BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Keita Nakao - From Words to Disease Trajectories: Exploring Parkin
 son’s Disease - ICJ
DTSTART:20261012T083000Z
DTEND:20261012T093000Z
DTSTAMP:20261010T001900Z
UID:indico-event-16812@indico.math.cnrs.fr
DESCRIPTION:Can everyday medical records tell us how a disease develops be
 fore it is diagnosed? In this talk\, I will introduce an approach to explo
 ring early signs of Parkinson’s disease using electronic health records.
  These records contain descriptions of symptoms collected over many years\
 , but turning this text into information that we can analyze and interpret
  is challenging.\nTo address this\, we developed a representation called H
 CVR that captures two aspects of each clinical entry: which symptom catego
 ry it belongs to and how favorable or unfavorable its meaning is. This all
 ows us to turn a person’s medical history into a series of numerical pro
 files while retaining clinically meaningful information. We then use a hid
 den Markov model to infer underlying health states and explore how people 
 move between them over time.\nUsing records from approximately 20\,000 peo
 ple with Parkinson’s disease and 19\,000 controls\, we found that people
  followed several different pathways before diagnosis. Even among people a
 ssigned to the same underlying state\, their clinical records contained di
 fferences that helped distinguish those who later received a Parkinson’s
  diagnosis from controls. I will discuss how combining language models wit
 h an interpretable statistical model can help us explore the early stages 
 of disease and understand why the path to diagnosis differs between indivi
 duals.\n\nhttps://indico.math.cnrs.fr/event/16812/
URL:https://indico.math.cnrs.fr/event/16812/
END:VEVENT
END:VCALENDAR