Self-information from a next-token EHR foundation model identifies clinically surprising tokens and events whose counts predict mortality and long length-of-stay, and whose removal degrades representation-based prognostic models.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Quantifying surprise in clinical care: Detecting highly informative events in electronic health records with foundation models
Self-information from a next-token EHR foundation model identifies clinically surprising tokens and events whose counts predict mortality and long length-of-stay, and whose removal degrades representation-based prognostic models.