Artificial intelligence in pediatric nephrology: current applications and emerging frameworks for evidence generation

Comprehensive Summary

Artificial intelligence (AI) is being explored in pediatric nephrology through this narrative review because clinical evidence is often limited. Randomized trials are difficult to conduct due to small patient populations, ethical constraints, and logistical barriers. Clinical decisions about dialysis, prognosis, and long-term kidney function frequently rely on adult data or expert opinion. Large datasets from longitudinal studies, including the Chronic Kidney Disease in Children (CKiD) study and international registries, contain necessary clinical information but are difficult to analyze in practice. Machine learning approaches analyze high-dimensional data to identify patterns related to disease progression and treatment outcomes. Common approaches include supervised learning using labeled outcomes, unsupervised learning to detect patterns without predefined outcomes, reinforcement learning that evaluates decision strategies through feedback, and hybrid causal machine learning that compares treatment strategies. Current AI applications in pediatric nephrology involve prediction and classification. Examples include imaging and genetic analysis in congenital anomalies of the kidney and urinary tract (CAKUT) and chronic kidney disease (CKD) progression predictions using estimated glomerular filtration rate and clinical variables. Emerging research focuses on integrating multimodal models using clinical records, imaging, and genomic data to monitor transplant outcomes and predict dialysis complications. Automated kidney biopsy analysis can improve quantification of fibrosis and inflammation. Proposed future applications include patient-specific “digital twin” models that simulate treatment scenarios and AI frameworks that integrate genomic and multi-omics datasets to interpret uncertain variants in rare kidney diseases. Limitations include small pediatric datasets, class imbalance in rare diseases, and limited validation across institutions. Future work emphasizes standardized phenotypes, interoperable data systems, federated research networks, transparency, and inclusion of patient-reported outcomes for ethical AI use.

Outcomes and Implications

AI can support more informed decision making and more personalized, patient-centered care for pediatric patients and their families. Because AI-based predictions can update as patient conditions change, they may be particularly useful for monitoring transplant outcomes and treatment progression. In clinical practice, AI is intended to support rather than replace physician judgment. AI may also help refine disease subtypes, especially for rare kidney diseases that are currently classified using broad phenotype-based categories in nephrology. Advancing these approaches will require standardized phenotyping, extraction of longitudinal dataset information, and closer collaboration among clinicians, geneticists, and data scientists. In addition, AI can help inform treatment and prognosis decisions in situations where clinical trials are not feasible, allowing observational data to guide protocol development and decision making. More broadly, AI research may encourage greater collaboration among stakeholders, including clinicians, geneticists, data scientists, ethicists, and patients and families.