Correlation of triglyceride-glucose index with the incidence and prognosis of hyperglycemic crises in critically ill patients with diabetes mellitus - a machine-learning-based multicenter retrospective cohort study

Aug 5, 2026

Hyperglycemic crisis events (HCEs), including diabetic ketoacidosis and hyperosmolar hyperglycemic state, drive high mortality in intensive care patients with diabetes mellitus. This study evaluates whether the triglyceride-glucose (TyG) index predicts HCE occurrence and adverse clinical outcomes in critically ill diabetics, and develops a machine-learning risk model. A multicenter retrospective cohort with TyG calculated within 24 hours post-admission analyzes HCE incidence, in-hospital death, and ICU death, using regression, weighting, matching, and SHAP-based variable importance. Elevated TyG is independently associated with higher death risk.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages14

Reading Time35 min

AccessFree to download

Background

Methods

Results

Conclusion

Keywords

OPEN ACCESS EDITED BY Mithun Rudrapal, Vignan’s Foundation for Science, Technology and Research, India REVIEWED BY Samiksha Garse, DY Patil Deemed to be University, India Kratika Singh, Centre of Bio-Medical Research (CBMR), India *CORRESPONDENCE Rui Zhang [zztg@163.com](zztg@163.com)[ ](zztg@163.com)Jianhua Cheng [cjh19940317@163.com](cjh19940317@163.com)[ ](cjh19940317@163.com)Jian Xu [xujianqdmc@126.com](xujianqdmc@126.com) †These authors have contributed equally to this work RECEIVED 18 June 2025 ACCEPTED 20 August 2025 PUBLISHED 04 September 2025 CITATION Xie M, Zhang Y, Wu H, Wu Z, Han H, Xie X, Zhang R, Cheng J and Xu J (2025) Correlation of triglyceride-glucose index with the incidence and prognosis of hyperglycemic crises in critically ill patients with diabetes mellitus: a machine-learning-based multicenter retrospective cohort study. Front. Nutr. 12:1649553 . doi: 10.3389/fnut.2025.1649553 COPYRIGHT © 2025 Xie, Zhang, Wu, Wu, Han, Xie, Zhang, Cheng and Xu. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. TYPE Original Research PUBLISHED 04 September 2025 DOI 10.3389/fnut.2025.1649553 Correlation of triglyceride-glucose index with the incidence and prognosis of hyperglycemic crises in critically ill patients with diabetes mellitus: a machine-learning-based multicenter retrospective cohort study Mingchen Xie 1,2,3†, Yahui Zhang 1,3†, Haitao Wu 1,3†, Zeyu Wu 1, Hao Han 1,3, Xun Xie 1,3, Rui Zhang 2*, Jianhua Cheng 1* and Jian Xu 1* 1 Department of Neurosurgery (Diabetes Critical Care Research Consortium), The Affiliated Hospital of Qingdao University, Qingdao, China, 2 Department of Neurological Intensive Care Unit, The Affiliated Hospital of Qingdao University, Qingdao, China, 3Qingdao University Medical College, Qingdao University, Qingdao, China Background: Hyperglycemic crisis events (HCEs)—encompassing diabetic ketoacidosis (DKA) and hyperosmolar hyperglycemic state (HHS)—constitute lethal determinants for patients with diabetic mellitus (DM) in intensive care. The triglyceride-glucose (TyG) index, an emergent insulin resistance surrogate, lacks rigorous investigation regarding HCE occurrence trajectories and prognostic sequelae among critically ill diabetics. This study aims to evaluate the relationship between the TyG index and HCE incidence/clinical outcomesin critically ill patients with DM and to construct a risk prediction model using machine-learning algorithms. Methods: This multi-center retrospective investigation leveraged clinical repositories from Medical Information Mart for Intensive Care IV (MIMICIV) and eICU Collaborative Research Database (eICU-CRD) . Inclusion criteria encompassed critically ill subjects with diabetes possessing computable TyG indices within 24 h post-admission. The main study endpoints included death occurring during hospitalization and death within the intensive care unit. TyG index-outcome interrelationships underwent interrogation via logistic regression, restricted cubic spline (RCS), correlation, and linear analytical methodologies. Overlap weighting (OW), inverse probability treatment weighting (IPTW), and propensity score matching (PSM) mitigated confounding influences. Stratified examinations occurred per determinant factors. Five machine-learning architectures constructed mortality prognostication frameworks, with SHapley Additive exPlanations (SHAP) delineating pivotal predictors. Results: Among 4,098 critically ill patients with DM, 328 developed HCE. Patients with HCE had significantly higher TyG levels [10 .2 (

FAQs

What clinical conditions are included as hyperglycemic crisis events in this study?

Hyperglycemic crisis events include diabetic ketoacidosis (DKA) and hyperosmolar hyperglycemic state (HHS). These events are treated as the primary clinical crises of interest in critically ill diabetes patients.

How is the triglyceride-glucose (TyG) index used in the study?

The TyG index is calculated within 24 hours after ICU or hospital admission and assessed for its ability to discriminate HCE occurrence and to predict in-hospital and ICU mortality outcomes.

Which machine-learning approach performed best for mortality prediction?

The XGBoost model achieved higher predictive accuracy among the machine-learning architectures, with the TyG index identified as an important component for prognosis.

Aug 5, 2026

Upload by Evangeline