A System for Sepsis Detection and Mortality Prediction Based on a Machine Learning Algorithm Using Common Features

Abstract

Background: Sepsis is caused by the immune system’s overreaction following infection. Delaying treatment for 1 hour causes organ damage and increased mortality. Therefore, early detection is essential. This study developed a system for sepsis detection and mortality prediction using machine learning algorithms by analyzing electronic health record data.

Methods: A machine learning-based system for sepsis detection and mortality prediction was established. The proposed system was developed using intensive care unit and general ward data provided by China Medical University Hospital in 2018. The data warehouse was verified in a completely anonymous population for challenge scoring.

Results: A total of 53 readily available features were extracted from the data warehouse. The impact of each feature on sepsis detection and mortality prediction was analyzed. The proposed sepsis detection model has 0.861 area under the curve. The proposed mortality prediction model has 0.797 area under the curve. Both two models in the proposed system only rely on commonly available features. Our model achieved excellent performance in detecting the risk of sepsis and predicting the risk of mortality.

Conclusions: The proposed system includes two models, a sepsis detection model and a mortality prediction model, and both models only use commonly available features. It has achieved excellent performance in detecting the risk of sepsis and predicting the risk of mortality. In addition, in terms of clinical contribution, the proposed system is suitable for rural hospitals or small hospitals with insufficient medical workers, which can reduce the burden on medical workers.

Note:

Funding Declaration: The study was funded by China Medical University and China Medical University Hospital, grant numbers: 111-2622-8-039-001-IE, 111-2321-B-039-005-, 110-2314-B-039-010-MY2, MOST 111-2321-B- 030-004 and DMR-111-227.

Conflict of Interest: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Keywords: sepsis detection, mortality prediction, Machine Learning, electronic health records, intensive care unit