Guiding the development of the tuberculosis screening target product profile using single-screen and multi-screen approaches: a modelling study.

BACKGROUND

Reducing tuberculosis incidence requires new, symptom-agnostic screening tools. Target product profiles guide their development by defining performance. In this study, we aimed to support the 2025 WHO tuberculosis screening target product profile update by modelling the minimum sensitivity and specificity thresholds needed across different prevalence levels and screening algorithms.

METHODS

In this modelling study, we used decision trees and parameter multiplication to simulate screening strategies across four levels of tuberculosis prevalence (0·10%, 0·25%, 0·50%, and 1·00%). Tests were considered acceptable if they achieved a post-screen prevalence of 5% or higher and diagnosed at least 60% of tuberculosis cases after confirmatory testing. We modelled high-sensitivity (defined as ≥90% sensitivity) and high-specificity tests (defined as ≥98% specificity) within one-screen algorithms (novel screening test followed by confirmatory testing) and two-screen algorithms (inclusive of a secondary screen approximated by chest x-ray performance).

FINDINGS: At 1·00% prevalence, a test with 90% sensitivity required specificities of at least 83% (one-screen) and at least 52% (two-screen). High-specificity tests (98%) required sensitivities of 68% (one-screen) or 80% (two-screen) across all prevalence levels, with the exception of 0·10% prevalence in a one-screen algorithm. At 0·10% prevalence, no acceptable one-screen tests were identified. At 0·10% prevalence, tests in two-screen algorithms required 90% sensitivity with at least 96% specificity or at least 80% sensitivity with 98% specificity.

INTERPRETATION: In higher tuberculosis prevalence settings, a test with moderate sensitivity and specificity is sufficient to achieve high case detection without overloading the confirmatory diagnostic cascade. Screening in very low prevalence settings remains challenging and requires a very high-sensitivity and high-specificity screening test even within two-screen algorithms.

FUNDING: Bill and Melinda Gates Foundation.