Early Detection of Dementia with Smart Devices

For 12 weeks in 2018, more than 100 people with varying states of cognitive decline—or none at all—used an iPhone, an Apple Watch, an iPad pro with a smart keyboard, and a Beddit sleep monitoring device. Each of these devices contains various sensors such as gyroscopes, pedometers, accelerometers, heart rate monitors, and sleep sensors. The iPad also administered language and motor control tests on a biweekly basis. Throughout the study period, participants talked, slept, worked, cleaned, and socialized as their digital biomarker data streams flowed to a cloud-based server viewed by researchers at the study headquarters at Evidation in San Mateo.

The project, which ultimately aims to improve diagnosis of cognitive decline and the diseases it often accompanies, addresses a pressing need. More than 5.8 million Americans live with dementia, costing the healthcare system and patients’ families about $305 billion per year. One in 10 Americans over the age of 65 now suffer from Alzheimer’s disease, deaths from which have increased 146 percent since 2000—it’s now the sixth leading cause of mortality in the US. There are no drug treatments on the market to improve the cognitive function of people who already have dementia.

We know nothing, or very, very little, about early-stage disease.

 —Carol Rout­ledge, Alzheimer’s Research UK

But diagnosing dementia, especially early on, can be difficult. Typically, doctors assess patients in their offices with tests that only effectively diagnose people who have already noticeably started losing cognitive ability. These tests must be administered by a health professional, and provide just a snapshot of the patient’s experience.

Thus, current diagnostic tools usually cannot identify people in the early stages of disease, nor determine whether someone will develop dementia years down the road. By the time patients receive a diagnosis, neurons have died and brain anatomy has changed. “The periods beforehand have been unknown territory,” says Arlene Astell, a dementia and technology researcher at the University of Reading in the UK and the University of Toronto. “We haven’t been able to collect those sorts of data in the past.” Carol Routledge, director of research at the nonprofit Alzheimer’s Research UK, agrees: “We know nothing, or very, very little, about early-stage disease.” 

Early diagnosis could not only aid intervention for people at high risk of developing dementia, but provide better opportunities to design new therapies, says Routledge. “If we were able to diagnose early, then that might start telling us what goes wrong in the very early stages,” she says, “which in turn might help us get insight into more appropriate targets that we could eventually develop treatments for.” Pharmaceutical companies might be able to work with populations of at-risk individuals to re-examine failed drugs that didn’t work in people with later stages of the disease. “We likely might even already have some of those treatments, it’s just that we are not stratifying [the patient population] at the precision level that allows them to work as effectively as they could,” says Rhoda Au, a neuropsychologist at Boston University School of Medicine.

That’s why researchers at Evidation and its collaborators, who presented the findings of their smart-device study in a conference paper last year, are so interested in harnessing the power of wearable or mobile technology. By passively gathering data from people not yet showing obvious clinical symptoms of cognitive decline, these devices could be used to create a digital phenotype that helps clinicians diagnose dementia early, before neuronal death.

Evidation isn’t the only company interested in harnessing these 21st-century communications technologies to try to address neurodegeneration. Pharma companies are incorporating digital technologies in their own research, trying to develop in-house solutions, says Au, and many venture firms and other funders are now backing such efforts. Last April, the Alzheimer’s Drug Discovery Foundation put out a call for proposals on digital biomarkers of the disease and related dementias, and the US National Institute on Aging is also funding research in this area. Earlier this year, Alzheimer’s Research UK launched EDoN (Early Detection of Neurodegenerative diseases), a global initiative that will develop “digital fingerprints” of conditions such as Alzheimer’s to “revolutionise the early detection of neurodegenerative diseases,” according to a press release.

With all the investment in digital biomarkers of early cognitive decline, says Au, “I think . . .  collectively, we are going to start to have these solutions emerge.” 

Early days for digitally monitoring cognitive decline

Fortunately for the research community, members of the public are keen to monitor their health, as evidenced by the more than 300,000 health-related apps and 340 wearable devices already available as of 2017, according to a report by health-focused data science company IQVIA. Many apps purport to detect cognitive decline using data on a user’s movement, cognition, and other factors that may begin to slide years before that person would fail a clinical test for dementia.

There’s science to support the idea that subtle changes can precede dementia. Studies have found, for example, that around 12 years before a clinical diagnosis
of mild cognitive impairment, a person’s gait begins to slow dramatically. Other research has shown that, compared with healthy controls, patients suffering from mild cognitive impairment have a higher blink rate and lower heart rate variability. Circadian rhythm disruptions also seem to occur in the very early stages of cognitive decline. But by themselves, these small changes are unreliable markers of neurodegenerative disease. Few of the apps and devices on the market have been validated by rigorous research; none are FDA-approved.

KEEPING TABS: Massachusetts-based digital health company Linus Health has developed tools for smartphones and tablets that aim to measure cognitive function.

COURTESY OF LINUS HEALTH

The study conducted by Evidation and its collaborators aimed to provide real predictive ability by aggregating data from the sensors in multiple devices, as well as basic device usage metrics—how often phones were locked and unlocked, and numbers of calls and texts—to evaluate cognitive status. Researchers looked at gross motor function using accelerometers, pedometers, and gyroscopes; heart rate using the heart rate monitor in a smartwatch; circadian rhythms using Beddit sleep sensors; various behavioral, social, and cognitive characteristics measured by app usage, phone use behavior, and text message and phone call metadata; fine motor control using an iPad assessment app for typing and dragging tasks; and language skills using the iPad app.

Through these devices, the researchers monitored 113 people between the ages of 60 and 75 years old—31 people with cognitive impairment (as determined by standard criteria) and 82 without. Once generated, participants’ data arrived encrypted at Evidation’s Study Platform, where they were time-stamped, stored, and analyzed. The team found some important differences between the groups of subjects. For example, participants with cognitive decline typed more slowly and had more pauses during typing, perhaps because of fine motor problems or language difficulties or both, the researchers reported last summer. Those with cognitive impairment also walked in a less regular pattern, and their first steps came later in the day. They sent fewer text messages, had a greater reliance on helper apps such as the Clock app, which tells the time and sets alarms, and were more likely to use Siri’s app suggestions.

The researchers used machine learning on the dataset to develop a model to distinguish which people had cognitive impairment and which were healthy, based solely on the pattern of digital data received from the participants’ devices and their responses to iPad tasks. The resulting model was able to distinguish between healthy individuals in that dataset, those who had mild cognitive impairment, and those with mild Alzheimer’s disease, with an accuracy similar to that of computerized cognitive tests administered in clinical settings. “[Eli] Lilly [working with Evidation] has done a lot of good work there,” says Graham Jones, director of innovation at Novartis Technical Research and Development, who has researched digital biomarkers and wearable devices for Alzheimer’s disease but wasn’t involved in the study.

The data collected is very sensitive, and privacy must be the first consideration.

—Nikki Marinsek, Evidation Health

Another group, venture capital–funded Linus Health, is working with Au and other neurocognitive experts to develop a brain health monitoring platform that is independent of any particular tech brand or company. The platform will analyze several aspects of a person’s behavior to glean insights into their brain, and merge it with medical health records, says David Bates, one of the company’s cofounders. Linus has already built its monitoring tool: a smartphone app that reminds users to do certain tasks and measures reaction time, voice and speech, gait, and other potential biomarkers. One task might be to assess changes in gait under a cognitive load: a user walks normally with a smartphone in her pocket; then, she is asked to count backwards by threes from 300 while still walking. Analysis of data generated by the device’s built-in sensors can assess heel strike, toe lift, balance, and walking speed, and differences seen during counting could reveal a declining mind.

Researchers at Linus Health are also developing an in-clinic assessment on a tablet. For example, in one task a person might be asked to describe what is going on in a picture while their voice is recorded and speech transcribed, “so it can be analyzed for neuromuscular and cognitive health,” says Bates. Other tasks could involve subjects playing games while the device tracks accuracy, response times, and eye movements. “All these different things combined give a holistic picture of what’s going on in someone’s brain,” he says.

Another initiative called the GameChanger project, led by digital phenotyping researcher Chris Hinds at the University of Oxford’s Big Data Institute in the UK, uses Mezurio, a phone app that administers game-like tasks to measure executive function, paired-association learning, and speech production, for 5 minutes per day for 30 days. Users also self-report mood, sleep, and any word-finding difficulties or disorientation. In this first phase of the project, anyone who hasn’t been diagnosed with cognitive impairment can download the app and take part. After a year, they are invited to take part in GameChanger again. Users’ responses provide information for researchers about brain function in people without dementia, and how the brain changes over time, according to the website of the UK-based Alzheimer’s Society, which partially funds the project. Since 2018, more than 16,500 people across the UK have used the app and contributed data. Hinds and colleagues note in a 2019 preprint about the app that they plan to study people with diagnosed cognitive decline, too.

Concerns about privacy and access to technology

For the digital detection of cognitive decline to work, researchers need huge amounts of personal data to be transmitted, stored, and analyzed, making the privacy of participating users an obvious concern. There are some regulations protecting consumer data, such as the General Data Protection Regulation in the European Union, which requires companies storing data over long periods to implement the “right of erasure,” allowing participants to delete their personal data. Several US states, including California, Massachusetts, and New York, have data privacy laws, and the US Health Insurance Portability and Accountability Act of 1996 (HIPAA) led to federal standards to protect health information in electronic form.

For studies such as the one by Evidation and collaborators, this could mean setting up the personal devices in such a way as to limit the data sent to the computing facility. Another solution could be setting up local data storage instead of a centralized computing center. “The ethical implications of any further development run deep,” writes Nikki Marinsek, a data scientist at Evidation Health, in an email to The Scientist. “The data collected is very sensitive, and privacy must be the first consideration when dealing with this kind of data.”

Some people with cognitive impairment may also have difficulty understanding the consent they need to give to share their personal data with a company. Many studies enroll a partner or family member for each person with cognitive decline, to help them use the technology appropriately and ensure they’re well cared for.

Another concern is how to provide access to the technologies needed to do the monitoring: personal devices such as iPhones and Apple Watches aren’t cheap and may be difficult to use for some people, even with assistance. “The question is, can you deploy them at scale, economically?” says Novartis’s Jones. He recommends
researchers use something economical such as a smart speaker—some of which cost as little as $30—to collect speech and other data on participants, rather than a several-hundred-dollar iPhone or Apple Watch. Researchers at Dartmouth-Hitchcock Medical Center in New Hampshire and the University of Massachusetts, Boston, recently won a grant worth more than $1 million from the National Institutes of Health to study whether voice assistants such as Alexa (used by Amazon Echo) or Google Assistant (Google Home) could be used to detect early cognitive impairment.

See “Computers Extract Disease Clues from Speech”

Overall, researchers are enthusiastic about the potential for digital technology to improve early detection of dementia. Au estimates that it’ll be less than five years before there’s a well-validated digital phenotype that will be able to identify people who are at a higher risk of developing dementia over the following decade or so. “We have technologies that allow us to now track behaviors in much more continuous, granular ways, so we can sort people out into various subgroups with much greater precision,” Au says. “On top of that, we have more advanced analytic capabilities that are allowing us to look at multi-dimensional sources of information. These are all advances that are happening simultaneously . . . we are getting closer, faster. I’m quite optimistic.” 

Clarification (May 6): This story has been updated to note that the Evidation study findings were published in a conference paper rather than a peer-reviewed journal article.