Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rare Event Early Detection: A Dataset of Sepsis Onset for Critically Ill Trauma Patients

Yin Jin, Tucker R. Stewart, Deyi Zhou, Chhavi Gupta, Arjita Nema, Scott C. Brakenridge, Grant E. O'Keefe, Juhua Hu

TL;DR

This work creates a publicly accessible, trauma-focused dataset for post-traumatic sepsis onset by re-labeling MIMIC-III with a trauma-specific Sepsis-3 adaptation and aligning detection to daily ICU workflows, framing it as a rare-event early detection problem. It introduces careful preprocessing of blood cultures and antibiotics, a modified SOFA score excluding confounding trauma factors, and explicit onset criteria to label sepsis days, resulting in S and N datasets with thousands of labeled nighttime instances. The authors propose a two-stage, reconstruction-based representation learning benchmark using MAE pretraining, with oversampling and augmentation (including masking and reconstruction) to tackle severe class imbalance, and they show that pretraining on a larger general ICU population improves performance over trauma-only pretraining. Through extensive experiments, the paper demonstrates the value of trauma-specific labels, highlights the challenges of rare-event detection in this setting, and provides a reproducible pipeline for future research and benchmarking in post-trauma sepsis detection.

Abstract

Sepsis is a major public health concern due to its high morbidity, mortality, and cost. Its clinical outcome can be substantially improved through early detection and timely intervention. By leveraging publicly available datasets, machine learning (ML) has driven advances in both research and clinical practice. However, existing public datasets consider ICU patients (Intensive Care Unit) as a uniform group and neglect the potential challenges presented by critically ill trauma patients in whom injury-related inflammation and organ dysfunction can overlap with the clinical features of sepsis. We propose that a targeted identification of post-traumatic sepsis is necessary in order to develop methods for early detection. Therefore, we introduce a publicly available standardized post-trauma sepsis onset dataset extracted, relabeled using standardized post-trauma clinical facts, and validated from MIMIC-III. Furthermore, we frame early detection of post-trauma sepsis onset according to clinical workflow in ICUs in a daily basis resulting in a new rare event detection problem. We then establish a general benchmark through comprehensive experiments, which shows the necessity of further advancements using this new dataset. The data code is available at https://github.com/ML4UWHealth/SepsisOnset_TraumaCohort.git.

Rare Event Early Detection: A Dataset of Sepsis Onset for Critically Ill Trauma Patients

TL;DR

This work creates a publicly accessible, trauma-focused dataset for post-traumatic sepsis onset by re-labeling MIMIC-III with a trauma-specific Sepsis-3 adaptation and aligning detection to daily ICU workflows, framing it as a rare-event early detection problem. It introduces careful preprocessing of blood cultures and antibiotics, a modified SOFA score excluding confounding trauma factors, and explicit onset criteria to label sepsis days, resulting in S and N datasets with thousands of labeled nighttime instances. The authors propose a two-stage, reconstruction-based representation learning benchmark using MAE pretraining, with oversampling and augmentation (including masking and reconstruction) to tackle severe class imbalance, and they show that pretraining on a larger general ICU population improves performance over trauma-only pretraining. Through extensive experiments, the paper demonstrates the value of trauma-specific labels, highlights the challenges of rare-event detection in this setting, and provides a reproducible pipeline for future research and benchmarking in post-trauma sepsis detection.

Abstract

Sepsis is a major public health concern due to its high morbidity, mortality, and cost. Its clinical outcome can be substantially improved through early detection and timely intervention. By leveraging publicly available datasets, machine learning (ML) has driven advances in both research and clinical practice. However, existing public datasets consider ICU patients (Intensive Care Unit) as a uniform group and neglect the potential challenges presented by critically ill trauma patients in whom injury-related inflammation and organ dysfunction can overlap with the clinical features of sepsis. We propose that a targeted identification of post-traumatic sepsis is necessary in order to develop methods for early detection. Therefore, we introduce a publicly available standardized post-trauma sepsis onset dataset extracted, relabeled using standardized post-trauma clinical facts, and validated from MIMIC-III. Furthermore, we frame early detection of post-trauma sepsis onset according to clinical workflow in ICUs in a daily basis resulting in a new rare event detection problem. We then establish a general benchmark through comprehensive experiments, which shows the necessity of further advancements using this new dataset. The data code is available at https://github.com/ML4UWHealth/SepsisOnset_TraumaCohort.git.
Paper Structure (30 sections, 8 figures, 7 tables)

This paper contains 30 sections, 8 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Publicly Available Sepsis Definition
  4. Early Sepsis Detection
  5. Imbalance and Rare Event Challenge
  6. Dataset Construction
  7. Cohort Extraction: Critically Ill Trauma Patients
  8. Post-trauma Sepsis Definition
  9. Pre-processing: blood culture
  10. Pre-processing: antibiotic
  11. Pre-processing: modified SOFA score
  12. Post-trauma sepsis onset criteria
  13. Early Sepsis Detection Setup
  14. Feature extraction
  15. Instance construction
  16. ...and 15 more sections

Figures (8)

  • Figure 1: Data flow diagram of study inclusion and exclusion criteria.
  • Figure 2: Timeline visualization of post-trauma sepsis onset assignment criteria.
  • Figure 3: Subset of physiological trends in the four days before sepsis onset (x = 0 marking sepsis onset).
  • Figure 4: Overview of the Two-Stage Training Pipeline
  • Figure 5: Sepsis ratio vs. number of ventilation days.
  • ...and 3 more figures