Early Prediction of Sepsis Using Multi-Feature Fusion Based XGBoost Learning and Bayesian Optimization

Abstract

Early prediction of sepsis is critical in clinical practice since each hour of delayed treatment has been associated with an increase in mortality due to irreversible organ damage. This study aimed to develop an algorithm for accurately predicting the onset of sepsis in the proceeding of six hours. Thus, a multi-feature fusion based XGBoost classification model was developed and was further improved by a Bayesian optimizer and an ensemble learning framework. Analysis was performed on the PhysioNet/Computing in Cardiology Challenge 2019, which provided a publicly available sepsis data sourced from 40,336 ICU patients. The proposed algorithm was evaluated on the selected independent inner test data. When applying an optimized predicted risk threshold of 0.525, the best performance was achieved, with an area under receiver operating characteristic (AUROC) of 0.847, classification accuracy (ACC) of 0.812 and í µí±ˆ í µí±›í µí±œí µí±Ÿí µí±ší µí±Ží µí±™í µí±–í µí± §í µí±’í µí±‘ score defined by the organizers of 0.430. Finally, we applied the developed model on the hidden test set and obtained a í µí±ˆ í µí±›í µí±œí µí±Ÿí µí±ší µí±Ží µí±™í µí±–í µí± §í µí±’í µí±‘ score of 0.430, which showed a good generalization ability on the test hidden data.