Passive Digital Marker-enabled identification and engagement for Alzheimer’s disease and related dementias care in primary care: the INCLUDE study protocol

Abstract

Introduction

More than half of older adults living with Alzheimer’s disease and related dementias (ADRD) never receive a formal diagnosis, and when a diagnosis occurs, it is often years after symptom onset. Primary care clinicians are ideally positioned to detect ADRD early; however, current workflows lack scalable tools that support systematic identification and follow-up. The Passive Digital Marker (PDM), a machine learning model that uses structured electronic health record (EHR) data, can identify patients at elevated risk for ADRD without adding burden to clinicians. This protocol outlines a feasibility study to develop and evaluate a patient-informed secure messaging intervention paired with PDM-based risk stratification to enhance patient engagement in cognitive assessment in primary care settings.

Methods and analysis

This will be a non-randomised pilot study conducted across 12 single health system primary care clinics. The PDM will be applied to EHR data to identify patients aged ≥65 years who are at high risk for ADRD. High-risk patients will receive a co-designed secure message prior to and after upcoming primary care visits encouraging follow-up evaluation with a trained nurse, the Brain Health Navigator (BHN). The primary objectives are to: (1) determine the feasibility of applying the PDM to EHR data across 12 primary care clinics; (2) assess the feasibility of engaging patients identified as positive on the PDM through secure text messaging prior to a primary care encounter and (3) evaluate engagement with the BHN following secure text messaging. Study outcomes will assess the feasibility of implementing the PDM and secure messaging workflow, including identification of high-risk patients using the PDM, message delivery and patient engagement measured through message open rates, completion of cognitive concern questions and appointments scheduled with the BHN. Quantitative data will be analysed using descriptive statistics.

Ethics and dissemination

This study was deemed exempt as part of enhanced patient care. The findings will be disseminated through peer-reviewed publications, professional conferences, health system reports and public-facing communications.

Trial registration number

NCT07016178.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study applies a validated Passive Digital Marker (PDM) into routine primary care electronic health data, enabling systematic, low-burden identification of patients at elevated risk for undiagnosed Alzheimer’s disease and related dementias without requiring additional clinician screening time.

• The intervention uses a patient-informed, co-designed, secure messaging strategy delivered through an existing health system platform, enhancing real-world feasibility, scalability and alignment with routine care processes.

• This study uses feasibility metrics to assess patient engagement across diverse primary care clinics.

• The non-randomised, single-health system design may limit the generalisability to settings with different electronic health record or patient portal infrastructure, or care navigation resources.

• Reliance on secure messaging and patient portal access may under-represent patients with limited digital access or literacy, potentially introducing selection bias in engagement and follow-up.

Introduction

Background and rationale

In 2021, an estimated 57 million people worldwide were living with dementia, with nearly 10 million new cases diagnosed each year.¹ In the USA, approximately 7.2 million adults aged 65 years and older are currently living with Alzheimer’s disease.² More than 50% of Americans living with Alzheimer’s disease and related dementias (ADRD) never receive a formal diagnosis^3–11; and when they do, diagnosis often occurs 2–5 years after symptom onset.^{8 12} These delays or missed diagnoses are associated with higher healthcare utilisation, misdiagnosed comorbidities, unmanaged behavioural symptoms and missed opportunities for treatment and advance care planning.^{2 3 8 13 14} Late diagnosis also increases caregiver stress and burden and limits preparedness for caregiving,¹³ compounding the burden across patients, families and the healthcare system.^{2 15}

Early detection, however, creates opportunities for prevention and risk reduction. The 2024 Lancet Commission on ADRD¹⁶ estimates that up to 45% of cases could be theoretically prevented by addressing modifiable risk factors, many of which, including hypertension, diabetes, depression, hearing loss and physical inactivity, are routinely managed in primary care. Detecting ADRD early enables clinicians to intervene during a critical period when interventions can potentially affect the trajectory of cognitive decline, slow down progression and improve quality of life.^16–18

Despite these benefits, primary care clinicians face persistent barriers to early detection, including time constraints, limited training and confidence, lack of validated tools integrated into the electronic health record (EHR) and inadequate reimbursement for cognitive assessment.^19–26 These barriers help explain why, despite Medicare requirements to assess cognition during the Medicare Annual Wellness Visit, fewer than one-third of older adults report receiving such evaluation.^27–29 At the same time, recent therapeutic advances have increased the urgency of timely diagnosis. Anti-amyloid monoclonal antibodies are the first to demonstrate disease-modifying potential by lowering brain amyloid levels and slowing the rate of cognitive and functional decline by approximately 30% when initiated in the earliest stages of the disease.^30–35 Early detection, therefore, enables improved management of comorbid conditions and care planning and access to emerging treatments, making timely identification of ADRD a clinical and operational priority for health systems.

Interventions to improve ADRD detection and diagnosis have primarily focused on clinician education, dementia awareness and screening initiatives.^{20 36–41} The use of clinician-administered paper or digital cognitive screening tools presents challenges for scalability and sustainability in primary care settings due to time and workflow constraints.^{5 11 42 43}

Digital tools that leverage routinely collected EHR data represent a promising approach for scalable identification.^44–48 Over the past 5 years, Indiana University (IU) researchers have developed and validated a Passive Digital Marker (PDM), a random forest machine learning algorithm that analyses structured elements (eg, demographics, diagnoses, prescriptions) and unstructured clinical notes to identify patients at risk for ADRD with approximately 80% accuracy at a 1-year prediction horizon.^{44 45} Unlike traditional approaches that rely on resource-intensive cognitive testing or manual chart review, the PDM operates on existing clinical documentation, requiring no additional time from clinicians or patients.

The PDM was validated using dementia cases divided into training (80%) and testing (20%) data sets. The model performance was evaluated using fivefold cross-validation, in which the dataset was divided into five groups, and the training and testing processes were repeated across five iterations. Generalisability was assessed using data from 15 institutions for dementia cases and 25 institutions for the controls. Using structured EHR data alone, the PDM demonstrated 1-year accuracy of 68%–69%.⁴⁴

A recent randomised clinical trial demonstrated that when paired with a patient-reported outcome tool, such as the Quick Dementia Rating System (QDRS), the PDM significantly increased the incidence of ADRD-related diagnostic follow-up compared with usual primary clinical care without a screening intervention.⁴⁵ The combined PDM+ QDRS approach strengthened the clinical signal by linking objective EHR-based prediction with patient-reported concerns, increasing the likelihood of clinicians initiating diagnostic assessment.⁴⁵ These findings highlight patient engagement as a key link between risk identification and clinical action.

In addition to these innovations, health systems have responded to the need for enhanced follow-up by introducing clinical roles that support primary care teams. One such model is the Brain Health Navigator (BHN), a trained nurse who evaluates cognitive impairment, identifies potential addressable causes, facilitates shared decision-making, connects patients and their caregivers with community and research resources and supports entry into the treatment pathways.^{42 49 50} By providing this support, the BHN reduces the burden on primary care clinicians without overburdening specialty care and improves the diagnostic workup for patients with positive cognitive screenings. Despite these benefits, the limited upfront identification of patients at high risk for ADRD constrains the full utilisation of this model.^{42 49 51}

Thus, there is a critical need for a novel, pragmatic and scalable approach that can identify individuals at risk for ADRD within primary care while minimising additional burden on already constrained clinical workflows. To address these gaps, we propose conducting a pilot feasibility study to evaluate a patient-informed text-based messaging intervention aimed at increasing patient engagement with the BHN programme among patients identified as high-risk for ADRD through the PDM.

Objectives

The primary objectives of this study are as follows:

1. Determine the feasibility of applying the PDM to EHR data across 12 rural primary care clinics within a single health system.

2. Assess the feasibility of engaging primary care patients who are identified as positive on the PDM (PDM+) via secure text messages prior to a primary care encounter.

3. Evaluate engagement with the BHN among PDM+ patients following secure text messaging.

Study design

This study is a single-arm, non-randomised embedded clinical study.⁵² We will follow established pilot feasibility criteria⁵² and will report our protocol and findings in accordance with the Standard Protocol Items: Recommendations for Interventional Trials (SPIRIT) guidelines.⁵³

Methods and analysis

Study setting

This study will be conducted at IU Health, a statewide academic health system that serves diverse urban, suburban and rural populations across Indiana. The intervention will be implemented across 12 participating primary care clinics in diverse areas that use a shared EHR platform.

Eligibility criteria

Participants will be eligible if they: (1) are aged 65 years and older; (2) are established patients of one of the 12 selected health system’s primary care clinics participating in the study and (3) are identified by the PDM as high risk for developing or having undetected mild cognitive impairment (MCI) or ADRD. Participants will be excluded if they: (1) are younger than 65 years of age; (2) have <2 years of available EHR data; (3) their primary language is not English and (4) have a documented diagnosis of ADRD or MCI, schizophrenia or bipolar disorder (identified via International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes).

Intervention

The intervention applies the PDM to EHR data to identify patients at high risk for ADRD and delivers a previously developed, patient-informed, secure message to identified patients. The project will be conducted in the following sequential steps: (1) implementation of the PDM within the IU Health EHR across 12 participating primary care clinics; (2) deployment of secure messages to PDM+ patients and (3) evaluation of patient engagement with text messages or the BHN for further cognitive assessment (figure 1).

Figure 1. Overall study workflow. ADRD, Alzheimer’s disease and related dementias; EHR, electronic health record.

Key components of the intervention

PDM for ADRD

We will implement a validated PDM across the primary care panel of 12 clinics that are not currently participating in any ADRD screening trial or brain health practice improvement initiatives, have a high volume of patients aged 65 years and older, serve racially and ethnically diverse patients, include rural/underserved communities and have dual Medicare/Medicaid representation. The PDM algorithm was designed to analyse EHR data to identify primary care patients aged ≥65 years who are at elevated risk of undetected ADRD. Our application of the PDM will use structured data from the EHR, including diagnoses and medications. Model performance will not be re-evaluated in this feasibility pilot study. The health system’s information technology (IT) team will extract EHR data for patients aged 65 years and older within the secure institutional environment. Medication data will be standardised using Anatomical Therapeutic Chemical classification codes prior to analysis. A limited dataset will be securely transferred to the engineer responsible for running the PDM algorithm. The PDM will generate a risk score based on relationships among diagnoses and medication patterns. The health system’s IT team will subsequently link the PDM results back with patient records to identify patients at elevated risk for ADRD who are eligible for outreach. All study procedures will comply with institutional privacy and security policies and federal regulations related to patient data protection, including Health Insurance Portability and Accountability Act (HIPAA).

Patient engagement via secure text messaging

We will use the Twistle platform (Health Catalyst), a commercially available cloud-based software platform used in our health system, to deliver the co-designed message to PDM+ patients. Messages will be delivered 10 days prior to a scheduled primary care appointment. If patients do not open the initial message, the system will resend the message 3 days before the appointment and again 2 days after the visit.

The messages will include instructions for patients to complete three validated subjective cognitive questions assessing memory concerns (“Are you concerned about your memory?”, “Are your loved ones concerned about your memory?” and “Is your mind as clear as it used to be?”)^{54 55} and will provide options to contact the BHN and visit the BHN programme website (figure 2).

Figure 2. Secure messaging workflow. Workflow for identifying and digitally engaging PDM-positive primary care patients in early dementia detection. Patients entering the workflow <10 days before their appointment receive only messages scheduled after entry. BHN, Brain Health Navigator; PDM, Passive Digital Marker.

Patient engagement with BHN

Patients may initiate contact with the BHN after receiving any of the secure messages using the phone number provided. Patients who do not open the messages will receive an additional invitation to the BHN programme 4 days after their primary care appointments. In addition, the BHN may initiate contact with patients who submit the subjective cognitive questions, by phone within 2 weeks of submission. If the BHN is unable to reach the patient, a portal message or a mailed letter will be sent. At the conclusion of the study, the BHN will notify primary care clinicians of PDM+ patients and contact patients by phone or letter to invite them to participate in brain health screening.

Once a patient agrees to an evaluation, the BHN will follow an established, evidence-based protocol that includes reviewing medications and assessing anticholinergic burden; administering the Montreal Cognitive Assessment; evaluating potentially reversible causes of cognitive impairment and assessing cardiovascular risk factors.^{42 49 51}

The BHN will document all findings and communicate recommendations to the patient’s primary care clinician through the EHR, including proposing laboratory testing, imaging or referrals to specialists, including geriatrics, neurology and neuropsychology. In addition, the BHN will perform patient education on brain health lifestyle modifications and offer community and research resources.

Outcomes

The study outcomes assess reach and patient engagement pre-index and post-index primary care encounter. Reach measures include the number of patients eligible for the PDM, the number identified as high risk (PDM+), the number eligible for secure messaging and the number who receive a secure message. Patient engagement measures include secure message open rates, completion of subjective cognitive concern questions and the number of patients who schedule an appointment with the BHN (table 1).

Table 1. Study outcomes.

Outcome domain	Outcome/Measure	Data source
Reach/Identification	Patients eligible for PDM	EHR
	Number and proportion of PDM+ patients	EHR; PDM output
	PDM+ patients eligible for secure messaging	EHR
	Patients receiving ≥1 secure message during study window	Dashboard
Patient engagement	Patients opening message 1, 2 or 3	Dashboard
	Patients completing SCC questions	Dashboard
	Patients with ≥1 positive SCC	Dashboard
	Number of appointments scheduled with BHN	BHN record

BHN, Brain Health Navigator; EHR, electronic health record; PDM+, patients identified as high risk by the PDM; PDM, Passive Digital Marker; SCC, subjective cognitive concern questions.

Study timeline

The study began in April 2025 and is expected to be completed by December 2026. The study timeline is presented in figure 3.

Figure 3. Study timeline.

Recruitment and sample size

Primary care clinics within the health system that are not currently engaged in dementia screening initiatives will be approached for participation. Practices expressing interest will be invited to an online session to describe the study procedures. Clinics will be enrolled consecutively until the recruitment target of 12 clinics is reached. All eligible patients within the enrolled primary care clinics will be included, yielding an anticipated sample size of approximately 10 000 patients. Since the intervention will be embedded within routine clinical care, individual patient informed consent will not be required.

Assignment of interventions

This study does not involve the allocation of participants to intervention groups. The intervention is delivered as part of a pragmatic, system-level implementation within routine primary care workflows. Assignment to intervention components is determined by the identification of high ADRD risk based on the PDM.

Patient and public involvement

Patients and members of the public were not involved in the design of this study and will not be involved in its conduct or dissemination. The text messaging intervention evaluated in this protocol incorporates messaging content informed by prior patient co-design sessions, which will be described separately.

Data collection

Message reach and engagement metrics will be captured using the Twistle dashboard, additional data will be obtained from BHN records and the EHR. We will collect data on patient engagement with the messaging system, BHN interaction and clinical outcomes (including testing and diagnosis) from EHR. All study data will be stored and managed using Research Electronic Data Capture, with ongoing quality checks to ensure accuracy, completeness and timely availability for analysis. The principal investigators will oversee all aspects of data collection and management.

Data analysis

Quantitative analyses will focus on feasibility and patient outcomes. Descriptive statistics, including means, medians, proportions and 95% CIs, will be reported as appropriate. Feasibility outcomes will be evaluated during implementation, and patient outcomes at 90 days and 12 months following the index primary care encounter.

Data monitoring

A formal Data Monitoring Committee will not be required for this study. Study oversight will be provided by the investigative team in collaboration with the IU Health clinical and informatics leadership. Any concerns related to data integrity, participant safety or unintended consequences of implementation will be reviewed using institutional processes. Decisions regarding continuation or modification of study procedures will follow institutional review board (IRB) requirements and health system governance processes.

The embedded intervention poses minimal risk to participants and consists of existing tools for patient outreach delivered within routine primary care workflows. No experimental treatments will be administered, and no randomisation or comparative clinical interventions will be performed.

Harms

The intervention poses minimal risk. Potential unintended effects include patient confusion or anxiety related to secure messaging about cognitive health risks and increased clinical workload from patient inquiries. No physical harm is anticipated from this study. Patient concerns will be addressed through usual clinical care. Events meeting institutional reporting criteria will be reported to the IRB in accordance with policy.

Auditing

The research team will monitor the study’s conduct, data integrity and adherence to approved procedures in collaboration with the health system’s operational and informatics leadership.

The study remains subject to audit or review by the IRB or institutional compliance office, if requested. Any protocol deviations identified through these processes will be documented and addressed per institutional requirements.

Ethics and dissemination

The IU IRB determined that this study is exempt as part of enhanced patient care activities delivered during routine care. Because the intervention is embedded within routine clinical care, individual informed consent is not required. The PDM uses routinely collected EHR data within the health system’s secure infrastructure. All study procedures comply with institutional policies and federal regulations related to patient privacy and data security, including the HIPAA. The study is registered with ClinicalTrials.gov (NCT07016178).

Study findings will be disseminated through peer-reviewed publications and presentations at national conferences in geriatrics, neurology, primary care and implementation science. Findings will also be shared with health system leadership, community partners, advocacy organisations and the public through press releases and blog communications.

Confidentiality

All participant information will be collected and stored within secure institutional systems. Access to identifiable information will be limited to authorised clinical and study personnel in accordance with institutional policies. Study data used for analysis will be extracted from limited datasets. Data will be maintained on secure, password-protected servers and will comply with institutional and regulatory requirements.

Discussion

This study will advance the understanding of how digital technologies can support the early detection of ADRD in primary care. The intervention integrates the application of a machine learning algorithm, the PDM, with structured patient outreach and BHN evaluation offer, creating a scalable pathway to identify and engage high-risk older adults.

This single-arm, pragmatic, embedded, clinical study will assess the feasibility of implementing the PDM within a healthcare system EHR and evaluate the feasibility and acceptability of a secure patient messaging intervention. At present, early detection within primary care is limited by the lack of integrated digital screening tools and direct patient communication systems that support timely engagement.^19–21 Findings from this study will inform the design of future comparative effectiveness and hybrid-implementation trials aimed at improving early identification and care pathways for ADRD.

Primary care settings vary in EHR infrastructure, secure messaging capabilities and access to brain health resources, which may influence the feasibility of implementation across healthcare environments. However, this intervention was intentionally designed to leverage routinely collected EHR data and existing secure messaging systems commonly available in many health systems. Given the study design, the following limitations will be considered when interpreting study findings. First, the non-randomised selection of clinics introduces the possibility of selection bias, which may limit the generalisability of the results. Second, the messaging intervention was available only in English and non-English-speaking patients were excluded, limiting the cultural and linguistic applicability of the findings. Third, the intervention requires an active mobile phone number linked to the EHR; consequently, patients without text-enabled phones or landline-only contact information will not be included, potentially excluding lower-income and older patients. In addition, variability in EHR completeness and documentation practices, as well as differences in comorbidities, socioeconomic factors, health literacy and access to technology, may influence patient identification, engagement and study outcomes. Finally, the implementation of the PDM requires access to EHR data and technical infrastructure to support data extraction, processing and algorithm deployment. Future research should evaluate the implementation of the PDM across diverse health systems with varying EHR platforms, technical infrastructure, patient populations and language outreach capabilities. Additional studies are needed to determine the effectiveness of PDM-based risk identification coupled with patient outreach messaging in increasing completion of diagnostic evaluations and detection of previously unrecognised MCI and ADRD among high-risk patients.

The funders had no role and will have no role in the study design; data collection, management, analysis or interpretation; manuscript preparation or the decision to submit the manuscript for publication.