Wearable sensors can tell you more than just step count: measuring nighttime scratching and sleep

A more holistic picture of life with atopic dermatitis

Like Comment
Read the paper

Key takeaways 

  • Atopic dermatitis, which is sometimes referred to as eczema, consists of dry skin, flaky or scaly patches, itching, and sores that may weep. 
  • People suffering from atopic dermatitis often experience increased episodes of scratching during the night as well as sleep disturbances, resulting in negative impact on their and their families’ quality of life. 
  • The traditional episodic, subjective, and burdensome assessments of disease state provide a limited view of the patient’s experience with the condition. 
  • Quantitative digital measurements of nighttime scratching and sleep can provide continuous, objective assessments of disease state, ultimately allowing for a more complete picture of the impact of the condition on daily life and better evaluations of therapeutic interventions 

Motivation 

As a common chronic or recurrent inflammatory skin diseaseAtopic Dermatitis (AD) affects approximately 20% of children and up to 10% of adults from developed countries and is ranked 15th among nonfatal diseases and 1st amongst skin diseases with regards to disease burden (disability-adjusted life-years; DALYs) [1]. Recent work has shown that pruritus (itch) is regarded by patients as the most problematic and burdensome symptom of AD [2]. Itch is a complex feeling that often results in the action of scratching. In AD, scratching can perpetuate and exacerbate the condition, creating open sores, oozing, and potential infections. Furthermore, both itch and scratch can have a marked effect on sleep, causing sleep disturbances and making sleep less restful. All of these can have a significant negative impact on the quality of life of patients, caregivers, and patients’ families. As one patient expressed it:

The symptoms compound. I itch so much that I wake up scratching. The scratching leads to raw, open wounds. All of this impacts my sleep and general health. […] All of this also impacts my other physical health because exercising is painful, and sleep is not restful” [2]

Current treatment paradigms for AD often rely on data captured using traditional assessment tools such as clinical outcome assessments (COAs) and patient reported outcomes (PROs). While COAs, performed in-clinic by a dermatologist, are designed to assess lesion severity, they provide data points only at infrequent intervals (at in-person visits) and do not capture day-to-day variability of symptoms such as itch severity or scratch quantity experienced outside the clinic. PROs (surveys taken in the home by the patient) do provide insight into itch severity and sleep quality as experienced by the patient; however, they are subject to recall bias, subjective, and often suffer from issues with compliance due to the high degree of effort they require from the patient [3]. As a result, data obtained from these assessments provides an incomplete picture of all the symptoms that accompany the disease.  

Advancements in digital technology, including wearable devices with embedded sensors, have started to allow for objective measurements of human health during daily life. Originally, these tools were primarily used to measure activity levels (e.g. step counts), but recent research efforts have explored utilizing these tools for passive, unsupervised monitoring of disease symptoms as well as to provide more reliable indicators of treatment efficacy, and better inform treatment paradigms. In this paperwe highlight a method for processing raw accelerometer data from wrist-worn device to passively derive digital measures of nighttime scratch activity (total number of scratch episodes and duration) and sleep quantity (including total sleep time and percent time asleep) in patients with ADWe foresee that using a tool like this will complement current clinical assessments used to measure disease progression by providing quantitative behavioral insights into treatment efficacy and the ability to tailor treatment paradigms to individual patients. Figure 1inspired by recent work done to support the inclusion of mobile sensors in clinical trials [4], illustrates this concept below. 

Figure 1. Concept of use for wearable-based measurements in atopic dermatitis. 
Figure 1. Concept of use for wearable-based measurements in atopic dermatitis. 

wearable-based scratch and sleep monitoring system with model interpretability 

Leveraging learnings from prior research efforts in this space, we utilized a hierarchical paradigm when developing the analytical system to derive continuous measures of nighttime scratch and sleep quantity. The proposed framework first determines when a person intends to sleep (i.e., context detection) and then derives the sleep quantity and nighttime scratch activity within that window (illustrated in Figure 2). Our goal was to develop a method that requires no input from the patient (ex. inputting sleep start and end times). The proposed method should be able to achieve continuous monitoring for days or weeks without requiring any action from the patient 

Figure 2. Analytical system design. We used a hierarchical approach for detection and assessment of sleep and nighttime scratch using accelerometer data from a wrist-worn device. Raw accelerometer data is first sliced into a 24-hour window (12:00pm – 12:00pm), then segmented to the total sleep opportunity (TSO) window, and finally measures of sleep and scratch are computed during the TSO window.

We took a pragmatic approach for algorithm development, using machine learning for more complicated tasks (ex. scratch detection) and heuristic algorithms for simpler tasks. Along with ensuring our method enables continuous monitoring, it was equally important to us that we build highly functional algorithms. When deploying tools to monitor health, it is important to understand the accuracy of the derived measures. Our results show that the digital measures derived from this method perform equally well (and better in some cases) to prior work and are highly correlated with reference measurements of nighttime scratch and sleep. In this case, as part of the qualification process we used annotated video-based recordings to quantify and teach the algorithm nighttime scratch activity; in a similar fashion we used polysomnography to provide ground truth that the subject was sleeping. 

When designing our scratch detectormodel interpretability was important because providing clarity and transparency into how algorithms make decisions is critical within highly regulated environmentsWe employed a supervised learning feature-based approach, using time and frequency domain features to parameterize scratching and restless (non-scratching) movements during the night. This approach provides greater insight into how the model determines what types of hand movements are specific to scratchingwhich we believe is advantageous compared to black box prediction algorithms [5] and may allow for further research into the impact of scratching movements (and severity of scratching movements) on AD disease progression.   

Looking ahead 

Quantitative digital measures of nighttime scratch and sleep, captured continuously during a patient's daily life, can enable clinicians, researchers, and patients to objectively track a new dimension of the disease that they haven’t been able to explore before. From this new information, we can not only better characterize the disease, but also quantitatively evaluate treatment-induced changes with a known level of accuracy. Moreover, scratch and sleep disturbances are symptoms of several other conditions in addition to AD. We believe utilizing a solution like this can continue to strengthen our understanding of these conditions, hopefully allowing for improved treatments and outcomes for patients.   

Authors:

Nikhil Mahadevan, Junrui Di, Carrie Northcott and Yiorgos Christakis

References 

  1. Laughter MR, Maymone MBC, Mashayekhi S, Arents BWM, Karimkhani C, Langan SM, Dellavelle RP, and Flohr C. The global burden of atopic dermatitis: lessons from the Global Burden of Disease Study 1990-2017.  British Journal of Dermatology.  2020 Oct 2. 
  2. More Than Skin Deep, Understanding the Lived Experience of Eczema. https://www.aafa.org/media/2628/more-than-skin-deep-voice-of-the-patient-report.pdf.  
  3. Murray, C. & Rees, J. Are Subjective Accounts of Itch to be Relied on? The Lack of Relation between Visual Analogue Itch Scores and Actigraphic Measures of Scratch. Acta Derm. Venereol. 91, 18–23 (2011). 
  4. Walton MK, Cappelleri JC, Byrom B, Goldsack JC, Eremenco S, Harris D, Potero E, Patel N, Flood E, Daumer M. Considerations for development of an evidence dossier to support the use of mobile sensor technology for clinical outcome assessments in clinical trials. Contemp Clin Trials. 2020 Apr;91:105962. doi: 10.1016/j.cct.2020.105962. Epub 2020 Feb 20. PMID: 32087341. 
  5. Moreau A, Anderer P, Ross M, Cerny A, Almazan TH, Peterson B, Moreau A, Anderer P, Ross M, Cerny A, Almazan TH, Peterson B. Detection of Nocturnal Scratching Movements in Patients with Atopic Dermatitis Using Accelerometers and Recurrent Neural Networks. IEEE J Biomed Health Inform. 2018 Jul;22(4):1011-1018.  

 

Yiorgos Christakis

Data scientist, Pfizer, Inc.