Stress hyperglycaemia ratio and post-operative delirium after cardiac surgery: a retrospective cohort study of the...

Post-operative delirium is a serious neurocognitive complication of cardiac surgery. The stress hyperglycaemia ratio (SHR), which adjusts acute glucose levels to baseline glycaemia using glycated haemoglobin, has demonstrated superior prognostic value in critical care; however, its association with delirium remains under-investigated.

This study's primary aim was to investigate the association between the SHR and post-operative delirium. The secondary aim was to develop machine-learning models for early risk stratification.

This retrospective study analysed 8346 adult patients from the Medical Information Mart for Intensive Care IV database. The SHR was calculated using the peak glucose level within the first 24 h after intensive care unit admission. Delirium was assessed using the Confusion Assessment Method for the Intensive Care Unit. Multivariable Cox regression and restricted cubic splines evaluated associations, while six machine-learning models were constructed using Boruta-selected features.

Delirium incidence was significantly higher in the high-ratio group than in the low-ratio group (18% vs 11%; P < 0.001). After multivariable adjustment, each one-unit increase in the ratio was associated with a 60% increased delirium risk (hazard ratio, 1.60; 95% confidence interval, 1.29-1.99). A high ratio independently predicted delirium (hazard ratio, 1.54; 95% confidence interval, 1.31-1.82), with restricted cubic splines confirming a linear dose-response relationship (P for non-linearity = 0.203). The support vector machine model achieved the best discriminative performance (area under the curve = 0.8042).

An elevated SHR is an independent risk factor for post-operative delirium. These machine-learning models may facilitate timely perioperative risk stratification.