4-CAGE-TB

The challenge

In sub-Saharan Africa, an estimated 2.5 million new cases of tuberculosis occur each year, of which about 1 million are undiagnosed. This means that many patients do not receive timely treatment, and there are more opportunities for the disease to spread.

Case finding is therefore an important aspect of TB control, but presents multiple challenges. Identification of TB cases is not straightforward, as symptoms are non-specific and traditional culture methods are very time-consuming. Screening large numbers of people for TB is therefore very difficult to implement. Molecular platforms have simplified case detection; however, their use in routine population screening would be prohibitively expensive and overwhelm health systems.

A possible solution could be triage, whereby a simple and inexpensive test is used to distinguish those most likely to have TB, who could then be followed up using more sensitive molecular testing.

The project

The 4-CAGE-TB project has been exploring an innovative approach based on an analysis of cough vocalisations. Coughs are a sensitive way to detect tuberculosis disease (most patients with TB have a cough), but they are not very specific (patients may have a cough for a wide variety of other reasons).

To address this issue, the project has been using machine learning technology to identify the features of a ‘TB cough’ that would differentiate it from coughs of other causes. The coughs of a cohort of possible TB patients have been recorded, and their clinical history is then tracked, including the results of gold-standard TB diagnostic tests. Using this data, a ‘cough classifier’ has been trained to identify the acoustic features of coughs associated with a positive TB test.

The cough classifier thus developed will be integrated into a simple-to-use mobile phone app, which analyses a patient’s cough and assesses its similarity to a TB cough. A health worker is presented with a simple binary readout – that TB is likely or unlikely. If TB is likely, a patient can be referred for further testing. This approach would be quick and simple, making it possible to introduce into primary care settings.

The project was disrupted by the COVID-19 pandemic, which affected recruitment and led multiple project team members to be diverted to COVID-19-related work. The additional funding will enable the project to complete its work, including consultations with health workers on the design of the app and integration into practice, and on the acceptability of the approach. 

The project aims to identify key factors that would affect implementation and generate an initial indication of cost implications. The outcome of the project will be a prototype device and supportive evidence to justify its testing in a formal clinical trial.

Impact

The 4-CAGE-TB project could have a major impact on tuberculosis case detection in sub-Saharan Africa. It will:

• Demonstrate whether a cough classifier has the potential to distinguish between TB and other coughs, meeting the sensitivity and specificity criteria for triage set out by WHO.
• Produce a practical prototype for clinical evaluation.
• Provide an indication of its likely impact on TB case detection and costs of use; modelling suggests that a simple triage test could reduce costs by 40% compared with molecular testing of all presumptive cases. 

Ultimately, the project could deliver a simple, widely applicable health tool able to identify possible TB cases for follow-up, helping to reduce the huge number of new cases that are still being missed every year.