Research Signal – Source date 2026-04-07
The available ClinicalTrials.gov description of Predicting Pre-dementia proposes testing a focused question: among adults aged 55 and older who carry at least one ApoE4 allele and are cognitively normal or only very mildly impaired, can a combined risk model anticipate a transition to early Alzheimer’s-related cognitive impairment within 24 months?
The proposed endpoint is more specific than a genetic risk estimate or a positive blood test. Participants would need to meet criteria for mild cognitive impairment, or MCI, meaning measurable cognitive difficulty that is not the same as dementia, while also having plasma pTau217 above a validated threshold associated with Alzheimer’s pathology. The study therefore asks whether several kinds of information can work together to predict a clinically meaningful transition, rather than whether any single measurement can diagnose Alzheimer’s disease.
The prediction is built from different kinds of information
The planned Progression and Risk model combines inherited susceptibility, molecular measurements, cognitive assessments and observations collected between clinic visits. Each category contributes a different kind of information.
Inherited risk is summarized partly through a polygenic risk score, or PRS. A PRS combines the effects of many genetic variants associated with Alzheimer’s risk into one estimate. In this protocol, the APOE region is excluded from the PRS so that it is not counted twice; APOE genotype, including ApoE4 status, is entered separately. The score is standardized against an ancestry-matched reference group and divided into risk categories for recruitment and analysis.
Blood testing adds measurements that may change over time. The protocol describes repeated plasma pTau217 testing alongside other possible Alzheimer’s-related markers and proteomic or methylation panels. Cognitive assessments provide the functional part of the picture, including standardized testing, the Clinical Dementia Rating and digital cognitive tasks. Those assessments are intended to identify changes consistent with MCI rather than treating a biological marker as a complete clinical outcome.
Digital monitoring supplies another layer. Wearables and smartphone tools may collect sleep, activity, heart-rate metrics, mobility or location patterns, and features from brief speech-linked or cognitive tasks. These observations could reveal changes between scheduled visits, but the study is testing whether they improve prediction beyond information that is easier to collect in ordinary clinical settings. The measurements describe patterns; they do not make a wearable device an independent Alzheimer’s diagnostic tool.
The model must earn the extra data
The study is designed around a prespecified model rather than an algorithm that keeps changing to fit the new cohort. Candidate models are developed and calibrated using existing data, after which one P&R model is intended to be selected and locked before it is evaluated prospectively. The locked rules can then produce risk and staging estimates as new participant information becomes available.
The full model is to be compared with reduced versions, such as approaches based on PRS alone, biomarkers alone, standard clinical risk factors or models without continuous digital monitoring. This comparison matters because a larger collection of measurements is not automatically a better prediction system. It is useful only if the additional information improves the separation of people who do and do not reach the endpoint while remaining practical and acceptable to collect.
The planned analyses use several ways to judge that question. Discrimination, often summarized with AUC or ROC analysis, asks how well a model separates participants who convert from those who do not during the study period. Calibration asks whether predicted risks agree with the frequencies actually observed. Time-to-conversion analysis asks whether the model helps identify when a transition occurs, not merely whether it occurs within a broad window. The registry also describes comparisons of different monitoring and blood-sampling strategies, including their cost and the number of conversion events detected.
Because the cohort is modest and the analyses are described as exploratory and hypothesis-generating, these are planned tests of added predictive value. They are not reported performance results.
Two years of measurements, not a treatment
The protocol describes an in-person baseline visit followed by visits around months 6, 12, 18 and 24, with an allowance of roughly two months around each follow-up. Participants would review their health and medications, repeat cognitive and functional assessments, provide blood samples and discuss adherence to digital monitoring. The risk and staging estimates could be updated as new cognitive and biomarker data arrive.
Between visits, participants would use an Oura Ring and smartphone-based tools for continuous or frequent monitoring. If the participant or the monitoring system suggests a concerning change, the team may arrange a triggered visit or remote assessment. That could include focused cognitive testing and repeat blood biomarkers. If the criteria for biomarker-positive MCI are met, the described pathway includes referral to a neurologist and possible discharge from intensive research follow-up.
No experimental treatment is given. The study is observational, so its purpose is to gather measurements and test a prediction model. EEG, retinal hyperspectral imaging, MRI, amyloid PET or sharing of clinically indicated cerebrospinal-fluid samples are described as optional, sub-cohort or clinically indicated components rather than requirements for everyone.
The information is also unusually sensitive. The registry describes consent discussions covering genetic and biomarker findings, data sharing and the voluntary nature of participation. Digital consent is intended to explain the collection of location, movement and speech-related features, including that speech features are not the same as storing the spoken content itself. The planned safeguards include unique study identifiers, separation of direct identifiers from research data, encrypted transmission and role-based access. Participants may be able to pause or decline particular digital components where the protocol allows. Anxiety about genetic risk, biomarker status or subtle cognitive change is another stated consideration.
What this cohort can and cannot tell us
The proposed population is deliberately selected. Participants must already have documented ApoE4 status, existing whole-genome sequencing or suitable high-density genotyping data, the ability to consent and a willingness to use digital devices. The initial plan describes roughly 90 participants, with expansion toward 100 as the study matures, and representation across PRS-based risk categories. Those figures are design plans, not evidence of enrollment.
This selection helps the investigators test a defined higher-risk group, but it narrows what the results could mean. Findings from ApoE4-positive adults who already possess genomic data and accept frequent monitoring would not automatically describe people without ApoE4, people in a general screening population or groups not represented in the cohort. The digital requirements may also shape who can participate and how consistently measurements are collected.
The main observation period is two years. The registry also describes conversion estimates spanning two to five years and modeled projections up to ten years. Estimates beyond the two-year follow-up would depend on extrapolation; the ten-year projections would use the observed two-year data and external cohort information, not observed ten-year outcomes.
The supplied study description is a plan for a prospective cohort, not a report of completed follow-up. It does not establish that recruitment has begun, that the model has achieved a particular accuracy or calibration, or that wearable monitoring improves clinical decisions. Even a promising prediction result would leave a separate question: whether clinicians and patients can act on it in a way that improves care.
What better prediction could mean for earlier care
A useful model could eventually help clinicians identify which genetically susceptible older adults need closer cognitive follow-up or more timely specialist assessment. Information gathered between scheduled visits might make it easier to notice a concerning pattern early enough to review it rather than waiting for the next routine appointment.
That possibility depends on several conditions. The model would need to perform prospectively in an appropriate population, the burden and privacy costs of repeated monitoring would need to be acceptable, and acting on the predictions would need to improve care rather than simply increase testing or anxiety. The registered study is designed to examine the predictive and practical foundations of that possibility. It does not establish prevention, treatment success or clinical benefit.
Key Sources
- Primary source: Predicting Pre-dementia
Government or Regulator · Clinical Trial · 2026-04-07
Keep exploring