Study design and participants
A total of 56 participants (33 females, mean age 74.2 ± 3.9 years, years of education 11.7 ± 2.0 years) were enrolled among a population of lonely older adults living in the community of Usuki, Oita Prefecture, Japan. In the present study, the inclusion criteria, meet by all participants, were as follows: 1) aged 65 and over, 2) lived in Usuki, 3) physically and psychologically healthy, 4) no dementia, and 5) had an independent function in the activities of daily living. All subjects were instructed to wear wristbands (Silmee™ W20, TDK Corporation Tokyo, Japan) throughout the day, except when showering, for a week during September 2020. These subjects had previously participated in a prospective cohort study with a continuous follow-up that analyzed the association between objectively quantified lifestyle factors and mental function from August 2015 to March 2019. During that cohort study, we assessed lifestyle factors using the wearable sensors every 3 months for 3 years and calculated an average annual data [16]. Overall, the average annual data for 2016 and 2018 and the average weekly data for 2020 were collected for analysis.
Data obtained by the wearable sensors
For movement detection, we used a tri-axis accelerometer that can measure acceleration in three perpendicular axes. Accelerometers are based on the principle of differential capacitance, which results from the movement of the sensing element due to acceleration. Data were recorded continuously and analyzed every minute. We defined non-wearing time as the time when heart rate counts were zero. We analyzed the data for participants who had at least three days of valid accelerometer monitoring during the measurement period and at least three hours of valid accelerometer monitoring per day. We used our wearable sensors to assess various lifestyle parameters, including moving steps, metabolic equivalents of tasks (METs), total sleep time (TST), sleep effectiveness, wake time after sleep onset (WASO), number of awakenings, and conversation time. We summarized the measured values of these parameters into the sum of the sensor data collected on each day and their average values over the entire length period. Each parameter was represented as an average value per day. Before our study, we validated the accuracy of our measurements of hiking steps, talking time, and sleep time by comparing them with data from video observations of healthy elderly subjects.
Physical activity
We defined walking step as a movement in the frequency range of 2-3 Hz of acceleration detected by the sensor. The device computed the intensity of activities as METs using algorithms developed by the product designer. Sedentary behavior was defined as activities that involved ≤1.5 METs such as sitting, lying down, or watching television. Light physical activity (LPA) was defined as activities that involved 1.5–3.0 METs such as slow walking, laundry, cooking food, washing dishes, or vacuuming, whereas moderate and vigorous physical activities (MVPA) were defined as activities that involved ≥3 METs such as walking, jogging, or ascending and descending stairs.
Sleep
We evaluated sleep-wake parameters using the magnitude of acceleration and cumulative energy synthesized by a tri-axial accelerometer. The data were verified and corrected visually by a qualified technician. The bedtime was determined according to the number of activities logged by the wristband sensor. Sleep factors including TST, WASO, and sleep efficiency, and the awakening count were measured between 6:00 pm and 5:59 am (the following day). The time of sleep onset was defined as the time when the resting state began, with no movement for more than 20 minutes. We evaluated sleep fragmentation by using WASO, sleep effectiveness, and awakening counts. We defined nightly awakening as 5 to 90 minutes of continuous movement during a continuous sleep period. Sleep effectiveness was determined as the percentage of TST over bedtime. In this study, sleep diaries were not utilized but measured the total bedtime spent using TST and WASO.
Conversation time
This sensor was able to detect whether an adult or someone nearby had made utterances. Although we were not able to exclude the speech of someone nearby from the audio data, the contribution of the participant in the conversation itself was judged to be a valuable sign of social activity. The microphone on the wearable sensor cannot detect the substance of the chat, but it can collect data as sound every minute. We analyzed sound data to evaluate the time of the conversation. Our wearable sensor can detect the sound pressure level of utterances within a 2m radius of the device. The vibration level, considered as utterances at this distance, was 55-75 dBA. Furthermore, the incidence band corresponding to the human voice was extracted as a signal frame from the sound data within the vibration range. A chat was defined as a period of one min during which there were more than four sound frames.
Statistical analysis
The annual average data for 2016 and 2018 and the weekly average data for 2020 before and during the pandemic, respectively, were used for statistical analyses. A repeated-measures analysis of variance (rANOVA) was conducted to compare nine variables of lifestyle factors (walking steps, sedentary time, LPA, MVPA, chat time, TST, WASO, sleep effectiveness, and waking time count) before and during the pandemic. All statistical analyses were conducted using SPSS statistical software (version 25.0, IBM Corporation, USA) and Prism (version 7.00, GraphPad), and all p-values of <0.05 were considered statistically significant.