From Risk Factors to Real-World Prevention: Operationalizing Brain Health at Population Scale
As dementia rates climb, health systems have an opportunity to turn existing EHR, claims and risk-factor data into scalable prevention workflows that identify high-risk patients earlier and connect them to brain health interventions.
The global dementia burden is expanding at a pace that outstrips any single intervention's capacity to keep up. Cases worldwide grew roughly 161% between 1990 and 2021, from an estimated 21.8 million to 56.9 million, and projections built on the same Global Burden of Disease dataset point toward a range as high as 191 million cases by 2050. In the United States, an estimated 7.2 million Americans age 65 and older are living with Alzheimer's dementia today, a figure expected to nearly double to 13.8 million by 2060. Population aging explains most of this growth, but it is not the whole story — and the part that is not explained by aging is where health data infrastructure has the most immediate role to play.
The Modifiable Fraction
The 2024 update to the Lancet Commission on Dementia Prevention, Intervention, and Care identified 14 modifiable risk factors spanning early life, midlife, and late life, including hearing loss, hypertension, obesity, physical inactivity, social isolation, and untreated vision loss, with a combined population attributable fraction of approximately 45%. A separate analysis using Global Burden of Disease 2021 data arrived at a similar estimate of 42.5%, identifying high BMI, hypertension, and physical inactivity as the three largest individual contributors. The same analysis found that a 20% reduction in these modifiable risks could avert an estimated 12 to 18 million dementia cases globally by 2050 — a population-scale opportunity that depends less on any single breakthrough therapy and more on systematically identifying and managing risk across entire patient populations.
Notably, each of the leading contributors — BMI, blood pressure, physical activity, hearing status — is already captured, at least partially, in structured EHR fields, vitals data, or claims records that most health systems generate as a byproduct of routine care. The prevention opportunity is less a data-collection problem than a data-utilization one: the signal often already exists inside the enterprise data warehouse; it simply is not yet organized into a risk-stratification and outreach workflow.
Turning Structured Data Into Case-Finding
Several EHR-embedded tools already demonstrate that this translation is achievable at scale. The EHR Risk of Alzheimer's and Dementia Assessment Rule (eRADAR), developed and validated across Kaiser Permanente Washington and UCSF, uses 31 routinely collected predictors - age, BMI, blood pressure, diabetes status, emergency department utilization, and medications, to flag patients with likely undiagnosed dementia, achieving predictive accuracy in the range of 0.79 to 0.84 (AUC). In real-world deployment, patients scoring above the 90th percentile were more than four times as likely to receive a new dementia diagnosis within a year compared with average-risk patients. The tool is now embedded in a pragmatic trial across 11 primary care clinics, where high-risk flags trigger a structured "brain health visit" combining functional assessment, depression screening, and cognitive testing, a working template for converting a population-level risk score into an individual clinical action.
eRADAR is not an isolated example. A scoping review published in the Journal of the American Medical Informatics Association catalogued 19 distinct EHR-based phenotyping algorithms for Alzheimer's disease and related dementias, 12 of which are specifically designed to flag high-risk, likely-undiagnosed patients before a formal diagnosis occurs, evidence that this is a maturing, multi-vendor field rather than a single proprietary tool. For payers and accountable care organizations without direct EHR access, claims-based approaches offer a parallel path: a machine-learning model built on more than 125 million patient records in a large administrative claims dataset achieved 0.77 to 0.81 accuracy (AUC) in predicting incident Alzheimer's disease and related dementias four to five years in advance. Emerging digital phenotyping research, capturing behavioral signals from smartphone sensors such as gait, typing patterns, and voice, is being piloted across global cohorts exceeding 230,000 participants in 19 countries, offering a lower-cost, continuous data stream that could extend risk detection into populations with less frequent clinical contact.
Payment Models Are Catching Up to the Data
Detection tools only translate into population health impact if payment structures support the resulting care pathways. Medicare already reimburses a foundational building block: providers are required to screen for cognitive impairment during the Annual Wellness Visit, and can bill a separate, more detailed cognitive assessment and care-planning visit under CPT code 99483 when impairment is detected.
A more structural example arrived in July 2024 with the launch of the CMS Innovation Center's GUIDE Model — the first CMS payment model built specifically around dementia care. GUIDE pays participating health systems a per-beneficiary-per-month Dementia Care Management Payment for delivering standardized, interdisciplinary care coordination and caregiver support, with patients assigned to one of several risk tiers based on a comprehensive needs assessment. More than 330 organizations were selected to participate at launch, and the model includes a health equity adjustment tied to area deprivation and dual-eligibility status. Structurally, GUIDE mirrors the same stepped-care, risk-stratified design principles that population health programs use for diabetes or heart failure — evidence that payers are beginning to build financial infrastructure around exactly the kind of risk-tiered data workflows that EHR-based tools like eRADAR are designed to feed.
Designing for the Population That Actually Bears the Burden
Any population-scale prevention strategy has to account for who actually carries the disease burden, and the data here are unambiguous: women represent an estimated 58% to 62% of dementia cases globally, with a female-to-male prevalence ratio projected to persist near 1.7 through 2050. In the United States, roughly 4.4 million of the 7.2 million Americans living with Alzheimer's are women, nearly two-thirds of the total. Because sex is already a standard structured field in virtually every EHR and claims dataset, incorporating it as a risk-adjustment variable in population health stratification models requires no new data infrastructure — only a decision to use the field that is already there. Research on the APOE4 gene, the strongest known genetic risk factor for late-onset Alzheimer's, suggests the case for sex-specific calibration goes beyond simple longevity effects: APOE4 carriers who are women show substantially higher associated risk elevation than male carriers, indicating that a uniform, sex-blind risk model may under- or over-estimate risk in ways that matter clinically.
Building the Workflow, Not Just the Model
None of these tools function as prevention infrastructure on their own. Value comes from connecting risk-stratification output to a defined clinical workflow — a brain health visit, a care-management enrollment, a referral pathway — and from measuring whether that connection actually changes outcomes at scale, not just accuracy at the algorithm level. The eRADAR pragmatic trial's randomized design, comparing providers who receive risk flags against usual care, is instructive precisely because it tests the full pathway rather than the prediction model in isolation. For health system and health IT leaders, the strategic question is not whether population-scale dementia risk stratification is technically feasible — the evidence base above suggests it clearly is — but whether the organizational workflow, payment alignment, and governance exist to act on the signal once it is produced. With a 20% reduction in modifiable risk factors carrying the potential to avert millions of cases worldwide, the operational plumbing connecting data to action may end up mattering as much as any single new algorithm or biomarker.
Kenneth R. Deans, Jr. DHA, MBA is the President and CEO of Health Sciences South Carolina.