21 Mar 2025

Cognitive slippage is a normal part of aging, but what distinguishes this inevitable slowdown from the type of decline that foreshadows neurodegenerative disease? And might remote, digital measurements detect these changes before symptoms are noticeable? Such is the dream of scientists studying cognition using the mobile devices that by now most people in the world use day to day. In the largest project of its kind, scientists at Biogen, Apple, and academic institutions joined forces to put such digital measures to the test. As reported March 4 in Nature Medicine, the Intuition Brain Health study deployed a custom research app to survey health and demographics, conduct cognitive assessments, and passively monitor all manner of phone usage and physical activity among more than 23,000 enrollees over two years. The volunteers used their own iPhones, Apple watches, and web-based cognitive tests.

The study halted just shy of completion when Biogen pulled funding. Even so, it managed to amass a mountain of data. The sheer size of this undertaking is the most exciting aspect of this first publication to come from the project, commented Jason Hassenstab of Washington University in St. Louis. Still, he noted that the findings he was hoping to see were not included in the study, namely data linking mobile assessments and passive monitoring to cognition.

Kathryn Papp of Brigham and Women’s Hospital in Boston echoed this sentiment. “The main takeaway from this work is that large-scale cognitive monitoring is truly feasible,” she wrote. “No other study has been able to collect such detailed cognitive data remotely on such a large sample with such geographic, ethnoracial, and educational representation.”

Fair warning for those hoping to see a plethora of new connections between digital measurements and faltering cognition: the paper contains none of that. Instead, it describes the people in the cohort and how well they adhered to monitoring, offering but a taste of how well digital assessments track with cognition. Specifically, the authors report that a combination of baseline demographics, subjective cognitive complaints, and objective cognitive tests can identify people with mild cognitive impairment. By far, subjective measures were the strongest indicators of MCI. “This paper presents the study design and establishes that remote cognitive tests correlate with clinical ones,” said first and senior author Paul Butler. Now a neurologist at Brigham and Women’s, Butler worked at Biogen during the study. “There’s more to come,” Butler told Alzforum. Now that the groundwork has been laid, future studies will sift through petabytes of data for signals linking various active and passive measurements to cognitive decline.

Starting in 2021, Intuition deployed a multipronged approach to recruit participants, including targeted email campaigns, word-of-mouth referrals, advertisement on Apple’s AppStore, social media posts, traditional referrals from clinics, and at community brain health events. The process started when potential participants downloaded the Intuition study app on their iPhone. Eighteen months later, 23,004 people between the ages of 21 and 86 had enrolled. To be eligible, recruits needed to be current iPhone users, be willing to wear an Apple watch for at least four hours every day, and provide sufficient health and demographic data. The last step was completion of the Cambridge Neuropsychological Test Automated Battery (CANTAB)—a 30-minute, web-based cognitive assessment. The score didn’t matter. All volunteers then received an Apple watch, which they were allowed to keep if they stayed in the study. They also accrued points for completing assessments over the two-year study, which could be redeemed for a total of up to $280.

Trust Your Intuition? Starting with 126,640 people who downloaded the Intuition study app, recruits went through consent, eligibility screening, study orientation, and an initial CANTAB assessment, resulting in 23,004 enrollees. [Courtesy of Butler et al., Nature Medicine, 2025.]

Enrollees hailed from all 50 U.S. states, and 64 percent were women. Non-Hispanic whites made up 75 percent, while 10 percent identified as Asian, 8 percent as black/African American, and 7.7 percent as Hispanic or Latino. Nearly a quarter of the cohort earned less than $50,000 per year, and a third had less education than a bachelor’s degree. Dyslipidemia, namely high plasma cholesterol and/or triglycerides, reportedly affected 41 percent of enrollees, while 38 percent had hypertension, 30 percent were obese, and 28 percent reported a family history of dementia.

The scientists separated these participants into seven groups based on age and cognitive status. People in “late adulthood,” aged 60-86, were divided into five groups: two control groups with either high or low risk of dementia based on family history; one comprising people with subjective cognitive concerns, who scored four or higher on the 14-item cognitive function instrument (CFI-14); another who self-reported as having been diagnosed with MCI by a clinician; and a group with clinically confirmed diagnosis of MCI. The latter group was referred from clinics. Participants in early/midlife (EM), i.e., between the ages of 21-59, were split into a control group and a self-reported MCI group.

Telehealth visits were offered to all participants in late adulthood, as well as to a subset of those who were clinically confirmed. Around half of those invited took up the offer. During these visits, participants could elaborate on their MCI diagnosis, and take the tele-Montreal Cognitive Assessment. Nearly all of the clinically diagnosed MCI participants also had MCI based on the teleMoCA. For the self-reported groups, about 73 percent also had tele-MoCA-based MCI.

The MCI-EM group, average age 35, raised some eyebrows among dementia researchers, including Hassenstab, who view MCI as an age-related indicator of neurodegenerative disease. Why do these young people think they have MCI? Overall, the reasons were varied, but unlikely to reflect neurodegenerative disease, Butler said. Intuition telehealth visits with some of these participants revealed that about a third had received an MCI diagnosis while being assessed or treated for a psychiatric or medical condition, Butler said. As such, they were more likely than the late adulthood group to have been diagnosed with MCI by a primary care doctor. Compared to controls in the EM age group, those with MCI had less education, lower incomes, and more were black or Hispanic.

In addition to the baseline CANTAB test, volunteers were subject to regular active and passive monitoring over the next two years. Interactive tests included a monthly web-based CANTAB assessment, as well as quarterly two-week stints of high-frequency “Cam-Cog bursts,” in which participants completed at least seven two-minute tests on their iPhones. They also used the Intuition iPhone app to take surveys of subjective cognitive complaint twice a year. Meanwhile, running in the background on their iPhones and Apple watches, a slew of passive measurements were taken, including physical activity, heart rate, sleep patterns, device and application usage, message and phone use, tapping and typing speed, voice and speech characteristics, and so on.

Stick With It. Adherence to daily Apple watch wear (left), monthly CANTAB (middle), and quarterly high frequency tests (right) varied by group (EM, early/mid-life; L, late adulthood; SCC, subjective cognitive concerns). [Courtesy of Butler et al., Nature Medicine, 2025.]

Compliance varied substantially by group (image above). On the monthly CANTAB, all of the older participants stuck it out more than either of the early/mid-life groups. Within each age category, controls tested more than did those with cognitive complaints or MCI. Late-adulthood control groups did the best, with at least 80 percent dutifully taking a monthly CANTAB over a year. By comparison, fewer than half of the ficklest group—early/mid-lifers with MCI—kept up the tests. Similar trends emerged on the high-frequency bursts, or in the number of participants who wore their Apple watches for at least four hours per day.

The findings in the paper were mostly limited to those from active cognitive tests, including the CANTAB, and subjective self-surveys of cognition, namely the 12-item everyday cognition scale (E-Cog-12 and CFI-14). The scientists found that performance on these tests differed between groups. Among older participants, people with MCI—whether self-reported or clinically confirmed—scored lower on cognitive tests than age-matched controls or people with subjective cognitive concerns. Interestingly, the same happened in early/mid-life, with the MCI group consistently scoring at or near the bottom of the cohort on subjective and objective measures.

For an initial proof-of-concept experiment with this data, the scientists used machine learning to see if baseline cognitive and demographic data could identify MCI. They limited their analysis to those aged 50-86, and included controls, people with clinically confirmed MCI, and people with self-reported MCI whose diagnosis was later confirmed via telehealth assessment or medical records. These MCI groups were pooled, as were controls and people with only subjective cognitive concerns. A model based on a combination of demographics, baseline CANTAB scores, and the subjective E-Cog-12 and CFI-14 tests distinguished those with MCI from the control/SCC group with 85 percent accuracy. Notably, the CANTAB picked out MCI with 66 percent accuracy, while the subjective surveys did so 79 percent of the time.

Butler wasn’t surprised that subjective tests outperformed the CANTAB. For example, people who score within the mid-range on CANTAB might appear to have normal cognition, when in fact they are scoring much lower than they would have done in their prime. “Subjective measures do a good job at pulling out those mid-range individuals,” Butler said.

Hassenstab noted circular logic with this result. “If you use self-report to make the groups, then of course the subjective measure is going to be best predictor,” he said.

Butler said that now that they have established that remote, active cognitive assessments can classify MCI, the next step will be to correlate passive measures with these active ones. He will also compare the performance of repeated, high-frequency cognitive assessments to infrequent but longer tests like the CANTAB.

David Berron of the German Center for Neurodegenerative Diseases in Magdeburg said that the data collected by this massive study holds the potential to test whether passive monitoring works as well as active assessments do. This would address a major unanswered question in the field, which has yet to convincingly show that passive monitoring can reliably pick up cognitive changes, he said.—Jessica Shugart

Comments

1. • Kate Papp
     Harvard Medical School
   • Posted: 21 Mar 2025

   I think the main takeaway from this work is that large-scale cognitive monitoring is truly feasible. No other study has been able to collect such detailed cognitive data, remotely, on such a large sample with such geographic, ethnoracial, and educational representation. This is an impressive project!

   Their ability to maintain high levels of participation over time was formidable, and indicates that incentivizing participants with smartwatch ownership was a winning strategy. This specific approach of incentivization may not be generalizable to other studies or clinical practice, but it does inform us about motivating factors that could be leveraged in the future.

   While this study was remarkable in showing feasibility, I do think it is premature to conclude that MCI can be diagnosed fully remotely with digital tests and surveys—but this is clearly a step in that direction. It was striking that subjective concerns by the participants were the best predictors of MCI diagnosis, rather than more commonly observed measures of objective cognition and age, which raises questions about the robustness of the MCI designation across the different MCI groups. That said, only a small portion of what appears to be an enormous and rich dataset has been published, and we hope that future learnings, if not raw data, from this study will be shared in the name of scientific advancement.  

   View all comments by Kate Papp

Make a Comment

To make a comment you must login or register.

References

No Available References

Further Reading

No Available Further Reading