Interpretable machine learning model for predicting kidney failure among CAKUT children in multicenter large-scale study
Liu, T.; Wang, H.; Liu, J.; Zhao, X.; Xia, Y.; Wang, X.; Kang, Y.; Liu, C.; Gao, X.; Jiang, X.; Mao, J.; Li, Y.; Zhang, A.; Wang, M.; Bai, H.; Shen, T.; Dang, X.; Wang, D.; Zhang, R.; Lu, Y.; Shen, Q.; Nie, S.; Chen, Y.; Xu, H.; Zhai, Y.
Show abstract
Congenital anomalies of the kidney and urinary tract (CAKUT) are the leading cause of pediatric kidney failure, but predicting individual progression remains challenging. This multicenter study developed and validated POCC, a machine learning model for predicting kidney failure risk at 1, 3, and 5 years post-diagnosis in CAKUT patients. Two versions were created using data from 2,249 children. The general model achieved internal AUCs of 0.93-0.99 and external AUCs of 0.90-0.98 and 0.81- 0.90 in two independent validations at pediatric and general hospitals, respectively. The specialized model, integrating congenital-hereditary features, achieved internal AUCs of 0.93-0.99 and external AUCs of 0.91-0.96 in pediatric hospitals. Deployed online, POCC demonstrated 90.7% accuracy in real-world validation, with the specialized model reaching 100% sensitivity and specificity for 5-year predictions. As the first tool for multi-timepoint risk prediction across diverse CAKUT subphenotypes per patient, POCC has strong potential to support personalized management.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.