Authors (including presenting author) :
CHU ACH(1), LEE CFT(2), YIP LM(2), LEUNG L(3), LEE H(3), LOH A(3)
Affiliation :
(1) Department of Neurosurgery, Kwong Wah Hospital (2) Nursing Services Department, Kwong Wah Hospital (3) IT Department, Kwong Wah Hospital
Keyword 1: :
Fall prevention
Keyword 2: :
thermographic sensor
Keyword 3: :
artificial intelligence
Keyword 4: :
bed-exit detection
Keyword 5: :
implementation science
Keyword 6: :
patient safety
Introduction :
Inpatient falls remain the most common adverse event in acute care hospitals, occurring at a rate of 0.44 per 1,000 inpatient bed days in Kwong Wah Hospital (KWH). It could result in serious injury, prolonged hospitalization, and substantial costs. Current fall prevention relies on pressure-sensitive bed alarms (PSBA), which are ineffective due to high false alarm rates (72–99%). Emerging thermographic sensor systems combined with artificial intelligence (AI) algorithms offer promise for "predictive" detection of patients' pre-exit behavioural patterns, potentially providing 5–15 seconds of lead time for intervene. Moreover, thermographic systems use anonymized heat silhouettes, enabling 24-hour continuous monitoring while maintaining patient privacy.
Objectives :
The primary objective is to determine the sensitivity of the thermographic AI system for identifying bed-exit events among hospitalized adult patients. Secondary objectives are to evaluate the false alarm rate, system utilization, and nursing staff acceptance and satisfaction with the system.
Methodology :
This was a prospective observational cohort study conducted in 7 acute-care wards at KWH. Thermographic AI devices were installed above individual beds for patients with fall risks. The AI model classified patient postures and movements and issued an alarm when a bed-exit attempt was predicted. Secondary data sources include system utilization logs and nursing staff survey.
Result & Outcome :
Over the research period, the device monitored 1,145 hours, achieving over 95% sensitivity for bed-exit detection (84 valid alarms) and a 24.1% false alarm rate. Valid alarms occurred at 1.845 per bed per day; false alarms at 0.587 per bed per day. The system captured 50 bed-exit attempts. Among 102 nursing staff (92.7% response rate), the system achieved very high acceptance: usability (4.51/5), interface user-friendliness (4.36/5), minimal training required (3.01/5). Strong support for hospital-wide implementation was reported (4.29/5). Overall satisfaction was 4.36/5; Workflow disruption and alert workload were minimal (1.99/5 and 1.74/5). The thermographic AI system achieved over 95% sensitivity for bed-exit detection with manageable false alarm rate. Nursing staff demonstrated very high acceptance with minimal workflow disruption. The device's favourable diagnostic accuracy, usability profile, and strong staff endorsement warrant hospital-wide implementation and larger prospective studies measuring fall incidence reduction.
