Interpretable early mortality prediction in oncology ICU patients: A dual-cohort validation of a LASSO–XGBoost–SHAP framework
-
By
-
Xinyi Chen
-
Lu Wang
-
Wan Qin
-
Mu Yang
-
Yuanmei Yan
-
Xiaoxiao Luo
-
Xianglin Yuan
-
July 7, 2026
-
Objective:
To develop and validate a machine learning framework for early mortality prediction in oncology ICU patients.
Approach:
- Machine Learning Framework: Utilized LASSO, XGBoost, and SHAP for mortality prediction based on early ICU data.
- Data Sources: Leveraged MIMIC and eICU databases for model development and external validation.
- Validation Standards: Followed TRIPOD+AI and PROBAST+AI guidelines for transparency and auditability in reporting.
Key Findings:
- Machine learning models demonstrated superior performance compared to traditional scoring systems in predicting mortality.
- Calibration drift in existing models leads to misalignment with contemporary oncology ICU populations.
- Integration of high-dimensional data from early ICU stays significantly improves prediction accuracy.
Interpretation:
Limitations:
- The scope of external validation is limited to the datasets used in the study.
- There is variability in how calibration and decision-analytic utility are reported across different studies.
Conclusion:
A robust machine learning framework for early mortality prediction in oncology ICU patients can enhance clinical decision-making.
Sources:
