Scholarlink

Background <p>The triglyceride-glucose (TyG) index and triglyceride-glucose-body mass index (TyG-BMI) are emerging surrogate markers of insulin resistance and metabolic risk. However, their comparative prognostic value in critically ill older adults with atherosclerotic cardiovascular disease (ASCVD) remains unclear. We aimed to compare the prognostic performance of the TyG index and TyG-BMI for 90-day all-cause mortality in this high-risk population.</p> Methods <p>We conducted a retrospective cohort study of critically ill patients aged ≥ 65 years with ASCVD from the MIMIC-IV database. The TyG index and TyG-BMI were measured at intensive care unit admission, and 90-day all-cause mortality was the primary outcome. Kaplan–Meier analysis, multivariable Cox proportional hazards models, and restricted cubic splines (RCS) were used to assess associations with mortality. Five machine learning algorithms, including Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT), and Gaussian Naive Bayes (GNB), were adopted to construct mortality risk prediction models, with discrimination assessed by the area under the receiver operating characteristic curve (AUC) and area under the precision-recall curve (AUPRC). Additionally, the SHapley Additive exPlanations (SHAP) approach was used to identify the key predictors of mortality.</p> Results <p>A total of 2,368 critically ill older adults with ASCVD were included, and 18.5% died during follow-up. In multivariable Cox analysis, the TyG index was independently associated with higher 90-day mortality risk (HR 1.114, 95% CI 1.014–1.225; <i>P</i> = 0.025), whereas TyG-BMI was not significantly associated with mortality (<i>P</i> = 0.883). Kaplan–Meier analysis showed progressively worse survival across increasing TyG index quartiles. In machine learning models, the TyG index consistently outperformed TyG-BMI. XGBoost showed the best discrimination (AUPRC = 0.455; specificity = 0.912). Adding the TyG index improved AUC from 0.763 to 0.794 (<i>P</i> = 0.005), whereas adding TyG-BMI did not significantly improve AUC (0.783; <i>P</i> = 0.101). SHAP analysis identified the TyG index as the most important metabolic predictor.</p> Conclusion <p>In elderly critically ill patients with ASCVD, the TyG index showed better prognostic performance than TyG-BMI and was an independent predictor of 90-day mortality. These findings suggest that the TyG index may be a useful and cost-effective biomarker for risk stratification in this high-risk population.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Comparative prognostic value of TyG index and TyG-BMI for 90-day all-cause mortality in critically ill older adults with ASCVD: a machine learning analysis of the MIMIC-IV database

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月17日

查看全文
  • Shuangmei Zhao,
  • Qianqian Wang,
  • Xingyu Li,
  • Shiyin Ma,
  • Chucheng Jiao,
  • Guangdong Wang,
  • Huihui Cao

摘要

Background

The triglyceride-glucose (TyG) index and triglyceride-glucose-body mass index (TyG-BMI) are emerging surrogate markers of insulin resistance and metabolic risk. However, their comparative prognostic value in critically ill older adults with atherosclerotic cardiovascular disease (ASCVD) remains unclear. We aimed to compare the prognostic performance of the TyG index and TyG-BMI for 90-day all-cause mortality in this high-risk population.

Methods

We conducted a retrospective cohort study of critically ill patients aged ≥ 65 years with ASCVD from the MIMIC-IV database. The TyG index and TyG-BMI were measured at intensive care unit admission, and 90-day all-cause mortality was the primary outcome. Kaplan–Meier analysis, multivariable Cox proportional hazards models, and restricted cubic splines (RCS) were used to assess associations with mortality. Five machine learning algorithms, including Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree (DT), and Gaussian Naive Bayes (GNB), were adopted to construct mortality risk prediction models, with discrimination assessed by the area under the receiver operating characteristic curve (AUC) and area under the precision-recall curve (AUPRC). Additionally, the SHapley Additive exPlanations (SHAP) approach was used to identify the key predictors of mortality.

Results

A total of 2,368 critically ill older adults with ASCVD were included, and 18.5% died during follow-up. In multivariable Cox analysis, the TyG index was independently associated with higher 90-day mortality risk (HR 1.114, 95% CI 1.014–1.225; P = 0.025), whereas TyG-BMI was not significantly associated with mortality (P = 0.883). Kaplan–Meier analysis showed progressively worse survival across increasing TyG index quartiles. In machine learning models, the TyG index consistently outperformed TyG-BMI. XGBoost showed the best discrimination (AUPRC = 0.455; specificity = 0.912). Adding the TyG index improved AUC from 0.763 to 0.794 (P = 0.005), whereas adding TyG-BMI did not significantly improve AUC (0.783; P = 0.101). SHAP analysis identified the TyG index as the most important metabolic predictor.

Conclusion

In elderly critically ill patients with ASCVD, the TyG index showed better prognostic performance than TyG-BMI and was an independent predictor of 90-day mortality. These findings suggest that the TyG index may be a useful and cost-effective biomarker for risk stratification in this high-risk population.

Graphical abstract