# Glycemic Variability as a Predictor of Mortality in Sepsis Patients With Concurrent Persistent Inflammation, Immunosuppression, and Catabolism Syndrome

## TL;DR

High blood sugar variability predicts higher death rates in sepsis patients with a specific syndrome called PICS.

## Contribution

Glycemic variability coefficient (GVC) is shown to be a strong predictor of mortality in sepsis patients with PICS, validated using deep learning models.

## Key findings

- Higher GVC tertiles correlate with increased 28-day and 180-day mortality in sepsis patients with PICS.
- A deep learning model using GVC achieved an AUC of 0.960 for predicting 28-day mortality.
- GVC was identified as a key predictor of mortality risk by the Boruta algorithm.

## Abstract

Sepsis is a life‐threatening condition caused by infection, which triggers dysregulated systemic inflammatory responses. Among sepsis patients, those who concurrently develop persistent inflammation, immunosuppression, and catabolism syndrome (PICS) have a significantly poorer prognosis. It has been demonstrated that associations exist between elevated glycemic variation coefficient (GVC) levels and the development of PICS in septic populations. However, the association between GVC and adverse clinical outcomes in the subgroup of septic patients with PICS requires further investigation.

The study analyzed data from the Medical Information Mart for Intensive Care IV (MIMIC‐IV) database, which comprised 1353 critically ill septic patients who developed nosocomial infections during hospitalization. Based on the 2024 Critical Care Medicine Guidelines on Glycemic Control, the patients included in this study will be divided into GVC < 20 group, 20 ≤ GVC ≤ 36 group, and GVC > 36 group. The primary outcome measure was 28‐day all‐cause mortality, with secondary outcomes comprising in‐hospital mortality and 180‐day mortality. Cox proportional hazards regression and Kaplan–Meier analysis were utilized to examine the relationship between GVC and adverse outcomes. The Boruta algorithm evaluated the predictive capacity of GVC, followed by the development of prognostic models through machine learning (ML) and deep learning (DL) algorithms, externally validated using an independent cohort of 116 patients from the Emergency Department of Tianjin Medical University General Hospital.

The analysis included 1353 septic patients. Kaplan–Meier analysis indicates that the highest GVC tertile has significant differences in 28‐day and 180‐day mortality rates. Cox regression analysis revealed that patients in the highest GVC tertile had a significantly elevated 28‐day mortality risk. (OR = 1.60, 95% CI: 1.11–2.32, p < 0.05). The Boruta algorithm identified GVC as a key predictor for mortality risk. The 28‐day mortality prediction model developed using tabular prior‐data fitted network (TabPFN) achieved an area under the curve (AUC) of 0.960.

GVC demonstrated significant correlations with 28‐day and 180‐day mortality in sepsis patients complicated by PICS. DL models confirm the utility of GVC as a robust prediction tool for septic patients, providing valuable references for clinical decision‐making.