Comprehensive Summary
In this study, Taejin et al. conducted a systematic review on the application of artificial intelligence (AI) in pediatric healthcare, examining its clinical potential, providing diverse perspectives, and exploring implications. This review evaluated studies across various domains, including diagnostic, treatment support, administrative applications, and ethical concerns. AI-driven models had valuable time-saving benefits in the early detection and treatment of pediatric conditions. Some models used biosignal data to predict possible deterioration and guide timely interventions, which might improve survival rates. Deep learning models trained on genetic, neuroimaging, and clinical data were found to have supplemental diagnostic precision for more complex illnesses, such as pediatric brain tumors. In addition to serving as an assistive device for clinicians, AI has been integrated into patient treatment devices, such as smart insulin pumps, capable of providing real-time patient data and treatment. In cardiology, they found that deep learning models could detect cardiac dysfunction with more ease than electrocardiograms (ECGs). Machine learning models have also used optimal classification trees to predict clinical outcomes of patients with congenital heart surgery. Large Language Models (LLMs) can help streamline administrative tasks and reduce the administrative burden within healthcare. While LLMs have good diagnostic performance comparable to pediatricians, there is still a great need to fine-tune these AI systems, which is imperative for proper clinical applications. Some of the challenges that should be considered are that these devices can analyze large amounts of medical data, which may lead to ethical issues regarding the data collection of patient health information (PHI) on public AI systems.
Outcomes and Implications
This systematic review also identified several medical implications and limitations. Bias within LLMs and other AI systems presents a significant risk, resulting in further health disparities in diagnostics, treatments, and resource allocation. Additional challenges also included false positive alerts from AI-based diagnostic and monitoring tools and decreased vigilance towards interventions due to the machine's responsiveness to clinical emergencies. Many researchers are also concerned that the “biomarkup” created as a diagnostic threshold for AI systems will lead to the machine detecting multiple conditions at once, which can be billable and unnecessary. Patient-centered care may be at risk and AI technologies should serve pediatric populations in an effective way rather than prioritizing costs over patient health. While AI demonstrates a strong future in improving diagnostic accuracy, streamlining clinical tasks, and enhancing treatment methods, current issues of bias, reliability, and ethical regulation still require further research before clinical implementation.