient selection procedures within its design warranting separate quality assessments. A total of 19 quality assessments were completed.

The QUADAS-2 assessment identified notable concerns regarding both bias and applicability, largely due to gaps in transparency and reporting. The risk of bias was most frequently rated as “unclear” across the domains of patient selection, index test, and flow and timing, affecting 13 out of 19 assessments for patient selection and nine for index test blinding (figure 2). This lack of detail, especially regarding recruitment methods, blinding procedures and participant follow-up, limited our ability to fully assess potential sources of bias in many studies. Only a minority of studies were rated as having a “low” risk of bias across all domains, indicating that overall methodological transparency was insufficient in much of the sample (figure 2).

FIGURE 2.

Risk of bias assessment using QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies 2).

In terms of applicability, the risk was generally low, with most studies using appropriate patient populations, index tests and reference standards for TB diagnostic and prognostic evaluation (figure 2). However, a few studies received “unclear” ratings, particularly concerning reference standards, due to either variations in protocols or limited information on the test's clinical applicability (figure 2). These “unclear” ratings underscore a common challenge across studies included in the systematic review, namely insufficient detail that hinders a robust assessment of study rigor and real-world relevance. The full QUADAS-2 assessment for each study is provided in appendix 5.

Performance of biomarker-based tests

Our review identified 64 unique biomarker-based tests for predicting progression to active TB, including 23 gene signatures (table 2), 14 protein signatures (table 3) and 33 cytokines/chemokines (table 4).

TABLE 2.

Summary of diagnostic performance of host RNA biomarkers for predicting progression to active tuberculosis (ATB) (n=10 studies)

First author [ref.], year	Index test name	Reference standard used	Group 1	Group 2	Timing	Sensitivity % (95% CI)	Specificity % (95% CI)	PPV % (95% CI)	NPV % (95% CI)	AUC (95% CI)
Vargas [27] 2023	Gene signature (comprising 42 genes)#	NR	Progressors (n=NR) No group characteristics reported	Nonprogressors (n=NR) No group characteristics reported	<3 months to disease	87.3 (81.5–93.1)	91.7 (89.9–93.5)	17.7 (14.5–20.9)	99.7 (99.6–99.8)	NR
					<12 months to disease	78.6 (73.4–83.9)	80.3 (77.7–83.0)	7.5 (6.6–8.5)	99.5 (99.4–99.5)	NR
					<18 months to disease	77.3 (72.3–82.3)	78.3 (75.6–81.0)	6.8 (6.0–7.6)	99.4 (99.3–99.5)	NR
					<24 months to disease	74.7 (69.7–79.6)	78.3 (75.6–81.0)	6.6 (5.8–7.3)	99.3 (99.3–99.4)	NR
					<6 months to disease	82.1 (76.4–87.8)	83.8 (81.3–86.2)	9.3 (8.1–10.6)	99.6 (99.5–99.7)	NR
					<30 months to disease	74.2 (69.3–79.2)	78.3 (75.6–81.0)	6.5 (5.8–7.3)	99.3 (99.3–99.4)	NR
	Gene signature (comprising 18 genes)¶	NR	Progressors (n=NR) No group characteristics reported	Progressors (n=NR) No group characteristics reported	<3 months to disease	83.3 (76.8–89.8)	91.6 (89.8–93.4)	16.8 (13.8–19.9)	99.6 (99.5–99.7)	NR
					<6 months to disease	74.0 (67.5–80.5)	86.9 (84.7–89.2)	10.4 (8.8–11.9)	99.4 (99.3–99.5)	NR
					<12 months to disease	74.8 (69.2–80.4)	81.0 (78.4–83.6)	7.4 (6.5–8.4)	99.4 (99.3–99.5)	NR
					<18 months to disease	78.4 (73.5–83.4)	74.9 (72.0–77.8)	6.0 (5.3–6.6)	99.4 (99.3–99.5)	NR
					<24 months to disease	77.1 (72.2–81.9)	74.9 (72.0–77.8)	5.9 (5.3–6.5)	99.4 (99.3–99.5)	NR
					<30 months to disease	77.3 (72.5–82.0)	74.9 (72.0–77.8)	5.9 (5.3–6.5)	99.4 (99.3–99.5)	NR
Xin [28] 2022	hsa-miR-16-5p	NR	ATB (n=73) More than 50% ≥60 years old Predominately male	LTBI (n=77) More than 50% ≥60 years old Predominately male Higher BMI than ATB group	Followed-up to 5 years	68.49 (57.14–78.00)	58.44 (47.29–68.79)	NR	NR	NR
	hsa-miR-16-5p+co-variables					73.97 (62.89–82.66)	90.91 (82.40–95.53)	NR	NR	NR
	hsa-miR-16-5p+hsa-miR-451a					38.36 (28.05–49.83)	90.91 (82.40–95.53)	NR	NR	NR
	hsa-miR-451a					38.36 (28.05–49.83)	90.91 (82.40–95.53)	NR	NR	NR
	hsa-miR-451a+co-variables					80.82 (70.34–88.22)	79.22 (68.88–86.78)	NR	NR	NR
	hsa-miR-451a+hsa-miR-16-5p+co-variables					76.71 (65.83–84.92)	83.12 (73.23–89.86)	NR	NR	NR
Long [29] 2021	Gene signature (comprising 10 genes)+	NR	Nonprogressor (within 0–1 year) (n=NR) No group characteristics reported	Progressor (within 0–1 year) (n=NR) No group characteristics reported	At exposure	NR	NR	NR	NR	0.66
					6 months after exposure	NR	NR	NR	NR	0.72
			Nonprogressor (within 1–2 years) (n=NR) No group characteristics reported	Progressor (within 1–2 years) (n=NR) No group characteristics reported	At exposure	NR	NR	NR	NR	0.48
					6 months after exposure	NR	NR	NR	NR	0.52
					18 months after exposure	NR	NR	NR	NR	0.81
de Araujo [35] 2021	NPC2	Smear microscopy, culture (nonspecified) and/or Xpert	HHC (nonprogressors) (n=208) No group characteristics reported	ATB (progressors) (n=13) No group characteristics reported	Time to TB ≤3 months	92.3 (64–99.8)	75 (68.5–80.7)	NR	NR	NR
				ATB (progressors) (n=34) No group characteristics reported	Time to TB 4–6 months	50 (32.4–67.6)	75 (68.5–80.7)	NR	NR	NR
				ATB (progressors) (n=19) No group characteristics reported	Time to TB 7–12 months	63.2 (38.4–83.7)	75 (68.5–80.7)	NR	NR	NR
				ATB (progressors) (n=32) No group characteristics reported	Time to TB 13–18 months	34.4 (18.6–53.2)	75 (68.5–80.7)	NR	NR	NR
Roe [30] 2020	Three-gene SVM model (BATF2, GBP5, SCARF1) BATF2	Culture (nonspecified), chest radiography	Progressors (n=12)	Nonprogressors (n=48)	Time to TB within 90 days	0.83 (0.52–0.98)	0.96 (0.83–0.99)	23 (NR)	NR	0.96 (0.92–1)
					Time to TB within 90 days	0.83 (0.52–0.98)	0.92 (0.83–0.99)	13	NR	0.93 (0.86–1)
			Progressors (n=29)		Time to TB 91–360 days	0.52 (0.36–0.74)	NR	NR	NR	NR
			Progressors (n=25)		Time to TB >360 days	0.24 (0.12–0.49)	NR	NR	NR	NR
Penn-Nicholson [31] 2020	RISK6	NR	Progressor (ACS cohort) (n=46)	Nonprogressors (ACS cohort) (n=284)	0–12 months before TB diagnosis	82.8 (NR)	82.8 (NR)	NR	NR	87.6% (82.8–92.4)
					12–24 months before TB diagnosis	NR	NR	NR	NR	74.0% (66.0–82.0)
			Progressors (GC6-74 cohort) (n=50)	Non-progressors (GC6-74 cohort) (n=372)	0–12 months before TB diagnosis	50.0 (NR)	56.0 (NR)	NR	NR	70.6% (61.6–79.5)
					12–24 months before TB diagnosis	NR	NR	NR	NR	67.6% (58.2–76.9)
Leong [32] 2020	ACS-CoR	Clinical signs, culture (nonspecified), chest radiography	Progressors (n=11) Nine of the 11 progressors were culture-proven TB Predominately female (11/16) Median age: 22	Nonprogressors (n=21) Age- and sex- matched	Up to 5 years: five were diagnosed within 2 years (early progressors) and six were diagnosed after 2 years (late progressors) of sample collection	51.5 (46.0–57.1)	77.4 (72.8–82.0)	10.6 (5.7–16.5)	96.8 (94.6–98.2)	0.670 (0.640–0.700)
	Jacobsen3					55.3 (48.9–61.6)	63.3 (57.0–69.5)	7.2 (3.9–11.3)	96.4 (93.8–98.0)	0.575 (0.546–0.605)
	Kaforou					66.4 (60.3–72.5)	56.7 (51.3–62.0)	7.3 (4.6–11.5)	97.0 (94.2–98.5)	0.629 (0.602–0.656)
	PREDICT29					74.2 (70.4–78.0)	84.8 (81.6–88.0)	20.2 (13.1–29.4)	98.5 (96.9–99.3)	0.911 (0.894–0.928)
	RISK4					41.3 (33.5–49.1)	66.5 (58.8–74.3)	6.1 (3.5–11.4)	95.6 (93.3–97.6)	0.461 (0.434–0.488)
	Sambarey10					63.2 (57.1–69.3)	58.6 (52.5–64.7)	7.3 (4.5–11.6)	96.8 (94.4–98.5)	0.623 (0.595–0.651)
	Sweeney3					42.7 (35.3–50.1)	74.6 (67.6–81.6)	8.0 (4.2–13.6)	96.1 (93.9–97.6)	0.590 (0.560–0.620)
Warsinske [33] 2018	Three-gene TB score	Smear microscopy, culture (nonspecified)	Progressors (ATB) (n=43) Adolescents: 12–18 years old	Nonprogressors (LTBI) (n=101) Adolescents: 12–18 years old	0–7 days	89.47 (NR)	63.37 (NR)	NR	99.13§ (NR)	0.86 (0.77–0.96)
					8–180 days	86 (NR)	84 (NR)	NR	98.63 (NR)§	0.86 (0.70–1.00)
					181–360 days	NR	NR	NR	NR	0.62 (0.55–0.69)
					361–540 days	NR	NR	NR	NR	0.86 (0.49–0.65)
					541–720 days	NR	NR	NR	NR	0.86 (0.51–0.75)
Zak [14] 2016	Zak16	Culture (nonspecified), smear microscopy, clinical signs	Progressors (n=46)	Healthy controls (n=107)	1–360 days before TB	66.1 (63.2–68.9)	80.0 (78.6–81.4)	NR	NR	0.779 (0.761–0.798)
			Progressors (n=73)	Nonprogressors (n=301)		53.7 (42.6–64.3)	82.8 (78.7–86)	NR	NR	0.718 (0.637–0.800)
Gupta [34] 2022	Nine-gene signature	NR	Progressor (n=NR)	Nonprogressor (n=NR)	Up to 1 year prior to sputum conversion	76 (NR)	83 (NR)	NR	NR	NR
					8–180 days prior to sputum conversion	NR	NR	NR	NR	0.934 (0.841–1)
					181–360 days prior to sputum conversion	NR	NR	NR	NR	0.841 (0.734–0.949)
					361–540 days prior to sputum conversion	NR	NR	NR	NR	0.59 (0.418–0.761)
					541–720 days prior to sputum conversion	NR	NR	NR	NR	0.44 (0.213–0.667)

ACS: Adolescent Cohort Study; AUC: area under the curve; BATF2: basic leucine zipper ATF-like transcription factor 2; BMI: body mass index; GBP5: guanylate binding protein 5; GC: Grand Challenges; HHC: household contacts; hsa-miR: Homo sapiens microRNA; LTBI: latent tuberculosis infection; NPC2: Niemann–Pick type C2; NPV: negative predictive value; NR: not reported; PPV: positive predictive value; SCARF1: scavenger receptor class F member 1; SVM: support vector machine; TB: tuberculosis. ^#: ADM, AIM2, ANKRD22, APOL6, BATF2, C1QB, CASP5, CD19, CD274, CD5, DUSP3, GBP1, GBP4, GBP5, FBXO6, FCGR1B, GK, HP, IFIT2, IFIT3, JAK2, KCNJ15, LAP3, LHFPL2, LMNB1, LRRK2, LY96, MAPK14, NELL2, P2RY14, PSTPIP2, RSAD2, RTP4, SAMD9L, SERPING1, SLC6A12, SPOCK2, STAT1, TIMM10, TLR5, VAMP5 and ZNF43. ^¶: ADM, ANKRD22, APOL6, BATF2, CD274, CD5, DUSP3, GBP4, GBP5, FBXO6, FCGR1B, LMNB1, LRRK2, IFIT2, NELL2, SPOCK2, VAMP5 and ZNF438. ⁺: CD274, KIF1B, IL15, TLR1, TLR5, FCGR1A, GBP1, NOD2, GBP2 and EGF. ^§: At 4% prevalence.

TABLE 3.

Summary of diagnostic performance of host protein biomarkers for predicting progression to active tuberculosis (ATB) (n=4 studies)

First author [ref.], year	Index test name	Reference standard used	Group 1	Group 2	Timing	Sensitivity % (95% CI)	Specificity % (95% CI)	PPV % (95% CI)	NPV % (95% CI)	AUC (95% CI)
Burel [36] 2024	IgG glycosylation profiling	No reference standard used	LTBI-Risk (as determined by Zak16) (n=10)	LTBI-Other (n=46)	Pre-treatment	NR	NR	NR	NR	0.71
	AbGlyc score		ATB (n=19) Mean age 33 years 79% male	LTBI-R (n=10) Mean age 29 years 54% male	Pre-treatment	NR	NR	NR	NR	0.82
Araujo [37] 2018	p-29878/MMP9	Chest radiography	Incipient TB (n=28)	Healthy controls (n=14)	NR	61.54 (31.25–91.83)	50.00 (14.01–85.99)	61.55 (31.25–91.83)	50.00 (14.01–85.99)	NR
	p-29878/uPAR				NR	46.43 (26.17–66.69)	86.67 (66.13–100.0)	86.67 (66.13–100.0)	46.43 (26.17–66.69)	NR
Singer [39] 2023	Eight-protein signature (comprising CD14, CD166, DSG2, LAMP1, LRP1, NCAM2, R4RL2, VASN)	NR	ATB (n=24) Mean age 38 years USA cohort	Non-TB (n=48) Mean age 39 years	0–6 months prior to TB diagnosis	NR	NR	NR	NR	0.74
	12-protein signature (comprising AMPN, CA2D1, CBPQ, CNTN1, COL11, DPP4, HYOU1, ICAM1, LAMP1, LUM, PCOC1, PIGR)				18–24 months prior to TB diagnosis	NR	NR	NR	NR	0.8
	Seven-protein signature (comprising APOA1, APOA4, GP1BA, LUM, NID1, PLSL, SCTM1)				6–12 months prior to TB diagnosis	NR	NR	NR	NR	0.76
	Seven-protein signature (comprising APOA4, CATA, COL11, LUM, MEGF8, PLSL, SCTM1)				12–18 months prior to TB diagnosis	NR	NR	NR	NR	0.84
	Five-protein signature (comprising CD248, HYOU1, NID1, PON1, VCAM1)	Culture (nonspecified), chest radiography, smear microscopy	ATB (n=30) Females 73% South African cohort	Non-TB (n=62) Female 76%	0–6 months prior to TB diagnosis	NR	NR	NR	NR	0.93
	Six-protein signature (comprising CD248, HYOU1, NID1, PON1, VCAM1, APOA1)				6–12 months prior to TB diagnosis	NR	NR	NR	NR	0.86
	Eight-protein signature (comprising APOA4, CBPN, CD14, CPN2, LCAT, LUM, LYAM1, PNPH)				12–18 months before diagnosis	NR	NR	NR	NR	0.75
	11 protein signature (comprising CD248, CSTN1, FUCO2, HYOU1, MINP1, MMP2, MYOC, NID1, PON1, VASN, VCAM1)				18–24 months prior to TB diagnosis	NR	NR	NR	NR	0.87
Penn-Nicholson [38] 2019	3PR	Smear microscopy, culture (nonspecified)	Progressor (ACS training+test cohort) (n=37)	Nonprogressor (ACS training+test cohort) (n=106)	Time to TB 0–12 months	75 (61.05–85.97)	70.69 (65.09–75.87)	NR	NR	0.80 (0.74–0.86)
					Time to TB 0–6 months	90.91 (70.84–98.88)	70.69 (65.09–75.87)	NR	NR	0.89 (0.84–0.95)
					Time to TB 7–12 months	63.33 (43.86–80.07)	70.69 (65.09–75.87)	NR	NR	0.72 (0.64–0.81)
					Time to TB 13–24 months	57.14 (37.18–75.54)	70.69 (65.09–75.87)	NR	NR	0.71 (0.63–0.80)
		Culture (nonspecified), smear microscopy, chest radiography, clinical signs	Progressor (GC6 validation cohort) (n=34)	Nonprogressor (GC6 validation cohort) (n=115)	Time to TB 0–12 months	46.34 (30.66–62.58)	75.0 (68.2–80.96)	NR	NR	0.65 (0.55–0.75)
					Time to TB 0–6 months	47.83 (26.82–69.41)	75.0 (68.26–80.96)	NR	NR	0.64 (0.50–0.78)
					Time to TB 7–12 months	44.44 (21.53–69.24)	75.0 (68.26–80.96)	NR	NR	0.67 (0.55–0.79)
					Time to TB 13–24 months	42.11 (20.25–66.5)	75.0 (68.26–80.96)	NR	NR	0.62 (0.48–0.76)
	TRM5	Smear microscopy, culture (nonspecified)	Progressor (ACS training cohort) (n=24)	Nonprogressor (ACS training cohort) (n=70)	Time to TB 0–12 months	75.76 (57.74–88.91)	75.0 (67.22–81.75)	NR	NR	0.84 (0.75–0.92)
					Time to TB 0–6 months	100 (73.54–100)	75.0 (67.22–81.75)	NR	NR	0.96 (0.93–0.99)
					Time to TB 7–12 months	61.9 (38.44–81.89)	75.0 (67.22–81.75)	NR	NR	0.76 (0.65–0.87)
					Time to TB 13–24 months	37.5 (18.8–59.41)	70.27 (62.21–77.5)	NR	NR	0.55 (0.41–0.69)
		Smear microscopy, culture (nonspecified)	Progressor (ACS test cohort) (n=13)	Nonprogressor (ACS test cohort) (n=36)	Time to TB 0–12 months	78.95 (54.43–93.95)	75.27 (65.24–83.63)	NR	NR	0.80 (0.70–0.89)
					Time to TB 0–6 months	90 (55.5–99.75)	75.27 (65.24–83.63)	NR	NR	0.85 (0.75–0.96)
					Time to TB 7–12 months	66.67 (29.93–92.5)	75.27 (65.24–83.63)	NR	NR	0.73 (0.62–0.85)
					Time to TB 13–24 months	37.5 (8.52–75.51)	75.27 (65.24–83.63)	NR	NR	0.69 (0.53–0.84)
		Culture (nonspecified), smear microscopy, chest radiography, clinical signs	Progressor (GC6 validation cohort) (n=34)	Nonprogressor (GC6 validation cohort) (n=115)	Time to TB 0–12 months	48.78 (32.88–64.87)	75.0 (68.26–80.96)	NR	NR	0.66 (0.56–0.75)
					Time to TB 0–6 months	60.87 (38.54–80.29)	75.0 (68.26–80.96)	NR	NR	0.69 (0.57–0.82)
					Time to TB 7–12 months	33.33 (13.34–59.01)	75.0 (68.26–80.96)	NR	NR	0.61 (0.48–0.74)
					Time to TB 13–24 months	36.84 (16.29–61.64)	75.0 (68.26–80.96)	NR	NR	0.61 (0.47–0.75)

3PR: 3-protein pair-ratio; AbGlyc: antibody glycosylation; ACS: Adolescent Cohort Study; AMPN: aminopeptidase N; APOA1: apolipoprotein A-I; APOA4: apolipoprotein A-IV; AUC: area under the curve; CA2D1: calcium voltage-gated channel subunit alpha-2/delta-1; CATA: catalase; CBPN: carboxypeptidase N catalytic chain; CBPQ: carboxypeptidase Q; CD: cluster of differentiation; CNTN1: contactin 1; COL11: collectin-II; CoR: Correlation of Risk; CPN2: carboxypeptidase N subunit 2; CSTN1: calsyntenin 1; DSG2: desmoglein-2; DPP4: dipeptidyl peptidase-4; FUCO2: plasma alpha-L-fucosidase; GC6: Grand Challenges 6; GP1BA: platelet glycoprotein Ib alpha chain; HHC: household contacts; HYOU1: hypoxia upregulated protein 1; ICAM1: intercellular adhesion molecule 1; LAMP1: lysosomal associated membrane protein 1; LCAT: lecithin-cholesterol acyltransferase; LRP1: prolow-density lipoprotein receptor related protein 1; LTBI: latent tuberculosis infection; LUM: lumican; LYAM1: L-selectin; MEGF8: multiple epidermal growth factor like domains protein 8; MINP1: multiple inositol polyphosphate phosphotase 1; MMP2: matrix metalloproteinase-2; MYOC: myocilin; NCAM2: neural cell adhesion molecule 2; NID1: nidogen 1; NPV: negative predictive value; NR: not reported; PCOC1: procollagen C-endopeptidase enhancer 1; PIGR: polymeric immunoglobulin receptor; PLSL: plastin-2; PNPH: purine nucleoside phosphorylase; PON1: serum paraoxonase/arylesterase 1; PPV: positive predictive value; R4RL2: reticulon-4 receptor-like 2; SCTM1: secreted and transmembrane protein 1; TB: tuberculosis; TRM5: TB Risk Model 5; uPAR: urokinase-type plasminogen activator receptor; VASN: vasorin; VCAM1: vascular cell adhesion molecule 1.

TABLE 4.

Summary of diagnostic performance of pathogen cytokines or chemokines biomarkers for predicting progression to active tuberculosis (ATB) (n=3 studies)

First author [ref.], year	Index test name	Reference standard used	Group 1	Group 2	Timing	Sensitivity % (95% CI)	Specificity % (95% CI)	PPV % (95% CI)	NPV % (95% CI)	AUC (95% CI)
Daniel [40] 2023	CCL11	Chest radiography, liquid culture, smear microscopy, TST, IGRA	HHC (progressors to ATB) (n=14)	HHC (nonprogressors) (n=20)	2-year follow-up	100 (85.8–100)	75 (50.9–91.3)	82.8 (64.2–94.2)	100 (78.2–100)	88.6 (75.4–96.2) cutoff# >204.20
	CCL19					100 (85.8–100)	100 (83.2–100)	100 (85.8–100)	100 (83.2–100)	100 (92–100) cutoff# <628.98
	CCL3					91.7 (73–99)	100 (83.2–100)	100 (84.6–100)	90.9 (70.8–98.9)	95.5 (84.5–99.4) cutoff# >234.20
	CCL4					66.7 (44.7–84.4)	80 (56.3–94.3)	80 (56.3–94.3)	66.7 (44.7–84.4)	72.7 (57.2–85) cutoff# >862.18
	GM-CSF					91.7 (73–99)	90 (68.3–98.8)	91.7 (73–99)	90 (68.3–98.8)	90.9 (78.3–97.5) cutoff# >128.73
	IFN-α					79.2 (57.8–92.9)	80 (56.3–94.3)	82.6 (61.2–95)	76.2 (52.8–91.8)	79.5 (64.7–90.2) cutoff# >326.45
	IFN-β					79.2 (57.8–92.9)	100 (83.2–100)	100 (82.4–100)	80 (59.3–93.2)	88.6 (75.4–96.2) cutoff# >36.76
	IFN-γ					100 (85.8–100)	90 (68.3–98.8)	92.3 (74.9–99.1)	100 (81.5–100)	95.5 (84.5–99.4) cutoff# <491.20
	IL-1ra					100 (85.8–100)	85 (62.1–96.8)	88.9 (70.8–97.6)	100 (80.5–100)	93.2 (81.3–98.6) cutoff# <2405.48
	IL-1α					79.2 (57.8–92.9)	100 (83.2–100)	100 (82.4–100)	80 (59.3–93.2)	88.6 (75.4–96.2) cutoff# <305.80
	IL-2					62.5 (40.6–81.2)	95 (75.1–99.9)	93.8 (69.8–99.8)	67.9 (47.6–84.1)	77.3 (62.2–88.5) cutoff# <180.94
	IL-4					79.2 (57.8–92.9)	80 (56.3–94.3)	82.6 (61.2–95)	76.2 (52.8–91.8)	79.5 (64.7–90.2) cutoff# >141.14
	IL-5					83.3 (62.6–95.3)	80 (56.3–94.3)	83.3 (62.6–95.3)	80 (56.3–94.3)	81.8 (67.3–91.8) cutoff# >75.29
	IL-6					50 (29.1–70.9)	100 (83.2–100)	100 (73.5–100)	62.5 (43.7–78.9)	72.7 (57.2–85) cutoff# >227.00
	IP-10					100 (85.8–100)	100 (83.2–100)	100 (85.8–100)	100 (83.2–100)	100 (92–100) cutoff# >226.06
	QuantiFERON supernatants (IP10/CCL19)				<6 months	100 (73.5–100)	100 (83.2–100)	100 (73.5–100)	100 (83.2–100)	100 (89.1–100) cutoff >0.24
	QuantiFERON supernatants (IP10/CCL19)	TST, IGRA, liquid culture, smear microscopy, chest radiography			>6 months	100 (73.5–100)	100 (83.2–100)	100 (73.5–100)	100 (83.2–100)	100 (89.1–100) cutoff >0.24
Robison [41] 2021	Threshold-based reduced random forest feature analysis for the high-risk clinical designation – 14 features	IGRA, other	High risk of TB reactivation (n=24) Mean±sd age 45.7±18.6 54.2% females	Unexposed control group (n=23) mean±sd age 58.5±16.0 80% females Includes patients under treatment for LTBI	Baseline	NR	NR	NR	NR	0.855
	Full random plot analysis 143 features				Baseline	NR	NR	NR	NR	0.715
Zhang [42] 2020	Basic FGF	Smear microscopy, Xpert, culture (nonspecified)	ATB cases (untreated) (n=9)	Non-TB cases (untreated) i.e. QTF positive (n=18) Age and gender-matched to ATB cases	To further evaluate the performance on predicting active TB, the AUCs were calculated among untreated participants at T0 (baseline, 1 week before starting treatment)	88.89 (51.75–99.72)	94.44 (72.71–99.86)	NR	NR	0.96 (0.88–1.02) cutoff 17.49 pg·mL−1
	Eotaxin					88.89 (51.75–99.72)	94.44 (72.71–99.86)	NR	NR	0.95 (0.87–1.03) cutoff 50.54 pg·mL−1
	G-CSF					88.89 (51.75–99.72)	88.89 (65.29–98.62)	NR	NR	0.89 (0.76–1.02) cutoff 58.89 pg·mL−1
	IFN-γ					88.89 (51.75–99.72)	94.44 (72.71–99.86)	NR	NR	0.95 (0.87–1.03) cutoff 17.41 pg·mL−1
	IL-17A					100.00 (66.37–100.0)	72.22 (46.52–90.31)	NR	NR	0.80 (0.63–0.98) cutoff 3.48 pg·mL−1
	IL-1ra					100.00 (66.37–100.0)	72.22 (46.52–90.31)	NR	NR	0.85 (0.70–1.00) cutoff 114.70 pg·mL−1
	IL-1α					88.89 (51.75–99.72)	88.89 (65.29–98.62)	NR	NR	0.92 (0.82–1.02) cutoff 8.06 pg·mL−1
	IL-4					100.00 (66.37–100.0)	100.00 (81.47–100.0)	NR	NR	1.00 (1.00–1.00) cutoff 1.16 pg·mL−1
	IL-8					77.78 (39.99–97.19)	83.33 (58.58–96.42)	NR	NR	0.86 (0.73–1.00) cutoff 3.44 pg·mL−1
	IL-9					88.89 (51.75–99.72)	94.44 (72.71–99.86)	NR	NR	0.90 (0.77–1.04) cutoff 38.85 pg·mL−1
	MCP-1 (MCAF)					66.67 (29.93–92.51)	83.33 (58.58–96.42)	NR	NR	0.79 (0.59–0.99) cutoff 10.34 pg·mL−1
	MIP-1β					55.56 (21.20–86.30)	94.44 (72.71–99.86)	NR	NR	0.75 (0.55–0.94) cutoff 57.65 pg·mL−1
	SDF-1α					100.00 (66.37–100.0)	50.00 (26.02–73.98)	NR	NR	0.79 (0.61–0.97) cutoff 218.00 pg·mL−1
	TNF-α					100.00 (66.37–100.0)	61.11 (35.75–82.70)	NR	NR	0.82 (0.66–0.97) cutoff 10.48 pg·mL−1
	TRAIL					77.78 (39.99–97.19)	88.89 (65.29–98.62)	NR	NR	0.88 (0.74–1.01) cutoff 34.51 pg·mL−1
	VEGF-A					88.89 (51.75–99.72)	72.22 (46.52–90.31)	NR	NR	0.83 (0.66–1.00) cutoff 14.31 pg·mL−1

AUC: area under the curve; CCL11: C-C motif chemokine ligand 11 (eotaxin-1); FGF: fibroblast growth factor; G-CSF: granulocyte colony-stimulating factor; GM-CSF: granulocyte macrophage–colony-stimulating factor; HHC: household contacts; IFN: interferon; IGRA: interferon gamma release assay; IL: interleukin; IP10: interferon gamma-induced protein 10; LTBI: latent tuberculosis infection; MCP-1 (MCAF): monocyte chemoattractant protein-1 (monocyte chemotactic and activating factor, CCL2); MIP-1β: macrophage inflammatory protein-1 beta (CCL4); NPV: negative predictive value; NR: not reported; PPV: positive predictive value; QFT: QuantiFERON-TB test; SDF-1α: stromal cell derived factor-1 alpha (CXCL12); TB: tuberculosis; TNF-α: tumour necrosis factor alpha; TRAIL: TNF-related apoptosis inducing ligand; TST: tuberculin skin test; VEGF-A: vascular endothelial growth factor A. ^#: As determined by classification and regression tree (CART) model.

Transcriptomic signatures

Our systematic review identified several studies that investigated the potential of transcriptomic profiling as a biomarker approach predicting progression from TBI to active TB (10 studies) [14, 27–35]. Across these 10 studies, 23 unique gene signatures were identified, with data on sensitivity, specificity and predictive values at multiple time points prior to disease onset (table 2). Although many signatures demonstrated promising performance, significant heterogeneity in population characteristics, reference standards and reporting practices limited interpretability and comparability.

Short-term prediction

10 studies investigated gene signatures designed to predict progression to active TB within different time windows (table 2). For example, Vargas et al. [27] assessed a 42-gene signature and reported a sensitivity of 87.3% (95% CI 81.5–93.1) and specificity of 91.7% (95% CI 89.9–93.5) for predicting progression within three months of TB diagnosis by conventional methods. Performance declined over extended intervals (>18 months), suggesting that some transcriptomic markers may primarily support short-term prediction.

Penn-Nicholson et al. [31] evaluated RISK6, reporting an AUC of 87.6% (95% CI 82.8–92.4) for predictions within 0–12 months. Sensitivity decreased to 50% and specificity remained 82.8% (95% CI not reported) for predictions 12–24 months prior to disease onset. The study authors provided minimal details on the study participants’ TB exposure or TB infection status definitions, adding uncertainty about the test's applicability in broader cohorts.

Longer-term prediction

Leong et al. [32] assessed multiple previously reported gene signatures up to 5 years prior to disease progression, finding AUCs ranging from 0.62 (95% CI 0.59–0.65) to 0.91 (95% CI 0.89–93). Among most studies included in our review, reporting was limited for several important characteristics, including how reference standards were applied, and lacked detailed demographic data, complicating the applicability of results. A lack of reporting on comorbidities or sociodemographic factors in the study population limits the extrapolation of these findings to diverse settings.

Micro-RNA signatures

Xin et al. [28] evaluated microRNA markers (e.g. hsa-miR-16-5p and hsa-miR-451a), reporting a sensitivity of 80.8% (95% CI 70.3–88.2) and specificity of 79.2% (95% CI 68.9–86.8) over 5 years when adjusted for covariables (table 2). However, inconsistent reference standard definitions across studies hinder direct comparisons and underscore the need for harmonised methods and criteria when assessing microRNA-based markers.

Proteomic signatures

Four studies evaluated proteomic signatures for predicting TB progression [36–39] (table 3). Performance varied across markers and panels.

IgG glycosylation profiling

Burel et al. [36] evaluated IgG glycosylation features to differentiate higher-risk individuals (based on the Zak16 signature) from lower-risk TBI populations, reporting an AUC of 0.71 (95% CI not reported) without using a reference standard. An AbGlyc score distinguished active TB from higher-risk TBI with an AUC of 0.82 (95% CI not reported), indicating moderate but improved discrimination potential. Like studies focused on transcriptomic profiles, there was significant inter-study heterogeneity, which limits the interpretability and comparability of results.

Protein panels

Singer et al. [39] examined multiple protein panels at varying time intervals before active TB diagnosis (table 3), identifying a five-protein panel (cluster of differentiation 248, hypoxia upregulated protein 1, nidogen 1, paraoxonase 1 and vascular cell adhesion molecule 1) that performed best within 0–6 months of disease onset (AUC 0.93, 95% CI not reported) while a six-protein panel measured 6–12 months before diagnosis achieved an AUC of 0.86 (95% CI not reported). Additional panels assessed in this study demonstrated moderate performance, with AUCs ranging from 0.75 to 0.87 depending on the timing of sample collection (table 2). In Penn-Nicholson et al. [38], the tRNA (guanine-N¹)-methyltransferase 5 protein showed the strongest discriminatory ability, with an AUC of 0.96 (95% CI 0.93–0.99), 100% sensitivity (95% CI 73.5–100.0) and 75% specificity (95% CI 67.2–81.7) for predicting TB progression within 6 months in their training cohort, although performance declined substantially for longer prediction intervals (AUCs 0.55–0.76 up to 24 months). While proteomic markers demonstrate some promising short-term predictive potential, heterogeneity in study design, populations and reference standards limits direct comparability across studies and biomarkers.

Cytokines/chemokines

Our review identified three studies [40–42] that investigated pathogen-induced cytokine and chemokine biomarkers for predicting progression to active TB. The findings highlight varying diagnostic performance across different biomarkers (table 4).

Daniel et al. [40] evaluated multiple cytokines, including C-C motif chemokine ligand (CCL) 19 and interferon gamma-induced protein 10, both showing 100% sensitivity and specificity (AUC 1.00, 95% CI 0.92–1.00) for prediction within 2 years. However, confidence intervals were fairly wide (83.2–100.0%) and sample sizes small (<20 per group) (table 4). Notably, the CCL19 marker outperformed others, with a perfect sensitivity, specificity, PPV and NPV at optimal cutoff values, suggesting its high potential as a predictive biomarker. Other cytokines such as CCL11 and granulocyte macrophage–colony-stimulating factor also showed promising results with sensitivities and specificities over 90% but slightly lower AUC values.

Zhang et al. [42] examined various cytokines, including interferon (IFN)-γ, interleukin (IL)-17A and IL-4, in exposed individuals. IL-4 achieved the highest AUC of 1.00, indicating excellent diagnostic capacity, while IFN-γ and IL-1α had AUC values between 0.95 (95% CI 0.87–1.03) and 0.92 (95% CI 0.82–1.02), with sensitivities and specificities close to 90%. In contrast, some markers, such as macrophage inflammatory protein-1β and monocyte chemoattractant protein-1, displayed moderate AUCs (0.75 and 0.79) with lower sensitivity values, indicating more limited predictive utility for TB progression.

Robison et al. [41] applied a random forest model to cytokine data among individuals at high risk of reactivation, achieving an AUC of 0.855 (95% CI not reported).

Discussion

Our systematic review highlights the rapidly evolving landscape of biomarker-based diagnostics for predicting progression from TBI to active disease. Although multiple promising host- and pathogen-derived signatures have emerged since 2016, no single test has yet demonstrated sufficient accuracy, reproducibility or generalisability to support clinical implementation. This gap is particularly pressing given the global push toward TB elimination, the emerging interest in fighting asymptomatic TB [43] and the limited predictive utility of existing tests such as TSTs and IGRAs [4].

Among the various biomarkers evaluated, transcriptomic signatures have received the most attention, with several studies investigating whole-blood gene expression profiles. While some signatures, such as the 16-gene correlate developed by Zak et al. [14], have shown encouraging predictive performance, their accuracy tends to decline as the prediction window increases. For instance, signatures that perform well within a 6-month prediction interval often fall below acceptable thresholds (as defined by WHO target product profiles (TPPs)) at 12 or 24 months [38]. Biomarker performance possibly declines as the prediction window increases because early host responses may wane over time, biological signals become less specific and greater variability in progression timing introduces noise [38, 44]. Predictive performance is also shaped by pre-test probability; biomarkers tend to perform better in high-risk populations, while their positive predictive value declines in lower-risk settings, even if sensitivity and specificity remain unchanged. Thus, performance is context-dependent, with factors such as concurrent infections, genetic diversity and geographic variation potentially affecting test utility.

While several biomarkers evaluated in this review demonstrated sensitivity and specificity that meet or exceed WHO-recommended TPP thresholds, particularly within short prediction windows, none have progressed to late-stage clinical trials or implementation studies. This disconnect between promising accuracy estimates and limited translational advancement likely reflects a combination of factors. First, many studies were conducted under idealised conditions with highly selected populations and performance may not generalise to routine clinical settings. Second, studies often lacked critical information on test reproducibility, cost and operational feasibility, which are key considerations for developers and public health programmes. Third, incomplete reporting and methodological inconsistencies, including nonstandardised reference standards, limit regulatory confidence and hinder investment. Lastly, few biomarker platforms have been integrated into user-friendly diagnostic formats suitable for point-of-care use, further delaying the transition to clinical trial evaluation.

Protein-based signatures and cytokine/chemokine profiles offer potential advantages, such as lower cost and easier scalability for decentralised applications. In theory, these tests could support decentralised screening in community or primary care settings, especially in high-burden countries where laboratory infrastructure is limited. However, several barriers remain. Our review identified very few studies that evaluated test performance in primary care/decentralised settings or provided evidence on operational feasibility. Additionally, limited data exist on turnaround time, cost-per-test or integration into existing diagnostic workflows. Most importantly, the trade-offs between sensitivity and specificity observed across studies raise concerns about the risk of overtreatment or missed cases if these tools are implemented prematurely.

Multianalyte approaches that combine transcriptomic, proteomic and immunological markers may enhance predictive power. However, such strategies increase assay complexity, technical requirements and cost. Without advances in assay simplification and automation, the feasibility of implementing these platforms at the point-of-care level will likely remain aspirational.

The studies included in our review were highly heterogeneous in terms of design, population and reference standards. This variation precluded meta-analytic synthesis and underscores the need for harmonised approaches to study design and reporting. While transcriptomic, proteomic and immunological biomarkers offer theoretical and early empirical promise, their clinical utility and applicability hinge on validation across diverse settings and populations, particularly among high-risk groups such as people living with HIV, young children and close contacts of individuals with active TB.

Moreover, methodological limitations persist. Many studies exhibited unclear risk of bias due to incomplete reporting on sampling methods, reference standards or participant follow-up. This lack of transparency is not uncommon in diagnostic research [16] and compromises the translational potential of findings related to biomarker-based tests [22].

A major strength of this review is the notably sensitive search strategy used and its comprehensive analysis of biomarker tests across multiple domains, highlighting the potential of various biomarker-based approaches for predicting TB progression. However, several important limitations were identified, particularly concerning the transparency of reporting within the included studies. Many studies received “unclear” ratings for risk of bias and applicability, primarily due to insufficient details regarding patient selection, recruitment methods and the flow of participants through the study. This lack of transparency not only complicates the assessment of methodological quality but also limits the external validity of the findings, making it challenging to generalise results to broader populations. Furthermore, the observed heterogeneity in study designs and outcomes underscores the pressing need for standardised reporting practices in TB biomarker research, which would enhance the interpretability and applicability of findings.

Future research of promising biomarkers must prioritise large-scale, prospective validation studies using standardised protocols to evaluate biomarkers based on WHO-endorsed TPPs or similar criteria intended to determine the usefulness of such tests. Biomarker development should align with implementation science frameworks to assess feasibility, cost-effectiveness and scalability in high-burden settings [15]. Without such methodological and operational rigor, the potential for biomarkers to guide targeted TB preventive therapy and reduce disease incidence will remain unrealised.

Ultimately, our findings show that while the concept of disease progression is scientifically sound, the path to clinical adoption requires more robust evidence. Cross-disciplinary collaboration, global investment and the use of standardised evaluation frameworks will be essential to translate the promise of TB biomarkers into practice.

Points for clinical practice

• Current biomarker-based tests for predicting progression from TBI to active disease show promise but are not yet ready for clinical implementation.

• Most evaluated tests demonstrate only moderate predictive accuracy and lack validation in diverse, real-world populations.

• Existing immunological tests (e.g. TSTs and IGRAs) have low PPV, underscoring the need for more accurate tools to guide targeted preventive therapy.

• Clinicians should remain cautious in interpreting biomarker test results for TB progression until standardised, validated and cost-effective tools become available.

• Integration of biomarker-based tests into TB prevention programmes will require careful consideration of feasibility, cost and operational logistics, especially in high-burden settings.

Questions for future research

• Which biomarker signatures (e.g. transcriptomic, proteomic, cytokine-based) offer the best balance of sensitivity, specificity and feasibility for predicting TB progression?

• How can future studies improve methodological rigor and transparency to enhance reproducibility and comparability of biomarker performance?

• What are the optimal time windows for prediction and how does predictive accuracy vary across different populations and epidemiological contexts?

• Can multianalyte or combined biomarker approaches improve predictive performance without compromising scalability or affordability?

• What are the cost-effectiveness and implementation challenges of deploying biomarker-based tests in resource-limited settings?

Supplementary material

Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.

Supplementary material

ERR-0151-2025.SUPPLEMENT