BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Contributed Talk: Representing Time Series\, Missingness\, and Sem
 antics in Surgical Risk Prediction - Hugo Armando Guillen Ramirez (Univers
 ität Bern)
DTSTART:20260210T113000Z
DTEND:20260210T120000Z
UID:TALK243103@talks.cam.ac.uk
DESCRIPTION:Modern surgical care generates heterogeneous data spanning con
 tinuous intraoperative physiology\, postoperative laboratory measurements\
 , and complex procedural classifications. A key challenge is how to repres
 ent and integrate these data in a way that preserves clinically meaningful
  structure while enabling robust and interpretable learning.\nWe present a
  unified framework for perioperative risk modelling focused on postoperati
 ve infectious complications. Using large-scale electronic health record da
 ta\, we show that representations of intraoperative vital-sign dynamics de
 rived from irregular and noisy time series enable accurate prediction of p
 ostoperative infections already at the end of surgery\, substantially outp
 erforming models based on static preoperative variables alone. These repre
 sentations capture physiologic instability through distributional\, trend\
 , and entropy-based features and remain interpretable using explainable ma
 chine learning.\nWe further extend prediction into the postoperative phase
  by modelling laboratory time series with imputation strategies that accou
 nt for missing-not-at-random mechanisms driven by clinical decision-making
 . Integrating laboratory kinetics enables earlier and more accurate detect
 ion of infection than clinician-initiated treatment and supports procedure
 -specific laboratory testing recommendations.\nFinally\, we introduce MAP-
 CARE\, a multilingual embedding framework that aligns national surgical pr
 ocedure classifications into a shared semantic space using large language 
 models. This representation layer enables cross-language retrieval and sur
 gery-agnostic learning\, providing a foundation for generalizable perioper
 ative risk prediction.
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR