Phenotype discovery and mortality prediction in sepsis-induced myocardial dysfunction: a deep learning and stratified modeling approach
Machine learning reveals that sepsis-related organ failure has distinct biological subtypes with different survival outcomes, enabling personalized critical care strategies.
Autoencoder + UMAP + K-means clustering on MIMIC data identified 3 distinct SIMD phenotypes with markedly different 90-day mortality; phenotype-specific XGBoost modeling (AUC 0.880) outperformed global modeling, demonstrating that precision critical care ML approaches improve mortality prediction. SHAP analysis identified lactate, bilirubin, coagulation, and GCS as key discriminators.
What the study was
- Study design
- Retrospective MIMIC database analysis
- Population
- SIMD patients in MIMIC-III and MIMIC-IV databases
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- BMC Medical Informatics and Decision Making
Why it surfaced
MIMIC retrospective; strong ML methodology; SIMD is outside primary watchlist (sepsis not directly listed) but AI/ML in clinical diagnostics covers it. Marked as unsolicited_find=true as sepsis is outside watchlist. Scored 5.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.