Can a machine-learning model predict if a child will have coronary artery lesions?
Coronary artery lesions (CAL) are a serious complication of Kawasaki disease (KD). Predicting which children will develop CAL can help doctors start early treatment. Recent studies show that machine-learning models, which analyze routine lab tests and clinical features, can predict CAL with good accuracy. These models are not yet standard in all clinics, but they are a promising tool.
What the research says
Several studies have built machine-learning models to predict CAL in children with KD. One study using the CatBoost algorithm reported an area under the curve (AUC) of 0.953 in training and 0.874 in internal validation, with sensitivity of 0.908 and specificity of 0.860 2. Another study developed a column chart model based on LASSO regression, achieving AUCs of 0.879 in training and 0.859 in validation 7. A third study used logistic regression to create a nomogram from routine lab indicators, which also showed good predictive performance 1. A systematic review and meta-analysis confirmed that machine learning can accurately discriminate KD from other febrile illnesses and predict CAL, though the authors noted variability across studies 8. Overall, these models rely on factors like age, sex, fever duration, and lab values such as C-reactive protein (CRP) and platelet count 127.
What to ask your doctor
- Are there any machine-learning tools or risk scores used at your hospital to predict coronary artery lesions in Kawasaki disease?
- What lab tests or clinical signs do you monitor most closely to assess my child's risk of coronary artery problems?
- If a prediction model suggests high risk, what treatment options (like steroids or infliximab) might be considered?
- How often should my child have follow-up echocardiograms to check for coronary artery changes?
This question is drawn from common patient questions about Cardiology and answered using cited medical research. We do not provide individualized advice.