Interpretable Machine Learning for Early Prediction of Prognosis in Sepsis: A Discovery and Validation Study
Problem
Early prediction of in-hospital mortality in critically ill sepsis patients is difficult for clinicians.
Why it matters
Sepsis is life-threatening, and early mortality prediction could guide critical care decisions.
What the AI did
Seven machine-learning models were trained on clinical features from the MIMIC-IV database to predict in-hospital mortality, with SHAP used for interpretability; the best model was selected by accuracy and AUC in a validation cohort.
Results
8,817 sepsis patients enrolled; best model selected by accuracy and AUC in validation cohort, but metric values not stated in abstract.
Impact
A validated interpretable prediction model on retrospective ICU data; clinical deployment or external validation not shown.
Record
| Venue | Infectious Diseases and Therapy |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Chang Hu, Lu Li, Weipeng Huang, Tong Wu, Qiancheng Xu, Juan Liu, Bo Hu |
| Citations | 159 (OpenAlex, at import time) |
| References | 32 |
| Model or method | not stated in source (seven machine-learning methods compared; Lasso used for feature selection) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1007/s40121-022-00628-6 |
| Open access | gold |
| Keywords | sepsis, machine learning, mortality prediction, SHAP, MIMIC-IV, interpretability |
Importance rating
significant — thousands to millions affected
Early mortality prediction in critically ill sepsis patients could guide critical care; abstract calls sepsis critical but states no scale of affected population.