Generating a Novel Measure of Scratch Using Wrist-Worn Wearables

Researchers now have an objective, passive measure to assess itchy conditions and the benefit of potential treatments. We know the feeling of itch, often followed by scratch behavior, negatively affects the quality of life of millions of people living with a wide range of clinical conditions. Moreover, the action of scratching is a key component of several disease symptomologies. Scratch and itch are interdependent symptoms. Itch can, and often does, provoke the desire to scratch, but scratching can also be an unprovoked action. Scratching then leads to further problems such as inflammation and lesions that, in turn, can cause itch.1 Many people are caught in a vicious itch-scratch cycle.

Traditionally, such clinical conditions can only be assessed using physician assessments of lesions (often the result of the scratching) and patients’ own subjective views. This can be problematic as physician assessments only provide a periodic snapshot and patient-reported outcomes (PROs) can be considered burdensome and prone to recall bias.1 An objective, passive measure of scratching could help to reveal one of the outcomes that matters most to people and demonstrate the impact of therapeutic treatments to intervene.

In most cases, people use their hands to scratch. Accelerometers worn on the wrist can capture this motion right where the action is. To develop better therapies for atopic dermatitis (AD) and improve patient’s lives, scientists at Pfizer have been developing a novel digital measure that can measure scratching in patients with AD when they are sleeping, passively, quantitatively and over time, with wrist worn wearables. To learn more about what is involved in generating a novel digital outcome measure, I reached out to Yiorgos Christakis, a Senior Data Scientist at Pfizer’s Digital Medicine and Translational Imaging group (DMTI).

 

Q&A with Yiorgos Christakis

 

Mari Griffioen (MG)
Thank you for taking the time to meet today, Yiorgos. Can you tell me about your background and role at Pfizer?

Yiorgos Christakis (YC)
After I obtained my biomedical engineering degree, I realized that a degree in computer science and machine learning would help me become a better and more effective engineer in today’s quickly evolving environment. The combination helped prepare me to contribute to the development of new digital biomarkers. During my last semester, I joined Pfizer as a data scientist and focused on building algorithms, using them to develop new digital measures, and generating the evidence for regulators to accept digital endpoints for use in clinical trials. This is exciting work as I got to be a part of the research process from start to finish.

MG:
Can you tell me about how you became interested in using data from wearables as digital endpoints in clinical trials?

YC:
I’ve always wanted to combine my problem solving and analytical skills in a way that benefits people and improves their lives. As a data scientist at Pfizer, I get to do that by helping to determine what we can measure, and how we can measure it using objective digital data and the growing collection of digital health technologies. Our group is interested in minimizing participant burden and increasing the validity of the results of our clinical trials. An algorithm that could detect nighttime scratch in AD patients seemed to meet these criteria. More importantly, it would give us new information about what scratch looks like in the real-world – we could start to explore what the effects of new treatments or behavioral modifications are on patients’ day to day experience with AD. Essentially, this novel digital endpoint – scratch – could help us determine whether a treatment was effective in people's daily lives.


Highlights of developing an objective scratch measure

As we continued our conversation, Yiorgos outlined some of the main points of validation, analytical approach, and the impact of this work on science and patients. One question I had for him was how to validate a novel measure when a “gold standard” does not exist. Researchers might be asked to examine the correlation with PROs. Very often, however, PROs as measures of subjective experiences are not appropriate to serve as the validation gold standard of objective novel measures. It is important to identify the appropriate reference measure that can be accepted as the “ground truth.” In this case, the Pfizer team chose to use thermal camera recordings taken while subjects spent the night in a sleep lab. Seeing is believing!


But the challenge did not stop here. Even with recordings of the actual scratch behaviors, the team needs to come up with a guide that can be used to annotate scratches by human observers consistently. And the annotation of “scratching severity” is particularly sticky. The study team used a definition based on the number of joints involved in the scratching movements. This, however, might not be the best way to capture all types of scratches, for instance, deep scratches that involve greater pressure on the skin rather than high amplitude movements.


To derive the algorithms of scratch from the acceleration signals, the team chose to use machine learning as the analytical approach. I asked him why not deep learning. He said this was indeed a decision that the team discussed early on. Deep learning is a powerful technique, but the team ultimately chose a more interpretable, classical machine learning model and hand-crafted features. Regulatory acceptance is the ultimate goal for this work, and thus the interpretability of the algorithms is highly desirable.


Finally, we discussed the impact of this work. First, an objective, quantitative and passive measure to capture scratch as an endpoint could accelerate drug development and bring better treatments to patients. Second, a wrist-worn wearable is easy to deploy at home, has a low participant burden, and by quantifying scratch, participants have an objective measure to support their self-report.


In summary, I want to thank Yiorgos for the opportunity to learn about the development of a novel digital outcome measure. This objective, passive measure of scratch using a wrist-worn wearable has the potential to improve the lives of people living with the vexing problem of itch and scratch.


References

  1. Ständer, S. et al. Gender differences in chronic pruritus: women present different morbidity, more scratch lesions and higher burden. Br. J. Dermatol. 168, 1273–1280 (2013).
  2. Kapur, S., Watson, W. & Carr, S. Atopic dermatitis. Allergy Asthma Clin. Immunol. 14, 52 (2018).
  3. Mahadevan N, Christakis Y, Di J, et al. Development of digital measures for nighttime scratch and sleep using wrist-worn wearable devices. npj Digit Med. 2021;4(1):1-10. doi:10/gnk2v4

 

 

Do you know of a research study that involves the use of digital endpoints in a unique or innovative way? If so, please contact us at science@theactigraph.com.

 


 

Related White Paper

Actigraphy for Sleep Measurement: Three ActiGraph Use Cases 

Disturbed or impaired sleep is a growing area of research focus, both as a primary disorder and in cases where disturbed sleep is a symptom of another disease. Actigraphy monitoring with wearables provides a low-burden and remote approach to objectively quantify participants’ sleep behavior in their natural environment, often revealing meaningful insights that might not be available with polysomnography or self-report data. In this white paper, we present three use cases for collecting actigraphy-derived sleep measures during an interventional clinical trial. 

AG_Marketing_SocialMedia_WhitePapers_2021_ActigraphyForSleepMeasurements_FINAL_WEB


Interested in receiving more articles like this?

Subscribe to Our Blog

 

 

Back to Blog

Recent Articles

Bridging the Knowledge Gap on Digital Clinical Measures: A Technology Partner’s Perspective on ‘The Playbook’

ActiGraph was one of the very first commercial providers of wearable motion-sensing technology,...

Quantifying Sleep in the Real World

ActiGraph recently published a white paper presenting several examples of how wearables have been...

How Might Temperature Sensor Data Improve Digital Outcomes?

The temperature of the human body is tightly regulated and stays within a narrow range between 36...