BEGIN:VCALENDAR
PRODID:-//github.com/ical-org/ical.net//NONSGML ical.net 4.0//EN
VERSION:2.0
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:STANDARD
DTSTART:20241103T020000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20250309T020000
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
DESCRIPTION:SPEAKER : Tianxi Cai\, PhD\, John Rock Professor of Population
  and Translational Data Sciences Biostatistics\, Department of Biostatist
 ics\, Harvard T.H. Chan School of Public Health TITLE : “ From Real-World
  Data to Durable Clinical AI: Building Reliable Evidence Across Instituti
 ons and Over Time” ABSTRACT: The widespread adoption of electronic health
  record (EHR) systems has created unprecedented opportunities to harness 
 real-world clinical data for artificial intelligence\, translational rese
 arch\, and evidence generation. Yet realizing this potential requires add
 ressing a fundamental challenge: EHR data are inherently noisy\, heteroge
 neous\, and dynamic. Differences in coding\, clinical workflows\, patient
  populations\, and data modalities complicate integration across healthca
 re systems\, while temporal and institutional shifts can cause even well-
 performing models to degrade after deployment. In this talk\, I will pres
 ent a series of analytical frameworks for building reliable and durable c
 linical AI from multi-modal\, multi-institutional EHR data. I will first 
 discuss methods for integrating heterogeneous information across institut
 ions and across structured records and unstructured clinical narratives\,
  including recent opportunities enabled by large language models and repr
 esentation learning. I will then consider the challenge of developing mod
 els that remain reliable as populations and healthcare environments chang
 e. Moving beyond reactive retraining\, we formulate model maintenance thr
 ough robust optimization\, constructing uncertainty sets of plausible fut
 ure environments and proactively balancing performance in the current pop
 ulation against robustness to future distributional shifts. I will also d
 iscuss new inferential methods for quantifying uncertainty in robust opti
 mization\, an important but relatively underdeveloped area. Using example
 s from longitudinal EHR data across multiple healthcare systems\, I will 
 illustrate how these approaches can transform heterogeneous real-world da
 ta into reliable evidence and support clinical AI systems that are not on
 ly accurate today\, but also generalizable across institutions and durabl
 e over time. YSPH values inclusion and access for all participants. If yo
 u have questions about accessibility or would like to request an accommod
 ation\, please contact Charmila Fernandes at Charmila.fernandes@yale.edu 
 . We will try to provide accommodations requested by October 27\, 2026.\n
 \nSpeaker:\nTianxi Cai\, PhD\n\nAdmission:\nFree\n\nFood:\nRefreshments a
 t 3:45 PM\n\nDetails URL:\nhttps://medicine.yale.edu/event/ysph-biostatis
 tics-seminar-tba-11-02-26-copy/\n
DTEND;TZID=America/New_York:20261102T170000
DTSTAMP:20260929T082004Z
DTSTART;TZID=America/New_York:20261102T160000
GEO:41.303666;-72.932218
LOCATION:Yale School of Public Health (LEPH)\, 115\, 60 College Street\, N
 ew Haven\, CT\, United States
SEQUENCE:0
STATUS:Confirmed
SUMMARY:YSPH Biostatistics Seminar: “From Real-World Data to Durable Clini
 cal AI: Building Reliable Evidence Across Institutions and Over Time"
UID:fff69dd6-482b-4d05-8fbc-5d60a7ac7435
END:VEVENT
END:VCALENDAR