ARTIFICIAL INTELLIGENCE FOR RARE PEDIATRIC DISEASE DIAGNOSIS: GENOMIC ANALYSIS, PHENOTYPIC MATCHING, AND CLINICAL DECISION SUPPORT

Abstract

Background: Rare diseases affect 300 million people globally, with 70% presenting in childhood. The diagnostic odyssey—averaging 4.8 years for rare pediatric conditions—causes substantial suffering and delayed treatment. Objective: To evaluate AI approaches for accelerating rare pediatric disease diagnosis through genomic variant interpretation, phenotypic matching, and clinical decision support. Methods: A multicenter study evaluated three AI tools: a deep learning variant prioritization engine, an HPO-based phenotypic similarity algorithm, and an ensemble clinical decision support system in 1,847 unsolved pediatric cases. Results: AI tools identified the causative genetic variant in 31.2% of previously unsolved cases within 30 days. AI-assisted diagnosis reduced the diagnostic odyssey by a mean of 2.3 years. Combined AI genomic-phenotypic analysis outperformed either modality alone. Conclusion: AI substantially improves diagnostic yield and reduces diagnostic delay in rare pediatric diseases, with transformative implications for affected families and precision treatment access.