Medicine

Wearables are collecting a flood of data. An ambitious new study of pregnancy aims to prove that’s valuable for health

This post originally appeared on StatNews.

When Stephen Friend left Apple in 2017 after a stint helping to jumpstart its health efforts, the company’s marketing engine was already promoting a future that hadn’t arrived: Apple devices were going to help you live a healthier life.

In reality, iPhones, Apple Watches, and competitor products were collecting a flurry of data about activity, heart rate, and sleep. But the evidence they could detect disease or actually improve health outcomes was nascent.

Friend set out to change that.

advertisement

He is a fervent believer that the data collected continuously by consumer devices can be used to improve how care is delivered and give patients a far more personalized way to manage their own health. Rather than collecting snapshots of a few vital signs at an appointment and asking patients to recall symptoms that stretch back months, devices can record a steady stream of measured data and make it easier for patients to regularly keep tabs on how they’re feeling.

It was with that possibility in mind that Friend co-founded the nonprofit 4YouandMe to swing for the fences to prove the value of consumer devices in medical care.

advertisement

“We built this organization in order to take on high-risk questions and de-risk them so that others can move forward,” he said. “We’re not trying to make a product.”

That’s how Friend, who spent years as an executive at Merck advancing cancer treatments before turning to open-data efforts at Sage Bionetworks, found himself spearheading a study on pregnancy.

4YouandMe has launched an effort to follow the pregnancies of roughly 1,000 people, collecting a constant stream of hundreds of measures from devices as well as regular reports about symptoms like mood and fatigue. The goal, first and foremost, is to prove that so much data can be reliably collected. But beyond that, the researchers hope they can use some of these measures to describe the unique experience of an individual pregnancy and how much variability there is between individuals. While the study, called Better Understanding the Metamorphosis of Pregnancy, or BUMP, is only seeking to suss out the feasibility of methods, it could also pave the way for new technologies that can predict complications or better direct treatment for them.

Pregnancy is a prime target for what 4YouandMe wants to accomplish. It occurs over a finite period of time, brings a string of dramatic physical transformations, and should create a lush data set to mine for insights. That also makes it an all-in bet.

“If these types of consumer grade sensors can pick up things in real life, I think this study either makes or breaks them in many ways,” said Christine Lemke, co-CEO of Evidation Health, which built the study’s smartphone application and is responsible for data collection. “Not to put too much pressure on the back of this study, but if these sensors can’t be de-noised and contextualized to pick up some of the radical changes that happen in pregnancy, what good are they?”

The BUMP study is remarkable for the sheer amount of data it aims to collect before conception, during pregnancy, and postpartum. Participants will sport a Garmin smartwatch and an Oura smartring, step daily on a Bodyport cardiac scale, and field frequent questions about how they’re feeling in a smartphone app and interviews with the research team. They’ll complete walking and cognition tests and video diaries. Some participants will submit genetic data. Those who opt in will contribute anonymized details about their Instagram activity and phone usage.

Stress, fatigue, shortness of breath, sleep patterns, that telltale first trimester morning sickness — it’ll all be tracked.

From there, the researchers will go hunting for signal. To explain how, Friend pulled up dozens of charts mapping a symptom (shortness of breath) and measurement (respiratory rate) among individual participants. Some charts appeared to show strong correlation, while others could well have been random. The researchers plan to make similar comparisons between other variables, like a two-minute walk test and fatigue. For those measures that look worth probing, the Vector Institute for Artificial Intelligence in Toronto with help from a team at MIT will use machine learning to see if the data from the whole cohort can establish a robust correlation. They’ll then look for variations between individuals.

Stephen Friend Wikimedia Commons

At this point, Friend is hopeful they’ll be able to identify strong objective measures worth studying further for three or four of the pregnancy symptoms they’re studying.

4YouandMe’s focus on how a variety of measures add up to a person’s health is a significant evolution, said Jennifer Goldsack, CEO of the Digital Medicine Society. “I think there’s too much noise in a single measure. I also think that human physiology is complex, and the way that they are capturing multiple different data streams is absolutely the way the field needs to be moving.”

The data, which will be made available to other researchers, could also open the door to  future work on pregnancy. Catherine Spong, the chief of maternal fetal medicine and chair of the department of obstetrics and gynecology at the University of Texas Southwestern Medical Center, told STAT that while researchers have a general idea about rates of anxiety and depression in pregnancy, as well as an understanding of  physiological changes like increases in heart rate and blood volume, that understanding is at a high level over longer time periods.

“I think that this [study] would give more granular data across the time period to be able to answer some unique questions about how pregnancy affects a woman’s health,” she said.

One example: The study’s BUMP-C cohort, in which participants are followed for up to six months while trying to get pregnant, will give researchers a window into the early weeks of pregnancy about which there is very little data.

True to the study’s ambition, the researchers will be collecting outcomes and searching for patterns that could potentially hint at complications, though Sarah Goodday, 4YouandMe’s lead research scientist, has a hunch it’s unlikely that there are enough participants in this early study to arrive at that holy grail.

“The data, when you look at it at this high resolution, are so much more variable than we expected, it makes it really difficult to apply machine learning models to the data because it’s just so wildly different across different participants,” she said. “And so that’s an important finding that changes how we approach the use of these data.”

Be that as it may, Friend will walk you right to the edge of the study’s potential. When a question excites him, he fires off like a pinball into a forest of bumpers, bouncing off a description of a conversation with his collaborators at MIT, pulling up slides buried deep in presentation decks, and arriving finally at a comparison to the Framingham Heart Study, an evolving effort that has been collecting data on a single pool of people and their offspring since the 1940s.

He described one finding that caught the team’s attention in the early data:  a possible pattern emerging around what happens before someone goes into labor.

“The week before someone delivers, it looks as if the body is shifting to almost a post-delivery mood or post-delivery state,” he said. “And you must ask the question, how does the body say, it’s time to deliver? And so it’s not that we know that, but that’s the type of question we can ask…  Can we watch the body prepare for delivery in a way where you could say, this person is going to deliver in the next week?”

For all its potential health insights, BUMP is also an experiment in how to conduct a study with such intensive data collection. It requires a slew of partners to recruit the participants, develop the technology, and 4YouandMe also employs eight full-time clinical research engagement coordinators who do the hard work of shepherding participants through the study, troubleshooting technical issues, and collecting additional data as well feedback on what it’s like to use the devices.

Goodday said that though there’s an impulse to see digital health as a way to use technology in lieu of people for care, 4YouandMe’s early work has shown that “we cannot get rid of human contact. It’s really important. It needs to be paired with the digital health approach.”

What the researchers glean from running the BUMP study will inform the design of future work by 4YouandMe, including a larger study of pregnancy being planned. In addition to testing and tweaking efforts to keep participants actively engaged in the study, they’re also trying to make sure the project doesn’t run into the common problem of only recruiting well-off patients. The team is trialing several methods for finding a diverse pool of participants, such as recruiting in clinics from Connecticut’s Community Health Centers which target underserved populations and testing the waters with finding possible participants through Facebook groups and Reddit.

“I actually think that our hopes or our vision for digital health could fail if we don’t figure out this engagement piece,” said Goodday. “So we are testing out and trying to learn: How can you engage individuals from diverse backgrounds in the use of these tools and how can we ingrain them into their daily lives?”

The hand-holding is helpful because the study requires a substantial commitment from participants, who receive a small amount of compensation and are expected to complete daily surveys about their stress and symptoms, to wear their devices constantly, and to submit to a battery of other regular tests and questionnaires.

For Kirsten Kelley, a participant who was recruited to BUMP after completing prenatal genetic testing with study partner and funder Sema4, it wasn’t a tough sell — she joked that the prospect of getting free devices she could keep once the study wrapped was appealing enough on its own. “I was like, hell yeah, I’ll do whatever I need to do,” she said.

For the most part, Kelley said, the study tasks were doable during her breaks at work. She did her postpartum walking tests carrying her daughter, born in February. “She would not let me put her down,” said Kelley.

Kelley said it was helpful to have someone outside of her family support system that could talk to her and validate her experiences. Likewise, recording all her symptoms in the app allowed her to process the discomfort and stresses of pregnancy.

“It was good to be able to step back and just purge all of those into an app at the end of day, so that I’m like, OK, it’s down, somebody knows this is happening,” she said. That — along with the feeling she was contributing to a valuable effort — has made participating worthwhile for her.

“So much of women’s medicine is unfortunately not listening to women,” said Kelley. “And this study is literally using the voices of women, every bit of their experience, everything that they’re sharing about what is happening to their body. And that is so important.“

4YouandMe goes to great effort to both listen and involve the participants as much as possible. On an April Zoom call for the 300 participants who’d enrolled so far, the staff members all introduced themselves, and Goodday revealed a surprise: Based on feedback, the team had been working to design a cognition test that wouldn’t take them quite so long to finish — one minute versus four minutes. While ring and watch usage rates were hovering over 80% before birth, completion rates for the cognition tasks sat at about 50%. Understandably, participation in all study-related activities tended to fall off after delivery.

To give participants a flavor for the data, Goodday showed the participants a set of charts showing individual participants’ swelling and the peripheral fluid level as detected by the Bodyport scale. Friend reminded them that when results of the study were published, they would all be invited to join as co-authors.

“This is not our study,” he said. “This is your study.”

This post originally appeared on StatNews.