Joint impact of stress hyperglycaemic ratio and glycaemic variability in patients with ischaemic stroke and machine learning for mortality prediction

TL;DR

This study shows that combining stress hyperglycaemic ratio and glycaemic variability improves mortality prediction in ICU patients with ischaemic stroke.

Contribution

The novel contribution is the combined use of SHR and GV for mortality prediction in ischaemic stroke patients using machine learning.

Findings

Patients with high SHR and GV had the highest mortality risk across all timeframes.

Combining SHR and GV improved predictive performance compared to using either metric alone.

The combined model showed AUC values of 0.64–0.65 for 30-day to 360-day mortality prediction.

Abstract

The global burden of ischaemic stroke (IS) is high, which is potentially relevant to stress hyperglycemia ratio (SHR) and glycaemic variability (GV). This study aims to evaluate the combined effect of the SHR and GV with predict short-term, medium-term and long-term mortality outcomes in the intensive care unit (ICU).

This retrospective study utilised data from the MIMIC-IV database, including adult ICU patients diagnosed with ischaemic stroke. SHR and GV were calculated and categorised into tertiles, with combined effects grouped into four categories. Study outcomes included 30-day, 90-day, and 360-day mortality outcomes. Kaplan-Meier curves, restricted cubic splines and Cox proportional hazards models were used to assess the SHR and GV with mortality outcomes. Then, we further assessed the associations with subgroup analyses by diabetes, age, sex, and body mass index. Predictive performance was evaluated using receiver operating characteristic curves and area under the curve (AUC) comparisons.

In 749 patients with IS (age 72.9 [61.1–83.0] years; 47.3% male), and 30-day, 90-day, and 360-day ICU mortality rates of 23.2%, 29.6%, and 35.4%, respectively. Patients with both high SHR and high GV (G4 vs. G1) had the highest mortality risk in the overall population, with HRs of 2.43 (95% CI: 1.42–4.14) for 30-day mortality, 2.18 (95% CI: 1.36–3.06) for 90-day mortality, and 1.77 (95% CI: 1.14–2.74) for 360-day mortality. The combined effect of SHR and GV demonstrated superior predictive performance (AUC: 0.643 for 30-day, 0.652 for 90-day, and 0.640 for 360-day mortality) compared to SHR or GV alone. These findings highlight the prognostic utility of combining SHR and GV for mortality prediction in critically ill patients with IS.

The combination of SHR and GV is promising to facilitate early identifying IS critically ill patients at high-risk of mortality.

The online version contains supplementary material available at 10.1186/s12883-025-04510-z.

Figures