Multicenter Validation of a Machine Learning Algorithm for Mortality Prediction at Decision-Critical Timestamps in ICU Sepsis Patients

Abstract

Sepsisis a critical, life-threatening condition commonly encountered in Intensive Care Units (ICUs). Healthcare professionals face significant challenges not only from the vast volume of clinical data but also due to the condition's inherent complexity and systemic nature. These factors create a chal-lenging environment that complicates decision-making processes, especially when predicting patient mortality. Given the critical importance of early mortality prediction in improving patient outcomes, this research aims to predict mortality for ICU sepsis patients at 12, 24, and 48 hours in advance through machine learning models based on clinical data. The study was conducted with the Medical Information Mart for Intensive Care database, which includes data from 7,511 ICU sepsis patients from a single hospital, and the Electronic Intensive Care Unit Collaborative Research database, which contains data from 3,786 ICU sepsis patients across multiple hospitals. Eight supervised machine learning models were evaluated based on the area under the curve, where Light Gradient Boosted Machine demonstrated the best performance across all critical timestamps. It also outperformed the Sequential Organ Failure Assessment score in predicting mortality. This research underscores the potential of machine learning to advance mortality prediction for ICU sepsis patients, enabling timely decision-making and ultimately improving patient outcomes.

Ähnliche Arbeiten