Recent advancements in artificial intelligence (AI) are shedding light on its potential to enhance pediatric diagnostics, particularly for rare diseases. A study published in Pediatric Investigation, spearheaded by Dr. Cristian Launes from Hospital Sant Joan de Déu in Barcelona, reveals that sophisticated AI models surpass clinicians in diagnostic accuracy. This research utilized authentic clinical cases to assess AI’s performance, with findings indicating that a collaborative human-AI approach yields the highest success rates. The study underscores AI’s promising role as a complementary tool in improving diagnostic precision and patient outcomes.
Pediatric diagnosis can often be challenging due to the subtle or overlapping symptoms of rare diseases, which can lead to delayed treatment and complications. While AI has shown promise, previous studies have relied on simplified cases rather than real-world data, leaving a gap in understanding AI’s efficacy in everyday clinical settings. The study compared four advanced language models against 78 pediatric clinicians across 50 cases, encompassing both common and rare conditions. These assessments were based on patient summaries from the initial 72 hours of presentation, evaluating diagnostic accuracy and consistency.
The results highlighted that AI models demonstrated superior diagnostic accuracy over clinicians, particularly in identifying rare diseases. However, in complex scenarios, clinicians exhibited strengths that AI could not match, pointing to differing diagnostic reasoning approaches. The study assessed a hypothetical “human-plus-AI” workflow, estimating that the combination could achieve a 94.3% Top-5 union accuracy, suggesting clinicians and AI might offer different correct hypotheses in challenging cases. Dr. Launes emphasized that AI should be viewed as a clinician-supervised second opinion, especially with rare diseases, to broaden differential diagnoses and reduce missed diagnoses.
From a governance perspective, diagnostic decision-support systems are considered high-risk under the European Union AI Act, necessitating robust risk management, transparency, and human oversight. The study also found that additional clinical data, such as lab results, improved diagnostic performance, emphasizing the importance of integrating AI within continuous, information-rich clinical workflows. Dr. Launes noted that AI is most effective when part of an ongoing clinical process, where clinicians continually update and verify the clinical picture to feed the model.
This study highlights the potential for AI-assisted tools in pediatric healthcare to support earlier and more accurate diagnoses, particularly for rare diseases. Integrating AI into clinical workflows promises to foster more collaborative and data-driven decision-making, encouraging partnerships among clinicians, engineers, and policymakers. Although challenges remain, the findings point to AI’s promising role as a supportive tool in pediatric diagnostics, especially when used alongside human expertise.