Early clinical decision support using interpretable artificial intelligence in acute illness (sepsis and infection): A systematic review and meta-analysis.
Mohamed Ibraheem, Mosab Khalil, Somaia Khalil et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis evaluated the diagnostic and prognostic performance of interpretable AI models in sepsis and severe infections. Among 15 studies, the pooled sensitivity for sepsis detection was 0.685 and specificity was 0.898 (AUC = 0.879), while mortality prediction showed a sensitivity of 0.618 and specificity of 0.778 (AUC = 0.753).
Key findings
- Pooled sensitivity for sepsis detection was 0.685 and specificity was 0.898 (AUC = 0.879).
- Mortality prediction models had a pooled sensitivity of 0.618 and specificity of 0.778 (AUC = 0.753).
- Fourteen of the 15 studies had a high overall risk of bias.
- Sensitivity analysis showed stable performance for sepsis detection (sensitivity 0.751, specificity 0.873, AUC = 0.890).
Why it matters
The findings highlight the potential of interpretable AI in enhancing early diagnosis of sepsis, a critical factor in improving patient outcomes. However, the moderate performance in mortality prediction suggests that further validation is necessary before widespread clinical implementation.
What to keep in mind
The included studies predominantly exhibited a high risk of bias, and there was substantial heterogeneity among them.