Association between glycemic variability and delirium in critically ill older adults patients: a retrospective cohort study

Delirium affects 30–70% of critically ill older adults patients and is associated with poor outcomes. While glycemic control is essential in intensive care, the impact of glycemic variability (GV) on delirium remains understudied, particularly in older adults patients. This study investigated the association between GV and delirium in critically ill older adults patients. We analyzed 16,335 patients aged ≥ 65 years from the MIMIC-IV database. GV was calculated as the coefficient of variation of blood glucose measurements and categorized into quartiles. Delirium was assessed using Confusion Assessment Method for the Intensive Care Unit (CAM-ICU). Multivariate logistic regression evaluated the association between GV and delirium, adjusting for demographics, clinical parameters, and comorbidities. We assessed the added predictive value of GV using area under the receiver operating characteristic curve (AUC), net reclassification improvement (NRI), and integrated discrimination improvement (IDI). Random forest and Boruta algorithm determined variable importance. Delirium incidence increased across GV quartiles (28.7% in Q1 vs. 41.8% in Q4, p < 0.001). After full adjustment, the highest GV quartile showed 42% higher delirium risk compared to the lowest (OR 1.42, 95% CI 1.18–1.71, p = 0.003), with a significant dose-response relationship. Random forest and Boruta algorithm identified GV as the most important predictor of delirium. Adding GV to prediction models significantly improved risk reclassification, with significant NRI in the fully adjusted model ( p = 0.01) and significant IDIs across all models (all p < 0.05). The association remained consistent across all subgroups regardless of diabetes status. Higher GV independently predicts increased delirium risk in critically ill older adults patients. Monitoring and managing GV may represent a modifiable target for delirium prevention in this vulnerable population.