Electronic health records (EHRs), the digital charts clinics and hospitals use to track diagnoses, visits, and test results, already power most dementia research tools built in 2026. Researchers still need better data on lifestyle and caregiving factors, longer prediction windows, and more reliable ways to confirm who actually has dementia before these tools can be trusted at scale. For a reader managing memory concerns in their own family, this matters because EHR-based tools are increasingly the thing deciding whether a doctor orders a cognitive workup at all. Knowing what these systems can and cannot yet do helps separate a promising screening result from a confirmed diagnosis.
Medical information disclaimer: This article is for general educational purposes only and does not provide medical advice, diagnosis, or treatment. Always consult a physician or other qualified health professional about symptoms, medications, tests, or treatment decisions.
Table of Contents
- What EHR research has focused on, and what it has skipped
- How far ahead can these tools actually predict risk?
- Can researchers trust a dementia diagnosis code in the record?
- What EHRs miss about caregivers and daily life
- What's changed: combining patient reports with text mining
What EHR research has focused on, and what it has skipped
Most EHR-based dementia research to date has studied medical conditions and treatments, not the everyday factors that shape brain health. A 2024 scoping review of 80 studies conducted between 2010 and 2023 found that 87.5% used a retrospective design, meaning researchers looked backward through existing records rather than tracking people forward in time, according to the scoping review published on PMC. The same review found that lifestyle, socioeconomic, and environmental factors — things like diet, income, housing, or air quality — were rarely studied, even though these factors shape dementia risk outside a doctor's office.
That gap matters for research quality. A record built mostly from clinic visits and prescriptions captures what happened medically, not what happened at home. Researchers still need to close that gap before EHR data can tell a fuller story about who develops dementia and why.
How far ahead can these tools actually predict risk?
Most EHR-based prediction tools forecast risk within a short window, not years in advance. A 2025 systematic review that screened 8,312 studies and included 56 covering 461 separate EHR-based dementia prediction models found that most are prognostic, with prediction horizons under five years, according to the systematic review posted on medRxiv.
That means long-range forecasting — flagging risk a decade or more before symptoms appear — remains underdeveloped. A separate pilot from Indiana University, running across 12 primary care clinics between 2024 and 2026, reports roughly 80% accuracy predicting a dementia diagnosis one to three years in advance using routine EHR data, according to the USC Schaeffer Center's summary. Together, these findings show current tools are better suited to near-term flagging than to identifying risk far in advance.
Can researchers trust a dementia diagnosis code in the record?
Diagnostic codes in EHRs are not a reliable stand-in for a confirmed dementia diagnosis. A systematic review of UK dementia diagnostic codes found that reported validity estimates were relatively high, but every included study carried a significant risk of bias, and few reported the basic accuracy measures — sensitivity, specificity, or negative predictive value — needed to judge a code's trustworthiness, according to the systematic review on PMC.
A commentary on a separate code-based algorithm found about 70% sensitivity and positive predictive value, meaning roughly 30% of true dementia cases were missed entirely, according to the King's College London commentary in the Journal of Alzheimer's Disease. The patients most often missed were older adults, people with mild cognitive impairment rather than full dementia, and those with more coexisting health conditions — exactly the group where early detection matters most.
- A diagnosis code confirms a clinician recorded dementia — it does not confirm the diagnosis was accurate or complete.
- Missed cases skew toward older patients, those with mild cognitive impairment, and those with multiple other conditions.
- Few studies report the accuracy measures needed to judge how much a code-based identification can be trusted.
What EHRs miss about caregivers and daily life
Dementia care involves two people — the person with memory concerns and the person helping them day to day — but most EHR systems only capture one record at a time. A 2026 report in Annals of Family Medicine concluded that current electronic medical record systems do not capture the relational, interdependent nature of dementia care, because few systems support routine linkage between a patient's record and their caregiver's, according to the report in Annals of Family Medicine.
That gap limits both clinical care and research into how caregiver stress, support, or absence affects a patient's trajectory. Researchers studying dyads — the patient-caregiver pair as a unit — currently have to work around records that were never built to connect them. Closing this gap would let future studies track how caregiving circumstances change outcomes over time, rather than treating each patient's record as if the caregiving relationship doesn't exist.
What's changed: combining patient reports with text mining
One trial published in JAMA Network Open in November 2025 shows a working alternative to code-based detection alone. Researchers at Eskenazi Health ran a pragmatic randomized trial with more than 5,000 primary care patients, combining a 10-question patient-reported tool with a "passive digital marker," an NLP-based system that mines the free text already in a patient's chart. The combined approach increased new Alzheimer's and related-dementia diagnoses by 31% compared with usual care, with no added clinician time, according to the Regenstrief Institute's report.
That detection boost came with a dependency: the same trial found a 41% increase in follow-up diagnostic workups, such as imaging and cognitive testing, after a flagged result, according to Indiana University School of Medicine's coverage. In other words, the tool identifies more people worth evaluating, but a confirmed diagnosis still depends on a clinician ordering and completing the follow-up workup. A flagged chart is a starting point, not an endpoint.
- A positive screen from a passive digital marker is a prompt for further testing, not a diagnosis.
- The 31% increase in new diagnoses only materialized because follow-up workups also rose — the tool alone did not produce diagnoses.
- Ongoing eRADAR trials at Kaiser Permanente Washington and UCSF, including one marked completed as of September 2026, are testing whether a low-cost EHR-based flagging tool holds up across other health systems, according to ClinicalTrials.gov.
You Might Also Like
- UAS-CLEAR Dementia Care Research Data Priorities in 2026: What Researchers Still Need to Learn for Adults With Memory Concerns
- Home-Based Dementia Care Research Priorities in 2026: Service Gaps That Matter to Adult Children
- The $100 Million PROTECT-Cog Alzheimer’s Study Announced in 2026: What Is Established and What Remains Uncertain for Adults With Memory Concerns
