Replicating existing paper-and-pencil tests on tablets or smartphones is one thing (Part 13 of this series), but what about taking technology itself to the next level? What are the prospects for continuously collecting information on people’s daily behavior, and using that wealth of data to discover entirely new biomarkers for dementia? Jeffrey Kaye, Oregon Health and Science University, Portland, and his group have pursued this approach for nearly a decade (Dec 2012 news), and at the 2019 AAIC conference in Los Angeles, Kaye gave a progress update. In his center’s Collaborative Aging Research Using Technology (CART) Initiative, Kaye’s group has been using remote-monitoring technology to collect data on people in their homes. The hope is that it will detect cognitive decline and track the effects of interventions. In homes and cars, nonintrusive remote-sensing devices track a participant’s movement, computer use, medication adherence, driving habits, and more, with the data continuously being fed into a central database.

So far, Kaye said the project is monitoring 274 people in four communities, including low-income participants in Portland, Oregon, veterans in rural Oregon, African Americans in Chicago, and Hispanic participants in Miami. Kaye noted that all of these can be challenging groups to reach and recruit into traditional clinic-based studies.

À LA CART: The smart home project pairs multiple passive monitoring technologies with conventional assessments to discover digital biomarkers for dementia. [Courtesy of Jeffrey Kaye.]

One such study addressed whether changes in sleep could flag cognitive decline. Most sleep and dementia studies take place in sleep labs, over a night or two, with people Kaye called “advantaged adults.” His group tracked total sleep time per day for up to five months in 126 CART participants living on their own in Oregon, 84 of whom were cognitively healthy and 42 of whom had MCI. They were asked to wear an activity-tracking watch round the clock. They did not seem to mind the watch overly much, as compliance with this request came in at 86.5 percent. People with MCI wore the watch less, but still on more than 80 percent of study days, Kaye said.

All groups posted about seven hours of sleep per night. Unadjusted sleep duration was the same between the groups, but when the investigators controlled for age, sex, and education, shorter sleep was significantly associated with MCI status.

Importantly, monitoring sleep for only two weeks did not predict MCI, nor did self-reported sleep time. “By monitoring for longer times, we can start to see differences,” Kaye said. “Normally, monitoring might be two weeks. Our volunteers did months with high compliance.” Kaye said CART is now analyzing sleep in its other participating groups. In the coming months, CART will move to more passive methods of sleep assessment, such as embedded infrared sensors or bed mats, Kaye said.  

Other warnings of early cognitive impairment might come from tracking how people take their medications. Nora Mattek of OHSU collected data from 64 adults who used electronic pillboxes that reported by way of livestreaming each time a compartment was opened. Compared with cognitively intact older adults, people who were developing MCI started to take their pills later in the day, and varied their pill-taking time more. This in-person variability grew during the time up to and after the diagnosis of MCI. The authors suggest medication-taking behavior could be an early and predictive digital biomarker of emerging MCI.

Yet another harbinger of mild cognitive impairment can be when people start to move more slowly around the house, Kaye’s group had previously found (Dodge et al., 2012). New data from Antoine Piau of the University of Toulouse, France, now suggests that slowed walking also predicts impending falls. Working in Kaye’s lab, Piau analyzed data from 125 adults who were monitored continuously at home with a series of ceiling-mounted activity monitors. He discovered that people’s walking speed slowed significantly in the three months prior to a fall. Day-to-day variability in speed was also lower in the month and week prior to a fall (Piau et al., 2019). Such sensor-based monitoring could open up new opportunities for targeted and timely interventions to prevent falls, Piau said.

By this point in his longstanding research program, Kaye can tie in-home monitoring to neuropathology data. In LA, he showed a comparison of autopsy results and lifetime data from 41 participants who had lived with the activity-monitoring platform in their homes for more than a decade. Their average age at death was 92 years. Half of them passed away while still living in their sensor-equipped homes; for the others, the average amount of time between when they left their monitored home and when they passed away was one year. At autopsy, amyloid plaque load was scored as none, sparse, or moderate. An increasing plaque score was accompanied by stepwise decreases in markers of cognition (computer use), mobility (walking speed), and increased sleep time. Socialization waned at moderate levels of plaque burden, as measured by time spent out of the house. The same relationships held for Braak scores of tau pathology.

This is the first data showing a correlation between digital biomarkers and neuropathology in older people, Kaye said. Going forward, he said CART will continue to collect data from its diverse populations, to further define connections between the monitored behaviors and brain changes.—Pat McCaffrey

News Citations

  1. Technology Brings Dementia Detection to the Home
  2. Tech Revolution: Monitoring and the Power of Real-Time Data

Paper Citations

  1. .
    In-home walking speeds and variability trajectories associated with mild cognitive impairment.
    Neurology. 2012 Jun 12;78(24):1946-52.
    PubMed.
  2. .
    When will my patient fall? Sensor-based in-home walking speed identifies future falls in older adults.
    J Gerontol A Biol Sci Med Sci. 2019 May 16;
    PubMed.