NCT07455916

Short-Term Effects of an AI-Based Wearable Adherence Monitor in Outpatient Psychiatry

Short-Term Effects of an AI-Based Wearable Adherence Monitor in Outpatient Psychiatry: Randomized Controlled Trial.

• Status: Completed
• Phase: N/A
• Study type: Interventional
• Enrollment: 78 (actual)
• Sponsor: Wonkwang University Hospital · Academic / Other
• Sex: All
• Age: 12 Years
• Healthy volunteers: Not accepted

Summary

Medication nonadherence undermines treatment effectiveness in psychiatric care, yet objective, continuous measurement in routine practice is challenging. AI-enabled wearables may offer scalable monitoring but evidence from randomized evaluations remains limited.

Detailed description

We conducted a single-center, prospective, randomized controlled trial of an AI-based, wrist-worn adherence monitoring system in outpatient psychiatric care. Participants (age ≥12 years) were randomized 1:1 to smartwatch intervention or usual care for 4 weeks. Algorithm-derived adherence was calculated as the proportion of detected medication events relative to scheduled doses over prespecified 28-day baseline and post-intervention windows. The primary endpoint was change in adherence (Δ = post - pre). Analyses used complete cases with linear regression (adjusting for baseline adherence, age, sex, and medication covariates) and HC3 standard errors; ANCOVA served as a confirmatory model. Prespecified responder thresholds were Δ ≥10 and ≥20 percentage points (pp). Sensitivity analyses excluded benzodiazepine users.

Conditions

• Mental Disease
• Adherence to Care
• Digital Health

Interventions

Type	Name	Description
DEVICE	AI-enable smartwatch use	The device was designed to monitor medication-related behaviors in real-world settings (pill taking and, by design, use of eye drops, inhalers, and nasal sprays). A built-in camera remained in sleep mode and recorded brief \~20-second clips only when an electronic tag affixed to the medication container signaled three concurrent conditions: (1) container motion detected by the tag's accelerometer, (2) ambient light detected by the tag's light sensor, and (3) watch-tag proximity within approximately 10-15 cm via BLE ranging. After capturing a clip, the camera returned to sleep. Encrypted videos were transmitted to a secure server and linked to de-identified study IDs. Server-side algorithms then analyzed the full 20-second sequence, covering the continuous hand actions from opening to closing of the container, and returned a binary medication event (medication vs no medication). Participants were instructed to wear the smartwatch for 4 weeks

Timeline

• Start date: 2022-12-31
• Primary completion: 2025-08-31
• Completion: 2025-08-31
• First posted: 2026-03-06
• Last updated: 2026-03-06

Locations

1 site across 1 country: South Korea