Endocrinology

Interpretable random‑forest predicts in‑hospital hypoglycemia in Chinese adults with T1DM

Multicenter retrospective cohort of 1,498 adult inpatients with T1DM developed an interpretable random‑forest predicting in‑hospital hypoglycemia with external AUC 0.831. Model sensitivity was 0.793 and specificity 0.748 in external validation, supporting effective risk stratification into four quartiles. Top predictors were hemoglobin, potassium, sodium, LDL, and age at onset, with several showing U‑shaped risk relationships. Authors present decision‑curve analysis showing net clinical benefit, but the model applies to hospitalized Chinese adults and needs external geographic validation. 1

GLP‑1 receptor agonists provide trivial cognitive gains but worse verbal fluency in non‑diabetic AD/PD

Meta‑analysis of 14 RCTs (n=1,260) found a small improvement in global cognition (SMD 0.14) with GLP‑1RAs, judged unlikely to be clinically meaningful. Verbal fluency worsened (SMD −0.43), and GLP‑1RAs reduced weight while increasing gastrointestinal adverse events. In Parkinson disease subgroup, depression improved modestly (MD −2.09) but below minimal important difference. Overall evidence shows no convincing disease‑modifying benefit and tolerability concerns limit clinical utility. 2

Four‑cluster phenotyping approximates classical T2D subtypes but shows modest stability

Prospective Mediterranean cohort (n=991) replicated Ahlqvist architecture with SAID plus four clusters: MARD, SIRD, MOD, and SIDD. Cluster separation was modest and bootstrap stability low (Jaccard 0.555–0.718) for the four‑cluster model. An unconstrained three‑cluster solution showed substantially higher stability (Jaccard 0.799–0.863). Authors advise local validation before using cluster labels to guide prognosis or therapy in this population. 3