AI and the World's Hard Problems · Medicine & drug discovery · Nature (computational biology) · 2026-10-24
A machine learning-based web application aids Kawasaki disease diagnosis and predicts its outcomes
Kawasaki disease must be promptly treated with IVIG, or patients may become unresponsive to treatment or develop aneurysms. Using data from 628 KD and 155 KD-like cases, the authors built machine learning models to predict KD diagnosis, IVIG unresponsiveness and aneurysm formation; by day 4 the AUCs exceeded 0.7, and on day 5 the KD model reached a positive predictive value above 90%. The models were packaged as a web-based application for early diagnosis and risk stratification in both resource-limited and resource-rich settings.