BGPT: Paper Review: New technologies for diagnosing active TB: the VANTDET diagnostic accuracy study

Explore by Goal

Quick Explanation Copied

VANTDET (10.3310/eme08050) — critical takeaway

Multi-omics host-response signatures looked promising in discovery studies, but this prospective “real-life TB suspect” validation found that none achieved clinically useful accuracy for all active TB patients; however, specific subgroup utility emerged (notably phenotypic flow-cytometry signatures in IGRA-positive, smear-negative suspects).

Primary reference:

Long Explanation

Paper Review (Evidence-based, skeptical, visual): VANTDET diagnostic accuracy study

Reference:

Core biological/statistical claim under test

Whether previously proposed host-response signatures (transcriptomic, proteomic, cellular-immune phenotypes; plus chemokine downstream of IFNγ) retain diagnostic accuracy when evaluated prospectively in a clinically diverse set of TB suspects (including culture-unconfirmed cases).

What the report is doing (visual framework)

1) Transcriptomics (whole blood RNA; Illumina microarray)

Validated published gene-expression signatures using LDA/score methods; reported AUC for detecting all active TB vs other diseases/active-TB-excluded.

2) Proteomics (serum protein signatures; SELDI-TOF and LC-MS)

Checked cross-platform reproducibility (“bridging”); evaluated multiple previously published protein signatures via ROC/AUC and LDA.

3) Cellular immune signatures (flow cytometry; phenotypes after stimulation)

Validated six published flow signatures in IGRA-positive suspects using PBMC stimulation (PPD; RD1 antigen peptides) and antibody panels; phenotypic markers (e.g., %HLA-DR) performed best.

4) Candidate molecular “rule-out” tests (qRT-PCR & chemokines)

Tested chemokine downstream of IFNγ (CXCL9/CXCL10) and BATF2 gene expression after RD1 peptide stimulation; also tested CXCL10/IFNγ in QFT supernatants by MSD. Reported that universal rule-out thresholds were not met and assay failure rate was high.

Important skeptical note: AUC values above are not all from the same analytic setting (different cohorts, reference standards, and cut-off philosophies). The figure is therefore best read as a directional comparison of performance ceilings reported by VANTDET, not as a strict head-to-head meta-comparison.

Dataset & reference-standard structure (what could bias accuracy)

Source cohort: IDEA prospective multi-site recruitment of adult suspected TB in routine NHS care (12 sites in England).
Reference standard: Dosanjh composite categorization (culture-confirmed TB, highly probable TB, clinically indeterminate excluded, and active TB excluded/other diseases).
Spectrum and “culture unconfirmed”: OD included clinically adjudicated non-active TB, but some clinically classified OD groups may include culture-unconfirmed active TB or alternative diagnostic uncertainty; VANTDET notes that “culture negative” differs from “active TB excluded.”
Blinding: Laboratory staff were blinded to clinical data and reference standard during index testing.

Technology-by-technology critique (evidence-first)

A) Transcriptomics: published gene-expression signatures in adult NHS TB suspects

Headline result

Using LDA in the full transcriptomic cohort, the best published transcriptomic signatures reached AUCs around 0.81–0.84 for detecting all TB, but the report states that these were not clinically useful for all TB cases or for culture-confirmed TB when considered across the spectrum tested.

Skeptical blind spot: VANTDET reports that performance dropped substantially in HIV+ participants and that the number of HIV+ samples is small, so subgroup inference may be unstable/underpowered.

Cut-off & classifier validity concerns

VANTDET used cut-off schemes (e.g., sensitivity >95% for rule-out or specificity >90% for rule-in) and also applied LDA to ensure consistent evaluation across signatures. This reduces some “threshold mismatch” bias, but the report acknowledges that diagnostic performance depends on cohort and that “optimal thresholds” can shift between laboratories/cohorts.

B) Proteomics: cross-platform reproducibility problems matter biologically

SELDI-TOF platform discontinuation and bridging failure are not just “logistics”—they can change the measurable protein features.

VANTDET reports that SELDI-TOF arrays were discontinued by the manufacturer, and that cross-platform reproducibility was limited; consequently, only a four-protein in-house SELDI-TOF signature was meaningfully bridged to LC-MS, and concordance was poor.

Key limitation: Even the best proteomic results did not meet VANTDET’s clinically useful thresholds for all TB. The authors conclude none of the evaluated proteomic signatures achieved clinically useful accuracy in the IDEA-derived clinical-spectrum cohort.

C) Flow cytometry: phenotypes outperform “function-only” signatures

Headline result

Across six previously reported cellular immune signatures, VANTDET found that phenotypic markers (including %HLA-DR and %T EFF) produced the best diagnostic discrimination; functionality-only signatures (e.g., %TNF-α-only and SCORE) performed poorly for separating TB from OD/LTBI in some strata.

Subgroup signal (most clinically relevant finding): In smear-negative, IGRA-positive suspects, VANTDET reports high PPV for %HLA-DR and %T EFF (and especially for culture-positive TB).

Skeptical blind spot

VANTDET restricts cellular immune evaluation to IGRA-positive individuals (because signatures depend on detectable Mtb-specific T-cell responses). That means sensitivity in IGRA-negative TB suspects cannot be inferred from the flow-cytometry substudy.

D) Molecular “rule-out”: CXCL9/CXCL10 upshift sensitivity but specificity (and failures) block universal triage

Headline result: none met the universal rule-out threshold.

VANTDET reports that none of the evaluated candidate rule-out tests achieved >95% sensitivity and >65% specificity for active TB, and the assay had a high failure rate and complexity.

Skeptical interpretation: VANTDET describes increased chemokines downstream of IFNγ, but notes that these chemokines can also be upregulated in some OD/IGRA-negative settings, which compresses specificity.

E) Cost-effectiveness modelling: best-performing signatures still not cost-effective under assumed costs

VANTDET reports that none of the evaluated signatures was cost-effective for use in rule-in or rule-out scenarios under their modelling assumptions; they attribute this to combination of high test costs and suboptimal sensitivity/specificity for the full TB population.

Most important counterpoints / blind spots

Substudy restriction & spectrum mismatch: Flow signatures were evaluated only in IGRA-positive suspects because assay requires detectable antigen-specific responses; so performance in IGRA-negative TB suspects remains unknown.
Platform and measurement non-equivalence: Proteomic bridging across SELDI-TOF and LC-MS showed limited correlation, and SELDI-TOF consumable discontinuation limited sample numbers.
HIV subgroup uncertainty: VANTDET reports reduced performance in HIV+ groups and acknowledges that small numbers can underpower subgroup evaluation.
Artificial cohort enrichment for molecular rule-out: The GO/NO-GO design enriched for IGRA-failing situations, which is useful for a proof-of-principle but may not match routine prevalence/odds structures in all settings.
Confounding OD categories: VANTDET highlights confounding differential diagnoses (e.g., sarcoidosis impacting proteomic/transcriptomic discrimination).

Data quality & bias checks requested by a skeptical reviewer

Conflicts of interest disclosures (relevant to biomarkers)

The report abstract includes disclosed patent and royalty interests for Ajit Lalvani and related assay technologies; it also lists grants and patent relationships in the profile summary.

Skeptical implication: COIs don’t invalidate the methods, but they heighten the need to scrutinize pre-registration, analytic reproducibility, and independence of validation thresholds. VANTDET claims blinding during index testing and uses protocol governance, but the excerpt provided does not contain full prespecification artifacts.

What would falsify the study’s conclusions (in mechanistic/analytic terms)

Reproducibility: Independent labs should reproduce at least the subgroup performance (e.g., flow phenotypic signatures in smear-negative, IGRA-positive suspects) using the same gating and cut-off rules; failure would weaken the subgroup utility claim.
Reference-standard robustness: If the Dosanjh composite categorization used here is found to systematically misclassify TB/OD in a way that inflates AUC, recalibration may change conclusions.
Platform translation: If cross-platform bridging for proteomics and transcriptomics is improved (e.g., by better feature transfer or universal measurement units) and yields clinically useful accuracy, then “proteomics ceiling” may have been due to measurement incompatibility rather than biomarker biology. VANTDET documents limited bridging correlation as a likely driver.

Bottom-line (skeptical, evidence-weighted)

VANTDET provides a rare, prospectively-recruited, real-clinic-spectrum multi-omics validation demonstrating that published host-response signatures often fail to generalize to “all TB in routine suspects,” while phenotypic flow signatures can still show strong subgroup utility (not universal triage).

Explore more on BGPT (science-action buttons)

Author reviews (BGPT)

Feedback:

Updated: March 21, 2026