CAGE-TB: Cough Audio Classification as a TB Triage Test

University of Stellenbosch (Other)
Overall Status
Recruiting ID
Amsterdam Institute for Global Health and Development (Other), University of Göttingen (Other), Makerere University (Other)

Study Details

Study Description

Brief Summary

TB is the single biggest infectious cause of death (1.5 million died in 2018), killing more HIV-positive people than any other disease, and is arguably the most important poverty-related disease in the world. TB's estimated incidence in Africa has been declining over recent years but progress is slow and plateauing. To avert stagnation, truly innovative and ambitious technologies are needed, especially those that improve case finding and time-to-diagnosis as, in mathematical models based on the TB care cascade framework, interventions that accomplish this will have the most impact on disrupting population-level transmission, including when deployed at facilities where patients are readily accessible. Critically, these interventions (triage tests) must promote access to confirmatory testing (e.g., Xpert MTB/RIF Ultra) by enabling patients to be referred rapidly and efficiently during the same visit. The investigators will optimise and evaluate a technology that, aside from the investigators early case-controlled study to show feasibility, is hitherto not meaningfully investigated for TB. This gap is alarming given, on one hand, the enormity of the TB epidemic and the need for a triage test and, on the other hand, promising proofs-of-concept that demonstrate high diagnostic accuracy of cough audio classifier for respiratory diseases such as pneumonia, asthma. pertussis, croup, and COPD. In some cases, these classification systems are CE-marked, awaiting FDA-approval, and subject to late-stage clinical trials. This demonstrates the promise of the underlying technological principle. CAGE-TB's innovation is further enhanced by: applying advanced machine learning methods that the team have specifically developed for TB patient cough audio analysis, use of mixed methods research - drawing from health economics, implementation science, and medical anthropology - to inform product design and assess barriers and facilitators to implementation, and uniquely for a TB diagnostic test, its potential deployment as a pure mHealth (smartphone-based) innovation that mitigates many barriers that typically jeopardise TPP criteria fulfilment.

Condition or Disease Intervention/Treatment Phase
  • Diagnostic Test: Cough sounds

Detailed Description

CAGE-TB is a diagnostic evaluation study that assesses a TB cough audio signature's potential to be used in a smartphone application to detect potential TB from a cough sound to screen (triage) TB. The purpose of CAGE-TB is to promote the adoption of mobile health (mHealth) based cough audio triage testing for active pulmonary TB in health facilities located in high burden settings. The study is funded by the EDCTP2 programme supported by the European Union and involves four international partners. The study participants, older than 12 years, include participants who have a cough for a duration exceeding two weeks that present to healthcare clinics where the investigators have clinical recruitment infrastructure and permissions to conduct TB research. In this two-phase observational, cross-sectional study, each participant will be seen once only, at diagnosis, and no intervention is planned. In the first phase, the investigators will collect data from a discovery cohort, which will be used to train a machine learning algorithm. During the second phase, data will be collected from a validation cohort, comprising a larger number of participants from two geographically distinct study sites, which will be used to evaluate the performance of the algorithm. The aims of this study are to: (1) generate and separately validate a cough audio classifier that meets WHO triage test TPP sensitivity and specificity criteria. This aim lays the foundation for CAGE-TB by generating a classifier and a common public resource (cough sounds database) for potential later use in other studies. (2) Produce data on potential cost savings of cough audio app for triage by collecting primary data to demonstrate potential cost savings estimated using state-of-the-art methods to satisfy a key TPP criterion (<USD 2 per patient). This aim will support the further independent evaluation of the classifier, including in clinical trials focused on patient endpoints. (3) Package the technology into an easy-to-use smartphone app. Many TB tests offer improvements in accuracy and cost, but are not widely adopted. This aim is designed to mitigate this risk with the cough classifier and by using the advantages of mHealth, a product may be delivered that is readily usable by nurses, trusted by patients, and capitalised upon by healthcare providers. Accomplishing this requires incorporating important features into the mobile application, such as connectivity, automated reporting, build-in guidance, and quality control, which are important but often neglected components of the WHO triage test TPP profile. (4) Form a foundation for subsequent studies where the app will be evaluated to measure its impact on patient care, to build evidence for global policy change and adoption. The objectives of CAGE-TB include: (1.1) sampling (cough audio, sputum microbiology) all patients with a prolonged cough entering primary health facilities irrespective of reason for presentation (n=473 in Cape Town discovery cohort); (1.2) use machine learning to develop a cough audio classifier to differentiate between TB and non-TB coughs; (1.3) evaluation of the classifier for sensitivity and specificity in new validation cohorts (n=511 in Cape Town, n=767 in Kampala); (2) in both settings, the investigators will use validated tools to calculate potential provider costs averted and conduct mixed methods research to identify barriers and facilitators to inform development of the mHealth solution (smartphone app) intended for use by minimally trained health workers with the final product ideally functioning offline without necessarily needed to sync to an online server for processing done at Stichting-Amsterdam Institute for Global Health and Development (AIGHD); (3) the app will be further refined and made user-friendly, simple and visually intuitive based on study feedback towards a more final product which is free for any person to use; and (4) this mHealth solution packaged into a smartphone app will be sent for review to large international stakeholders such as the Foundation for Innovation and Diagnostics (FIND) and the World Health Organization (WHO) and lead by colleagues at AIGHD with Stellenbosch University and Makerere University input to make this TB triage application relevant for use in African settings and populations.

Study Design

Study Type:
Anticipated Enrollment :
1751 participants
Observational Model:
Time Perspective:
Official Title:
Automated Smartphone-based Cough Audio Classification for Rapid Tuberculosis Triage Testing (Cough Audio triaGE for TB; CAGE-TB)
Actual Study Start Date :
Apr 19, 2022
Anticipated Primary Completion Date :
Jun 1, 2024
Anticipated Study Completion Date :
Sep 30, 2024

Arms and Interventions

Arm Intervention/Treatment
Discovery Cohort

An anticipated number of 473 participants will be recruited in Cape Town, South Africa. Data (cough audio) will be collected and used to train a machine learning algorithm. The cough audio signal specific for TB will be refined. During the discovery phase, the ground truth obtained through biological testing of sputum specimens will be used to inform the machine learning.

Diagnostic Test: Cough sounds
The investigators will discover a cough audio signature and then validate it.

Validation Cohort

In the validation phase, the cough audio signature will have its sensitivity and specificity measured in new patients in Cape Town, South Africa (n=511) and Kampala, Uganda (n=767). The data will be used to evaluate the performance of the algorithm.

Diagnostic Test: Cough sounds
The investigators will discover a cough audio signature and then validate it.

Outcome Measures

Primary Outcome Measures

  1. Develop and validate algorithms that can distinguish between TB and non-TB coughs [24 months]

    Cough audio data will be collected and used to define the cough audio signal specific for TB. The optimised TB audio signature will then have its sensitivity and specificity measured in new patients to evaluate the performance of the algorithms.

  2. Finalised smartphone-based mHealth application [30 months]

    The best-performing algorithm will be incorporated into a smartphone app, which will be designed with human-centered approach, that can be used as a point-of-care triage test for TB.

  3. Avert unnecessary Ultra tests [24 months]

    The investigators will calculate potential cost savings that the application will be able to facilitate to avoid unnecessary tests.

Eligibility Criteria


Ages Eligible for Study:
12 Years and Older
Sexes Eligible for Study:
Accepts Healthy Volunteers:
Inclusion Criteria:
  • participant must be at least 12 years old

  • participant must have a prolonged cough (for at least two weeks)

  • participant must provide informed consent

  • participant shall have a known HIV status or be willing to undergo standard of care HIV testing and counseling

Exclusion Criteria:
  • individuals who refuse informed consent

  • individuals who have received treatment for TB in the 60 days prior to enrolment

  • individuals who are unable to provide a sputum specimen for microbiological testing

  • individuals who have haemoptysis or a bloody cough with any forced coughs for audio recordings

Contacts and Locations


Site City State Country Postal Code
1 Stellenbosch University Cape Town Western Cape South Africa 7505
2 Makerere University Kampala Uganda 7062

Sponsors and Collaborators

  • University of Stellenbosch
  • Amsterdam Institute for Global Health and Development
  • University of Göttingen
  • Makerere University


  • Principal Investigator: Grant Theron, PhD, University of Stellenbosch

Study Documents (Full-Text)

None provided.

More Information


None provided.
Responsible Party:
Grant Theron, Professor, University of Stellenbosch Identifier:
Other Study ID Numbers:
  • M21/10/022
First Posted:
Apr 7, 2022
Last Update Posted:
Jul 26, 2022
Last Verified:
Jul 1, 2022
Individual Participant Data (IPD) Sharing Statement:
Plan to Share IPD:
Studies a U.S. FDA-regulated Drug Product:
Studies a U.S. FDA-regulated Device Product:
Keywords provided by Grant Theron, Professor, University of Stellenbosch
Additional relevant MeSH terms:

Study Results

No Results Posted as of Jul 26, 2022