Predicting Pre-dementia

Overview

The goal of this observational study is to learn how well a multimodal "Progression and Risk" (PR) model can predict and stage early mild cognitive impairment (MCI) due to Alzheimer's disease in cognitively normal or very mildly impaired ApoE4-positive adults aged 55 and older. The main questions it aims to answer are: Can a prespecified proteogenomic PR model accurately predict conversion from cognitively normal (CN) or very mildly impaired status to pTau217-positive MCI Stage I within 24 months in ApoE4-positive adults? Does adding digital monitoring features (e.g., sleep, activity, speech), EMR-lifestyle risk scores, and plasma biomarkers to a polygenic risk score (PRS) meaningfully improve risk stratification and time-to-conversion prediction compared with simpler models (e.g., PRS alone or standard clinical risk factors)? If there is a comparison group: Researchers will compare performance of the full multimodal PR model (integrating PRS, plasma proteomics and other omics, digital monitoring, and EMR-lifestyle data) with simpler or reduced models (for example, PRS-only, biomarker-only, or models without continuous digital monitoring) to see if the full model provides higher discrimination (AUC/ROC), better calibration, and improved time-to-conversion prediction for CN to pTau217-positive MCI transitions. Participants will: Provide prior genomic data (ApoE genotype and whole-genome sequencing or high-density genotyping array data) for calculation of an ancestry- and sex-normalized Alzheimer's disease PRS and assignment to PRS-based risk strata. Attend an in-person baseline visit and follow-up visits at months 6, 12, 18, and 24 (±2 months) for clinical evaluation, neurocognitive testing (including CDR and digital cognitive batteries), and venous or capillary blood collection for plasma pTau217 and other AD biomarkers, proteomic and methylome panels, and routine safety labs when indicated. Use digital devices (e.g., Oura Ring and smartphone-based tools) for continuous or frequent remote monitoring of sleep, activity, heart rate metrics, mobility/location, and speech-linked digital cognitive tasks, with adherence checks at study visits. Undergo optional or sub-cohort procedures as clinically indicated or as resources allow, such as EEG, retinal hyperspectral imaging, MRI, or amyloid PET, and optionally allow clinically indicated lumbar puncture CSF samples and external clinical data to be shared with the study for exploratory biomarker analyses.

This is an observational, first-in-human (FIH) cohort study that follows ApoE4-positive adults over two years to see how well a multimodal "Progression and Risk" (P\&R) model can predict who will develop very early, biomarker-confirmed mild cognitive impairment (MCI) due to Alzheimer's disease. The study combines genetic risk, blood biomarkers, digital testing, and wearable data to stage risk in people who are currently cognitively normal or only very mildly affected. Background and motivation Alzheimer's disease usually develops slowly over many years, beginning with a long "preclinical" phase in which people are cognitively normal but silently accumulate disease-related changes in the brain. During this period, subtle cognitive shifts may occur before symptoms reach the level of MCI or dementia. Blood and spinal fluid markers of phosphorylated tau 217 (pTau217) have emerged as highly accurate indicators of underlying Alzheimer's pathology and predictors of progression from MCI to dementia, with performance (AUC values) in the 0.8-0.9 range in prior work. The Clinical Dementia Rating scale, particularly the Sum of Boxes (CDR-SB), is a well-validated way to stage people from normal cognition through very mild impairment and MCI, where even small changes reflect meaningful clinical transitions. However, tools like the CDR typically require an informant and are not routinely used in primary care for people who appear cognitively normal. At the same time, large genetic studies have shown that late-onset Alzheimer's is highly polygenic: many common genetic variants, along with the APOE ε4 allele, together shape an individual's inherited risk. Polygenic risk scores (PRS) summarize this inherited risk by combining information across thousands of genetic variants, weighted by their association with Alzheimer's in genome-wide association studies. These scores, particularly when considered alongside APOE status, can help identify people with much higher odds of developing Alzheimer's and earlier onset, and they distinguish cases from controls with AUCs often in the 0.70-0.80 range. Recent work has expanded PRS into integrative scores that incorporate genetic signals across neurodegenerative, vascular, and metabolic pathways, sometimes using deep-learning methods to capture non-linear effects. In parallel, large plasma proteomic studies and multi-omic analyses (including DNA methylation) have identified protein and molecular signatures that correlate with Alzheimer's pathology and progression. Together, these advances suggest that combining PRS with blood-based proteomic and other "omics" data, plus clinical and digital assessments, could provide a rich picture of near-term risk and disease stage in people who are still functionally normal. The present study builds on "in silico" (computer-based) modeling work using existing cohorts that already have whole-genome sequencing, plasma proteomics including pTau217, imaging, and longitudinal cognitive and clinical data. Those analyses are being used to develop and calibrate candidate P\&R models that estimate the short-term hazard of conversion from cognitively normal or very mildly impaired status to pTau217-positive MCI. The current protocol is a prospective test of one pre-specified P\&R model in a new clinical cohort. Overall goals and main questions This study's overarching aim is to validate a fixed proteogenomic P\&R model in a real-world group of ApoE4-positive older adults and to assess whether such a model can feasibly be implemented using a combination of existing genomic data, blood tests, and scalable digital monitoring. For a broad audience, the main questions the study asks are: Among ApoE4-positive adults aged 55 and older who are cognitively normal or only very mildly impaired at the start of the study, how well does a pre-specified P\&R model distinguish those who will convert to early, pTau217-positive MCI Stage I within 24 months from those who will not convert during that time? How practical and acceptable is it to carry out this kind of multimodal assessment-combining existing genetic data, repeated blood tests, digital cognitive and speech measures, and continuous wearable monitoring-over a two-year period in this population? How does the performance of the full P\&R model compare with simpler approaches, such as models based only on APOE plus PRS or standard clinical and cognitive measures, when predicting early MCI due to Alzheimer's? Do digital and wearable measures (such as Oura Ring-derived sleep and activity metrics, brief speech features, and app-based cognitive tasks) meaningfully improve prediction beyond what can be achieved with genetic, blood-based, and traditional clinical measures alone? What are the practical costs and logistics of different sampling strategies (for example, venous vs fingerstick blood, more in-person visits vs more remote monitoring), and how do these relate to the number of true conversion events correctly identified? Secondary and exploratory goals include estimating short-term (2-5-year) and modeled longer-term (up to 10-year) conversion probabilities from cognitively normal status to MCI in this enriched, higher-risk population, and examining how an Oura Ring-based exposome risk score relates to proteogenomic risk scores and cognitive-functional measures. Study design in plain language This is a longitudinal observational cohort study: researchers observe and measure participants over time without giving any experimental drugs or devices. The cohort will include up to roughly 90 ApoE4-positive adults at first, with plans to expand toward 100 participants as the study matures; participants are at least 55 years old and either cognitively normal or only very mildly impaired at baseline. The study follow-up lasts 24 months from the baseline visit. Each participant attends in-person clinic visits at baseline (month 0) and at months 6, 12, 18, and 24, with an allowable window of about two months around each follow-up. Between visits, participants are monitored continuously or frequently through digital tools, including an Oura Ring and a smartphone app, to track sleep, physical activity, heart rate metrics, mobility and location, and short digital cognitive or speech tasks. A subset of about ten participants will have especially intensive wearable monitoring. No experimental treatment is given. Instead, the focus is on gathering detailed data-genetic, blood-based, cognitive, and digital-from a well-characterized, higher-risk group in order to apply and test the P\&R model. If participants meet criteria for MCI with positive Alzheimer's biomarkers (for example, if pTau217 becomes elevated), they are referred to a neurologist, and they may be discharged from the study once they are considered "MCI positive." Who can take part? The target population consists of adults who already know they carry at least one copy of the ApoE4 variant and have existing genomic data from earlier testing. These people may come from prior clinical care, previous research studies, or commercial direct-to-consumer genetic services, but they must be willing to share their data with the study. Key inclusion criteria: Age 55 or older at enrollment. At least one ApoE4 allele documented by prior testing (for example, clinical ApoE testing, a genetic panel, or research genotyping). Whole-genome sequencing (WGS) or, if WGS is not available, a high-density genotyping array covering Alzheimer's risk loci, with willingness to provide the raw data files (e.g., VCF, FASTQ) for PRS calculation. Cognitively normal or very mildly impaired at baseline, as judged by digital cognitive testing and standardized scales (global CDR 0 or 0.5, no diagnosis of dementia). For cognitively normal (CN) and subjective cognitive decline (SCD) participants, the P\&R model staging criteria-which combine PRS, biomarker, and cognitive data-will be applied to assign risk levels. Ability to provide informed consent and to comply with study procedures. Willingness to use digital devices (such as a smartphone, wearable sensors, and a sleep device like the Oura Ring) for continuous or frequent monitoring. Important notes for potential participants: The study does not perform ApoE genotyping or whole-genome sequencing as part of the research; these must already have been done before enrollment. People without documented ApoE4 carrier status, or without available genomic data suitable for PRS calculation, are not eligible. Exclusion criteria include: A diagnosis of dementia (of any cause) at baseline. Major neurological conditions that could interfere with cognitive assessment, such as Parkinson's disease, stroke with ongoing deficits, or frequent seizures. Major psychiatric illness (for example, uncontrolled major depression or schizophrenia) that would interfere with participation or interpretation of data. Serious uncontrolled medical illness (such as unstable heart, liver, or kidney disease) likely to limit life expectancy to less than about three years. Use of investigational drugs or disease-modifying Alzheimer's therapies within the previous six months, if these could alter biomarker trajectories. Contraindications to optional procedures like EEG, MRI, or PET scans (for example, certain implanted devices or severe claustrophobia). Inability or unwillingness to use required digital monitoring tools, such as lack of smartphone access or severe sensory impairments that prevent use. Refusal to share existing ApoE/genomic data or electronic medical record (EMR) information as needed for the study. Participants who at screening already meet criteria for MCI with elevated pTau217 (that is, biomarker-positive MCI due to Alzheimer's) will not continue in the study and will instead be referred to a neurologist. The initial plan is to start with a small "vanguard" group of about ten participants who already use Oura Rings, then gradually expand to a total of around 100 participants over time, with roughly equal representation across low, average, and high PRS categories among cognitively normal ApoE4 carriers. What is the P\&R model and PRS? The polygenic risk score (PRS) used in this study is derived from genome-wide data using effect sizes from large, well-validated Alzheimer's GWAS and meta-analyses. To avoid double-counting, the APOE region is excluded from the PRS, and APOE genotype (such as number of ApoE4 alleles) is modeled separately as its own predictor. In simplified terms, PRS is calculated by: Selecting a set of common genetic variants (single nucleotide polymorphisms, or SNPs) that have been associated with Alzheimer's risk in prior GWAS. For each variant, multiplying the number of risk alleles a person carries (0, 1, or 2) by the corresponding effect size (log-odds ratio) from the discovery study. Summing these weighted contributions across all selected SNPs to obtain a single score per person. Standard quality control and ancestry checks are performed on the genetic data, and methods like linkage disequilibrium pruning and p-value thresholds help balance predictive performance against the complexity of the score. Once computed, the PRS is standardized (converted to a z-score) within an ancestry-matched reference group and divided into quantiles (for example, lowest \~25%, next 25%, next 25%, highest \~25%). These categories are used both for recruitment (to ensure a range of genetic risk) and for subgroup analyses of conversion risk. The broader P\&R model adds additional layers on top of PRS and APOE, potentially including: Plasma proteomic features, including pTau217 and other Alzheimer's-related markers like pTau181, Aβ42/40 ratio, GFAP, and neurofilament light (NfL). DNA methylation features linked to proteomic changes and cognitive or brain-related phenotypes. Clinical and neuropsychological measures of cognition and function. EMR-derived lifestyle and medical risk factors (forming a dementia risk score). Digital and wearable metrics from Oura and smartphone apps (sleep patterns, activity, heart rate, speech features, etc.). Using existing cohort data, researchers will test different time-to-event models (such as Cox proportional hazards models) that predict conversion to MCI or pTau217-positive MCI and compare them based on statistical performance (e.g., discrimination and calibration), cost, and practicality. One "best" P\&R model will be selected and then "locked" before analyzing data from this FIH cohort. For each participant in the study, the fixed model will generate a baseline risk estimate and assign them to risk levels; those predictions are then followed prospectively to see how well they match actual outcomes. What do participants actually do? For a participant, the study experience includes a combination of clinic visits, blood draws, digital assessments, and continuous or frequent use of wearable devices. Main activities: Screening and consent (may overlap with baseline) Review of eligibility, including confirmation of prior ApoE4 genotype and availability of WGS or suitable genotyping data. Signing of informed consent and data-sharing authorizations, including permission to obtain genomic data from prior labs, research biobanks, or direct-to-consumer companies. Optional permission to access past and future Oura Ring data. Medical, neurologic, and psychiatric history, medication review, physical and neurological examination, and vital signs (some of which can be derived from wearables). Review of any prior cognitive tests, imaging, and biomarker results if available. Baseline visit (Month 0) At the baseline clinic visit, participants undergo: Standard cognitive and functional assessments, including the CDR (global and Sum of Boxes) and a neuropsychological battery covering memory, executive function, and language. A digital cognitive battery (e.g., from Punto Health), designed to be sensitive to subtle changes between normal cognition and early MCI. Blood collection (via venipuncture or fingerstick) for: Plasma pTau217 and other Alzheimer's biomarkers (pTau181, Aβ42/40, GFAP, NfL). Proteomic and methylome profiling using validated laboratory assays. Clinical lab tests like complete blood count and metabolic panel, as needed. Optional clinically indicated lumbar puncture (not required for the study) if the person's own clinician recommends it to clarify diagnosis; leftover cerebrospinal fluid can be shared for exploratory analyses with a separate biobanking consent. Optional EEG (resting and task-based paradigms) and retinal hyperspectral imaging, if available in the clinical setting. Optional imaging (MRI or amyloid PET) if ordered by a treating clinician as part of standard care; the study can use these imaging results if participants agree. Initiation of digital monitoring: participants are trained to use the Oura Ring and smartphone apps that track smell tests, location/mobility, activity, speech patterns, and sleep. Based on all baseline information, including PRS, biomarkers, and cognitive data, the P\&R model is applied to assign each participant to a risk stage (such as cognitively normal, subjective cognitive decline, or MCI Stage I) and to a risk level within those stages. Participants who already meet criteria for pTau217-positive MCI at baseline are referred to a neurologist and typically do not remain in the main CN-focused cohort. Follow-up visits (Months 6, 12, 18, and 24) At each scheduled follow-up visit, participants: Provide an interval history, including any new symptoms, events, or medication changes. Repeat CDR and selected neuropsychological tests to detect changes consistent with MCI criteria or progression. Repeat digital cognitive assessments. Have blood drawn again (fingerstick or venous) for pTau217 and other biomarkers; additional proteomics or methylome panels are done as budget allows. Review adherence to digital monitoring and troubleshoot any device issues. In some cases, repeat EEG and retinal imaging at selected timepoints (for example, baseline and 24 months) or in a subset of participants. Undergo MRI or PET only if clinically indicated or available as standard-of-care, not as mandatory research procedures. At each visit, the P\&R risk scores and staging categories are updated to reflect new biomarker and cognitive data. Continuous digital monitoring and "triggered" visits Between clinic visits, participants wear the Oura Ring and use smartphone apps that collect information on sleep, activity, heart rate, location/mobility, and performance on brief app-based cognitive or speech tasks. If these systems, or the participant themselves, flag potential cognitive or functional decline (for example, changes in sleep pattern, reduced activity, altered speech metrics, or decline on digital tasks), the study team can schedule a "triggered" visit or remote assessment. Triggered assessments may include focused cognitive testing, repeat pTau217 and other blood biomarkers, and could lead to referral to a neurologist if criteria for MCI with positive biomarkers are met. Participants who convert to biomarker-positive MCI are referred for clinical care and may stop further intensive research follow-up according to the protocol. In total, each participant typically has at least five in-clinic visits and five planned biospecimen collections over two years, plus any additional visits prompted by triggered assessments. Outcomes and how they are analyzed The primary outcome is conversion from cognitively normal status to pTau217-positive MCI Stage I within 24 months. This is defined by: Meeting consensus criteria for MCI due to Alzheimer's disease, including objective cognitive decline on standardized tests and a CDR global score around 0.5 with CDR-SB in the "questionable impairment" or very mild range; and Having plasma pTau217 above a validated threshold for Alzheimer's pathology, using a clinically validated assay. CSF pTau217, if available from clinically indicated lumbar puncture outside the study, may be examined as an exploratory component but is not required. A subset analysis focuses on participants with especially high baseline risk according to the P\&R model, to examine time to conversion within that group. Key secondary outcomes include: Time to conversion from cognitively normal to MCI Stage I, regardless of pTau217 status, and time to biomarker conversion from pTau217-negative to pTau217-positive. Estimated ten-year conversion rates from cognitively normal to MCI Stage I, extrapolating from the observed two-year data and informed by external cohort data. Hazard ratios for late-onset Alzheimer's disease associated with P\&R risk scores, PRS categories, and biomarker trajectories. Cost per participant and per detected conversion event for different sampling and monitoring strategies (e.g., fingerstick vs venous blood, more vs fewer in-person visits, different intensities of digital monitoring). Incremental predictive value of digital monitoring features (smell, location, activity, speech, EEG, sleep) beyond more traditional risk factors and biomarkers, assessed using measures like AUC/ROC and net reclassification improvement. Statistical analyses are primarily exploratory and hypothesis-generating, given the modest sample size and FIH nature of the study. The team will estimate P\&R model performance (such as AUC/ROC) with confidence intervals, compare alternative models using tests like DeLong's test, and use survival models and logistic regression to study time-to-event and binary outcomes. Missing data will be handled with appropriate methods (for example, multiple imputation or mixed-effects models), and sensitivity analyses will check how robust the results are to different assumptions about missingness. Data handling, privacy, and safety All study data are captured in a secure, regulatory-compliant electronic data capture system such as REDCap or an equivalent platform. Each participant is assigned a unique study ID, and information that directly identifies individuals is stored separately from research data, linked through a secure key file. Digital device data are transmitted over encrypted channels and stored on secure servers with role-based access controls; only authorized study staff can access identifiable data, and most analyses use coded datasets. To protect confidentiality, all staff receive training in human subjects protections and HIPAA requirements. Reports and publications share only aggregated or de-identified information; data sharing with collaborators or sponsors uses de-identified or limited datasets under formal data use agreements that forbid re-identification and require appropriate safeguards. A HIPAA authorization is incorporated into the consent process, or a waiver is obtained when appropriate. Any future uses of stored biospecimens or data beyond the current protocol will either require explicit participant consent or appropriate IRB and HIPAA determinations. Potential risks to participants include: Physical risks from blood draws (brief discomfort, bruising, rarely infection or fainting) and from lumbar puncture if clinically performed (headache, back pain, very rare bleeding or infection). Minimal risks from EEG and retinal imaging (mild discomfort or fatigue). Radiation exposure from amyloid PET scans when clinically indicated, as well as possible anxiety about imaging results. Privacy risks from digital monitoring of location, activity, and speech-related features, including the possibility-however low with safeguards-of data breaches. Psychological risks, such as anxiety or distress related to learning about genetic risk, biomarker status, or early cognitive changes. Risk mitigation strategies include using trained personnel and standard procedures for sample collection, limiting imaging to what is clinically necessary, implementing robust cybersecurity measures, and offering counseling and referrals around genetic and biomarker information. Participants can opt out of specific optional components (for example, CSF sharing or certain digital features) while continuing in other parts of the study where feasible. Because the study is observational with minimal to moderate risk, a formal external Data Safety Monitoring Board is not initially planned, but could be added if requested by the IRB or sponsors; the principal investigator oversees ongoing safety monitoring and reports serious adverse events to the IRB according to institutional policy. Informed consent and sharing of results Before any study-specific procedures, qualified staff obtain informed consent in a private setting. The consent discussion covers the purpose of the study, what participation involves, potential risks and benefits, alternatives (including not taking part), how confidentiality will be protected, how data may be shared, and the voluntary nature of participation (including the right to withdraw without affecting clinical care). The consent process also explains how genetic and biomarker results may be handled and under what conditions individual results might be returned. For digital monitoring, the consent materials clearly describe what kinds of data are collected (for example, GPS-based location, accelerometry, and speech features but not the actual spoken content), how those data are used, and how participants can pause or discontinue certain digital components if they wish. The study does not enroll children, prisoners, or pregnant individuals; these groups are explicitly excluded. For older adults with subtle cognitive changes, the study will formally assess decision-making capacity when there is concern and may involve legally authorized representatives when allowed by institutional policy and approved by the IRB. At the end of the study, findings will be disseminated through scientific publications and conference presentations, using de-identified or aggregated data. Participants will be offered a lay-language summary of the overall results if they would like one, so they can understand what the study learned about early detection and risk staging in Alzheimer's disease.