Related Experiment Videos
Temporal interpretation of physiologic group contributions for ICU mortality prediction
Jinwoong Kim1, Yeeun Kim2, Sangjin Park3,4
1Department of Industrial Data Engineering, Hanyang University, Seoul, 04763, Republic of Korea.
Scientific Reports
|July 21, 2026
Summary
Predicting ICU mortality is improved by analyzing groups of physiologic data over time. This new framework reveals temporal patterns in vital signs, offering better clinical insights than focusing on single data points.
Area of Science:
- Critical Care Medicine
- Biomedical Informatics
- Machine Learning in Healthcare
Background:
- Accurate prediction of intensive care unit (ICU) mortality is crucial for patient management.
- Existing prediction models often analyze individual variables or static time points, failing to capture complex physiologic dynamics.
- There is a need for methods that interpret evolving physiologic states and temporal progression for improved mortality prediction.
Purpose of the Study:
- To develop and evaluate a group-based temporal interpretation framework for ICU mortality prediction.
- To characterize the evolving contributions of distinct physiologic groups preceding ICU mortality.
- To shift interpretation from static variable importance to dynamic temporal trajectories of physiologic states.
Main Methods:
- Applied GroupSegment-SHAP to the MIMIC-IV database for a group-based temporal interpretation framework.
- Utilized Mamba, a selective state-space model, as the interpretation backbone due to its performance.
- Reorganized clinical variables into physiologic groups and quantified temporal importance using Shapley values across 24-, 48-, and 72-hour lookback windows.
Main Results:
- Group-level interpretation identified hemodynamic, neurologic, cardiopulmonary, and temperature groups as key mortality predictors.
- Temporal analysis revealed distinct patterns: 24-hour windows highlighted acute cardiopulmonary and metabolic issues, while longer windows captured hemodynamic and renal differences.
- Cardiopulmonary importance preceded peak group importance, with non-survivors showing significantly lower normal-range proportions.
Conclusions:
- The developed framework enables a more nuanced understanding of ICU mortality prediction by analyzing temporal trajectories of physiologic groups.
- This approach moves beyond static variable importance to capture dynamic physiologic shifts preceding death.
- The findings provide clinically meaningful insights into the evolving nature of critical illness leading to mortality.