Sepsis mortality prediction with Machine Learning Tecniques - Random forest classification

Aug 2, 2026

Upload by Miles

Objective: develop an ICU sepsis death classification model using machine learning. The study uses a cross-sectional descriptive design, drawing 180 sepsis patients from three Murcia hospitals and 4,559 patients from the MIMIC III open-access database. Key variables include age, weight, vital signs, lactate, oxygenation measures, blood pressure, pH, urine output, and potassium. A random forest classifier is built and evaluated in both datasets. Results show high sensitivity, specificity, accuracy, and AUC. Conclusions identify lactate, urine output, and acid-base related variables as leading mortality predictors, with potassium more critical in MIMIC III.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Abstract

Objective and design

Data sources and variables

Model and results

Conclusions

Introduction

Sepsis definitions over time

Burden and diagnostic challenges

Existing scoring tools

Role of electronic medical records and AI

Methods

Sepsis mortality prediction with Machine Learning Tecniques Javier Carrillo Pérez-Tome a, Tesifón Parrón-Carreño a, Ana Belen CastañoFernández b, Bruno José Nievas-Soriano a, Gracia Castro-Luna a a Department of Nursing: Physiotherapy and Medicine, University of Almeria, 04120 Almeria, Spain b Department of Applied Mathematics, University of Almería, 04120 Almeria, Spain Abstract Objective: To develop a sepsis death classiﬁcation model based on machine learning techniques for patients admitted to the Intensive Care Unit (ICU) . Design: Cross-sectional descriptive study. Setting: The Intensive Care Units (ICUs) of three Hospitals from Murcia (Spain) and patients from the MIMIC III open-access database. Patients: 180 patients diagnosed with sepsis in the ICUs of three hospitals and a total of 4559 patients from the MIMIC III database. Main variables of interest: Age, weight, heart rate, respiratory rate, temperature, lactate levels, partial oxygen saturation, systolic and diastolic blood pressure, pH, urine, and potassium levels. A random forest classiﬁcation model was calculated using the local and MIMIC III databases. Results: The sensitivity of the model of our database, considering all the variables classiﬁed as important by the random forest, was 95.45%, the speciﬁcity was 100%, the accuracy was 96.77%, and an AUC of 95% . . In the case of the model based on the MIMIC III database, the sensitivity was 97.55%, the speciﬁcity was 100%, and the precision was 98.28%, with an AUC of 97.3% . Conclusions: According to random forest classiﬁcation in both databases, lactate levels, urine outputand variables related to acid.base equilibrium were the most important variable in mortality due to sepsis in the ICU. The potassium levels were more critical in the MIMIC III database than the local database. Keywords Sepsis, MIMIC III, Machine learning, Sepsis mortality Introduction The concept of sepsis began to be deﬁned in 1992, when the ﬁrst consensus on sepsis, Sepsis-1, was published. The concept gave rise to the Systemic Inﬂammatory Response (SIRS), deﬁning sepsis as a “systemic inﬂammatory response associated with a disease.” In this consensus, levels of severity were also added: severe sepsis and septic shock. In 2001, modiﬁcations were made: a group of experts met and called this meeting Sepsis-2. Then, some values used for diagnosis were adjusted; however, there were no signiﬁcant changes in the deﬁnition of sepsis. A working group met again in 2016 and published an update on sepsis called Sepsis-3, which determined the most current deﬁnition. It was published by The Sepsis Deﬁnitions Working Group and deﬁned sepsis as “a lifethreatening organ dysfunction caused by a dysregulated host response to infection.”In addition to the deﬁnition, it included tools to help diagnose sepsis.1 , 2 Concerning the loss of human life, sepsis entails a high economic cost for healthcare systems. In the United States, one-third of patients diagnosed with sepsis die at a cost of about $20,3 million a year. In Spain, the incidence is 100 cases per 100,000 people/year, and mortality is also between 20% and 43% . The estimated average cost is about $20,000 for each episode of severe sepsis.3 , 4 , 5 A considerable drawback in managing sepsis is the great diliculty in reaching adiagnosis.There is no speciﬁc test to establish the diagnosis, and the symptoms present very heterogeneously, making it more challenging to determine the onset of the disease and, consequently, start treatment as soon as possible. There is evidence that early diagnosis of sepsis and, therefore, Early initiation of treatment signiﬁcantly reduces morbidity and mortality. Initially, during the onset of sepsis, it is dilicult to ﬁnd symptoms or parameters that help us diagnose it. When we ﬁnd easily recognizable signs, the disease is usually in an advanced stage, which entails more complex treatment and a worse prognosis. The elorts of many researchers are foc

FAQs

What is the main objective of the study?

To develop a sepsis mortality (death) classification model for ICU patients using machine learning techniques.

Which data sources are used to build and test the model?

The study uses patient data from three hospitals in Murcia, Spain, and compares performance with the MIMIC III open-access database.

What machine learning approach is applied and what key outcomes are reported?

A random forest classification model is calculated in both datasets, with reported sensitivity, specificity, accuracy, and AUC values.

Which variables are most important for predicting sepsis mortality, and do they differ by dataset?

Lactate levels, urine output, and variables related to acid-base equilibrium are the most important predictors in both databases. Potassium levels are more critical in MIMIC III than in the local database.

Aug 2, 2026

Upload by Miles