A third of the world’s 10.6 million annual tuberculosis cases go undiagnosed, and in southern Africa, HIV-driven TB is the leading infectious cause of death. The problem is that confirming TB requires a sputum sample, but many patients are too ill to produce one. Chest X-rays interpreted by AI software can flag likely cases cheaply and quickly, but the follow-on sputum test needed for confirmation costs around $28 per patient in Malawi—a major barrier to widespread use. The researcher has shown that factors like age, HIV status, and symptom severity affect the AI’s accuracy. By adjusting the threshold at which a patient is referred for sputum testing based on these individual characteristics, far fewer tests may be needed without missing cases. If this works, TB programmes in high-burden countries could become dramatically more efficient and affordable. The team will model existing datasets to find optimal thresholds, then run a randomised trial in Malawi comparing adaptive thresholds against the current fixed standard. They will work with policymakers to turn findings into real-world policy change, potentially improving access to TB diagnosis for millions.
View original technical description
In 2022, globally 10.6 million people were estimated to develop tuberculosis (TB), with one-third going undiagnosed and untreated. Countries in southern Africa have been severely impacted by TB, driven by generalised HIV epidemics, with TB now the leading infectious cause of death. Diagnosis of TB is challenging, hindering efforts to reach global elimination targets. Microbiological confirmation (by smear, culture, or molecular testing) requires a sputum sample, but many people with TB can't expectorate sputum due to illness and debility. Chest radiography can show early signs of TB, including detecting 'subclinical TB', where patients are sputum-positive but don't report symptoms. However, in high TB-burden countries, radiologists and doctors skilled in X-ray interpretation are severely lacking. Recently, the emergency of affordable digital radiography units and artificial intelligence software to interpret chest X-ray (DCXR-CAD) has been transformative for TB diagnosis. DCXR-CAD gives a probabilistic TB score; patients with a high score require confirmation with sputum testing. In Malawi, I led the first randomised trial of DCXR-CAD, demonstrating a substantial increase in case detection and reduction in time to treatment initiation, prompting WHO to make a favourable recommendation for use in adults in 2021. DCXR-CAD is accurate, cheap, and highly accessible in a wide range of community and clinical settings. However, follow-on confirmatory sputum testing is expensive (~US$28 full economic cost in Malawi) and remains the major barrier to widespread implementation. We have recently shown that older age, HIV status, severity of TB symptoms, and history of previous TB have a major impact on the specificity of DCXR-CAD. Therefore, adapting sputum-referral thresholds for DCXR-CAD based on patient characteristics could substantially reduce the number of sputum tests required to confirm TB, without impacting sensitivity. I hypothesise that by exploiting these individual-level determinants of prevalence and DCXR-CAD accuracy to set 'adaptive thresholds' for sputum testing referral, we could substantially reduce programmatic costs, increasing the efficiency and affordability of TB programmes, driving improved access to TB diagnosis. In this NIHR Global Health Professorship, I will establish a world-leading team of epidemiologists, public health modellers, and health economists to investigate the effectiveness, cost-effectiveness, and public health impact of adaptive DCXR-CAD thresholds. We will model existing datasets identified through systematic review using advanced multilevel latent class Bayesian models, accounting for conditional-missingness of reference standard sputum tests, to determine optimal adaptive thresholds, and estimate their potential cost-benefit to TB programmes. We will subsequently undertake a pragmatic individually-randomised trial in Malawi to investigate the impact and cost-effectiveness of adaptive DCXR-CAD thresholds on follow-on confirmatory sputum testing compared to the current standard-of-care fixed DCXR-CAD threshold. We will work with affected communities, and local, regional, and global policymakers to translate research findings into meaningful community benefit and policy change.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know