Abstract
BACKGROUND: Open-access data challenges can accelerate innovation in artificial intelligence-based tools. In the Cough Diagnostic Algorithm for Tuberculosis (CODA TB) DREAM Challenge, we developed and independently validated cough sound-based artificial intelligence algorithms for tuberculosis screening. METHODS: We included data from 2143 adults with ≥2 weeks of cough from outpatient clinics in India, Madagascar, the Philippines, South Africa, Tanzania, Uganda, and Vietnam. A standard tuberculosis evaluation was completed, and ≥3 solicited coughs were recorded using a smartphone. We invited teams to develop models using training data to classify microbiologically confirmed tuberculosis disease using (1) cough sound features only and/or (2) cough sound features with routinely available clinical data. After 4 months, they submitted the algorithms for independent test set validation. Models were ranked by area under the receiver operating characteristic curve (AUROC) and partial AUROC (pAUROC) to achieve at least 80% sensitivity and 60% specificity. RESULTS: Eleven cough models and 6 cough-plus-clinical models were submitted. AUROCs for cough models ranged from 0.69 to 0.74, and the highest performing model achieved 55.5% specificity (95% confidence interval, 47.7%-64.2%) at 80% sensitivity. The addition of clinical data improved AUROCs (range, 0.78-0.83); 5 of the 6 models reached the target pAUROC, and the highest performing model had 73.8% specificity (95% confidence interval, 60.8%-80.0%) at 80% sensitivity. The AUROC varied by country and was higher among male and human immunodeficiency virus-negative individuals. CONCLUSIONS: In a short period, an open-access data challenge facilitated the development of new cough-based tuberculosis algorithms and demonstrated potential as a tuberculosis screening tool.