Related Experiment Videos
Artificial Intelligence Models for Mortality and Outcome Prediction in Intensive Care Unit Sepsis: A Systematic
Giuseppe Mazza1, Giuseppe Neri1, Helenia Mastrangelo2
1Department of Medical and Surgical Sciences, University "Magna Graecia" of Catanzaro, 88100 Catanzaro, Italy.
Journal of Personalized Medicine
|July 27, 2026
Summary
Artificial intelligence (AI) models show promise in predicting outcomes for intensive care unit (ICU) sepsis patients, often outperforming traditional scores. However, cautious interpretation is needed due to bias and limited validation.
Area of Science:
- Intensive Care Medicine
- Medical Informatics
- Artificial Intelligence in Healthcare
Background:
- Artificial intelligence (AI), machine learning (ML), and deep learning (DL) are increasingly applied to prognostic prediction in intensive care unit (ICU) sepsis.
- The clinical readiness and reliability of these AI/ML/DL models remain uncertain, necessitating a thorough evaluation.
Purpose of the Study:
- To systematically review AI-, ML-, and DL-based models for predicting mortality and clinically relevant outcomes in adult ICU patients with sepsis or septic shock.
- To assess the performance, validation, clinical utility, and risk of bias of these advanced prognostic models.
Main Methods:
- A systematic search of PubMed/MEDLINE, Scopus, and Cochrane Library up to April 2026.
- Inclusion of adult ICU sepsis/septic shock cohorts evaluating AI/ML/DL prognostic models, with independent screening, data extraction, and risk of bias assessment (PROBAST + AI).
- Synthesis of outcomes, model performance (discrimination, calibration), validation strategies, clinical utility, explainability, and reporting completeness (TRIPOD + AI).
Main Results:
- Seventy-five studies were included, with AI/ML models generally showing higher discrimination than conventional scores (median ΔAUROC +0.108).
- External validation was reported in 27 studies, with heterogeneous but clinically relevant discrimination across sepsis phenotypes.
- A significant proportion (45 studies) were judged at high risk of bias, primarily due to analytical limitations.
Conclusions:
- AI/ML models demonstrate a consistent signal for prognostic discrimination in ICU sepsis, often matching or exceeding conventional scores.
- Cautious interpretation is advised due to heterogeneity, limited prospective validation, incomplete calibration, and frequent high risk of bias.
- Future research should focus on prospective, calibrated, externally validated, and workflow-integrated AI decision-support tools for clinical implementation.