Estimating the number of incorrect TB diagnoses in LMIC

Who

• Population: Individuals evaluated for pulmonary tuberculosis (TB) through routine healthcare services.

• Setting: 111 low- and middle-income countries (LMICs), representing ~98% of global TB incidence.

• Data scale: ~22.9 million people evaluated for TB in 2023; ~5.7 million estimated to have TB.

• Key subgroups: HIV-positive individuals (median 5% across countries) and those tested with rapid diagnostic tests (RDTs).

• Data source: Aggregate national notification data from the World Health Organization (no individual-level human subjects).

---

What

• Main finding: Current TB diagnostic algorithms in LMICs have moderate sensitivity (82.6%) and specificity (88.0%), resulting in substantial diagnostic error.

• Estimated global burden (2023, assuming 25% TB prevalence among those evaluated):

  • ~1.00 million false-negative TB diagnoses (missed TB cases).

  • ~2.05 million false-positive TB diagnoses (TB diagnosed in people without TB).

• Clinical diagnosis accounted for:

  • ~22% of true-positive diagnoses.

  • ~75% of false-positive diagnoses.

• Implication: Large numbers of people are either missed or incorrectly treated, highlighting trade-offs between sensitivity and specificity in TB diagnosis.

• Counterfactual analyses: Improvements such as full RDT adoption, better clinical algorithms, or more sensitive RDTs could substantially reduce both false-negative and false-positive diagnoses.

---

When

• Study year: Diagnostic performance estimated for 2023.

• Underlying evidence: Sensitivity/specificity inputs drawn from previously published diagnostic accuracy studies (various years; exact dates not specified).

---

Where

• Geographic scope: 111 LMICs across all WHO regions.

• Regional variation: Diagnostic performance varied widely by region, reflecting differences in RDT coverage and reliance on clinical diagnosis.

---

Why

• Rationale: There is limited direct evidence on how well real-world TB diagnostic algorithms perform at scale.

• Problem addressed: The balance between missed TB cases and overdiagnosis is poorly quantified globally, especially where clinical diagnosis is common.

• Goal: To estimate the true magnitude of false-positive and false-negative TB diagnoses and evaluate how alternative diagnostic strategies could improve outcomes.

---

How

• Design: Bayesian mathematical modeling of the TB diagnostic cascade.

• Inputs:

  • National TB notification data (laboratory-confirmed vs. clinically diagnosed).

  • Published estimates of sensitivity and specificity for smear microscopy, Xpert Ultra RDTs, and clinical diagnosis.

  • Adjustment for HIV status and possible culture-negative TB.

• Key assumptions:

  • Initial TB prevalence among evaluated individuals ranged from 5–50% (25% base case).

  • No adjustment for under-reporting of TB diagnoses.

• Analysis:

  • Bayesian inference with Hamiltonian Monte Carlo (5,000 simulations).

  • Country-level estimates pooled to regional and global results.

  • Multiple counterfactual scenarios tested (e.g., full RDT adoption, improved clinical algorithms).

• Software: R and RStan.

---

Overall conclusion

The study demonstrates that current TB diagnostic practices result in millions of incorrect diagnoses each year, underscoring the urgent need for improved diagnostic tools and strategies that reduce both missed TB and overdiagnosis.

Source: van Lieshout Titan, A., Dodd, P.J., Cohen, T. and Menzies, N.A., 2026. Estimating the number of incorrect tuberculosis diagnoses in low-and middle-income countries. Nature Medicine, pp.1-8.