In the evolving landscape of healthcare, artificial intelligence (AI) is gaining traction as a potential aid in clinical decision-making, though its effectiveness in real-world pediatric diagnosis remains uncertain. A recent study published in Pediatric Investigation sheds light on this issue, revealing that AI models, when applied to genuine clinical scenarios, surpass clinicians in diagnostic precision, especially for rare diseases. The research suggests that a hybrid model, combining human expertise with AI, could yield the highest diagnostic success, underscoring AI’s potential to enhance diagnostic accuracy and patient care outcomes.

Diagnosing pediatric conditions can be particularly daunting due to the nuanced and overlapping symptoms of rare diseases. This complexity often leads to delays in treatment and heightened risk of complications. Although AI has shown promise in healthcare, prior studies predominantly utilized simplified or curated case data rather than actual clinical cases, leaving a gap in understanding AI’s performance in real-world settings where decisions are made under constraints. Addressing this gap, a team led by Dr. Cristian Launes from Hospital Sant Joan de Déu in Barcelona evaluated AI’s diagnostic capabilities using real pediatric clinical cases.

Dr. Launes, a Clinical Professor and pediatrician specializing in pediatric infectious diseases, spearheaded the study, which compared four advanced AI language models with the diagnostic performance of 78 pediatric clinicians across 50 cases. These cases included both common and rare diseases. The study’s findings revealed that AI models outperformed human clinicians, particularly in diagnosing rare conditions that are often missed initially. However, clinicians showed prowess in complex scenarios, highlighting a divergence in diagnostic reasoning between humans and AI.

The study did not simulate an interactive “human-plus-AI” diagnostic process but instead estimated potential outcomes by evaluating whether the correct diagnosis appeared in the top-five predictions of either the AI or the clinicians. This approach indicated that a combined human and AI effort could achieve a 94.3% top-five union accuracy. Dr. Launes emphasized that AI should be viewed as a clinician-supervised second opinion, particularly in challenging cases involving rare diseases, to broaden diagnostic possibilities and reduce missed diagnoses.

AI tools in medical diagnostics are classified as high-risk applications by the European Union AI Act, necessitating stringent governance measures including risk management, transparency, and oversight. The study also noted that additional clinical data, such as lab results, improved diagnostic accuracy for both AI and clinicians, highlighting the importance of integrating AI into ongoing clinical processes. This integration could foster collaborative, data-driven decision-making, enhancing the role of AI as a supportive tool in pediatric healthcare and encouraging collaboration between clinicians, engineers, and policymakers.