Integrated prognostic evaluation of TyG and TyG-BMI in critically ill patients with comorbid atrial fibrillation and ischemic stroke using machine learning.

backgroundThe purpose of this research was to explore the relationship between triglyceride-glucose (TyG) and triglyceride-glucose body mass index (TyG-BMI) indices and both short-and long-term all-cause mortality in intensive care unit (ICU) patients with comorbid atrial fibrillation and ischemic stroke.

methodsData were obtained from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, a large, publicly accessible critical care repository. The study population comprised critically ill patients with comorbid atrial fibrillation and ischemic stroke. In critically ill patients with comorbid atrial fibrillation and ischemic stroke, time-to-event analyses from ICU admission were conducted to evaluate the associations of TyG and TyG-BMI with all-cause mortality using Cox regression, Kaplan-Meier methods, and restricted cubic splines. The Boruta algorithm was employed for feature selection, and machine learning models were developed to explore predictive performance.

resultsIn the fully adjusted model, each one-unit increase in TyG was associated with a 59.7% increase in 30-day mortality (HR = 1.597, 95%CI 1.303–1.959, P < 0.001) and a 51.8% increase in 180-day all-cause mortality (HR = 1.518, 95%CI 1.280–1.801, P < 0.001). In fully adjusted analyses, participants in the highest tertile of TyG exhibited significantly greater risks of both 30-day (HR = 1.701, 95%CI: 1.237–2.340) and 180-day all-cause mortality (HR = 1.613, 95%CI: 1.247–2.087) compared with those in the lowest tertile (all P < 0.001). In fully adjusted models, each 10-unit increase in TyG-BMI was associated with higher all-cause mortality risk (30-day HR = 1.049, 95%CI 1.029–1.069; 180-day HR = 1.038, 95%CI: 1.026–1.062; both P < 0.001). Compared with the lowest tertile, participants in the highest tertile had significantly elevated risks of 30-day (HR = 1.935, 95%CI 1.420–2.638) and 180-day all-cause mortality (HR = 2.172, 95%CI 1.687–2.798) (all P < 0.001). Machine learning analyses consistently recognized TyG as an important predictor of mortality, with support vector machines achieving the highest area under the curve (AUC) of 0.736, and decision curve analysis showing favorable clinical utility.

conclusionIn ICU patients with comorbid atrial fibrillation and ischemic stroke, higher TyG and TyG-BMI indices were independently associated with increased 30-day and 180-day all-cause mortality.