Artificial intelligence in pediatric endocrinology: clinical applications, governance, and future directions.
Clinical Snapshot
PICO Framework
| P — Population | Pediatric patients with endocrine conditions including diabetes mellitus, obesity, thyroid disorders, growth and puberty disorders, and those requiring bone age assessment |
| I — Intervention | Artificial intelligence-based tools including machine learning algorithms, large language models, automated insulin delivery systems, digital twin modeling, federated learning, and explainable AI frameworks |
| C — Comparator | Standard clinical care without AI augmentation; conventional diagnostic and therapeutic approaches in pediatric endocrinology |
| O — Outcomes | Diagnostic accuracy, glycemic forecasting performance, complication screening efficacy, risk stratification, clinical workflow efficiency, algorithmic bias, health equity, and governance considerations |
Bottom Line
This narrative review from clinician-researchers at two major US pediatric academic centres provides a clinically oriented overview of AI applications in pediatric endocrinology, spanning diabetes management, obesity, thyroid disease, growth disorders, and bone age assessment. The review's principal value lies in its accessible synthesis of emerging technologies — including automated insulin delivery, glycemic forecasting, and explainable AI frameworks — and its frank acknowledgment of governance challenges including algorithmic bias, data privacy, and health equity. However, the absence of a systematic search strategy, formal quality appraisal, quantitative synthesis, or GRADE assessment means this cannot be treated as high-level evidence. It functions as an expert opinion piece and educational resource rather than a practice-changing systematic review. For Australian clinicians, the most immediately actionable content concerns TGA-registered automated insulin delivery systems and the equity implications of deploying AI tools trained on non-representative datasets in Indigenous pediatric populations. Clinicians should treat AI-derived outputs as decision-support adjuncts requiring critical clinical interpretation, consistent with the authors' own Human-in-the-Loop recommendation. AI literacy integration into APEG training curricula, as advocated by the authors, warrants serious consideration by Australian training program directors.
Key Findings
P Value: Not reported
Effect Size: Not reported — no pooled effect sizes presented; individual study performance metrics referenced narratively without aggregation
Primary Outcome: Narrative synthesis of AI applications across pediatric endocrinology domains: glycemic forecasting and automated insulin delivery in diabetes; diagnostic support in obesity, thyroid disease, and growth disorders; bone age assessment automation
Nnt Or Sensitivity: Not reported quantitatively; bone age assessment and diabetic complication screening are described as demonstrating enhanced diagnostic accuracy relative to conventional methods, but specific sensitivity/specificity values are not synthesised in the abstract
Confidence Interval: Not reported
Clinical Application
Feasibility varies substantially by AI application. Automated insulin delivery systems (e.g., closed-loop systems) are already commercially available and in clinical use. Bone age AI tools are increasingly integrated into radiology workflows. Glycemic forecasting and digital twin modeling remain largely in research or early implementation phases. Widespread adoption is constrained by data infrastructure requirements, clinician AI literacy, regulatory approval status, and health system interoperability. In Australia, several AI-relevant considerations apply. The TGA regulates AI-based Software as a Medical Device (SaMD) under its Digital Health framework, and clinicians should verify TGA registration status before clinical adoption of any AI diagnostic tool. The RACGP and APEG (Australasian Paediatric Endocrine Group) have not yet published specific AI governance guidelines for pediatric endocrinology as of the appraisal date. Automated insulin delivery systems including hybrid closed-loop devices are PBS-listed for eligible patients with type 1 diabetes, representing the most clinically mature AI-adjacent technology in this domain. Federated learning approaches discussed in the review are particularly relevant to Australia given geographic dispersion of pediatric endocrinology expertise and the potential to leverage data from distributed centres (e.g., Children's Hospital Network, Royal Children's Hospital Melbourne) without centralising sensitive patient data. Health equity concerns raised by the authors are directly applicable to Aboriginal and Torres Strait Islander children, who face disproportionate burdens of type 2 diabetes and obesity and are systematically underrepresented in AI training datasets predominantly derived from North American and European cohorts. Pediatric patients (age range not explicitly specified) with type 1 or type 2 diabetes mellitus, obesity, thyroid disorders (including thyroid nodule evaluation), growth hormone deficiency, pubertal disorders, and those requiring skeletal maturity assessment. Most directly applicable in tertiary pediatric endocrinology settings with access to digital health infrastructure.
Abstract
PURPOSE OF REVIEW: Artificial intelligence has transitioned from theoretical promise to practical implementation across medicine including pediatric endocrinology. This review examines artificial intelligence applications across diabetes care, obesity management, thyroid disorders, growth and puberty, and bone age assessment followed by discussion of limitations and future directions. RECENT FINDINGS: Artificial intelligence-driven tools ranging from machine learning algorithms to multimodal large language models enable personalized risk prediction, diagnostic support, and treatment optimization. Human-in-the-Loop frameworks enhance safety by integrating computational efficiency with clinician oversight. In diabetes, artificial intelligence supports glycemic forecasting, automated insulin delivery, complication screening, and digital twin modeling for therapeutic discovery. In obesity, thyroid disease, growth disorders, and bone age assessment, artificial intelligence enhances diagnostic accuracy, risk stratification, and clinical workflow efficiency. However, widespread implementation raises concerns regarding algorithmic bias, data privacy, transparency, regulatory oversight, and health equity. Addressing these limitations through explainable artificial intelligence, federated learning, and continuous performance monitoring is essential. SUMMARY: Artificial intelligence should be viewed as an augmentative tool that enhances clinical judgment in pediatric endocrinology. Preparing the next generation of pediatric endocrinologists will require intentional integration of artificial intelligence literacy into training programs, equipping clinicians to critically evaluate algorithms, interpret outputs responsibly, and participate in the ethical development of future artificial intelligence-enabled tools.
References
- 1.Sasidharan Pillai, S., & Ashraf, A. P. (2026). Artificial intelligence in pediatric endocrinology: clinical applications, governance, and future directions. Current Opinion in Pediatrics. Advance online publication. https://doi.org/10.1097/MOP.0000000000001564
Related Research
The Journal of clinical endocrinology and metabolism
Diazoxide Choline Extended-release Tablets in Prader-Willi Syndrome: A Randomized, Double-blind, Withdrawal Period Study
19 May 2026
JMIR research protocols
Evaluating Patient-Entered Electronic Health Data as a Strategy to Improve Quality of Care in a Diabetes Clinic: Protocol for a Randomized Controlled Trial
13 May 2026
The Journal of clinical endocrinology and metabolism
Effect of Vosoritide Therapy on IGF-I and Endogenous C-Type Natriuretic Peptide in Hypochondroplasia
24 Apr 2026
This content is for educational purposes for healthcare professionals only and does not constitute clinical advice. Clinical decisions should be based on individual patient assessment, current guidelines, and appropriate specialist consultation. Editorial Standards · Privacy Policy · Terms of Service