Artificial intelligence in clinical endocrinology: A narrative review of predictiveanalytics, therapeutic optimization, and diagnostic applications

Background: Endocrine disease involves non-linear, multi-organ feedback loops and longitudinal data, making it suitable for ML and DL. This narrative review examines AI across five domains: diabetes risk prediction, metformin response/gut microbiome, CGM with closed-loop insulin, thyroid nodule ultrasound, and adrenal lesion imaging.

Methods: PubMed, Scopus, and Web of Science were searched for studies/reviews published between 2020 and 2026, using "machine learning," "deep learning," "diabetes prediction," "metformin," "gut microbiome," "continuous glucose monitoring," "closed-loop insulin," "thyroid nodule," and "adrenal incidentaloma." Priority was given to meta-analyses, randomized controlled trials, and cohorts; single-center case series were included where evidence was unavailable. As a systematic review, no PRISMA flow diagram or record count is reported.

Results: Ensemble models (random forest, XGBoost) predicted incident type 2 diabetes with 98-99% accuracy in a longitudinal cohort; accuracy can be inflated in imbalanced data, so interpret this alongside positive predictive value, not reported, and external validation remains uncommon. Deep learning models for thyroid ultrasound reached sensitivities of 87-91% and specificities of 83-95% across meta-analyses covering 146,000 patients. Closed-loop insulin delivery improves time-in-range, while reinforcement-learning dosing remains at simulation. Gut microbiome pharmacokinetic models achieved AUROC approximately 0.75; adrenal CT radiomics models reported AUCs above 0.85 in cohorts of 19-40 patients.

Conclusions: AI shows strong, evidence-backed value in endocrine risk prediction and image-based diagnosis, though diabetes-prediction accuracy warrants cautious interpretation given class imbalance. Evidence for therapeutic personalization, particularly metformin response, is earlier-stage. Larger prospective trials, external validation, standardized outcome reporting (including PPV), and improved interpretability are needed for clinical adoption.