Overview
- Researchers at Icahn Mount Sinai and collaborators detailed the framework in Nature Communications on September 26.
- The approach links a patient's labs, medications, vital signs, and visits into a temporal network to produce patient‑specific diagnostic signals.
- In validation at Mount Sinai and UC Irvine, it exceeded clinical rules by 12–16x for culture‑negative neonatal sepsis and 4–7x for postoperative acute kidney injury.
- The system learns directly from deidentified records with few labeled examples and transferred across the two hospital systems.
- The team is making the code available to researchers and will explore adapting the method to personalize treatments using clinical‑trial data.