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

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

Keywords: Rare Event, Early Sepsis Onset Detection, Trauma Patients, Severe Imbalance, Dataset

Abstract: Sepsis is a major public health concern due to its high morbidity, mortality, and cost. The clinical outcome can be substantially improved through early detection and timely intervention. During the past decade, by leveraging publicly available datasets, machine learning (ML) has driven advances in both research and clinical practice. However, existing public datasets consider mainly the general ICU (Intensive Care Unit) population and neglect the difference that trauma patients are having. In critically ill trauma patients, injury-related inflammation and organ dysfunction can increase the risk of sepsis, while also masking the clinical signs of infection. Therefore, a targeted identification of post-traumatic sepsis is critical but challenging, which was rarely studied before. To address this gap, we introduce a publicly available standardized post-trauma sepsis onset dataset extracted, relabeled using standardized post-trauma clinical facts, and validated from MIMIC-III (a large database with more than 40,000 patients who stayed in critical care units). Furthermore, we frame our early detection problem of post-trauma sepsis onset according to the ICU's real running routine that was ignored before, which results in a daily sepsis onset early detection problem for each patient, and sepsis onset becomes a rare event. In this work, we also establish a general benchmark to address this rare event challenge through comprehensive experiments, which shows the necessity of further advancements using this new dataset for early detection of rare events.

Primary Area: datasets and benchmarks

Submission Number: 21054