Key Takeaways
- A study of 123 adults aged 65 and older tested whether repeated smartphone assessments could capture cognitive differences associated with early Alzheimer’s disease.
- Among people reporting subjective cognitive decline, some episodic-memory and working-memory measures differed between predefined amyloid-positive and amyloid-negative groups.
- Amyloid status was established with PET imaging, not with the smartphone assessments.
- The study had a 97.6% completion rate and 96% average adherence during 28 days of remote testing.
- Some digital cognitive measures were associated with MRI-derived patterns of regional brain atrophy.
- The study does not show that a smartphone can diagnose Alzheimer’s disease or determine an individual’s amyloid status.
- F. Hoffmann-La Roche funded the research. Many authors were Roche employees, and several were Roche shareholders.
Introduction
Researchers are studying whether repeated smartphone-based cognitive assessments can capture subtle memory differences associated with early Alzheimer’s-related brain changes.
Repeated memory assessments performed on a smartphone may capture subtle cognitive differences associated with Alzheimer’s-related brain changes, according to research published September 5, 2026, in npj Digital Medicine.
The study evaluated a smartphone-based system called the Alzheimer’s Disease Digital Assessment Suite, or AD-DAS, in 123 adults aged 65 to 89.
The most interesting finding involved participants who reported subjective cognitive decline. Certain episodic-memory and working-memory measures differed between predefined groups who were amyloid-positive or amyloid-negative on PET imaging (Taylor et al., 2026).
That comparison matters because Alzheimer’s-related biological changes can precede clinically apparent dementia. The National Institute on Aging notes that changes such as amyloid plaques and tau tangles may begin a decade or more before memory and thinking problems appear. Importantly, not everyone with these brain changes develops dementia (National Institute on Aging, n.d.).
The distinction is essential.
Amyloid positivity does not by itself mean that a person has dementia, and this study did not demonstrate that smartphone testing can independently diagnose Alzheimer’s disease. Participants had already been clinically and biologically characterized before researchers compared their digital cognitive performance.
What the Study Examined
Researchers enrolled 123 participants at three sites in the United States and two in Spain.
The study included 32 amyloid-negative healthy controls, 31 amyloid-negative participants with subjective cognitive decline, 30 amyloid-positive participants with subjective cognitive decline, and 30 amyloid-positive participants with early Alzheimer’s disease.
Participants ranged in age from 65 to 89, with a median age of 71.
They also represented a highly selected population. Participants had previous experience using smartphones or tablets, and 94.3% of the study sample was White.
That lack of racial and ethnic diversity is important because it limits confidence that the same results would be observed in more representative populations.
Participants did not use their personal smartphones. Each received a preconfigured study device.
The AD-DAS system contained nine active tasks designed to assess areas including episodic learning and memory, working memory, processing speed and attention. During the 28-day remote phase, each task was scheduled on at least nine different days.
Researchers examined whether participants could consistently complete the assessments, how reliable the measurements were when repeated, whether performance differed among the four predefined study groups, and whether digital cognitive measures were associated with established clinical assessments and structural brain characteristics.
Amyloid status was established using PET imaging.
The smartphone assessments did not determine whether participants were amyloid-positive or amyloid-negative.
Structural brain measurements were obtained using MRI.
What Researchers Found
Participation in the remote assessment program was high.
Overall, 97.6% of participants completed the study. One healthy participant withdrew because of circumstances related to COVID-19, while two participants in the early Alzheimer’s group voluntarily withdrew.
Average adherence to scheduled digital assessments was 96%.
Participants generally rated the experience favorably. More than 85% described their overall experience as “good” or “very good.” Six rated the experience “poor,” and none rated it “very poor” (Taylor et al., 2026).
Reliability was not identical across all digital measurements.
Test-retest intraclass correlation coefficients ranged from 0.53 to 0.91, spanning moderate to excellent reliability depending on the measure.
Several digital measures differentiated participants with early Alzheimer’s disease from other study groups.
The more notable comparison involved people with subjective cognitive decline.
Some episodic-memory and working-memory measures differed between the amyloid-positive and amyloid-negative subjective cognitive decline groups.
Both groups reported subjective cognitive concerns, but their PET scans placed them in different amyloid categories.
The result therefore provides preliminary group-level evidence that repeated smartphone cognitive assessments may capture subtle cognitive differences associated with amyloid status.
It does not show that an individual smartphone score can identify whether a particular person has elevated amyloid.
Smartphone Measures Were Also Associated With Brain Structure
Researchers also compared digital cognitive performance with structural MRI findings.
Some smartphone measures were associated with patterns of atrophy in brain regions relevant to the cognitive functions those tasks were intended to measure.
Together, the cognitive and imaging findings provide converging evidence that some smartphone measurements were associated with predefined clinical-group differences and MRI-derived regional atrophy patterns.
However, those associations do not establish causation or diagnostic accuracy.
They also do not prove that smartphone performance directly measures Alzheimer’s pathology.
Other characteristics of participants, including differences in technology experience, health, cognition and study selection, could potentially influence performance.
How It Compares With Earlier Research
The Taylor study joins a growing body of research exploring whether frequent, remotely administered cognitive assessments can capture subtle cognitive differences or changes that occasional clinic testing may miss.
In 2021, researchers evaluated the Boston Remote Assessment for Neurocognitive Health, or BRANCH, in 234 clinically normal adults aged 50 to 89.
Participants largely completed the web-based assessment using their own devices. Remote performance correlated with standard in-person cognitive testing, and the researchers also examined associations with amyloid and entorhinal tau PET measurements (Papp et al., 2021).
Another study examined 69 cognitively normal older adults whose amyloid PET status was already known.
A remotely administered smartphone episodic-memory task achieved an area under the receiver operating characteristic curve, or AUC, of 0.77 for distinguishing participants by amyloid status (Thompson et al., 2023).
An AUC of 0.77 indicates discrimination better than chance, but it does not mean that the assessment is accurate enough to diagnose amyloid positivity or Alzheimer’s disease in an individual.
More recently, researchers studied a remotely administered smartphone system called the Mobile Toolbox in 100 cognitively unimpaired older adults.
Measures of fluid cognition were associated with conventional cognitive performance and were negatively associated with tau deposition in medial temporal and neocortical brain regions (Jutten et al., 2025).
Research published in June 2026 examined whether frequent smartphone assessments could detect cognitive changes over a longer period.
That study involved 202 participants, including 152 cognitively unimpaired adults and 50 with mild cognitive impairment. Among participants with mild cognitive impairment, the amyloid-positive group showed greater changes than cognitively unimpaired participants in measures of object-memory precision and familiarity-dependent memory. Repeated remote testing detected some group differences over approximately 30 weeks (Polk et al., 2026).
Taken together, these studies support continued investigation of remote cognitive assessment.
They do not establish smartphone testing as a substitute for clinical evaluation or validated Alzheimer’s biomarkers.
What the Results May Mean
One potential advantage of smartphone-based cognitive assessment is the ability to measure performance repeatedly.
Traditional cognitive assessments typically provide a snapshot of performance during a particular clinic visit.
Cognitive performance can vary from day to day for many reasons, including fatigue, distraction, stress and sleep.
Repeated measurements could potentially give researchers a more precise picture of an individual’s typical performance and how it changes over time.
That could be useful in Alzheimer’s research and clinical trials, particularly when researchers want to follow people during relatively early stages of disease.
Remote testing could also reduce some of the logistical burden of repeatedly bringing research participants into clinics.
But there is a large difference between detecting statistically significant differences among carefully characterized research groups and accurately screening individual members of the public.
The Taylor study did not report the sensitivity, specificity, positive predictive value or negative predictive value needed to determine how accurately AD-DAS could classify individuals in routine practice.
Researchers already knew participants’ amyloid status from PET imaging.
The smartphone did not establish it.
The findings therefore should not be interpreted to mean that people can determine whether they have Alzheimer’s disease or elevated brain amyloid by completing memory exercises on a phone.
Limitations
The study’s relatively small sample is one of its most important limitations.
Only 31 participants were in the amyloid-negative subjective cognitive decline group and 30 were in the amyloid-positive subjective cognitive decline group.
A statistically detectable difference between groups of this size requires replication in substantially larger samples.
The study population was also highly selected.
Participants were at least 65 years old, had prior experience using a smartphone or tablet, could undergo MRI, had a study partner and met detailed clinical eligibility requirements.
Those characteristics distinguish them from the general population of older adults.
The sample was also 94.3% White.
That substantially limits the ability to assume that the same digital measures will perform similarly across more racially and ethnically diverse populations.
Participants used preconfigured research smartphones rather than their personal devices.
That design reduces technical variability during a study, but it differs from real-world use across phones with different screen sizes, hardware, operating systems and user settings.
The remote phase lasted only 28 days.
The results therefore provide evidence about short-term feasibility, adherence and repeated measurement. They cannot establish whether AD-DAS can reliably track clinically meaningful disease progression over several years.
Most importantly, this was a validation study rather than a prospective diagnostic-accuracy study in a representative clinical population.
Finding statistically significant differences between predefined groups is not the same as demonstrating that a test can accurately classify an individual patient.
Funding and Conflicts of Interest
F. Hoffmann-La Roche funded the study.
According to the paper’s competing-interest disclosures, 14 authors were current or former Roche employees when the research was conducted, and 12 authors were listed as Roche shareholders.
Two additional investigators reported advisory or clinical-trial relationships with multiple pharmaceutical companies (Taylor et al., 2026).
These disclosures deserve attention because the sponsor was closely involved with the technology being evaluated.
Industry funding does not make research invalid.
However, independent replication would strengthen confidence that the findings can be reproduced outside the sponsor’s research program.
Final Thoughts
The study provides encouraging evidence that repeated smartphone cognitive assessments are feasible in a selected group of older adults and that certain digital memory measures differ between predefined amyloid-positive and amyloid-negative groups among people reporting subjective cognitive decline.
Some of those measurements were also associated with MRI-derived patterns of regional brain atrophy.
Those are potentially useful research findings.
They are not evidence that a smartphone can diagnose Alzheimer’s disease.
The next step is larger, more diverse and preferably independent research. Investigators will need to determine whether repeated digital cognitive measurements can accurately classify individuals, predict clinically meaningful cognitive changes over time, and provide information beyond established clinical assessments and Alzheimer’s biomarkers.
Until those questions are answered, smartphone cognitive testing is better understood as a promising research approach than as a consumer Alzheimer’s test.
References
Jutten, R. J., Burling, J. E., Slade, E., Thompson, J. C., Fu, J. F., Birkenbihl, C., Properzi, M. J., Marshall, G. A., Amariglio, R. E., Papp, K. V., Johnson, K. A., Price, J. C., Sperling, R. A., & Rentz, D. M. (2025). The Mobile Toolbox for remote, self-administered cognitive assessment in older adults: Associations with in-clinic cognitive testing and Alzheimer’s disease biomarkers. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, 17(3), e70160. https://doi.org/10.1002/dad2.70160
National Institute on Aging. (n.d.). What are the signs of Alzheimer’s disease? U.S. Department of Health and Human Services. https://www.nia.nih.gov/health/alzheimers-symptoms-and-diagnosis/what-are-signs-alzheimers-disease
Papp, K. V., Samaroo, A., Chou, H.-C., Buckley, R., Schneider, O. R., Hsieh, S., Soberanes, D., Quiroz, Y., Properzi, M., Schultz, A., García-Magariño, I., Marshall, G. A., Burke, J. G., Kumar, R., Snyder, N., Johnson, K., Rentz, D. M., Sperling, R. A., & Amariglio, R. E. (2021). Unsupervised mobile cognitive testing for use in preclinical Alzheimer’s disease. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, 13(1), e12243. https://doi.org/10.1002/dad2.12243
Polk, S. E., Clark, L. R., Basche, K., Kleineidam, L., Glanz, W., Butryn, M., Perneczky, R., Buerger, K., Fliessbach, K., Laske, C., Spottke, A., Schneider, A., Wiltfang, J., Teipel, S., Bartels, C., Rostamzadeh, A., Janowitz, D., Rauchmann, B.-S., Kilimann, I., . . . Berron, D. (2026). Smartphone-based detection of subtle memory decline in prodromal Alzheimer’s disease. npj Digital Medicine, 9, Article 402. https://doi.org/10.1038/s41746-026-02731-1
Taylor, K. I., Wolfer, A. M., Kurniawan, I. T., Veloso, M., Keita, G., Hagenbuch, N., Shi, B., Orfaniotou, F., Aponte, E. A., Garcia Valdecasas Colell, M., Chatham, C. H., Holiga, Š., Ullmann, R., Abouelkheir, W., Rey-Riek, S., Poon, E., Watson, D., Boada, M., & Perumal, T. M. (2026). From feasibility to neuroanatomic validity of remote cognitive smartphone assessments in early Alzheimer’s disease. npj Digital Medicine. https://doi.org/10.1038/s41746-026-03108-0
Thompson, L. I., Kunicki, Z. J., Emrani, S., Strenger, J., De Vito, A. N., Britton, K. J., Dion, C., Harrington, K. D., Roque, N., Salloway, S., Sliwinski, M. J., Correia, S., & Jones, R. N. (2023). Remote and in-clinic digital cognitive screening tools outperform the MoCA to distinguish cerebral amyloid status among cognitively healthy older adults. Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, 15(4), e12500. https://doi.org/10.1002/dad2.12500
