Learning a Continuous Sepsis Severity Score Without Hour-by-Hour Supervision: A Two-Site Retrospective Study
A retrospective two-cohort study of 29,116 and 7,691 adult Sepsis-3 patients from two hospital systems developed a sepsis index using 43 routinely charted variables over a 72-hour treatment window. The model used mortality as a treatment-level ranking signal, redistributing credit non-uniformly across timesteps. Evaluation on a 20% test holdout showed non-survivors scored 1.19-1.64 points higher than survivors on a 0-10 scale within all strata of baseline SOFA-2, with similar results stratifying within lactate, MAP, and creatinine.
Researchers developed a continuous sepsis severity score from routinely charted variables without hour-by-hour supervision, using a ranking-based learning approach. The study involved 36,807 patients across two sites and demonstrated consistent separation between survivors and non-survivors across multiple clinical strata.
The approach uses a treatment-level ranking signal rather than per-state targets, allowing credit assignment to be redistributed non-uniformly across timesteps. This enables learning from mortality outcomes without explicit hourly labels, potentially capturing dynamic severity trajectories more effectively than fixed indices.
The method could lead to more adaptive and data-driven clinical severity scores that reflect contemporary patient populations, addressing limitations of decades-old indices like SOFA. Adoption would require validation across diverse health systems and integration into electronic health record workflows.
Improved sepsis severity scoring could enhance risk stratification, resource allocation, and quality benchmarking in hospitals. A validated continuous score might be licensed or integrated into EHR platforms, creating value for health systems and clinical decision support vendors.
Next observable signals include prospective validation studies, comparisons with existing sepsis scores in clinical settings, and exploration of the approach for other time-series clinical prediction tasks. Regulatory and interoperability considerations will influence clinical deployment.