Machine Learning Predictive Models for Sepsis Risk in ICU Patients With Intracerebral Hemorrhage

Central South University

Status

Not yet enrolling

Conditions

Intracerebral Hemorrhage

Sepsis

Treatments

Other: no intervention

Study type

Observational

Funder types

Other

Identifiers

NCT06326385

53244021

Details and patient eligibility

About

Patients with intracerebral hemorrhage (ICH) in the intensive care unit (ICU) are at heightened risk of developing sepsis, significantly increasing mortality and healthcare burden. Currently, there is a lack of effective tools for the early prediction of sepsis in ICH patients within the ICU. This study aims to develop a reliable predictive model using machine learning techniques to assist clinicians in the early identification of patients at high risk and to facilitate timely intervention.

The Medical Information Mart for Intensive Care (MIMIC) IV database (version 2.2) is an international online repository for critical care expertise. This database contains patient-related information collected from the ICUs of Beth Israel Deaconess Medical Center between 2008 and 2019. It includes a vast dataset of 299,712 hospital admissions and 73,181 intensive care unit patients.

The eICU Collaborative Research Database (eICU-CRD) comprises data from over 200,000 ICU admissions for 139,367 unique patients across 208 US hospitals between 2014 and 2015, providing a valuable resource for critical care research.

This study aims to establish and validate multiple machine learning models to predict the onset of sepsis in ICU patients with ICH and to identify the model with the optimal predictive performance.

Full description

Data Collection: This study utilized two public databases. The model leveraged clinical data obtained from the Medical Information Mart for Intensive Care (MIMIC) IV database (version 2.2) and selected corresponding patients for external validation from the eICU Collaborative Research Database (eICU-CRD). Data on ICH patients were extracted from the MIMIC IV public database, including baseline characteristics, clinical parameters, therapeutic interventions, and outcomes. The data were randomly divided into two groups, with 70% serving as the training set and 30% as the validation set.
Model Development: Feature selection was performed using Lasso regression to construct various machine learning models (such as Random Forest, Logistic Regression, and Neural Networks).
Model Validation: In addition to the internal validation set, external validation was also conducted on the eICU database to test the model's generalizability.
Statistical Analysis: The predictive performance of the model was evaluated using metrics including the area under the ROC curve (AUC), sensitivity, and specificity.
Clinical Applicability Assessment: The clinical utility of the model was assessed using Decision Curve Analysis (DCA).

Enrollment

1,800 estimated patients

Sex

All

Ages

18 to 89 years old

Volunteers

No Healthy Volunteers

Inclusion criteria