PGHD for Digital Phenotyping and Digital Biomarker Development

Last month, UCB, held its first Digital Health Roundtable on Leveraging Patient-Generated Health Data for Digital Phenotyping and Digital Biomarker Development. The panel included five digital health experts:

• Erin Rainaldi, MS (Head of Sensors Data Science, Verily)
• Jennifer Goldsack, MChem, MA, MBA, OLY (Chief Executive Officer, Digital Medicine Society (DiME))
• Lorene Nelson, PhD, MS (Associate Professor, Department of Epidemiology and Population Health, Stanford University)
• Dr. Francesca Rinaldo, MD, PhD (SVP, Clinical Product and Innovation Sharecare)
• And lastly, panel moderator Emily Kunka,MS, CCRP (Digital Business Transformation Lead, UCB Biopharma)

The panel discussed a wide range of topics spanning from the current state of processes of digital phenotyping to the future of digital healthcare, with a focus on the role that patient generated health data and digital biomarkers will play as clinical research transforms.

A few key themes emerged from the various questions the panel addressed: 

1. Digital Phenotyping is a more promising solution to the phenotyping challenge

It is no secret that patient health data is becoming increasingly more accessible with the advent of new mobile health technologies. The growing conversation around digital phenotyping can be partially attributed the growing ability to leverage data collected from smartphones, wearables, ambient devices, medical devices, etc. These new measures for data collection have afforded the research community a new perspective into a patient’s condition. In general, instead of collecting data at isolated arbitrary points in time, these technologies allow data to be collected in a multidimensional way that is more in tune with the natural history of a disease. As Dr. Francesca Rinaldo from Sharecare put it; “it’s the difference between getting a snapshot of a disease versus watching a full length motion picture of the disease.”

The panel went on to describe the importance of this development for therapeutic areas where symptoms do not always present themselves in the way physician's would expect. For instance, sleep, which is impacted by a variety of health conditions, is often collected through self-reported measures, such as surveys or questionnaires that ask participants to be remember their sleep quality. With the advent of mobile health technology, we can now more granularly understand sleep duration, sleep phases, interruptions to sleep, and when layered with respiratory rate, or ePROs on day-time fatigue, can start to build a more holistic picture of a patient's condition. 

A quote by Dr. Rinaldo can sum up the panel’s sentiments best:

 “If you are not incorporating this level of digital phenotyping early in your R&D process, you are missing the ability to granularly understand your patient population, understand the strongest signals that are patient centric around therapeutic efficacy and how to not only bring an asset to market but make sure it is having a clinically meaningful impact on those populations.”

2. How do we make sense of the data derived from digital phenotyping?

With the increase in patient health data available to research teams, it should not come as a surprise that data management and data collection need to be well thought out and aptly prepared for. Designing a study the right way from the beginning is increasingly complex. Clinical trials of tomorrow will demand that researchers are designing studies with patient data at the forefront, especially with increasing regulatory guidance to do so.

For Verily's Erin Rainaldi, “There are several big categories of steps involved in digital phenotyping, and the first is study design. When thinking about the data you want to collect, you want to understand first, what problem am I trying to solve.” 

More specifically, critical questions to ask could be:

• What biomarkers do we need to design and develop? What data do we need to develop those?
• What stakeholders need to be involved so we ensure we are collecting the right data? Do we need stakeholders across functions like regulatory, clinical scientists, clinical operations, etc.?
• Which technology is best suited to solving our problem or collecting necessary data? Is it a watch, phone, device, RWE, etc.?
• Once we have collected data, what’s the right model to build on top of all this data to get the biomarker we are looking for?
• How do you validate that what you are measuring is what you think you are measuring?

Another key insight from Erin was her approach to digital biomarker development. 

“I like to think of this activity of getting to a biomarker that measures what you want it to, by starting with these smaller physiological signal building blocks.

For example, suppose you have a smart watch collecting [pulse rate information like] PPG. From PPG you can get intra vi intervals, which is another building block that can get you pulse rate and pulse rate variability, but you can also get sleep staging. Then once you have that you might care about how much time someone spends in REM sleep. From there you might care about REM sleep behavior disorder in Parkinson’s disease. As you go down this path of raw data to specific measurements, you can start getting closer to the clinical questions you can answer with the data you have.”

3. Moving from ‘sick care’ to healthcare with digital phenotyping and remote patient monitoring

Traditionally, a great deal of the healthcare industry’s work has been allocated to treatment/intervention, monitoring, and outcomes. The panel agreed that most of the attention of digital phenotyping and digital biomarkers should be focused on illness prediction/prevention and diagnosis. With the rise of Bring Your Own Device (BYOD) studies, researchers can leverage data from devices that people already have, taking data and directing it towards efforts surrounding prediction, prevention, and diagnosis. We are starting to see some of it happen today as people aim to be more in tune with their health than ever. However, with data privacy laws and regulations at heightened complexity, it is certainly a bumpy road, but one very much worth traveling. Digital biomarkers and digital phenotyping are actively moving healthcare upstream from treatment/monitoring to prevention and diagnosis. 

As Jen Goldsack from the DiMe Society said in closing, “By definition, our healthcare system and research enterprise largely kicks in when you identify that you are ill. We can really start to deeply understand not just sickness, but health with these measures. We can reimagine what good healthcare looks like, which is not how good we are at dealing with you once you're sick, but how good we are at keeping you healthy.”