AI tuberculosis screening cost about $20 a person in a two-country trial

In Lesotho, a community tuberculosis screening team set up temporary clinics with tents, tables and heaters. Rented vehicles carried its equipment between dispersed mountain villages. South Africa’s team used purchased vans and caravans as consultation units.[2]

The TB TRIAGE+ programme screened 20,023 adults in the two countries between September 2022 and September 2024. Its AI-assisted pathway cost about $20 per person in 2023 US dollars. A detailed analysis published in PLOS Global Public Health on 21 September explains the spending behind cost totals first reported with the July trial.[1]

One input was inexpensive: the recurring CAD4TBv7 image-analysis charge was $0.80. Equipment, staffing, transport and confirmatory testing sat elsewhere in the budget.[2]

That distinction helps explain both the opportunity and the arithmetic of taking screening to people who may otherwise remain beyond its reach.

The extra test that failed to save money

Every participant received a digital chest X-ray scored by CAD4TBv7. In one pathway, a sufficiently high score led to an Xpert MTB/RIF Ultra molecular test. The other introduced a C-reactive protein blood test for an intermediate range of AI scores; a CRP result above 5 mg/L triggered molecular confirmation. High AI scores went directly to confirmation. The same participants were assessed under both pathways.[3]

Adding CRP reduced the number indicated for molecular confirmation from 4,491 to 2,772. It adds an inflammation signal, but also another test and another decision about who proceeds.[2]

The trial’s complete-pair comparison found 69 versus 60 cases among 73 identified by either route. Detection proportions differed by −12.3 percentage points for the CRP pathway (95% confidence interval −23.0 to −1.6), failing its prespecified −10-point non-inferiority margin. Those proportions describe the trial’s identified cases, not sensitivity among all tuberculosis in the population.[3]

The cost table includes three additional CAD-pathway cases whose CRP-pathway data were incomplete. They were excluded from the paired clinical comparison but retained in the fuller costing population. That is why the same trial can correctly report 69 in one analysis and 72 in another.[2]

The budget follows the road

Personnel accounted for more than half of South Africa’s TB-only costs. In Lesotho, bundled vehicle hire was the largest component, including fuel, insurance and drivers. Buying mobile clinical units and renting transport for temporary sites create different spending patterns; the two vehicle budgets are not like-for-like purchase prices.[2]

A simple reconstruction puts the recurring AI charge in perspective. Multiplying the reported $0.80 by 20,023 gives roughly $16,018, or 4.1% of the CAD pathway’s total. In an illustrative scenario holding every other input fixed, eliminating that fee would therefore save about 4% of the programme budget.

The study’s modelled expansion to 100,000 participants did not produce a major reduction in unit costs: much of the recurring work grew with the campaign. Targeting a population with higher TB prevalence had a larger effect on cost per case, because the same screening effort found more cases. That scenario is a projection, not an additional field trial.[1]

There is a reason to take that effort into communities. By day 56, 100 participants had microbiologically confirmed tuberculosis across the trial’s broader diagnostic follow-up; 46 had reported none of the four main TB symptoms. A service that waits for a familiar symptom can miss people a screening visit reaches.[3]

The useful economic question is therefore how to make each journey find and connect more people to care. The automated reading’s value depends on the travelling team, the confirmation pathway and the people the programme is trying to reach.

Report an error or suggest a correction