Machine learning predicts sepsis deterioration trajectories - PubMed
5 hours ago
- #machine learning
- #clinical deterioration
- #sepsis
- Sepsis has heterogeneous clinical trajectories, and dynamic prediction is crucial for personalized intervention.
- An ensemble machine-learning model identified three latent recovery patterns: rapid recovery (41.5%), slow recovery (36.4%), and clinical deterioration (22.1%).
- The model showed strong performance with AUROC up to 0.92 in development and maintained accuracy in external validation across multiple datasets.
- Reduced heart rate variability (SD < 10 bpm) was a significant predictor of mortality, with an adjusted hazard ratio of 2.17.
- Implementation of the model led to clinical improvements: reduced ICU stay by 1.8 days, mechanical ventilation by 2.3 days, and 28-day mortality by 5.7%.
- The study provided early warnings for deterioration with a median lead time of 17.6 hours, enabling proactive care strategies.