Transparency and Validity of Artificial Intelligence Applications in Pediatric Diabetes: A Systematic Review

Area of Science:

• Pediatric Endocrinology
• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Artificial intelligence (AI) offers potential advancements in pediatric diabetes management.
• Clinical adoption of AI tools is currently limited by concerns regarding transparency and validity.
• Systematic evaluations of AI's transparency and validity in this field are scarce.

Purpose of the Study:

• To systematically review the transparency and validity of AI applications used in pediatric diabetes management.
• To assess the methodological rigor, reporting standards, and clinical readiness of existing AI studies.
• To identify gaps and provide recommendations for future AI development and implementation in pediatric diabetes.

Main Methods:

• Systematic review following PRISMA 2020 guidelines.
• Searched major scientific databases (Scopus, PubMed, IEEE Xplore, Web of Science, Embase).
• Included 10 studies on AI in pediatric diabetes, assessing transparency, validation, and risk of bias (QUADAS-2).

Main Results:

• AI applications covered glucose prediction, hypoglycemia risk, and insulin dosing.
• Transparency varied, with only 60% disclosing algorithm details.
• External validation was present in only 30% of studies, and 40% had concerns regarding risk of bias.
• Algorithmic opacity and small validation cohorts were noted limitations.

Conclusions:

• AI holds significant potential for pediatric diabetes management.
• Inconsistent transparency and insufficient validation impede clinical translation.
• Future research should focus on standardized reporting, multicenter validation, and diverse populations to ensure AI reliability and equity.