ResApp Health Limited announced positive results for a new novel cough audio-based COVID-19 screening test that only requires a smartphone. In a pilot clinical trial of 741 patients (446 COVID-19 positive) recruited in the United States and India, ResApp's screening test, which uses machine learning to analyse the sound of a patient's cough, was found to correctly detect COVID-19 in 92% of people with the infection. ResApp's algorithm achieved an area under the curve (AUC) of 0.93 using cough audio and patient-reported symptoms across both trials.

AUC is a standard measure of how well a test distinguishes between two diagnostic groups, where a value of 1 represents a perfect test. A value greater than 0.9 is considered outstandingi. With this AUC, ResApp can select different operating points depending on the setting to achieve either high sensitivity, high specificity, or a balanced sensitivity and specificity.

For use as a screening test prior to a rapid antigen or polymerase chain reaction (PCR) test to rule out COVID- 19, an operating point that provides a 92% sensitivity and 80% specificity could be selected. This sensitivity exceeds the real-world measured sensitivity of rapid antigen testsii,iii. The combination of high sensitivity and 80% specificity results in 8 out of 10 people without COVID-19 being correctly screened as negative and not requiring a follow-on rapid antigen or PCR test.

ResApp will initially target use in settings where frequent COVID-19 testing is required, such as employee, healthcare worker and student screening, travel, sports, entertainment, and aged care. In these settings a high sensitivity test that only requires a smartphone would significantly reduce the number of rapid antigen or PCR tests required, improving availability, reducing costs, and reducing environmental impact. A smartphone-based test also has the ability to improve security and reporting of results using biometric identification such as facial recognition.

To ensure that the algorithm is specific to COVID-19 it was tested against the Breathe Easy dataset. The Breathe Easy dataset was collected prior to the COVID-19 pandemic and was used to train and validate ResApp's existing regulatory-approved (Australian TGA and CE Mark) ResAppDx product for acute respiratory disease diagnosis. This dataset includes 1,007 patients with a variety of non-COVID-19-related respiratory conditions including upper respiratory tract infections, asthma exacerbations, COPD exacerbations and other viral lung infections including pneumonia.

The algorithm achieved greater than 90% specificity for these patients. This important result demonstrates that the algorithm is identifying COVID-19 and not general respiratory illness. Consistent performance was found in analysis of a range of subgroups, including study arm and location, age, gender, and vaccination status.

While genomic sequencing was not available, analysis of the data over two time periods, one where Delta was the dominant variant and another where Omicron was dominant, demonstrated consistent performance. As expected, and similarly to rapid antigen tests, the algorithm showed lower performance in asymptomatic patients, although only a small number (14) of asymptomatic patients were recruited in the trials. The performance of the algorithm was obtained using K-fold cross-validation to provide an estimate of performance on unseen data.

ResApp intends to submit the results for publication in a peer-reviewed journal in the coming weeks.