A Systematic Review of Machine Learning Models For Sepsis Prediction: An Appraise-Ai Approach.
Tingrui Wang, Qinqin Li, Zhangyi Wang et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review assessed machine-learning models for sepsis prediction and the quality of the supporting research. It included 53 studies, mainly retrospective cohort studies published from 2020 to 2024, and evaluated them using the APPRAISE-AI tool. Reported model area under the curve values ranged from 0.64 to 0.98; LightGBM and multilayer perceptron models reached 0.98. The review found that AI models generally outperformed non-AI methods, while most studies were of moderate quality and had methodological, robustness, or reproducibility concerns.
Key findings
- Across the 53 included papers, model area under the curve values ranged from 0.64 to 0.98.
- LightGBM and multilayer perceptron models had the highest reported area under the curve, at 0.98.
- AI models generally outperformed non-AI methods in sepsis prediction.
- Most studies were rated moderate quality, with concerns about methodology, robustness, and reproducibility.
Why it matters
Sepsis prediction models may support earlier recognition, but reported performance alone does not establish reliable clinical usefulness. The review highlights variation in study quality and gaps in robustness and reproducibility, which matter when interpreting model accuracy and considering whether predictions can be applied across clinical settings.
What to keep in mind
The review reports that most included studies were retrospective and of moderate quality, with methodological, robustness, and reproducibility flaws.