SepsisLens: Structure-Preserving Sequence Modelling for Decomposable Early Sepsis Warning
The model is built to keep sepsis alerts tied to the variables and organ-level signals behind them.
SepsisLens handles irregular ICU measurements without collapsing everything into one patient-level representation. Its risk head composes multi-horizon warnings from explicit variable-level and organ-level components. The authors report strong discrimination across three public ICU cohorts and one private-hospital cohort, plus lower alert burden at matched event recall on MIMIC-IV. Structural ablations and masking tests are presented as evidence that its ranked components track influential inputs. ArXiv · AI/CL/LG's note
SepsisLens handles irregular ICU measurements without collapsing everything into one patient-level representation. Its risk head composes multi-horizon warnings from explicit variable-level and organ-level components. The authors report strong discrimination across three public ICU cohorts and one private-hospital cohort, plus lower alert burden at matched event recall on MIMIC-IV. Structural ablations and masking tests are presented as evidence that its ranked components track influential inputs. ArXiv · AI/CL/LG's note
score 4