BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Los_Angeles
X-LIC-LOCATION:America/Los_Angeles
BEGIN:DAYLIGHT
TZOFFSETFROM:-0800
TZOFFSETTO:-0700
TZNAME:PDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0700
TZOFFSETTO:-0800
TZNAME:PST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20261012T210904Z
LOCATION:Summit Pavilion
DTSTART;TZID=America/Los_Angeles:20261021T173000
DTEND;TZID=America/Los_Angeles:20261021T183000
UID:HFESAM_ASPIRE 2026_sess257_POST334@linklings.com
SUMMARY:Confound-Aware Evaluation of Smart-Home Digital Biomarkers: Auditi
 ng Observation Intensity Dominance in Cognitive Status and Agitation
DESCRIPTION:Andrew Choi, Daniel Choi, and Junho Park (University of Calgar
 y)\n\nConfounding by observation intensity (“coverage”) is a fundamental r
 isk in smart-home digital biomarkers: coverage can correlate with labels a
 nd dominate engineered feature sets. We introduce a confound-aware protoco
 l with three models (Cov3, CovPP, All). This protocol is demonstrated on t
 wo datasets without cross-dataset transfer. On Technology Integrated Healt
 h Management (TIHM) agitation detection, the area under the receiver opera
 ting characteristic curve (AUC) increases from 0.8090 (Cov3) to 0.9138 (Al
 l; ΔAUC = +0.1048, p = 5 × 10−5), while CovPP is statistically indistingui
 shable from All (p = 0.9055). In contrast, on the Center for Advanced Stud
 ies in Adaptive Systems (CASAS) scripted cognitive tasks—dementia (DX1) an
 d mild cognitive impairment (MCI)—AUC gains beyond Cov3 are modest and sta
 tistically uncertain (DX1: +0.0771, p = 0.2989; MCI: +0.0586, p = 0.3421).
  These tasks serve as a low-power stress test (DX1: 20 positives) in which
  coverage can dominate discrimination. Results use participant-disjoint ou
 t-of-fold prediction with cluster-aware swap-null permutation tests; we re
 port both AUC and area under the precision–recall curve (AUPRC) to reflect
  class imbalance. Coverage baselines and paired incremental inference can 
 support more robust claims about cognitive specificity in ambient-sensing 
 digital biomarker studies while treating coverage as an explicit, auditabl
 e variable in deployed monitoring systems.\n\n
END:VEVENT
END:VCALENDAR