Physical Activity Monitoring in Older Adults in Simulated Free-Living Conditions

This prospective observational study was conducted at the eHealth House (eHH) of the TechMed Simulation Centre, University of Twente (Enschede, The Netherlands). The eHH is a controlled environment designed to simulate free-living conditions (FLC) and includes a living room, kitchen, bedroom, and bathroom. All experimental sessions were recorded using five fixed cameras installed throughout the eHH. Participants were eligible if they were aged 80 years or older and physically able to participate independently. Individuals with cognitive impairments or mobility disorders that impeded task performance were excluded. According to Dutch law, and based on a ruling from the Medical Research Ethics Committee (MREC) Arnhem–Nijmegen, the study was exempt from the Medical Research Involving Human Subjects Act. Ethical approval was granted by the Ethics Committee of the Natural Sciences and Engineering Sciences of the University of Twente. All participants provided written informed consent prior to participation. Each participant visited the eHH once for a 75-minute session. During the visit, participants performed a set of activities of daily living (ADL) at their own pace and in their preferred order. These tasks reflected functional discharge criteria for geriatric hip fracture patients and included: (1) walking inside the eHH (between the living room, kitchen, bedroom, and front door), (2) visiting the toilet once, (3) getting in and out of bed once (sit-to-lie, lying, and lie-to-sit), (4) preparing a meal (walking to the kitchen, cutting food, and returning to the living room), and (5) preparing a drink (walking to the kitchen, pouring a drink, and returning to the living room). Participants’ movements and postures were recorded using two wearable devices: the MOX activity monitor (Maastricht Instruments, The Netherlands) and the APDM activity tracker (Hankamp Rehab BV, The Netherlands). The MOX is a waterproof device containing a single triaxial accelerometer and was attached to the upper thigh, approximately 10 cm above the knee, using a medical plaster. Data were recorded at 25 Hz. This placement was selected based on evidence that upper-leg accelerations exhibit low inter-person variability, facilitating better generalization in HAR. The APDM comprises a triaxial accelerometer, gyroscope, and magnetometer, and was worn on the lower back using a strap. Data were recorded at 128 Hz. This location provides robust measurements of sedentary behaviors with low sensitivity to postural differences, and captures representative whole-body motion due to its proximity to the body’s center of mass. The combination of thigh and lower-back sensors was considered necessary to reliably distinguish all static and dynamic activities, as a single sensor location was insufficient. To obtain gold-standard labels, two independent reviewers annotated the video recordings into the following activity classes: walking, standing, sitting, lying (supine, left lateral recumbent, right lateral recumbent), and transfers (stand-to-sit, sit-to-stand, sit-to-lie, and lie-to-sit). Disagreements were resolved by a third reviewer. The activity annotations were synchronized with the sensor recordings by aligning the timestamps of the first annotated sit-to-stand transfer observed in the videos with the corresponding acceleration signals. After synchronization, all subsequent analyses for synthetic data generation were performed using accelerometer data only. The gyroscope and magnetometer signals were excluded, as previous studies showed that accelerometers alone outperform multimodal sensor combinations for comparable HAR tasks, with no significant performance gains from sensor fusion. All data processing was conducted in MATLAB (R2022a, MathWorks Inc., Natick, MA, USA).