A recent study highlights the potential of artificial intelligence (AI) to enhance diagnostic accuracy in pediatric healthcare, especially in identifying rare diseases. Research published in Pediatric Investigation indicates that AI models demonstrate superior diagnostic accuracy compared to clinicians, particularly in cases involving rare conditions. However, the most effective approach combines AI with human expertise, achieving the highest success rates. This finding underscores AI’s role as a complementary tool to boost diagnostic precision and improve patient outcomes.
Diagnosing pediatric patients is often challenging, especially when symptoms of rare diseases are subtle or overlap with more common conditions. The study, led by Dr. Cristian Launes of Hospital Sant Joan de Déu in Barcelona, Spain, evaluated the effectiveness of AI models using real-world clinical cases. Published on March 25, 2026, the research involved comparing the performance of four advanced language models against 78 pediatric clinicians across 50 cases, encompassing both common and rare diseases. The study’s results revealed that AI models generally outperformed clinicians in diagnostic accuracy, particularly in rare disease cases where AI models identified correct diagnoses that clinicians missed.
Despite AI’s advantages, the study noted that clinicians excelled in complex or context-dependent scenarios, highlighting the differences between human diagnostic reasoning and AI capabilities. The research did not involve a real-time, interactive human-AI diagnostic workflow but estimated the potential benefits of combining human and AI input. This “union” approach, where the correct diagnosis appeared in the Top-5 list of either clinicians or AI models, achieved a 94.3% accuracy rate. This suggests that AI can serve as a valuable second opinion under clinician supervision, especially in difficult cases involving rare diseases.
The study also emphasized the importance of integrating AI into continuous, information-rich clinical workflows. Additional clinical data, such as laboratory or imaging results, improved diagnostic performance for both AI models and clinicians. This finding underscores the need for ongoing clinical assessment and suggests that AI systems are most effective when part of an evolving process, where clinicians iteratively gather, verify, and curate information to feed into the models. The study advocates for AI-assisted tools to support earlier, more accurate diagnoses, particularly for rare diseases where specialized expertise may be scarce.
While the study demonstrates AI’s potential in pediatric diagnostics, it also highlights the need for robust oversight and governance frameworks. Medical diagnostic decision-support systems are considered high-risk applications under the European Union AI Act, necessitating careful risk management and human oversight. The researchers stress that AI should remain an advisory tool, with clear accountability and safeguards to address variability and mitigate the risk of misleading outputs. The integration of AI in clinical workflows could foster collaborative decision-making, encouraging closer partnerships between clinicians, engineers, and policymakers to enhance pediatric healthcare outcomes.