외래 정신과 환자에서 AI 기반 착용형 순응도 모니터의 단기 효과

실험적: AI-enabled smartwatch

Participants continued their usual psychiatric care and wore the AI-enabled smartwatch medication-monitoring system for four weeks. Electronic tags were attached to their active medication containers. The device did not provide medication reminders, adherence feedback, or additional counseling, and device-derived data were not returned to the treating clinician during the randomized comparison period.

장치: AI-enabled smartwatch

The smartwatch was designed to detect medication-taking behavior using a built-in camera and electronic tags attached to medication containers. The camera remained inactive under ordinary conditions and captured a brief sequence only when container motion, ambient light, and close proximity between the watch and tag were detected concurrently. The captured sequence was processed by an artificial intelligence algorithm to classify whether a medication-taking event had occurred. Participants assigned to the intervention were instructed to wear the device for four weeks. The device provided monitoring only and delivered no reminders, alerts, counseling, or treatment recommendations.

간섭 없음: Waitlist Usual Care

Participants continued usual psychiatric care for four weeks without smartwatch monitoring or additional medication-adherence counseling. After completing the randomized comparison period, participants were offered optional use of the smartwatch for an additional four weeks. Data from this optional period were analyzed exploratorily.

Mean Change From Baseline in Korean Brief Adherence Rating Scale Score at Week 4

Medication adherence was assessed using the clinician-rated Korean Brief Adherence Rating Scale (K-BARS). The clinician assessed the number of prescribed doses, the number of doses reportedly taken, and the frequency of missed doses over the preceding 28 days and integrated this information into a score ranging from 0% to 100%, with higher scores indicating greater medication adherence. For each participant, the change score was calculated as the Week 4 score minus the baseline score. The primary outcome was the difference in mean change score between the randomized groups.

Δ ≥ 10 퍼센트 포인트에서의 반응자 비율

기준선 대비 BARS 변화 점수가 10% 포인트 이상인 참가자의 비율. 로지스틱 회귀분석을 사용한 그룹 간 비교로 오즈비와 95% 신뢰 구간을 추정.

Δ ≥ 20%포인트에서 반응자 비율

기준시점에서 BARS 변화 점수가 20% 포인트 이상인 참가자의 비율. 로지스틱 회귀분석을 이용한 군 간 비교로 오즈비 및 95% 신뢰구간 추정.

장치 참여도와 BARS 변화 점수 간의 관계

중재군 참가자들 사이에서, 장치 참여율(장치가 기록한 ≥1회 약물 이벤트가 있는 중재 일수 비율)과 BARS 변화 점수 간의 연관성, Pearson 상관 및 선형 회귀를 사용하여 분석됨.

대기자 교차 코호트 내 피험자 내 순응도 변화

이후 스마트워치 기간을 완료한 대조군 참가자 중 대기자 관찰 단계 동안 측정된 순응도와 이후 스마트워치 기간의 비교. t-검정을 사용한 쌍체 비교.