Sleep 1.1 Overview of Series and Making Consumer Devices Research-Worthy

• My name is Andrew Ahn. I’m chief medical officer at Labfront. I am a physician trained in internal medicine, also a practicing hospitalist at the Boston VA Medical System, and also a clinical researcher with a background in physiology, math, and signal processing.
• The title of today's talk is Labfront's Vision for Advancing Sleep Research, Catalyzing Discovery and Healing. And it presents an overview of our vision on how we could best support sleep researchers such as yourselves, and this is the first in a series of talks oriented around wearables and sleep research.
• And to give you a preview of what types of topics will be discussed in this series, I'm listing here the planned topics for future webinars. These topics include actigraphy, PPG, circadian rhythm, sleep irregularity, sleep staging, and sleep disorders, all with a particular focus placed on wearables.
• And note that each of these subcategories will actually be a separate webinar talk. And so the goal of this series is actually threefold. One, to help you understand the science, particularly from a wearable perspective. Two, to demonstrate features within Labfront that support these efforts. And three, to obtain your feedback and help us to optimize various functionalities and to help us help you in your research work.
• And hopefully through these webinar talks, you will become increasingly appreciative of the complexity of wearable sciences and recognize the importance of physiology and signal processing. And some of the things we will address include how heart rate and heart rate variability, respiration, blood pressure varies with sleep stages, why PPG with green light is less susceptible to motion artifact compared to red light, what limitations exist for reflective PPG in assessing oxygen saturation, how does data differ between ring and wrist-based devices based on the physiology?
• And why skin tone does not typically change heart rate or heart rate variability quality but can diminish the quality of oxygen saturation? And specifically for the next talk, the focus will be on sleep actigraphy. And as you know, this technology has been around for decades, and the presumption, I think, out there, is that there's little room for scientific advancement.
• Basically, the idea that all that's been left had been discovered has been discovered already. But the, the truth is that, you know, sleep actigraphy has found new use cases. For instance, for sleep irregularity and circadian scientists. And nevertheless, the sleep research community has been stopped relying on analytical methodologies designed more than 30 years ago.
• The Cole-Kripke and Sadeh method, the two most commonly used sleep wake algorithms out there, used the AMI Motionlogger, which had a piezoelectric transducer accelerometer. Modern devices now use MEMS capacitive accelerometers. Which provide three axis accelerometer data, whereas the PAS electric accelerometer is a uni axial accelerometer, which provides only one axis acceleration information.
• So these modern accelerometers have some advantages, but there are also some disadvantages too, which I plan to go over in the next talk. So here's the outline for today's talk, which is divided to six sections, making consumer devices research worthy, data integration across devices, implementation science, performance evaluation, research transparency, and exploration of novel approaches and analytics.
• But before going too far into the nitty gritty, I think it's important to sort of establish the key measures frequently used in sleep research. And this is proposed by Dr. Daniel Buysse, from the University of Pittsburgh in his highly cited paper published in Sleep in 2014.
• And we focus on several important metrics. Sleep duration, the total amount of sleep an individual gets. Sleep efficiency. This includes how effectively one sleeps and often incorporates analysis of sleep stages. Sleep timing, this is the chronological aspect of when sleep occurs, such as whether one is a morning lark or a night owl.
• And I might also incorporate circadian rhythm within this category too. Fourth is sleep regularity. In fact, Dr. Buysse didn't include this as a category, but later on he did. And it's the consistency of sleep patterns over time. And then there's sleep satisfaction and daytime alertness, both of which are qualitative in nature.
• These measures can be represented visually, and again, this is adapted from Dr. Buysse Grand Rounds, which he presented in 2022 and this is kind of, sort of a representation of normal sleep from midnight to 8 a. m. , and you could have, you know, a rating of good satisfaction after you wake up, you could have low alertness, where during the day you have periods of sleep or naps.
• You have short duration sleep, long duration sleep, late timing, high efficiency, low efficiency, where there's a lot of periods of wake and then high regularity, where you have consistently, you know, regular patterns of time to go to bed and wake up. And you also have low regularity. And you can have a combination of these factors as well.
• Short duration, late timing, low efficiency, poor quality. So, you know, moving forward, you can then begin to imagine how wearables might be able to capture these measures on a day to day in the wild settings.
• So the first section of this talk is making consumer devices research worthy. And before jumping into consumer devices specifically, I wanted to give you the lay of the land when it comes to wearables overall.
• Because many of you have probably struggled trying to figure out which wearables should be incorporated in your research. And I think, well, I have to move this thing. So I think Max deZambotti did a great job at synthesizing the available options in their recent state of the science for wearables in sleep research publication in sleep earlier this year. And I believe it was endorsed by the Sleep Research Society. And this group classified wearables into three classes. Traditional, research grade actigraphy, consumer grade devices, and contemporary research grade devices.
• So the traditional research grade actigraphies, we know this, you know, by the CamNtech Motion Watch, the Philips Actiwatch 2, which was discontinued, as we know, in January 2023, AMI Motionlogger, and these have been extensively studies, they have accelerometer as well as some of them have ambient light detectors and also a button for event marker.
• And you also have raw data accessibility. There's some dichotomized sleep classification between sleep and versus wake, but there's no classification sleep staging because they can't do this, , with just accelerometer data alone. The disadvantage is that it does require a USB connection for downloading data.
• So the devices have to be shipped back or, you know, somehow physically, transferred from the study participants to, to the lab team. And as a result of that, there's really no real time data acquisition. So you can't sort of, assess for adherence in real time, except for after post hoc when you get the devices back and you see whether the data is there or not.
• There's a long battery life, sometimes up to 30 days, and the disadvantage is the cost and, and, but this includes not only the device, but also the software costs as well. And, there is some expectations that the personnel have some training in terms of data cleaning, interpretation of that data.
• For consumer devices, we are aware of all the ones that are out there. There's Fitbit Oura Ring, Apple Watches, and Garmin. And so they have diversity in hardware and software, and these devices have accelerometer and importantly, the addition of PPG compared to the traditional actigraphies. And because of the addition of PPGs, it's able to classify sleep and wake, but also sleep stages.
• The problem is that there is limited to no raw data access in general for the consumer devices. They also use undisclosed proprietary algorithms. We have no idea how they actually got their reports. And there's the issue of firmware and software updates. And so these things, the goalpost gets continually changed, as they're making updates.
• There's a shorter battery life, these hardware is not necessarily research grade calibrated, and the data is processed in the firmware, firmware in cloud, and the advantage is that they're lower cost and have a higher adoption rate. And then the third class are these contemporary research devices that are out there.
• We know GeneActive , Empatica, Sleep Image and Actigraph Leap as an example, they all have pretty diverse functionalities. The GeneActive is, in my opinion, just a high end actigraphy, but it has some temperature information. Empatica has EDA or GSR to assess sympathetic nervous activity. Sleep image assesses sleep stages as well as evaluation for sleep apnea.
• Actigraph has PPG temperature barometers as well. And they're research grade collaborated, calibrated, and they're able to classify sleep, wake, and sleep stages. The raw data is accessible, and the data can be processed in firmware or the cloud, and in some cases a Wi Fi where you don't have to use the cell phone.
• As a data relay, and the algorithms are can be open and sometimes proprietary as well. They tend to have longer battery life. And again, the disadvantage is the cost, you know, for not only the devices themselves for the software as well. And at Labfront, however, you know, we recognize that consumer grade wearables do have their limitations.
• And it's listed here and it's discussed previously in my slides, like two or three slides before. But there are also some undeniable strengths, which make it highly appealing to certain research groups. Consumer devices are laboriously designed by manufacturers to make it user friendly. And this includes the form factor, the watch interface, the mobile app design, and this really enhances adherence.
• There's also multiple models from which you can select and so this, you could choose the option that best suits your needs for, for instance, the VivoSmart 5s have an adjustable wrist size that can go very small and so can be suitable for children and adolescents. The low cost is probably the key factor here.
• The low cost enables us to get preliminary data for pilot studies and also becomes financially feasible for larger epidemiological studies. And I do feel that the hardware has gotten advanced enough that there is parity in terms of quality compared to research level devices. And importantly, we have found partner companies like Garmin who are willing to work with us to address these limitations, to open up the data, and to really sort of advance research.
• So how are we making consumer grade devices research worthy? First, we're making raw data from consumer devices accessible to the researcher. And take for instance Garmin, what we have is we have heart rate variability and this means continuous beat to beat interval, 24 hours a day, 7 days a week. There's, , it's continually obtained, like the green light that usually is used to turn on to indicate this is on, it's always on.
• And Garmin, to my knowledge is the only consumer wearable that gives us access to this continuous data. We also have, or the ability to access raw accelerometer data, 25 to a hundred Hertz. But again, this has significant effects on the battery life for Movesense, our EKG and IMU device, we also have accessibility to the raw data.
• This includes the nine axis and inter initial measurement unit, which includes three axis accelerometer, gyroscope, as well as magnetometer. And they're obtained at multiple sampling frequencies. And then we also can get our intervals or because of EKG functionalities, , and the raw EKG itself, the waveform, which you can get up to 500 Hertz.
• But acquiring raw data does entail a lot of data and particularly from the perspective of a consumer device, which has limited memory and battery storage capacity. And also limited upload rates via Bluetooth. And Labfront, we do give you the option of either streaming or logging the data. So streaming the data allows real time uploading to the cloud using a cell phone or relay device.
• But this does require like the cell phone or the tablet. To be in close proximity to the patient at all times, we also to get the option of logging the data, which temporarily stores the data on the watch until it connects to a cell phone to deliver it to the cloud. But again, the uploading syncing time could be quite long.
• I mean, it can conserve the battery for both these functionalities and for instance, a 24 hour acquisition of 25 hertz raw accelerometer data would take about 2 hours to upload all that data.
• So to streamline the data's transfer, Garmin has worked with us to create customizable actigraphy settings. And this gives the researchers the option to configure actigraphy settings to suit your specific needs. This means that you could adjust the Epoch window size, anything larger than one second, so it could be 30 seconds, it could be a minute to five minutes.
• You could give some options of the type of filter, the filter path, the process data, which includes zero crossing, energy and time above threshold, and even the threshold settings. These are things that I will talk about in the next talk of the webinar series. The other really interesting thing, the really cool things that Garmin helped to establish was that you could have three settings and each of them with different parameters.
• So different window size, different Butterworth or different type of filters, filter path. You could do zero crossing, more energy, a time above threshold and thresholds, and you could acquire it all simultaneously. Using our Labfront and Garmin sort of devices. So again, there's these advantages that sort of we've partnership with these companies that really help us to help your cause, to do research in this way, we also provide assessment of data quality and take for instance, beat to beat interval, which is BBI Garmin provides a confidence score to determine whether the PPG derived BBI is of good quality or not. And you could see here, RRI, which is derived from the gold standard EEG, the EKG, which is in red, and also the blue BBI which is derived from the PPG. And you could see that the, the times that the RRI does not sort of agree with the BBI.
• The confidence score is zero, so it's either one or zero, and it does a pretty outstanding job of telling us when the BBI data cannot be reliably used for our analysis.
• We also provide data analytical capabilities and we will spend some time in future webinars to discuss this. We provide one night reports for an individual, , for a sleep actigraphy, and we also hope to provide some data conversion formats to AWD file formats for, you know, Actiware's software, which we know is available for free ever since Philips inactivated or did away with the Actiwatch series.
• This software is available till December 2027. We also provide heart rate variability analysis as well, which include time domain. Frequency main and nonlinear heart rate variability metrics as well.

‍