Diagnostic accuracy of the Xpert MTB Host-Response assay for tuberculosis disease in adolescents and adults: a systematic review and individual patient data meta-analysis protocol

Abstract

Introduction

Tuberculosis (TB) remains a major global health challenge, with an estimated 10.8 million new cases and 1.25 million deaths in 2023. Despite advances in molecular detection of Mycobacterium TB (MTB), significant diagnostic gaps remain: in 2023, only 48% of newly diagnosed TB cases received rapid diagnostic testing, far below the 100% target. These challenges are intensified in high-burden settings, where sputum collection and distinguishing TB from other illnesses are difficult. The Xpert MTB Host Response (Xpert-HR) assay measures host immune gene expression from blood, shows promise but variable accuracy across studies. Hence, this study will perform an Individual Patient Data Meta-Analysis (IPDMA) to evaluate the diagnostic accuracy, subgroup performance, predictive values and clinical benefit of Xpert-HR compared with conventional sputum-based testing.

Methods and analysis

This systematic review and IPDMA will follow Preferred Reporting Items for Systematic Reviews and Meta-Analyses for Diagnostic Test Accuracy guidelines. Prospective studies including adolescents (>12 years) or adults with presumed TB tested using the Xpert MTB Host-Response assay will be identified through PubMed, Embase and Web of Science. Study quality will be assessed using an adapted diagnostic accuracy tool. Diagnostic accuracy will be pooled using random-effects models, with subgroup analyses where applicable. Decision curve analysis will evaluate clinical utility. Predictive values will be estimated across TB prevalences of 1-10%. Both one-stage and two-stage IPDMA approaches will be explored, and the proportion of unevaluable samples will be reported.

Ethics and dissemination

The review will be based on deidentified individual patient data to be obtained upon request from the corresponding authors of studies fulfilling all the data sharing agreement. Ethical approval has been obtained from the Ethical Committee of the Medical Faculty of Heidelberg University (Approval No. S-043/2026). The results will be disseminated through publication in a peer-reviewed journal, and through presentations at academic conferences.

PROSPERO registration number

CRD420251071857.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• Uses an individual patient data (IPD) meta-analysis to generate robust and up-to-date evidence on the diagnostic accuracy of the Xpert Mycobacterium tuberculosis Host-Response assay in adults.

• The use of IPD details subgroup exploration improving the precision and generalisability of findings across high-burden settings.

• Inclusion of data from prospective studies minimises recall and selection biases.

• Not all authors of eligible studies may agree to share their IPD, and among those that do, missing covariates or outcomes could bias the pooled estimates or limit certain analyses.

• Applying a threshold derived from research settings to diverse real-world populations, where the case-mix may differ, represents a potential limitation due to differences in population characteristics.

Introduction

Tuberculosis (TB) remains a leading cause of morbidity and mortality worldwide, with an estimated 10.8 million new cases and 1.25 million deaths in 2023.¹ Early and accurate diagnosis is critical to initiating treatment, reducing transmission and improving patient outcomes. Despite advances in molecular detection of Mycobacterium TB (MTB), significant diagnostic gaps remain. In 2023, only 48% of newly diagnosed TB cases received rapid diagnostic testing, which is far below the 100% target set for 2027 at the 2023 United Nations high-level meeting.¹ Challenges are also pronounced in differentiating TB from other systemic diseases in patients presenting with similar symptoms² and in identifying TB where sputum is unavailable.³ Analyses of the TB care cascade show major attrition even before diagnosis and at the linkage-to-diagnosis step.⁴ Evidence from high-burden settings further highlights persistent gaps: TB care cascades found that diagnostic losses were especially large in these settings.⁵ In East Africa, a retrospective study among presumptive TB patients documented substantial drop-offs between screening and diagnosis.⁶ In a study in Northwest Ethiopia, 12% of individuals with TB‐suggestive symptoms did not receive sputum or chest X-ray evaluation and were thus missed for diagnosis.⁷ Collectively, these highlight the need for accessible, accurate and patient-friendly TB diagnostic strategies, especially in high-burden, resource-constrained settings.

Rapid scale-up of improved TB diagnostic tools is a priority to achieve the End TB Strategy targets by 2035.⁸ To guide development, WHO developed a series of Target Product Profiles (TPP),^{9 10} with one of the highest priorities being a non-sputum-based screening test to ‘rule out’ TB disease (i.e., having a high NPV). Such a test with a minimum sensitivity of 90%, specificity of 60%-80% and that also meets the operational targets could potentially help reduce underdiagnosis and misdiagnosis of TB, reduce delays in starting TB treatment and lower the cost and infrastructure requirements of current diagnostic evaluation for TB in high-burden settings.¹¹ However, there are no screening tests that currently meet all these targets, and WHO currently suggests screening for TB disease among individuals aged 15 years or older in populations may be conducted using the symptom screen, chest X-ray and/or selected molecular rapid diagnostic tests.¹² Non-sputum-based screening tests for TB remain a priority in public global health efforts to End TB.

The Xpert MTB Host Response (HR) assay (Cepheid, Sunnyvale, California, USA, Xpert-HR) is a cartridge-based molecular assay that measures the expression of a host immune response genes from capillary or whole blood samples.¹³ It targets guanylate binding protein (GBP5) and dual specificity phosphate 3 (DUSP3), which are increased in TB, and TATA binding protein (TBP) and krupple-like factor 2 (KLF2), which are usually decreased. GBP5 is activated by interferon and helps trigger the body’s first-line immune defence while promoting inflammatory signalling.¹⁴ DUSP3 regulates a key cellular pathway that influences how immune cells respond to infection and inflammation.¹⁵ In contrast, KLF2 reduces inflammatory responses and is usually less active during TB.¹⁶ TBP serves as a stable reference gene, allowing accurate measurement and comparison of gene activity. By quantifying these transcriptional signatures associated with TB disease, the test provides a rapid diagnosis of TB without the need for a sputum sample. This approach has the potential to address critical limitations of current TB diagnostics. It may facilitate diagnosis in patients with difficulty producing sputum, such as children, people living with HIV or people with extrapulmonary TB. Additionally, it may enable screening in settings where confirmatory MTB detection methods may be inaccessible, delayed or too costly.

Several studies spanning diverse populations, geographic settings and clinical presentations have evaluated the diagnostic accuracy of the Xpert-HR test. However, the reported performance characteristics vary significantly, reflecting differences in study design, inclusion criteria, patient demographics, prevalence of co-morbidities or co-infections, TB incidence and reference standards used.17,19 These variations make it challenging to provide a robust, generalisable estimate of test performance for clinical and policy decision-making. An Individual Patient Data meta-analysis (IPDMA) offers a powerful approach to address these gaps. By obtaining and harmonising raw data from multiple studies, IPDMA enables standardisation of inclusion/exclusion criteria and reference standards across datasets, exploration of how patient-level factors (eg, age, HIV status, TB risk profile, symptoms) influence diagnostic accuracy, and provides more precise estimation of sensitivity, specificity and predictive values across clinically relevant subgroups.

Given the global TB diagnostic pipeline’s need for reliable, non-sputum-based tools, synthesising the available evidence on the Xpert-HR test through IPDMA will provide critical insights into the role of such host-response tests for use cases related to detection of TB disease. In this study, we will present the first IPDMA on diagnostic accuracy assessment of the Xpert-HR in a large cohort of adolescents and adults with presumptive TB disease. Hence, this study intends to evaluate the diagnostic accuracy of the Xpert-HR assay in adolescents and adults with presumed TB, including within key subpopulations.

Research aim and questions

Aim

To explore the diagnostic accuracy and potential clinical utility of Xpert Host-Response for TB.

Primary question

What is the diagnostic accuracy (sensitivity and specificity) of Xpert MTB Host-Response assay in adolescents and adults with presumed TB compared with mycobacterial culture or WHO recommended molecular diagnostic?

Secondary questions

1. What is the diagnostic accuracy of Xpert-HR in predefined subgroups (including by age, sex, HIV status, previous TB history, geographical setting and blood collection method)?

2. Does the Xpert-HR assay provide a higher net clinical benefit when used as a screening test compared with testing all individuals with sputum-based tests?

3. What proportion of non-positive non-negative (NPNN) test results are observed when for Xpert HR, and what is the accuracy of Xpert Host-Response using an intention to diagnose analysis?

4. What are the positive predictive values (PPV) and negative predictive values (NPV) of the Xpert MTB Host-Response assay across a range of prevalences applicable to different settings and use cases?

Methods

Development of the review method

The study follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses for diagnostic test accuracy (PRISMA-DTA) guidelines for systematic reviews of diagnostic test accuracy studies.²⁰ The PRISMA-DTA guideline will be followed during the systematic review. This protocol is registered on PROSPERO, with a registration number CRD420251071857.

Eligibility criteria

All prospective studies in adolescents (aged >12 years) or adults with presumed TB disease and tested with the Xpert MTB Host-Response assay will be included. Individuals who are receiving TB treatment at or in the 12 months prior to study entry will be excluded (those with previous treatment >12 months ago will be included). Patients with missing reference standard (sputum mycobacterium culture or PCR) results or incomplete assay results will be excluded from primary analyses. We will also exclude studies conducted on retrospective (biobanked) samples or case–control studies.

Search strategy

We will search PubMed, Embase and Web of Science for studies published anytime with the last intended update of the search being in May 2026. The search will be limited to studies published in English, as we are not able to interpret or extract data from non-English languages. Accordingly, the PubMed search engine, which uses a combination of MeSH terms and title/abstract keywords to identify relevant studies, will be customised to search relevant studies in the other databases. The search strategy combines terms for (1) The Index test (Xpert-HR assay) and (2) Target condition (TB) (table 1). The full list of search strategies for all databases searched is available (online supplemental table S1). Reference lists of the included articles will also be reviewed to identify any additional eligible studies.

Table 1. PubMed search strategy using a combination of terms.

No.	Search concept	Search terms
#1 Index test	Xpert Host-response assay	((Xpert(tiab))OR Cepheid(tiab)) AND (host(tiab))OR immune(tiab))OR gene(tiab))OR 3-gene(tiab))OR transcriptom*(tiab))OR expression(tiab)) AND (response(tiab))OR signature(tiab))OR assay(tiab))OR biomarker(tiab))) OR (“non sputum-based” (tiab))AND (“point-of-care” (tiab))OR “triage test” (tiab))) OR “finger-stick” (tiab))OR “blood test” (tiab)
#2 Target Condition	Tuberculosis	Tuberculosis (MeSH))OR “Mycobacterium tuberculosis” (MeSH))OR TB (tiab))OR MTB (tiab))OR “tuberculosis” (tiab))OR “Mycobacterium tuberculosis” (tiab)
Final combined query		#1 AND #2

Study selection

All references searched will be imported into a web-based software Covidence. Duplicates will be removed automatically by this software. Two reviewers (KFM and JKW) will independently screen titles and abstracts for relevance, with any discrepancies resolved by a third reviewer (AG-W). Potentially eligible articles will then be reviewed in full by the same two reviewers, with any discrepancies resolved by a third reviewer AG-W.

Quality assessment and risk of bias

The quality of included studies will be assessed using the revised quality assessment of diagnostic accuracy studies (QUADAS-2) tool, adapted to the context of this review, to inform the analysis and interpretation.²¹ The QUADAS-2 tool gives an assessment of quality based on four domains: patient selection, index test, reference standard and flow and timing. Quality assessment will be performed independently by two reviewers, with discrepancies resolved by a third reviewer.

Data extraction, acquisition and management

First, study-level data will be extracted from papers by a single reviewer into a customised and standardised data collection form. A final version of this extraction form will be created after 1–4 data extractions are completed, allowing for optimisation of this form. Study authors will be contacted in cases of unclear eligibility or missing data. In addition to study level data extraction, deidentified IPD will be requested from the corresponding authors. This includes studies identified through our systematic search and additional studies not yet published that the investigators are already aware of (eg, data from the Feasibility of New Diagnostics for Tuberculosis consortium). Patient-level variables requested will include basic data on demographic and clinical variables, as well as details of the methods for testing results from the index (Xpert-HR) and reference standard tests (eg, Xpert MTB/RIF Ultra PCR or other molecular assay, mycobacterial culture, composite or clinical reference standard). All data will be held securely, accessible only to the immediate project team. Authors who are able and willing to share the IPD will be invited to join the IPD group as two co-authors if able to contribute to data analysis and interpretation.

Data cleaning and harmonisation will be performed on copies of the IPD dataset to ensure consistency in variable definitions across datasets;²² original data will not be edited. The received IPD will be checked in relation to their published results to confirm agreement and that datasets were received in full. Any inconsistencies will be checked with the relevant authors before proceeding on the IPDMA. Missing data will be handled using multiple imputation techniques where appropriate.²³ Data extraction and risk of bias assessment of included studies will be performed using Covidence software.²⁴

For the IPD, the primary reference standard for TB will be defined using a composite of mycobacterial culture or WHO-recommended molecular tests. Accordingly, TB will be defined as either positive culture or a positive WHO-recommended molecular test. Conversely, ‘Not TB’ will be defined as cases with both negative culture and negative molecular test results when both tests are available or as a negative molecular test result when culture is not available.²⁵ For the index test (the Xpert HR), the four target genes (GBP5, DUSP3, KLF2 and TBP) will be combined to compute the TB score, where no prespecified cut-off value was provided by the manufacturer. The receiver operating characteristic (ROC) will be performed to identify the optimal cut-off value for the Xpert HR to achieve the target sensitivity of ≥90% while maximising specificity.

Data synthesis and statistical analysis

Descriptive measures including counts and proportions will be presented in tables and text to describe all included studies in the review. The overall and subgroup level diagnostic accuracy will be examined using ROC analysis of random effect models accounting for between study heterogeneity. Decision curve analysis (DCA) will be applied to assess net clinical benefit of using Xpert MTB HR assay as a screening tool. All analyses will be performed in Stata/SE 19.5 and R.

The ROC analysis of the Xpert HR TB Score will be performed against the reference standard to identify the optimal cut-off value to achieve the sensitivity of 90% or more. These cut-off values will be used to estimate diagnostic accuracy (including sensitivity, specificity, PPV, NPV and 95% CIs/credible intervals compared with reference standards. Prevalences ranging from 1% to 10%, in 1% increments, will be considered in modelling the NPV and PPV. Both (if applicable) one-stage and two-stage IPD meta-analyses²⁶ will be considered in estimating pooled diagnostic accuracy.

Where possible, we will perform random-effects IPDMA to estimate the pooled diagnostic accuracy of the Xpert MTB Host-Response assay using a bivariate framework for sensitivity and specificity implemented either through the frequentist generalised linear mixed model or through a Bayesian hierarchical logistic model.^{27 28} To test accuracy across numerous cut-offs aligned with the WHO TPP targets,¹⁰ we will employ Bayesian hierarchical summary ROC models and threshold modelling as needed.

We will explore heterogeneity using meta-regression on prespecified study-level covariates. Summary estimates of the study questions will be presented together with evaluation and presentation of heterogeneity. Additional analysis will determine the proportion of samples that could not be evaluated (ie, NPNN). Subgroup analysis should also be considered based on country, study, sampling method (venous or finger prick), HIV status, diabetes status, sex, previous TB history, type of TB (pulmonary vs extra-pulmonary) and age as these subgroups are known to be related to risk of TB disease.

The DCA, a method to evaluate diagnostic tests and predictive models,^{29 30} will be used to evaluate clinical usefulness of the Xpert HR TB Score test results across a range of decision thresholds, calculating clinical net benefit compared with treat all (stands for referring all for confirmatory test) and treat none (stands for referring no one for confirmatory testing). Where appropriate, net benefit estimates will be computed within each contributing study and combined using random-effects models to account for between-study heterogeneity.

We anticipate that IPD may not be available from all eligible studies. The primary analysis will be based on studies contributing IPD. Studies where IPD cannot be obtained will not be excluded a priori. Instead, study-level aggregate data will also be meta-analysed to ensure all available evidence is included. We will also undertake sensitivity analysis comparing results from the IPD-only pool versus a pool that includes aggregate data from studies not providing IPD. Further, sensitivity analysis will be undertaken using only a culture-based reference standard (ie, TB if a positive culture, not TB if negative culture). A secondary composite reference standard which includes ‘clinical’ TB diagnosis or TB treatment initiation as ‘TB’ even in the absence of a positive culture or molecular test.

Patients and public involvement

No patient or public will be involved.

Ethics and dissemination

The review will be based on deidentified IPD to be obtained on request from the corresponding authors of studies fulfilling all the data sharing agreements. Ethical approval has been obtained from the Ethical Committee of the Medical Faculty of Heidelberg University (Approval No. S-043/2026). The results will be disseminated through publication in a peer-reviewed journal, and through presentations at academic conferences.

Discussion

This IPDMA aims to provide a comprehensive and up-to-date assessment of the diagnostic accuracy and potential clinical utility of the Xpert MTB Host-Response (Xpert-HR) assay in adolescents and adults with presumed TB. By pooling IPD from multiple studies, this study will enable harmonised analyses across diverse populations and settings, allowing exploration of how patient-level characteristics—including age, sex, HIV status, prior TB history and sample collection method—impact test performance. Such analyses are critical for determining whether the Xpert-HR assay can serve as an effective screening tool, particularly in high-burden, resource-constrained settings where sputum collection is challenging and rapid diagnosis is essential.

A key strength of this study is the use of prospective data, which minimises selection and recall biases compared with retrospective or case-control designs. Furthermore, the IPDMA approach enables consistent definitions of index and reference tests, more precise assessment of sensitivity, specificity and predictive values, and robust evaluation of subgroup performance and clinical net benefit using DCA. By considering a range of TB prevalences and accounting for unevaluable or non-positive/non-negative results, this study will generate evidence that is directly applicable to real-world clinical and public health settings. Importantly, this work will provide the first IPD-based synthesis supporting the role of a host-response assay as a next-generation approach to TB diagnosis, with implications for early detection and screening strategies in high TB burden populations.

Nonetheless, several potential limitations should be considered. Not all eligible studies may be able to share IPD, which could introduce selection bias if the included datasets are not representative of the broader evidence base. In addition, among studies that provide IPD, missing covariates or incomplete outcome data could limit certain analyses or introduce bias. We have excluded children under 12 years of age, but we are aware of a separate paediatric meta-analysis of Xpert-HR. Finally, the results of this IPDMA will reflect the performance of the Xpert-HR assay under study conditions and may differ from routine programme settings.

Overall, this IPDMA will address an important evidence gap by generating robust, patient-level evidence on the diagnostic accuracy and clinical utility of the Xpert-HR assay. The findings of this study are expected to inform TB screening and early detection strategies, support the development of policy and clinical guidelines, and highlight areas where further research or assay optimisation may enhance its clinical and public health impact.