1. Machine Learning Predictive Model for Septic Shock in Acute Pancreatitis with Sepsis
- Author
-
Xia Y, Long H, Lai Q, and Zhou Y
- Subjects
machine learning ,acute pancreatitis ,sepsis ,septic shock ,Pathology ,RB1-214 ,Therapeutics. Pharmacology ,RM1-950 - Abstract
Yiqin Xia,1– 3 Hongyu Long,4 Qiang Lai,1– 3 Yiwu Zhou1– 3 1Emergency Department, West China Hospital, Sichuan University, Chengdu, Sichuan, People’s Republic of China; 2Laboratory of Emergency Medicine, West China Hospital, Sichuan University, Chengdu, Sichuan, People’s Republic of China; 3Disaster Medical Center, Sichuan University, Chengdu, Sichuan, People’s Republic of China; 4Department of Critical Care Medicine, Chengdu First People’s Hospital, Chengdu, Sichuan, People’s Republic of ChinaCorrespondence: Yiwu Zhou, Emergency Department, West China Hospital of Sichuan University, Chengdu, Sichuan, People’s Republic of China, Email 456zyw@163.comObjective: Acute pancreatitis (AP) progresses to septic shock can be fatal. Early identification of high-risk patients and timely intervention can prevent and interrupt septic shock. By analyzing the clinical characteristics of AP with sepsis, this study uses machine learning (ML) to build a model for early prediction of septic shock within 28 days of admission, which guided emergency physicians in resource allocation and medical decision-making.Methods: This retrospective cohort study collected data from the emergency departments (EDs) of three tertiary care hospitals in China. The dataset was randomly divided into a training dataset (70%) and a testing dataset (30%). Ten ML classifiers were utilized to analyze characteristics of AP with sepsis in the training dataset upon admission. Results were evaluated through cross-validation analysis. The optimal model was then tested on the testing dataset without any parameter modifications. The ML model was evaluated using the receiver operating characteristic curve (ROC) and compared to scoring systems through the DeLong test.Results: A total of 604 AP patients with sepsis were included in this study. The auto-encoder (AE) model based on mean normalization, Pearson correlation coefficient (PCC), and recursive feature elimination (RFE) selection, achieved the highest Area Under the Curve (AUC) on the validation dataset (AUC 0.900, accuracy 0.868), with the AUC of 0.879 and accuracy of 0.790 on the testing dataset. Compared to the Sequential Organ Failure Assessment (AUC 0.741), quick Sequential Organ Failure Assessment (AUC 0.727), Acute Physiology and Chronic Health Evaluation II (AUC 0.778), and Bedside Index of Severity in Acute Pancreatitis (AUC 0.691), the AE model showed superior performance.Conclusion: The AE model outperforms traditional scoring systems in predicting septic shock in AP patients with sepsis within 28 days of admission. This assists emergency physicians in identifying high-risk patients early and making timely medical decisions.Keywords: machine learning, acute pancreatitis, sepsis, septic shock
- Published
- 2024