AI-Based Sepsis Prediction in Hospitalized Adults: Systematic Review, Subgroup Meta-Analysis, and Contextual Analysis of Clinical Burden.
Gyeong Min Lee, Joo-Yun Won, Eun Young Cho et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis evaluated machine learning (ML) and deep learning (DL) models for sepsis prediction in hospitalized adults, analyzing 34 studies. The pooled area under the receiver operating characteristic curve (AUROC) was 0.913 (95% CI 0.887-0.933). The study highlighted significant variability in model performance across different prediction windows and validation maturity levels.
Key findings
- The pooled AUROC for sepsis prediction models was 0.913 (95% CI 0.887-0.933).
- Models predicting sepsis within 4 hours had a pooled AUROC of 0.894 (95% CI 0.829-0.936).
- Models predicting more than 4 hours before onset achieved a pooled AUROC of 0.926 (95% CI 0.897-0.948).
- Sepsis-related episodes resulted in longer hospital stays and higher medical costs compared to general inpatient episodes.
Why it matters
Understanding the performance of ML and DL models for sepsis prediction is crucial for clinicians, as it informs the potential for early intervention and improved patient outcomes. However, the variability in model performance across different settings necessitates cautious interpretation before implementation in clinical practice.