A dementia study built on insurance records will most often miss the mildest cases — people in the early stages of the disease who are still managing daily life, who have never had a dementia code entered into a claim. In a comparison of Medicare claims against rigorous annual clinical evaluations, claims caught 79% of confirmed dementia cases, meaning roughly one in five people with clinically confirmed dementia had no dementia diagnosis anywhere in their insurance records, according to Grodstein and colleagues in the Journals of Gerontology.
The misses are not random. They fall disproportionately on people with milder impairment, on Black and Hispanic older adults, on people with less income and education, and on anyone with mild cognitive impairment rather than full dementia. That pattern shapes what claims-based research can and cannot tell you.
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
- Why insurance records undercount dementia
- Mild and early-stage disease is missed most
- The misses fall unevenly by race and income
- The insurer itself changes the count
- How to read a claims-based dementia study
- Frequently Asked Questions
Why insurance records undercount dementia
Insurance claims record billing codes, not cognition. A dementia code appears only when someone sees a clinician, the clinician recognizes and documents the condition, and the diagnosis is coded on a bill. Each step can fail, and each failure looks identical in the data: no code, no case. The Grodstein study quantified this by following five research cohorts at the Rush Alzheimer's Disease Center, where every participant received a thorough annual clinical evaluation regardless of whether they ever sought care for memory problems.
Matching those evaluations against 2014–2018 Medicare claims showed 79% sensitivity and 88% specificity. The clinical evaluations serve as the gold standard; the 21% gap is dementia that exists but is invisible to any study reading only the claims. Stricter code definitions widen the gap. A separate analysis found that the Dementia DataHub's ICD-10 "likely-or-higher" claims definition caught only 50% of dementia cases identified by Health and Retirement Study survey-based classification, per Gianattasio and colleagues in Alzheimer's & Dementia. Precision comes at the cost of missing half the disease.
Mild and early-stage disease is missed most
The clearest pattern is severity. In the Rush comparison, people whose dementia went uncoded averaged a Mini-Mental State Exam score of 22.0 — mild impairment — versus 15.4 for detected cases. Nearly half of the missed cases (45%) had no limitations in daily activities, compared with 28% of detected cases. This makes sense mechanically.
Someone still cooking, driving, and paying bills is less likely to be brought to a doctor for memory concerns, and a clinician is less likely to document dementia when a patient presents well in a short visit. The disease is there; the paper trail is not. Mild cognitive impairment — measurable cognitive decline that has not yet reached the dementia threshold — is captured even less reliably. Lee and colleagues found that only about half of patients clinically diagnosed with MCI had any corresponding dementia-spectrum code in their Medicare claims. A claims-based study of "early detection" is therefore working with the least complete slice of the data.
The misses fall unevenly by race and income
Claims detection is racially skewed. In the Rush cohorts, White participants made up 90% of the dementia cases claims correctly detected but only 75% of the cases claims missed — Black participants were disproportionately among the false negatives. A separate Rush cohort study of 710 people with incident dementia found only 54% received a timely claims diagnosis, defined as a code within three years before or one year after clinical onset, and Black older adults had twice the odds of underdiagnosis. Socioeconomics drives much of this.
In Health and Retirement Study data, household income and education explained the largest shares — 17.9% and 12.7% respectively — of the Black–White gap in timely diagnosis. Non-Hispanic Black and Hispanic older adults are more likely to develop dementia yet less likely to have it recorded promptly. The research consequence is subtle and important: a disparities study built on claims measures who gets *diagnosed*, not who *has* dementia. Because the groups with more dementia are also the groups whose dementia is least likely to be coded, claims-based estimates understate the true gap.
The insurer itself changes the count
Which insurance a person carries also shapes whether their dementia appears in the data. Researchers have flagged underreporting in Medicare Advantage encounter data, and a natural experiment made the coding incentive visible: after a 2020 risk-adjustment change began paying Medicare Advantage plans for dementia codes, annual incident dementia diagnosis rates in those plans rose 11.5% relative to traditional Medicare.
The people in those plans did not suddenly develop more dementia. The payment rules changed, and the codes followed. Any study spanning that period, or comparing populations with different insurance mixes, is partly measuring billing behavior rather than brain health.
How to read a claims-based dementia study
When a headline reports dementia prevalence, risk factors, or trends from insurance data, a few checks put the numbers in context: For families, the same gap has a practical edge: a loved one can have clinically meaningful impairment with nothing in their medical record to show for it. The 54% timely-diagnosis figure means the record trailing reality is the norm, not the exception — if you see persistent memory or functional changes, ask directly for a cognitive assessment rather than assuming a doctor would already have coded a problem.
- Expect prevalence to skew toward severe disease. Early-stage and high-functioning cases are the ones the records miss, so claims-based counts describe a sicker-than-average population.
- Treat MCI findings from claims with extra caution — roughly half of clinical MCI never reaches the records at all.
- Read racial and socioeconomic comparisons as diagnosis gaps layered on top of disease gaps, which means the reported disparity is a floor, not the full picture.
- Check whether the study spans the 2020 Medicare Advantage risk-adjustment change or mixes payer types, since coding incentives shifted diagnosis rates independent of cognition.
Frequently Asked Questions
Does a missing dementia code mean a person doesn't have dementia?
No. In the Rush comparison, 21% of clinically confirmed cases had no dementia code in their Medicare claims, and those missed cases skewed toward milder, higher-functioning disease.
Are claims records better at ruling dementia in or ruling it out?
Ruling it in. Specificity ran 88–97% across studies, so a recorded dementia code is usually real; the absence of a code is much weaker evidence of absence.
Why would Medicare Advantage data show more dementia than traditional Medicare?
A 2020 risk-adjustment change began paying plans for dementia codes, and incident diagnosis rates rose 11.5% relative to traditional Medicare — a coding effect, not a change in who has the disease.
