Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
Not provided
| Name | Class |
|---|---|
| Michigan State University | OTHER |
Not provided
Not provided
Not provided
Not provided
Sleep is often a challenge for nightshift workers because their work and sleep schedules are inverted. Sleep is commonly measured using actigraphy, which is the standard measure of objective sleep in the general population; however, this method has substantial limitations for nightshift workers because the standard legacy algorithms only correctly identify 50.3% of daytime sleep. This significantly reduces the validity for nightshift workers. The purpose of this study is to test a novel method to expand actigraphy by using 1) a multi-sensor approach that 2) uses machine learning (ML) algorithms to increase the accuracy of detecting daytime sleep.
The first aim of this study is to establish an open-source machine learning algorithm for sleep tracking that outperforms legacy actigraphy algorithms in detecting daytime sleep periods. The second aim is to enhance tracking of sleep continuity variables by adding multiple sensors. The final aim is to identify facilitators and barriers of at-home implementation of multi-sensor sleep tracking. Our central hypothesis is that a multi-sensor ML approach will outperform legacy algorithms against gold-standard polysomnography (PSG).
This study will be type I hybrid effectiveness-implementation trial that 1) validates the proposed multi-sensor ML approach using in-lab polysomnography, and 2) examines implementation of the multi-sensor ML approach in an ecologically valid setting via an at-home implementation for four weeks. A sample of nightshift workers will be enrolled in the in-lab validation portion of the study and will be hooked-up to PSG with continuous data collection for the duration of the lab visit to capture five planned sleep opportunities at varying lengths (4 hr, 2 hr, 1.5 hr, and two 30-minute naps; 8 hrs total). For each participant, sensor data will be processed using two separate methods. For the legacy actigraphy algorithm method, only raw accelerometer data will be processed. For the multi-sensor machine learning method, accelerometer data from the watch along with additional sensors will be processed using a machine learning algorithm. Some participants who complete the in-lab portion of the study will be asked to complete the at-home portion of the study, which includes 4 weeks of at-home sleep tracking using the multi-sensor approach. Participants will receive the sensor kit and will have an at-home appointment with study staff to aid with sensor set-up, which will then be collected again at the end of the 4-week period. Daily sleep diaries will also be collected during the 4 weeks to enable data quality check.
Not provided
Not provided
Not provided
Not provided
| Label | Type | Description | Intervention Names |
|---|---|---|---|
| Single vs Multi-Sensor Sleep Tracking In-Lab | Experimental | In Part 1 of the study, all participants' data will undergo two separate methods for analyzing sleep. The legacy actigraphy algorithm methods will use only raw accelerometer data from a single sensor collected and processed using legacy actigraphy algorithms. The legacy algorithm is comprised first of reducing accelerometer data into activity counts per epoch, which will then be categorized into sleep or wake in accordance with the Cole-Kripke algorithm. The multi-sensor machine learning (ML) method will use raw accelerometer data in addition to data from additional sensors from the watch, phone, and other smart sensors in the sleeping environment. These data will be processed using a machine learning algorithm. |
|
| Multi-Sensor Sleep Tracking At-Home | Other | This condition includes 4 weeks of at-home sleep tracking using the multi-sensor approach. Daily sleep diaries will also be collected to enable data quality check. Once collected, all data will be processed with the same machine learning algorithm used in the in-lab experimental condition. |
|
| Name | Type | Description | Arm Group Labels | Other Names |
|---|---|---|---|---|
| Single-Sensor Tracking (In-Lab) | Other | In-lab sleep tracking using only raw accelerometer data from a single sensor collected and processed with legacy actigraphy algorithms. |
|
| Measure | Description | Time Frame |
|---|---|---|
| Sleep Continuity- Time in Bed | The amount of time (in minutes) a participant spends in bed from lights out to their final awakening time. All PSG variables will use standard American Academy of Sleep Medicine (AASM) sleep scoring rules. Data from the Apple Watch will have non-PSG inputs from the multi-sensor system to inform this sleep continuity variable. | Throughout study completion, up to 6 weeks |
| Sleep Continuity- Sleep Onset Latency | The amount of time (in minutes) a participant takes to fall asleep, from the time of lights out, or the amount of time spent awake but attempting sleep from lights out. All PSG variables will use standard AASM sleep scoring rules; indicated with "lights out" marker on a PSG, EEG scored as wake, accompanied with a prototypical sleep posture (e.g. supine) with eyes closed. Data from the Apple Watch will have non-PSG inputs from the multi-sensor system to inform this sleep continuity variable including dim lights or darkness with lux near zero, presence in bed, rare/interspersed motion from phone and watch. | Throughout study completion, up to 6 weeks |
| Sleep Continuity- Wake After Sleep Onset | The amount of time (in minutes) a participant spends awake from the time they initially falling asleep, and excluding their final wake up. All PSG variables will use standard AASM sleep scoring rules; indicated with "lights out" marker on a PSG, electroencephalography (EEG) scored as wake, accompanied with a prototypical sleep posture (e.g. supine) with eyes closed. Data from the Apple Watch will have non-PSG inputs from the multi-sensor system to inform this sleep continuity variable. | Throughout study completion, up to 6 weeks |
| Sleep Continuity- Sleep Efficiency | The proportion of the total amount of time a participant is asleep of the total amount of time in bed [(Total Sleep Time in minutes) / (Time in Bed in minutes)]. All PSG variables will use standard AASM sleep scoring rules. Data from the Apple Watch will have non-PSG inputs from the multi-sensor system to inform this sleep continuity variable, including dim lights or darkness, presence in bed, prolonged low motion from phone and watch, breathing rate changes, and heart rate (sleep staging). |
| Measure | Description | Time Frame |
|---|---|---|
| Interviews | These interviews will be semi-structured using the Consolidated Framework for Implementation Research (CFIR). The moderator guide will solicit in-depth feedback on participants' experiences, challenges, and suggestions for improvement. Key themes to be explored in the interviews include ease of use, comfort, perceived accuracy, and any barriers to regular use. | Within one month of the at-home intervention |
Not provided
Inclusion Criteria:
Exclusion Criteria:
Not provided
Not provided
Not provided
Not provided
Not provided
| Name | Role | Phone | Extension | |
|---|---|---|---|---|
| Philip Cheng, PhD | Contact | 248-344-7361 | pcheng1@hfhs.org | |
| Elle M Wernette, PhD | Contact | 2483442409 | ewernet1@hfhs.org |
Not provided
| Facility | Status | City | State | ZIP | Country | Contacts |
|---|---|---|---|---|---|---|
| Henry Ford Columbus Medical Center | Recruiting | Novi | Michigan | 48377 | United States |
Request for data sharing will be evaluated on a case-by-case basis
Not provided
Not provided
Not provided
Not provided
Not provided
Type I hybrid effectiveness-implementation trial with two steps. Step 1: In-lab trial (with a sample of participants balanced on degree of technological literacy) comparing data processing using legacy actigraphy to multi-sensor machine learning. Step 2: At-home implementation of multi-sensor machine learning approach (balanced by technological literacy).
Not provided
Not provided
Not provided
| Multi-Sensor Sleep Tracking (In-Lab) | Other | In-lab sleep tracking using raw accelerometer data and additional sensors collected and processed with machine learning. |
|
|
| Multi-Sensor Sleep Tracking (At-Home) | Other | At-home sleep tracking using raw accelerometer data and additional sensors collected and processed with machine learning. |
|
| Throughout study completion, up to 6 weeks |
| Wake | The amount of time (in minutes) a participant is awake [or the absence of any type of sleep- Stage 1 (N1), Stage 2 (N2), Stage 3 (N3), Rapid Eye Movement (REM)]. All PSG variables will use standard AASM sleep scoring rules; represented on PSG by activities prior to "lights out" marker or video monitoring (eg, video monitoring showing scrolling on social media in bed). Data from the Apple Watch will have non-PSG inputs from the multi-sensor system to inform these sleep continuity variables including motion, lights on, high heart rate. | Throughout study completion, up to 6 weeks |
| Detection of Daytime Sleep Periods | Any sleep periods between 6a and 6p will be designated as daytime sleep. A daytime sleep period from the Apple Watch will be considered successfully detected if it falls within ±30 minutes of the PSG start and end times, and is at least 50% the length of the actual sleep period. | Throughout study completion, up to 6 weeks |
| User experience | This will be indexed with the User Experience Questionnaire (UEQ) that has been validated for evaluation of new products and has clear and well-established benchmarks. The UEQ includes items along six domains: 1) Attractiveness (overall likability or appeal), 2) Perspicuity (learning curve and ease of use), 3) Efficiency (speed and efficiency of interactions), 4) Dependability (predictability of system behaviors), 5) Stimulation (how exciting and motivating the product is), 6) Novelty (innovation and creativity of the product). | Within two days of the at-home intervention |
| Digital health technology literacy | This will be measured using the Digital Health Technology Literacy scale (DHTL). This validated scale assesses degree of experience and skills in using digital health technology and services. The DHTL has strong internal consistency (Cronbach's α = 0.95) and strong validity with completion of ten digital tasks such as connecting a device to Wi-Fi and Bluetooth, downloading and installing an app, and entering weight data into an app. | During screening before the in-lab intervention |