Interpretable Model Predicts VTE in Sepsis
Researchers developed and externally validated an interpretable LightGBM model to predict venous thromboembolism (VTE) in ICU patients with sepsis, using 25,197 MIMIC‑IV patients for training and 328 Changshu Hospital patients for validation. The model attained an internal AUC of 0.956 and external AUC of 0.786 (0.816 in severe sepsis), showed good calibration and net benefit, and SHAP highlighted catheterization, chloride, bicarbonate, and prolonged PTT as top predictors, supporting personalized prophylaxis.
Key Points
- 1Developed an interpretable LightGBM model using 25,197 MIMIC‑IV and 328 external Changshu patients.
- 2Demonstrated high discrimination with internal AUC 0.956 and external AUC 0.786, improved to 0.816 in severe sepsis.
- 3Enables personalized VTE prophylaxis and early diagnostic decisions by explaining key predictors via SHAP.
Scoring Rationale
Strong external validation and explainability support practical relevance, though novelty is incremental within ML-for-health applications.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
