Blood pressure history improves near-term hypotension prediction and reassessment prioritisation in sepsis

A pressure-history-informed gradient boosting model predicts recurrent hypotension within four hours in sepsis, achieving AUROCs of 0.840–0.862 versus 0.791–0.832 for current information.

• Why it matters: Clinicians need better near-term detection of recurrent hypotension after target pressure is reached, and better prioritisation of reassessments to guide safe treatment.
• What they did: Using cross-fitting at the patient level, the model used 24 predictors across 370,447 windows from 22,485 patients in MIMIC-IV, eICU, and SICdb, then was externally evaluated and tested in US chronological replay.
• The result: In replay, it generated 9.0% and 11.8% fewer prompts while identifying 551 and 89 additional episodes before the first low-pressure record, and target-trial emulation linked early norepinephrine reduction to less drug exposure but greater hypotension risk in 4,180 patients.