AI could bring tuberculosis screening closer to people who struggle to reach a specialist. Its public value will depend on whether they can afford to complete the journey that follows.
On 23 October, Google described its work with Nexus AI to bring AI-assisted tuberculosis screening into African healthcare facilities. This is a company account of implementation, not proof that patients are recovering sooner or losing less income. Yet the possibility matters: faster interpretation of chest X-rays could help scarce clinical teams reach more people. The test of success should be completed care and the burden on households, alongside the speed of the scan.
Imagine a daily-wage worker attending a mobile screening camp. An abnormal result may be the beginning of timely help. It may also mean another journey, another missed shift and an anxious wait for further tests. These are hypothetical paths from the same initial service. The difference lies in what the health system has arranged next.
Screening identifies people who need further investigation; it does not, by itself, settle a diagnosis. WHO's screening guidance explicitly connects early detection with subsequent confirmatory testing. A faster first step can be valuable while leaving the most difficult part of a patient's journey untouched.
There is a serious case for moving quickly. WHO reported in June 2025 that six evaluated software products met its performance standards for TB screening in people aged 15 and older. That is evidence for specific tools and uses, not a universal endorsement of AI. But withholding useful screening until every weakness in healthcare is repaired would also impose a human cost.
Google says its mine-site deployment can already combine screening and confirmatory testing in one visit. That is a promising design claim; it still needs evaluation through patient outcomes.
The choice is how to expand it. If more people are referred, laboratories need capacity, samples need transport and someone must explain results and arrange care. Nurses and community workers may inherit this additional work. A budget that counts only the machine's savings can make their labour disappear from the calculation without making it disappear from the clinic.
India offers a concrete warning about what such accounting misses. A 2024 study of 1,482 patients with drug-susceptible TB across four states found substantial household costs, including income losses. Interviews took place between 2019 and 2022. Depending on how indirect costs were calculated, 30–61% faced costs reaching at least one-fifth of their pre-TB annual household income. This is neither a current national estimate nor an AI evaluation. It shows why the patient's economic experience belongs in the definition of better care.
The economic case for screening can still be strong. An August 2026 modelling study for South Africa found that computer-assisted chest radiography followed by confirmatory testing could be cost-effective under its assumptions. Its value depended partly on people reaching care. These are simulated outcomes, not results from Google's programme. They support a practical principle: purchasing detection and purchasing an effective route to treatment are inseparable investment decisions.
Who captures the gains will depend on contracts. If a supplier is paid for each scan, rising volume can reward the supplier before patients benefit. That is an incentive risk, not an allegation about this partnership. Health systems should measure referrals completed, time to treatment and household travel and income costs. Payment arrangements should also discourage providers from favouring people who are easiest to follow up.
Three futures are plausible. In the better one, screening comes with sample collection, reachable clinical support and help with transport; patients gain both time and a clearer route to care. Clinicians use their judgment where it matters, supported by staff paid to manage follow-up.
In an uneven future, well-funded clinics connect these steps while poorer districts build longer waiting lists. Geography continues to decide who benefits. In a harmful future, impressive screening totals conceal people lost between services, while families absorb extra journeys and worry.
Agency matters here too. People need an explanation they understand, accessible arrangements for disability or distance, and a named place to turn next. A digital notification should not become the only bridge to care for someone without reliable connectivity or confidence reading it.
Questions for founders and investors
Does the product reduce the cost of a completed patient journey, or only one institution's workload? Who funds the work after a positive screen? Can patients and clinics obtain useful records and continue care if the supplier changes?
Governments and health providers should fund diagnosis and follow-up alongside screening, publish where patients drop out, and involve patients in redesigning those handoffs. Faster detection deserves investment. The investment should carry people through the next door.
Sources